draft-ietf-ipv6-dns-discovery-07.txt

bind-3.2.

Network Working Group                                 Alain Durand
INTERNET-DRAFT                              SUN Microsystems, inc.
October 25, 2002                          Jun-ichiro itojun Hagino
Expires April 2002                         IIJ Research Laboratory
                                                       Dave Thaler
                                                         Microsoft




                Well known site local unicast addresses
               to communicate with recursive DNS servers
                 <draft-ietf-ipv6-dns-discovery-07.txt>



                          Status of this memo

   This document is an Internet-Draft and is in full conformance with
   all provisions of Section 10 of RFC2026.

   Internet-Drafts are working documents of the Internet Engineering
   Task Force (IETF), its areas, and its working groups.  Note that
   other groups may also distribute working documents as Internet-
   Drafts.

   Internet Drafts are valid for a maximum of six months and may be
   updated, replaced, or obsoleted by other documents at any time.  It
   is inappropriate to use Internet Drafts as reference material or to
   cite them other than as a "work in progress".

   To view the list Internet-Draft Shadow Directories, see
   http://www.ietf.org/shadow.html.



Abstract

   This documents specifies 3 well known addresses to configure stub
   resolvers on IPv6 nodes to enable them to communicate with recursive
   DNS server with minimum configuration in the network and without
   running a discovery protocol on the end nodes.  This method may be
   used when no other information about the addresses of recursive DNS
   servers is available.  Implementation of stub resolvers using this as
   default configuration must provide a way to override this.



Copyright notice

   Copyright (C) The Internet Society (2002).  All Rights Reserved.


1. Introduction

   RFC 2462 [ADDRCONF] provides a way to autoconfigure nodes with one or
   more IPv6 address and default routes.

   However, for a node to be fully operational on a network, many other
   parameters are needed, such as the address of a name server that
   offer recursive service (a.k.a. recursive DNS server), mail relays,
   web proxies, etc.  Except for name resolution, all the other services
   are usually described using names, not addresses, such as
   smtp.myisp.net or webcache.myisp.net.  For obvious bootstrapping
   reasons, a node needs to be configured with the IP address (and not
   the name) of a recursive DNS server.  As IPv6 addresses look much
   more complex than IPv4 ones, there is some incentive to make this
   configuration as automatic and simple as possible.

   Although it would be desirable to have all configuration parameters
   configured/discovered automatically, it is common practice in IPv4
   today to ask the user to do manual configuration for some of them by
   entering server names in a configuration form. So, a solution that
   will allow for automatic configuration of the recursive DNS server is
   seen as an important step forward in the autoconfiguration story.

   The intended usage scenario for this proposal is a home or enterprise
   network where IPv6 nodes are plugged/unplugged with minimum
   management and use local resources available on the network to
   autoconfigure. This proposal is also useful in cellular networks
   where all mobile devices are included within the same site.

   The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
   "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
   document are to be interpreted as described in [KEYWORDS].




2. Well known addresses vs discovery

   Some of the discussions in the past around DNS server discovery have
   been trying to characterize the solution space into stateless versus
   stateful or server oriented versus severless.  It is not absolutely
   clear how much state if any needs to be kept to perform DNS server
   discovery, and, although the semantic differences between a router
   and a server are well understood from a conceptual perspective, the
   current implementations tend to blur the picture.  In another attempt
   to characterize different approaches, one can look at how much
   intelligence a client needs to have in order to use the service.

   One avenue is to ask the IPv6 node to participate in a discovery
   protocol, such as SLP or DHCP, learn the address of the server and
   send packets to this server. Another one is to configure the IPv6
   node with well known addresses and let the local routing system
   forward packets to the right place.  This document explores this
   later avenue of configuration using well known addresses that does
   not require participation of the end node in any discovery mechanism.





3. Reserved prefix and addresses

   The mechanism described here is:
      - intended for ongoing use and not not just for bootstrapping
      - intended to populate a stub resolver's list of available
      recursive servers only if that list is otherwise unpopulated
      - providing reliability through redundancy using three unicast
      addresses.


3.1 Stub resolver configuration

   This memo reserved three well known IPv6 site local addresses.

   In the absence of any other information about the addresses of
   recursive DNS servers, IPv6 stub-resolvers MAY use any of those three
   IPv6 addresses in their list of candidate recursive DNS servers.


3.2 Recursive DNS servers configuration

   Within sites, one or more recursive DNS server SHOULD be configured
   with any of those three addresses. It is RECOMMENDED that large sites
   deploy 3 recursive DNS servers, one for each reserved address. Small
   site could use only one recursive DNS server and assign the 3
   addresses to it.


3.3 Rationale for the choice of three addresses

   Three was chosen based on common practice in many places in the
   industry.  While it's true that if the first one fails, that it's
   unlikely the second one will succeed (due to there really being no
   DNS server at all), using multiple addresses is important so that
   when ones do exist, the host can fail over to a second server more
   quickly than routing converges. Three servers is a compromise between
   extra reliability and increased complexity (maintaining additional
   servers, having multiple entries in the routing system, additional
   delays before the stub resolver returns an error,...).

   Another reason to have multiple addresses is to avoid the need to use
   of anycast addresses to achieve reliability through redundancy. On
   top of the classic problems (TCP sessions, ICMP messages,...) using
   an anycast address would hide the real locations of the recursive DNS
   servers to the stub resolver, prohibiting it to keep track of which
   servers are performing correctly. For this particular matter, using
   well known addresses is no different than configuring the stub
   resolver with regular addresses taken from the local site.


3.4 Implementation considerations

   Stub resolver implementation MAY be configured by default using those
   addresses. However, implementing only the mechanism described in this
   memo may end up causing some interoperability problems when operating
   in networks where no recursive DNS server is configured with any of
   the well known addresses.  Thus, stub resolvers MUST implement
   mechanisms for overriding this default, for example: manual
   configuration, L2 mechanisms and/or DHCPv6.




4. Routing

   A solution to enable the stub resolvers to reach the recursive DNS
   servers is to inject host routes in the local routing system.
   Examples of methods for injecting host routes and a brief discussion
   of their fate sharing properties are presented here:

      a) Manual injection of routes by a router on the same subnet.
      If the node running the recursive DNS server goes down, the router
      may or may not be notified and keep announcing the route.

      b) Running a routing protocol on the same node running the DNS
      resolver.
      If the process running the recursive DNS server dies, the routing
      protocol may or may not be notified and keep announcing the route.

      c) Running a routing protocol within the same process running the
      recursive DNS server.
      If the recursive DNS server and the routing protocol run in
      separated threads, similar concerns as above are true.

      d) Developing an "announcement" protocol that the recursive DNS
      server could use to advertize the host route to the nearby router.
      Details of such a protocol are out of scope of this document, but
      something similar to [MLD] is possible. However, the three first
      mechanisms should cover most cases.

   An alternate solution is to configure a link with the well known
   prefix and position the three recursive DNS servers on that link.
   The advantage of this method is that host routes are not necessary ,
   the well known prefix is advertised to the routing system by the
   routers on the link. However, in the event of a problem on the
   physical link, all resolvers will become unreachable.

   IANA considerations for this prefix are covered in Section 6.



5. Site local versus global scope considerations

   The rationales for having a site local prefix are:

      -a) Using a site local prefix will ensure that the traffic to the
      recursive DNS servers stays local to the site. This will prevent
      the DNS requests from accidentally leaking out of the site.
      However, the local resolver can implement a policy to forward DNS
      resolution of non-local addresses to an external DNS resolver.

      -b) Reverse DNS resolution of site local addresses is only
      meaningful within the site. Thus, making sure that such queries
      are first sent to a recursive DNS server located within the site
      perimeter increase their likelihood of success.




6. Examples of use

   This section presents example scenarios showing how the mechanism
   described in this memo can co-exist with other techniques, namely
   manual configuration and DHCPv6 discovery.

   Note: those examples are just there to illustrate some usage
   scenarios and in no way do they suggest any recommended practices.


6.1 Simple case, general purpose recursive DNS server

   This example shows the case of a network that manages one recursive
   DNS server and a large number of nodes running DNS stub resolvers.
   The recursive DNS server is performing (and caching) all the
   recursive queries on behalf of the stub resolvers.  The recursive DNS
   server is configured with an IPv6 address taken from the prefix
   delegated to the site and with the 3 well known addresses defined in
   this memo.  The stub resolvers are either configured with the "real"
   IPv6 address of the recursive DNS server or with the well known site
   local unicast addresses defined in this memo.

            --------------------------------------------
            |                                          |
            |                  ---------------------   |
            |                  |DNS stub resolver  |   |
            |                  |configured with the|   |
            |                  |"real" address of  |   |
            |                  |the recursive DNS  |   |
            |                  |server             |   |
            |                  ---------------------   |
            |  -----------          |                  |
            |  |recursive|          |                  |
            |  |DNS      |<----------                  |
            |  |server   |<----------------            |
            |  -----------                |            |
            |                  ----------------------  |
            |                  |DNS stub resolver   |  |
            |                  |configured with 3   |  |
            |                  |well known addresses|  |
            |                  ----------------------  |
            |                                          |
            --------------------------------------------

            (The recursive DNS server is configured to listen both on
            its IPv6 address and on the well known address)


6.2 Three recursive DNS servers

   This is a similar example as above, except that three recursive DNS
   resolvers are configured instead of just one.

            -------------------------------------------
            |                                          |
            |                  ---------------------   |
            |                  |DNS stub resolver  |   |
            |                  |configured with the|   |
            |                  |"real" address of  |   |
            |                  |the recursive DNS  |   |
            |                  |server             |   |
            |                  ---------------------   |
            |                       |                  |
            |  -----------          |                  |
            |  |recursive|          |                  |
            |  |DNS      |<---------|                  |
            |  |server 1 |<---------|------            |
            |  -----------          |     |            |
            |                       |     |            |
            |  -----------          |     |            |
            |  |recursive|          |     |            |
            |  |DNS      |<---------|     |            |
            |  |server 2 |<---------|-----|            |
            |  -----------          |     |            |
            |                       |     |            |
            |  -----------          |     |            |
            |  |recursive|          |     |            |
            |  |DNS      |<----------     |            |
            |  |server 3 |<---------------|            |
            |  -----------                |            |
            |                  ----------------------  |
            |                  |DNS stub resolver   |  |
            |                  |configured with 3   |  |
            |                  |well known addresses|  |
            |                  ----------------------  |
            |                                          |
            -------------------------------------------

            (The recursive DNS server is configured to listen both on
            its IPv6 address and on the well known address)


6.3 DNS forwarder

   A drawback of the choice of site local scope for the reserved
   addresses for recursive DNS server is that, in the case of a
   home/small office network connected to an ISP, DNS traffic cannot be
   sent directly to the ISP recursive DNS server without having the ISP
   and all its customers share the same definition of site.