Network Working Group                                           B. Aboba
Request for Comments: 5505                                     D. Thaler
Category: Informational                                     L. Andersson
                                                             S. Cheshire
                                             Internet Architecture Board
                                                                May 2009

Principles of Internet Host Configuration

Status of This Memo

   This memo provides information for the Internet community.  It does
   not specify an Internet standard of any kind.  Distribution of this
   memo is unlimited.

Copyright Notice

   Copyright (c) 2009 IETF Trust and the persons identified as the
   document authors.  All rights reserved.

   This document is subject to BCP 78 and the IETF Trust's Legal
   Provisions Relating to IETF Documents in effect on the date of
   publication of this document (http://trustee.ietf.org/license-info).
   Please review these documents carefully, as they describe your rights
   and restrictions with respect to this document.

Abstract

   This document describes principles of Internet host configuration.
   It covers issues relating to configuration of Internet-layer
   parameters, as well as parameters affecting higher-layer protocols.

   1. Introduction ....................................................3
      1.1. Terminology ................................................3
      1.2. Internet Host Configuration ................................4
           1.2.1. Internet-Layer Configuration ........................4
           1.2.2. Higher-Layer Configuration ..........................6
   2. Principles ......................................................7
      2.1. Minimize Configuration .....................................7
      2.2. Less Is More ...............................................7
      2.3. Minimize Diversity .........................................8
      2.4. Lower-Layer Independence ...................................9
      2.5. Configuration Is Not Access Control .......................11
   3. Additional Discussion ..........................................12
      3.1. Reliance on General-Purpose Mechanisms ....................12
      3.2. Relationship between IP Configuration and Service
           Discovery .................................................13
           3.2.1. Fate Sharing .......................................14
      3.3. Discovering Names versus Addresses ........................15
      3.4. Dual-Stack Issues .........................................15
      3.5. Relationship between Per-Interface and Per-Host
           Configuration .............................................16
   4. Security Considerations ........................................17
      4.1. Configuration Authentication ..............................18
   5. Informative References .........................................19
   Appendix A. Acknowledgments .......................................24
   Appendix B. IAB Members at the Time of This Writing ...............24

1. Introduction

   This document describes principles of Internet host [STD3]
   configuration.  It covers issues relating to configuration of
   Internet-layer parameters, as well as parameters affecting higher-
   layer protocols.

   In recent years, a number of architectural questions have arisen, for
   which we provide guidance to protocol developers:

   o The protocol layers and general approaches that are most
     appropriate for configuration of various parameters.

   o The relationship between parameter configuration and service
     discovery.

   o The relationship between per-interface and per-host configuration.

   o The relationship between network access authentication and host
     configuration.

   o The desirability of supporting self-configuration of parameters or
     avoiding parameter configuration altogether.

   o The role of link-layer protocols and tunneling protocols in
     Internet host configuration.

   The role of the link-layer and tunneling protocols is particularly
   important, since it can affect the properties of a link as seen by
   higher layers (for example, whether privacy extensions [RFC4941] are
   available to applications).

1.1. Terminology

   link

      A communication facility or medium over which nodes can
      communicate at the link layer, i.e., the layer immediately below
      IP.  Examples are Ethernets (simple or bridged), Point-to-Point
      Protocol (PPP) links, X.25, Frame Relay, or ATM networks as well
      as Internet- or higher-layer "tunnels", such as tunnels over IPv4
      or IPv6 itself.

   on link

      An address that is assigned to an interface on a specified link.

   off link

      The opposite of "on link"; an address that is not assigned to any
      interfaces on the specified link.

   mobility agent

      Either a home agent or a foreign agent [RFC3344] [RFC3775].

1.2. Internet Host Configuration

1.2.1. Internet-Layer Configuration

   Internet-layer configuration is defined as the configuration required
   to support the operation of the Internet layer.  This includes
   configuration of per-interface and per-host parameters, including IP
   address(es), subnet prefix(es), default gateway(s), mobility
   agent(s), boot service configuration and other parameters:

   IP address(es)

      Internet Protocol (IP) address configuration includes both
      configuration of link-scope addresses as well as global addresses.
      Configuration of IP addresses is a vital step, since practically
      all of IP networking relies on the assumption that hosts have IP
      address(es) associated with (each of) their active network
      interface(s).  Used as the source address of an IP packet, these
      IP addresses indicate the sender of the packet; used as the
      destination address of a unicast IP packet, these IP addresses
      indicate the intended receiver.

      The only common example of IP-based protocols operating without an
      IP address involves address configuration, such as the use of
      DHCPv4 [RFC2131] to obtain an address.  In this case, by
      definition, DHCPv4 is operating before the host has an IPv4
      address, so the DHCP protocol designers had the choice of either
      using IP without an IP address, or not using IP at all.  The
      benefits of making IPv4 self-reliant, configuring itself using its
      own IPv4 packets, instead of depending on some other protocol,
      outweighed the drawbacks of having to use IP in this constrained
      mode.  Use of IP for purposes other than address configuration can
      safely assume that the host will have one or more IP addresses,
      which may be self-configured link-local addresses [RFC3927]
      [RFC4862], or other addresses configured via DHCP or other means.

   Subnet prefix(es)

      Once a subnet prefix is configured on an interface, hosts with an
      IP address can exchange unicast IP packets directly with on-link
      hosts within the same subnet prefix.

   Default gateway(s)

      Once a default gateway is configured on an interface, hosts with
      an IP address can send unicast IP packets to that gateway for
      forwarding to off-link hosts.

   Mobility agent(s)

      While Mobile IPv4 [RFC3344] and Mobile IPv6 [RFC3775] include
      their own mechanisms for locating home agents, it is also possible
      for mobile nodes to utilize dynamic home agent configuration.

   Boot service configuration

      Boot service configuration is defined as the configuration
      necessary for a host to obtain and perhaps also to verify an
      appropriate boot image.  This is appropriate for disk-less hosts
      looking to obtain a boot image via mechanisms such as the Trivial
      File Transfer Protocol (TFTP) [RFC1350], Network File System (NFS)
      [RFC3530], and Internet Small Computer Systems Interface (iSCSI)
      [RFC3720] [RFC4173].  It also may be useful in situations where it
      is necessary to update the boot image of a host that supports a
      disk, such as in the Preboot Execution Environment [PXE]
      [RFC4578].  While strictly speaking, boot services operate above
      the Internet layer, where boot service is used to obtain the
      Internet-layer code, it may be considered part of Internet-layer
      configuration.  While boot service parameters may be provided on a
      per-interface basis, loading and verification of a boot image
      affects behavior of the host as a whole.

   Other IP parameters

      Internet-layer parameter configuration also includes configuration
      of per-host parameters (e.g., hostname) and per-interface
      parameters (e.g., IP Time-To-Live (TTL) to use in outgoing
      packets, enabling/disabling of IP forwarding and source routing,
      and Maximum Transmission Unit (MTU)).

1.2.2. Higher-Layer Configuration

   Higher-layer configuration is defined as the configuration required
   to support the operation of other components above the Internet-
   layer.  This includes, for example:

   Name Service Configuration

      The configuration required for the host to resolve names.  This
      includes configuration of the addresses of name resolution
      servers, including IEN 116 [IEN116], Domain Name System (DNS),
      Windows Internet Name Service (WINS), Internet Storage Name
      Service (iSNS) [RFC4171] [RFC4174], and Network Information
      Service (NIS) servers [RFC3898], and the setting of name
      resolution parameters such as the DNS domain and search list
      [RFC3397], the NetBIOS node type, etc.  It may also include the
      transmission or setting of the host's own name.  Note that link-
      local name resolution services (such as NetBIOS [RFC1001], Link-
      Local Multicast Name Resolution (LLMNR) [RFC4795], and multicast
      DNS (mDNS) [mDNS]) typically do not require configuration.

      Once the host has completed name service configuration, it is
      capable of resolving names using name resolution protocols that
      require configuration.  This not only allows the host to
      communicate with off-link hosts whose IP addresses are not known,
      but, to the extent that name services requiring configuration are
      utilized for service discovery, also enables the host to discover
      services available on the network or elsewhere.  While name
      service parameters can be provided on a per-interface basis, their
      configuration will typically affect behavior of the host as a
      whole.

   Time Service Configuration

      Time service configuration includes configuration of servers for
      protocols such as the Simple Network Time Protocol (SNTP) and the
      Network Time Protocol (NTP).  Since accurate determination of the
      time may be important to operation of the applications running on
      the host (including security services), configuration of time
      servers may be a prerequisite for higher-layer operation.
      However, it is typically not a requirement for Internet-layer
      configuration.  While time service parameters can be provided on a
      per-interface basis, their configuration will typically affect
      behavior of the host as a whole.

   Other service configuration

      This can include discovery of additional servers and devices, such
      as printers, Session Initiation Protocol (SIP) proxies, etc.  This
      configuration will typically apply to the entire host.

2. Principles

This section describes basic principles of Internet host
configuration.

2.1. Minimize Configuration

   Anything that can be configured can be misconfigured.  Section 3.8 of
   "Architectural Principles of the Internet" [RFC1958] states: "Avoid
   options and parameters whenever possible.  Any options and parameters
   should be configured or negotiated dynamically rather than manually."

   That is, to minimize the possibility of configuration errors,
   parameters should be automatically computed (or at least have
   reasonable defaults) whenever possible.  For example, the Path
   Maximum Transmission Unit (PMTU) can be discovered, as described in
   "Packetization Layer Path MTU Discovery" [RFC4821], "TCP Problems
   with Path MTU Discovery" [RFC2923], "Path MTU discovery" [RFC1191],
   and "Path MTU Discovery for IP version 6" [RFC1981].

   Having a protocol design with many configurable parameters increases
   the possibilities for misconfiguration of those parameters, resulting
   in failures or other sub-optimal operation.  Eliminating or reducing
   configurable parameters helps lessen this risk.  Where configurable
   parameters are necessary or desirable, protocols can reduce the risk
   of human error by making these parameters self-configuring, such as
   by using capability negotiation within the protocol, or by automated
   discovery of other hosts that implement the same protocol.

2.2. Less Is More

   The availability of standardized, simple mechanisms for general-
   purpose Internet host configuration is highly desirable.
   "Architectural Principles of the Internet" [RFC1958] states,
   "Performance and cost must be considered as well as functionality"
   and "Keep it simple.  When in doubt during design, choose the
   simplest solution."

   To allow protocol support in many types of devices, it is important
   to minimize the footprint requirement.  For example, IP-based
   protocols are used on a wide range of devices, from supercomputers to
   small low-cost devices running "embedded" operating systems.  Since

   the resources (e.g., memory and code size) available for host
   configuration may be very small, it is desirable for a host to be
   able to configure itself in as simple a manner as possible.

   One interesting example is IP support in preboot execution
   environments.  Since by definition boot configuration is required in
   hosts that have not yet fully booted, it is often necessary for pre-
   boot code to be executed from Read Only Memory (ROM), with minimal
   available memory.  Many hosts do not have enough space in this ROM
   for even a simple implementation of TCP, so in the Preboot Execution
   Environment (PXE) the task of obtaining a boot image is performed
   using the User Datagram Protocol over IP (UDP/IP) [RFC768] instead.
   This is one reason why Internet-layer configuration mechanisms
   typically depend only on IP and UDP.  After obtaining the boot image,
   the host will have the full facilities of TCP/IP available to it,
   including support for reliable transport protocols, IPsec, etc.

   In order to reduce complexity, it is desirable for Internet-layer
   configuration mechanisms to avoid dependencies on higher layers.
   Since embedded devices may be severely constrained on how much code
   they can fit within their ROM, designing a configuration mechanism in
   such a way that it requires the availability of higher-layer
   facilities may make that configuration mechanism unusable in such
   devices.  In fact, it cannot even be guaranteed that all Internet-
   layer facilities will be available.  For example, the minimal version
   of IP in a host's boot ROM may not implement IP fragmentation and
   reassembly.

2.3. Minimize Diversity

   The number of host configuration mechanisms should be minimized.
   Diversity in Internet host configuration mechanisms presents several
   problems:

   Interoperability

      As configuration diversity increases, it becomes likely that a
      host will not support the configuration mechanism(s) available on
      the network to which it has attached, creating interoperability
      problems.

   Footprint

      For maximum interoperability, a host would need to implement all
      configuration mechanisms used on all the link layers it supports.
      This increases the required footprint, a burden for embedded
      devices.  It also leads to lower quality, since testing resources

      (both formal testing, and real-world operational use) are spread
      more thinly -- the more different configuration mechanisms a
      device supports, the less testing each one is likely to undergo.

   Redundancy

      To support diversity in host configuration mechanisms, operators
      would need to support multiple configuration services to ensure
      that hosts connecting to their networks could configure
      themselves.  This represents an additional expense for little
      benefit.

   Latency

      As configuration diversity increases, hosts supporting multiple
      configuration mechanisms may spend increasing effort to determine
      which mechanism(s) are supported.  This adds to configuration
      latency.

   Conflicts

      Whenever multiple mechanisms are available, it is possible that
      multiple configurations will be returned.  To handle this, hosts
      would need to merge potentially conflicting configurations.  This
      would require conflict-resolution logic, such as ranking of
      potential configuration sources, increasing implementation
      complexity.

   Additional traffic

      To limit configuration latency, hosts may simultaneously attempt
      to obtain configuration by multiple mechanisms.  This can result
      in increasing on-the-wire traffic, both from use of multiple
      mechanisms as well as from retransmissions within configuration
      mechanisms not implemented on the network.

   Security

      Support for multiple configuration mechanisms increases the attack
      surface without any benefit.

2.4. Lower-Layer Independence

"Architectural Principles of the Internet" [RFC1958] states,
"Modularity is good. If you can keep things separate, do so."

   It is becoming increasingly common for hosts to support multiple
   network access mechanisms, including dialup, wireless, and wired
   local area networks; wireless metropolitan and wide area networks;
   etc.  The proliferation of network access mechanisms makes it
   desirable for hosts to be able to configure themselves on multiple
   networks without adding configuration code specific to each new link
   layer.

   As a result, it is highly desirable for Internet host configuration
   mechanisms to be independent of the underlying lower layer.  That is,
   only the link-layer protocol (whether it be a physical link or a
   virtual tunnel link) should be explicitly aware of link-layer
   parameters (although those link-layer parameters may be configured by
   general Internet-layer mechanisms).  Introduction of lower-layer
   dependencies increases the likelihood of interoperability problems
   and adds Internet-layer configuration mechanisms that hosts need to
   implement.

   Lower-layer dependencies can be best avoided by keeping Internet host
   configuration above the link layer, thereby enabling configuration to
   be handled for any link layer that supports IP.  In order to provide
   media independence, Internet host configuration mechanisms should be
   link-layer protocol independent.

   While there are examples of Internet-layer configuration within the
   link layer (such as in PPP IPv4CP [RFC1332] and "Mobile radio
   interface Layer 3 specification; Core network protocols; Stage 3
   (Release 5)" [3GPP-24.008]), this approach has disadvantages.  These
   include the extra complexity of implementing different mechanisms on
   different link layers and the difficulty in adding new higher-layer
   parameters that would require defining a mechanism in each link-layer
   protocol.

   For example, "PPP Internet Protocol Control Protocol Extensions for
   Name Server Addresses" [RFC1877] was developed prior to the
   definition of the DHCPINFORM message in "Dynamic Host Configuration
   Protocol" [RFC2131]; at that time, Dynamic Host Configuration
   Protocol (DHCP) servers had not been widely implemented on access
   devices or deployed in service provider networks.  While the design
   of IPv4CP was appropriate in 1992, it should not be taken as an
   example that new link-layer technologies should emulate.  Indeed, in
   order to "actively advance PPP's most useful extensions to full
   standard, while defending against further enhancements of
   questionable value", "IANA Considerations for the Point-to-Point
   Protocol (PPP)" [RFC3818] changed the allocation of PPP numbers
   (including IPv4CP extensions) so as to no longer be "first come first
   served".

   In IPv6, where link-layer-independent mechanisms such as stateless
   autoconfiguration [RFC4862] and stateless DHCPv6 [RFC3736] are
   available, PPP IPv6CP [RFC5072] configures an Interface-Identifier
   that is similar to a Media Access Control (MAC) address.  This
   enables PPP IPv6CP to avoid duplicating DHCPv6 functionality.

   However, Internet Key Exchange Version 2 (IKEv2) [RFC4306] utilizes
   the same approach as PPP IPv4CP by defining a Configuration Payload
   for Internet host configuration for both IPv4 and IPv6.  While the
   IKEv2 approach reduces the number of packet exchanges, "Dynamic Host
   Configuration Protocol (DHCPv4) Configuration of IPsec Tunnel Mode"
   [RFC3456] points out that leveraging DHCP has advantages in terms of
   address management integration, address pool management,
   reconfiguration, and fail-over.

   Extensions to link-layer protocols for the purpose of Internet-,
   transport-, or application-layer configuration (including server
   configuration) should be avoided.  Such extensions can negatively
   affect the properties of a link as seen by higher layers.  For
   example, if a link-layer protocol (or tunneling protocol) configures
   individual IPv6 addresses and precludes using any other addresses,
   then applications that want to use privacy extensions [RFC4941] may
   not function well.  Similar issues may arise for other types of
   addresses, such as Cryptographically Generated Addresses [RFC3972].

   Avoiding lower-layer dependencies is desirable even where the lower
   layer is link independent.  For example, while the Extensible
   Authentication Protocol (EAP) may be run over any link satisfying its
   requirements (see Section 3.1 of [RFC3748]), many link layers do not
   support EAP and therefore Internet-layer configuration mechanisms
   that depend on EAP would not be usable on links that support IP but
   not EAP.

2.5. Configuration Is Not Access Control

   Network access authentication and authorization is a distinct problem
   from Internet host configuration.  Therefore, network access
   authentication and authorization is best handled independently of the
   Internet and higher-layer configuration mechanisms.

   Having an Internet- or higher-layer protocol authenticate clients is
   appropriate to prevent resource exhaustion of a scarce resource on
   the server (such as IP addresses or prefixes), but not for preventing
   hosts from obtaining access to a link.  If the user can manually
   configure the host, requiring authentication in order to obtain
   configuration parameters (such as an IP address) has little value.
   Network administrators who wish to control access to a link can
   better achieve this using technologies like Port-Based Network Access

   Control [IEEE-802.1X].  Note that client authentication is not
   required for Stateless DHCPv6 [RFC3736] since it does not result in
   allocation of any limited resources on the server.

3. Additional Discussion

3.1. Reliance on General-Purpose Mechanisms

   Protocols should either be self-configuring (especially where fate
   sharing is important), or use general-purpose configuration
   mechanisms (such as DHCP or a service discovery protocol, as noted in
   Section 3.2).  The choice should be made taking into account the
   architectural principles discussed in Section 2.

   Taking into account the general-purpose configuration mechanisms
   currently available, we see little need for development of additional
   general-purpose configuration mechanisms.

   When defining a new host parameter, protocol designers should first
   consider whether configuration is indeed necessary (see Section 2.1).

   If configuration is necessary, in addition to considering fate
   sharing (see Section 3.2.1), protocol designers should consider:

   1. The organizational implications for administrators.  For example,
      routers and servers are often administered by different sets of
      individuals, so that configuring a router with server parameters
      may require cross-group collaboration.

   2. Whether the need is to configure a set of interchangeable servers
      or to select a particular server satisfying a set of criteria.
      See Section 3.2.

   3. Whether IP address(es) should be configured, or name(s).  See
      Section 3.3.

   4. If IP address(es) are configured, whether IPv4 and IPv6 addresses
      should be configured simultaneously or separately.  See Section
      3.4.

   5. Whether the parameter is a per-interface or a per-host parameter.
      For example, configuration protocols such as DHCP run on a per-
      interface basis and hence are more appropriate for per-interface
      parameters.

   6. How per-interface configuration affects host-wide behavior.  For
      example, whether the host should select a subset of the per-
      interface configurations, or whether the configurations are to
      merged, and if so, how this is done.  See Section 3.5.

3.2. Relationship between IP Configuration and Service Discovery

   Higher-layer configuration often includes configuring server
   addresses.  The question arises as to how this differs from "service
   discovery" as provided by Service Discovery protocols such as
   "Service Location Protocol, Version 2" (SLPv2) [RFC2608] or "DNS-
   Based Service Discovery" (DNS-SD) [DNS-SD].

   In Internet host configuration mechanisms such as DHCP, if multiple
   server instances are provided, they are considered interchangeable.
   For example, in a list of time servers, the servers are considered
   interchangeable because they all provide the exact same service --
   telling you the current time.  In a list of local caching DNS
   servers, the servers are considered interchangeable because they all
   should give you the same answer to any DNS query.  In service
   discovery protocols, on the other hand, a host desires to find a
   server satisfying a particular set of criteria, which may vary by
   request.  When printing a document, it is not the case that any
   printer will do.  The speed, capabilities, and physical location of
   the printer matter to the user.

   Information learned via DHCP is typically learned once, at boot time,
   and after that may be updated only infrequently (e.g., on DHCP lease
   renewal), if at all.  This makes DHCP appropriate for information
   that is relatively static and unchanging over these time intervals.
   Boot-time discovery of server addresses is appropriate for service
   types where there are a small number of interchangeable servers that
   are of interest to a large number of clients.  For example, listing
   time servers in a DHCP packet is appropriate because an organization
   may typically have only two or three time servers, and most hosts
   will be able to make use of that service.  Listing all the printers
   or file servers at an organization is a lot less useful, because the
   list may contain hundreds or thousands of entries, and on a given day
   a given user may not use any of the printers in that list.

   Service discovery protocols can support discovery of servers on the
   Internet, not just those within the local administrative domain.  For
   example, see "Remote Service Discovery in the Service Location
   Protocol (SLP) via DNS SRV" [RFC3832] and DNS-Based Service Discovery
   [DNS-SD].  Internet host configuration mechanisms such as DHCP, on
   the other hand, typically assume the server or servers in the local
   administrative domain contain the authoritative set of information.

   For the service discovery problem (i.e., where the criteria varies on
   a per-request basis, even from the same host), protocols should
   either be self-discovering (if fate sharing is critical), or use a
   general-purpose service discovery mechanism.

   In order to avoid a dependency on multicast routing, it is necessary
   for a host to either restrict discovery to services on the local link
   or to discover the location of a Directory Agent (DA).  Since the DA
   may not be available on the local link, service discovery beyond the
   local link is typically dependent on a mechanism for configuring the
   DA address or name.  As a result, service discovery protocols can
   typically not be relied upon for obtaining basic Internet-layer
   configuration, although they can be used to obtain higher-layer
   configuration parameters.

3.2.1. Fate Sharing

   If a server (or set of servers) is needed to get a set of
   configuration parameters, "fate sharing" (Section 2.3 of [RFC1958])
   is preserved if those parameters are ones that cannot be usefully
   used without those servers being available.  In this case,
   successfully obtaining those parameters via other means has little
   benefit if they cannot be used because the required servers are not
   available.  The possibility of incorrect information being configured
   is minimized if there is only one machine that is authoritative for
   the information (i.e., there is no need to keep multiple
   authoritative servers in sync).  For example, learning default
   gateways via Router Advertisements provides perfect fate sharing.
   That is, gateway addresses can be obtained if and only if they can
   actually be used.  Similarly, obtaining DNS server configuration from
   a DNS server would provide fate sharing since the configuration would
   only be obtainable if the DNS server were available.

   While fate sharing is a desirable property of a configuration
   mechanism, in some situations fate sharing may not be possible.  When
   utilized to discover services on the local link, service discovery
   protocols typically provide for fate sharing, since hosts providing
   service information typically also provide the services.  However,
   this is no longer the case when service discovery is assisted by a
   Directory Agent (DA).  First of all, the DA's list of operational
   servers may not be current, so it is possible that the DA may provide
   clients with service information that is out of date.  For example, a
   DA's response to a client's service discovery query may contain stale
   information about servers that are no longer operational.  Similarly,
   recently introduced servers might not yet have registered themselves
   with the DA.  Furthermore, the use of a DA for service discovery also
   introduces a dependency on whether the DA is operational, even though
   the DA is typically not involved in the delivery of the service.

   Similar limitations exist for other server-based configuration
   mechanisms such as DHCP.  Typically DHCP servers do not check for the
   liveness of the configuration information they provide, and do not
   discover new configuration information automatically.  As a result,
   there is no guarantee that configuration information will be current.

   Section 3.3 of "IPv6 Host Configuration of DNS Server Information
   Approaches" [RFC4339] discusses the use of well-known anycast
   addresses for discovery of DNS servers.  The use of anycast addresses
   enables fate sharing, even where the anycast address is provided by
   an unrelated server.  However, in order to be universally useful,
   this approach would require allocation of one or more well-known
   anycast addresses for each service.  Configuration of more than one
   anycast address is desirable to allow the client to fail over faster
   than would be possible from routing protocol convergence.

3.3. Discovering Names vs. Addresses

   In discovering servers other than name resolution servers, it is
   possible to either discover the IP addresses of the server(s), or to
   discover names, each of which may resolve to a list of addresses.

   It is typically more efficient to obtain the list of addresses
   directly, since this avoids the extra name resolution steps and
   accompanying latency.  On the other hand, where servers are mobile,
   the name-to-address binding may change, requiring a fresh set of
   addresses to be obtained.  Where the configuration mechanism does not
   support fate sharing (e.g., DHCP), providing a name rather than an
   address can simplify operations, assuming that the server's new
   address is manually or automatically updated in the DNS; in this
   case, there is no need to re-do parameter configuration, since the
   name is still valid.  Where fate sharing is supported (e.g., service
   discovery protocols), a fresh address can be obtained by re-
   initiating parameter configuration.

   In providing the IP addresses for a set of servers, it is desirable
   to distinguish which IP addresses belong to which servers.  If a
   server IP address is unreachable, this enables the host to try the IP
   address of another server, rather than another IP address of the same
   server, in case the server is down.  This can be enabled by
   distinguishing which addresses belong to the same server.

3.4. Dual-Stack Issues

   One use for learning a list of interchangeable server addresses is
   for fault tolerance, in case one or more of the servers are
   unresponsive.  Hosts will typically try the addresses in turn, only
   attempting to use the second and subsequent addresses in the list if

   the first one fails to respond quickly enough.  In such cases, having
   the list sorted in order of expected likelihood of success will help
   clients get results faster.  For hosts that support both IPv4 and
   IPv6, it is desirable to obtain both IPv4 and IPv6 server addresses
   within a single list.  Obtaining IPv4 and IPv6 addresses in separate
   lists, without indicating which server(s) they correspond to,
   requires the host to use a heuristic to merge the lists.

   For example, assume there are two servers, A and B, each with one
   IPv4 address and one IPv6 address.  If the first address the host
   should try is (say) the IPv6 address of server A, then the second
   address the host should try, if the first one fails, would generally
   be the IPv4 address of server B.  This is because the failure of the
   first address could be due to either server A being down, or some
   problem with the host's IPv6 address, or a problem with connectivity
   to server A.  Trying the IPv4 address next is preferred since the
   reachability of the IPv4 address is independent of all potential
   failure causes.

   If the list of IPv4 server addresses were obtained separately from
   the list of IPv6 server addresses, a host trying to merge the lists
   would not know which IPv4 addresses belonged to the same server as
   the IPv6 address it just tried.  This can be solved either by
   explicitly distinguishing which addresses belong to which server or,
   more simply, by configuring the host with a combined list of both
   IPv4 and IPv6 addresses.  Note that the same issue can arise with any
   mechanism (e.g., DHCP, DNS, etc.) for obtaining server IP addresses.

   Configuring a combined list of both IPv4 and IPv6 addresses gives the
   configuration mechanism control over the ordering of addresses, as
   compared with configuring a name and allowing the host resolver to
   determine the address list ordering.  See "Dynamic Host Configuration
   Protocol (DHCP): IPv4 and IPv6 Dual-Stack Issues" [RFC4477] for more
   discussion of dual-stack issues in the context of DHCP.

3.5. Relationship between Per-Interface and Per-Host Configuration

   Parameters that are configured or acquired on a per-interface basis
   can affect behavior of the host as a whole.  Where only a single
   configuration can be applied to a host, the host may need to
   prioritize the per-interface configuration information in some way
   (e.g., most trusted to least trusted).  If the host needs to merge
   per-interface configuration to produce a host-wide configuration, it
   may need to take the union of the per-host configuration parameters
   and order them in some way (e.g., highest speed interface to lowest
   speed interface).  Which procedure is to be applied and how this is
   accomplished may vary depending on the parameter being configured.
   Examples include:

   Boot service configuration

      While boot service configuration can be provided on multiple
      interfaces, a given host may be limited in the number of boot
      loads that it can handle simultaneously.  For example, a host not
      supporting virtualization may only be capable of handling a single
      boot load at a time, or a host capable of supporting N virtual
      machines may only be capable of handling up to N simultaneous boot
      loads.  As a result, a host may need to select which boot load(s)
      it will act on, out of those configured on a per-interface basis.
      This requires that the host prioritize them (e.g., most to least
      trusted).

   Name service configuration

      While name service configuration is provided on a per-interface
      basis, name resolution configuration typically will affect
      behavior of the host as a whole.  For example, given the
      configuration of DNS server addresses and searchlist parameters on
      each interface, the host determines what sequence of name service
      queries is to be sent on which interfaces.

   Since the algorithms used to determine per-host behavior based on
   per-interface configuration can affect interoperability, it is
   important for these algorithms to be understood by implementers.  We
   therefore recommend that documents defining per-interface mechanisms
   for acquiring per-host configuration (e.g., DHCP or IPv6 Router
   Advertisement options) include guidance on how to deal with multiple
   interfaces.  This may include discussions of the following items:

   1. Merging.  How are per-interface configurations combined to produce
      a per-host configuration? Is a single configuration selected, or
      is the union of the configurations taken?

   2. Prioritization.  Are the per-interface configurations prioritized
      as part of the merge process?  If so, what are some of the
      considerations to be taken into account in prioritization?

4.  Security Considerations

   Secure IP configuration presents a number of challenges.  In addition
   to denial-of-service and man-in-the-middle attacks, attacks on
   configuration mechanisms may target particular parameters.  For
   example, attackers may target DNS server configuration in order to
   support subsequent phishing or pharming attacks such as those
   described in "New trojan in mass DNS hijack" [DNSTrojan].  A number
   of issues exist with various classes of parameters, as discussed in
   Section 2.6, Section 4.2.7 of "IPv6 Neighbor Discovery (ND) Trust

   Models and Threats" [RFC3756], Section 1.1 of "Authentication for
   DHCP Messages" [RFC3118], and Section 23 of "Dynamic Host
   Configuration Protocol for IPv6 (DHCPv6)" [RFC3315].  Given the
   potential vulnerabilities, hosts often restrict support for DHCP
   options to the minimum set required to provide basic TCP/IP
   configuration.

   Since boot configuration determines the boot image to be run by the
   host, a successful attack on boot configuration could result in an
   attacker gaining complete control over a host.  As a result, it is
   particularly important that boot configuration be secured.
   Approaches to boot configuration security are described in
   "Bootstrapping Clients using the Internet Small Computer System
   Interface (iSCSI) Protocol" [RFC4173] and "Preboot Execution
   Environment (PXE) Specification" [PXE].

4.1. Configuration Authentication

   The techniques available for securing Internet-layer configuration
   are limited.  While it is technically possible to perform a very
   limited subset of IP networking operations without an IP address, the
   capabilities are severely restricted.  A host without an IP address
   cannot receive conventional unicast IP packets, only IP packets sent
   to the broadcast or a multicast address.  Configuration of an IP
   address enables the use of IP fragmentation; packets sent from the
   unknown address cannot be reliably reassembled, since fragments from
   multiple hosts using the unknown address might be reassembled into a
   single IP packet.  Without an IP address, it is not possible to take
   advantage of security facilities such as IPsec, specified in
   "Security Architecture for the Internet Protocol" [RFC4301] or
   Transport Layer Security (TLS) [RFC5246].  As a result, configuration
   security is typically implemented within the configuration protocols
   themselves.

   PPP [RFC1661] does not support secure negotiation within IPv4CP
   [RFC1332] or IPv6CP [RFC5072], enabling an attacker with access to
   the link to subvert the negotiation.  In contrast, IKEv2 [RFC4306]
   provides encryption, integrity, and replay protection for
   configuration exchanges.

   Where configuration packets are only expected to originate on
   particular links or from particular hosts, filtering can help control
   configuration spoofing.  For example, a wireless access point usually
   has no reason to forward broadcast DHCP DISCOVER packets to its
   wireless clients, and usually should drop any DHCP OFFER packets
   received from those wireless clients, since, generally speaking,
   wireless clients should be requesting addresses from the network, not

   offering them.  To prevent spoofing, communication between the DHCP
   relay and servers can be authenticated and integrity protected using
   a mechanism such as IPsec.

   Internet-layer secure configuration mechanisms include SEcure
   Neighbor Discovery (SEND) [RFC3971] for IPv6 stateless address
   autoconfiguration [RFC4862], or DHCP authentication for stateful
   address configuration.  DHCPv4 [RFC2131] initially did not include
   support for security; this was added in "Authentication for DHCP
   Messages" [RFC3118].  DHCPv6 [RFC3315] included security support.
   However, DHCP authentication is not widely implemented for either
   DHCPv4 or DHCPv6.

   Higher-layer configuration can make use of a wider range of security
   techniques.  When DHCP authentication is supported, higher-layer
   configuration parameters provided by DHCP can be secured.  However,
   even if a host does not support DHCPv6 authentication, higher-layer
   configuration via Stateless DHCPv6 [RFC3736] can still be secured
   with IPsec.

   Possible exceptions can exist where security facilities are not
   available until later in the boot process.  It may be difficult to
   secure boot configuration even once the Internet layer has been
   configured, if security functionality is not available until after
   boot configuration has been completed.  For example, it is possible
   that Kerberos, IPsec, or TLS will not be available until later in the
   boot process; see "Bootstrapping Clients using the Internet Small
   Computer System Interface (iSCSI) Protocol" [RFC4173] for discussion.

   Where public key cryptography is used to authenticate and integrity-
   protect configuration, hosts need to be configured with trust anchors
   in order to validate received configuration messages.  For a node
   that visits multiple administrative domains, acquiring the required
   trust anchors may be difficult.

5. Informative References

   [3GPP-24.008] 3GPP TS 24.008 V5.8.0, "Mobile radio interface Layer 3
                 specification; Core network protocols; Stage 3 (Release
                 5)", June 2003.

   [DNSTrojan]   Goodin, D., "New trojan in mass DNS hijack", The
                 Register, December 5, 2008,
                 http://www.theregister.co.uk/2008/12/05/
                 new_dnschanger_hijacks/

   [IEN116]      J. Postel, "Internet Name Server", IEN 116, August
                 1979, http://www.ietf.org/rfc/ien/ien116.txt

   [IEEE-802.1X] Institute of Electrical and Electronics Engineers,
                 "Local and Metropolitan Area Networks: Port-Based
                 Network Access Control", IEEE Standard 802.1X-2004,
                 December 2004.

   [DNS-SD]      Cheshire, S., and M. Krochmal, "DNS-Based Service
                 Discovery", Work in Progress, September 2008.

   [mDNS]        Cheshire, S. and M. Krochmal, "Multicast DNS", Work in
                 Progress, September 2008.

   [PXE]         Henry, M. and M. Johnston, "Preboot Execution
                 Environment (PXE) Specification", September 1999,
                 http://www.pix.net/software/pxeboot/archive/pxespec.pdf

   [RFC768]      Postel, J., "User Datagram Protocol", STD 6, RFC 768,
                 August 1980.

   [RFC1001]     NetBIOS Working Group in the Defense Advanced Research
                 Projects Agency, Internet Activities Board, and End-
                 to-End Services Task Force, "Protocol standard for a
                 NetBIOS service on a TCP/UDP transport: Concepts and
                 methods", STD 19, RFC 1001, March 1987.

   [RFC1191]     Mogul, J. and S. Deering, "Path MTU discovery", RFC
                 1191, November 1990.

   [RFC1332]     McGregor, G., "The PPP Internet Protocol Control
                 Protocol (IPCP)", RFC 1332, May 1992.

   [RFC1350]     Sollins, K., "The TFTP Protocol (Revision 2)", STD 33,
                 RFC 1350, July 1992.

   [RFC1661]     Simpson, W., Ed., "The Point-to-Point Protocol (PPP)",
                 STD 51, RFC 1661, July 1994.

   [RFC1877]     Cobb, S., "PPP Internet Protocol Control Protocol
                 Extensions for Name Server Addresses", RFC 1877,
                 December 1995.

   [RFC1958]     Carpenter, B., Ed., "Architectural Principles of the
                 Internet", RFC 1958, June 1996.

   [RFC1981]     McCann, J., Deering, S., and J. Mogul, "Path MTU
                 Discovery for IP version 6", RFC 1981, August 1996.

   [RFC2131]     Droms, R., "Dynamic Host Configuration Protocol", RFC
                 2131, March 1997.

   [RFC2608]     Guttman, E., Perkins, C., Veizades, J., and M. Day,
                 "Service Location Protocol, Version 2", RFC 2608, June
                 1999.

   [RFC2923]     Lahey, K., "TCP Problems with Path MTU Discovery", RFC
                 2923, September 2000.

   [RFC3118]     Droms, R., Ed., and W. Arbaugh, Ed., "Authentication
                 for DHCP Messages", RFC 3118, June 2001.

   [RFC3315]     Droms, R., Ed., Bound, J., Volz, B., Lemon, T.,
                 Perkins, C., and M. Carney, "Dynamic Host Configuration
                 Protocol for IPv6 (DHCPv6)", RFC 3315, July 2003.

   [RFC3344]     Perkins, C., Ed., "IP Mobility Support for IPv4", RFC
                 3344, August 2002.

   [RFC3397]     Aboba, B. and S. Cheshire, "Dynamic Host Configuration
                 Protocol (DHCP) Domain Search Option", RFC 3397,
                 November 2002.

   [RFC3456]     Patel, B., Aboba, B., Kelly, S., and V. Gupta, "Dynamic
                 Host Configuration Protocol (DHCPv4) Configuration of
                 IPsec Tunnel Mode", RFC 3456, January 2003.

   [RFC3530]     Shepler, S., Callaghan, B., Robinson, D., Thurlow, R.,
                 Beame, C., Eisler, M., and D. Noveck, "Network File
                 System (NFS) version 4 Protocol", RFC 3530, April 2003.

   [RFC3720]     Satran, J., Meth, K., Sapuntzakis, C., Chadalapaka, M.,
                 and E. Zeidner, "Internet Small Computer Systems
                 Interface (iSCSI)", RFC 3720, April 2004.

   [RFC3736]     Droms, R., "Stateless Dynamic Host Configuration
                 Protocol (DHCP) Service for IPv6", RFC 3736, April
                 2004.

   [RFC3748]     Aboba, B., Blunk, L., Vollbrecht, J., Carlson, J., and
                 H. Levkowetz, Ed., "Extensible Authentication Protocol
                 (EAP)", RFC 3748, June 2004.

   [RFC3756]     Nikander, P., Ed., Kempf, J., and E. Nordmark, "IPv6
                 Neighbor Discovery (ND) Trust Models and Threats", RFC
                 3756, May 2004.

   [RFC3775]     Johnson, D., Perkins, C., and J. Arkko, "Mobility
                 Support in IPv6", RFC 3775, June 2004.

   [RFC3818]     Schryver, V., "IANA Considerations for the Point-to-
                 Point Protocol (PPP)", BCP 88, RFC 3818, June 2004.

   [RFC3832]     Zhao, W., Schulzrinne, H., Guttman, E., Bisdikian, C.,
                 and W. Jerome, "Remote Service Discovery in the Service
                 Location Protocol (SLP) via DNS SRV", RFC 3832, July
                 2004.

   [RFC3898]     Kalusivalingam, V., "Network Information Service (NIS)
                 Configuration Options for Dynamic Host Configuration
                 Protocol for IPv6 (DHCPv6)", RFC 3898, October 2004.

   [RFC3927]     Cheshire, S., Aboba, B., and E. Guttman, "Dynamic
                 Configuration of IPv4 Link-Local Addresses", RFC 3927,
                 May 2005.

   [RFC3971]     Arkko, J., Ed., Kempf, J., Zill, B., and P. Nikander,
                 "SEcure Neighbor Discovery (SEND)", RFC 3971, March
                 2005.

   [RFC3972]     Aura, T., "Cryptographically Generated Addresses
                 (CGA)", RFC 3972, March 2005.

   [RFC4171]     Tseng, J., Gibbons, K., Travostino, F., Du Laney, C.,
                 and J. Souza, "Internet Storage Name Service (iSNS)",
                 RFC 4171, September 2005.

   [RFC4173]     Sarkar, P., Missimer, D., and C. Sapuntzakis,
                 "Bootstrapping Clients using the Internet Small
                 Computer System Interface (iSCSI) Protocol", RFC 4173,
                 September 2005.

   [RFC4174]     Monia, C., Tseng, J., and K. Gibbons, "The IPv4 Dynamic
                 Host Configuration Protocol (DHCP) Option for the
                 Internet Storage Name Service", RFC 4174, September
                 2005.

   [RFC4301]     Kent, S. and K. Seo, "Security Architecture for the
                 Internet Protocol", RFC 4301, December 2005.

   [RFC4306]     Kaufman, C., Ed., "Internet Key Exchange (IKEv2)
                 Protocol", RFC 4306, December 2005.

   [RFC4339]     Jeong, J., Ed., "IPv6 Host Configuration of DNS Server
                 Information Approaches", RFC 4339, February 2006.

   [RFC4477]     Chown, T., Venaas, S., and C. Strauf, "Dynamic Host
                 Configuration Protocol (DHCP): IPv4 and IPv6 Dual-Stack
                 Issues", RFC 4477, May 2006.

   [RFC4578]     Johnston, M. and S. Venaas, Ed., "Dynamic Host
                 Configuration Protocol (DHCP) Options for the Intel
                 Preboot eXecution Environment (PXE)", RFC 4578,
                 November 2006.

   [RFC4795]     Aboba, B., Thaler, D., and L. Esibov, "Link-local
                 Multicast Name Resolution (LLMNR)", RFC 4795, January
                 2007.

   [RFC4821]     Mathis, M. and J. Heffner, "Packetization Layer Path
                 MTU Discovery", RFC 4821, March 2007.

   [RFC4862]     Thomson, S., Narten, T., and T. Jinmei, "IPv6 Stateless
                 Address Autoconfiguration", RFC 4862, September 2007.

   [RFC4941]     Narten, T., Draves, R., and S. Krishnan, "Privacy
                 Extensions for Stateless Address Autoconfiguration in
                 IPv6", RFC 4941, September 2007.

   [RFC5072]     Varada, S., Ed., Haskins, D., and E. Allen, "IP Version
                 6 over PPP", RFC 5072, September 2007.

   [RFC5246]     Dierks, T. and E. Rescorla, "The Transport Layer
                 Security (TLS) Protocol Version 1.2", RFC 5246, August
                 2008.

   [STD3]        Braden, R., Ed., "Requirements for Internet Hosts -
                 Communication Layers", STD 3, RFC 1122, October 1989.

                 Braden, R., Ed., "Requirements for Internet Hosts -
                 Application and Support", STD 3, RFC 1123, October
                 1989.

Appendix A. Acknowledgments

   Elwyn Davies, Bob Hinden, Pasi Eronen, Jari Arkko, Pekka Savola,
   James Kempf, Ted Hardie, and Alfred Hoenes provided valuable input on
   this document.

Appendix B. IAB Members at the Time of This Writing

   Loa Andersson
   Gonzalo Camarillo
   Stuart Cheshire
   Russ Housley
   Olaf Kolkman
   Gregory Lebovitz
   Barry Leiba
   Kurtis Lindqvist
   Andrew Malis
   Danny McPherson
   David Oran
   Dave Thaler
   Lixia Zhang

Authors' Addresses

   Bernard Aboba
   Microsoft Corporation
   One Microsoft Way
   Redmond, WA 98052

   EMail: bernarda@microsoft.com

   Dave Thaler
   Microsoft Corporation
   One Microsoft Way
   Redmond, WA 98052

   EMail: dthaler@microsoft.com

   Loa Andersson
   Ericsson AB

   EMail: loa.andersson@ericsson.com

   Stuart Cheshire
   Apple Computer, Inc.
   1 Infinite Loop
   Cupertino, CA 95014

   EMail: cheshire@apple.com