Network Working Group T. Chown Request for Comments: 4477 University of Southampton Category: Informational S. Venaas UNINETT C. Strauf Clausthal University of Technology May 2006 Dynamic Host Configuration Protocol (DHCP): IPv4 and IPv6 Dual-Stack Issues 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) The Internet Society (2006). Abstract A node may have support for communications using IPv4 and/or IPv6 protocols. Such a node may wish to obtain IPv4 and/or IPv6 configuration settings via the Dynamic Host Configuration Protocol (DHCP). The original version of DHCP (RFC 2131) designed for IPv4 has now been complemented by a new DHCPv6 (RFC 3315) for IPv6. This document describes issues identified with dual IP version DHCP interactions, the most important aspect of which is how to handle potential problems in clients processing configuration information received from both DHCPv4 and DHCPv6 servers. The document makes a recommendation on the general strategy on how best to handle such issues and identifies future work to be undertaken.
Table of Contents 1. Introduction ....................................................3 2. Configuration Scenarios .........................................3 3. Dual-Stack Issues ...............................................4 3.1. Handling Multiple Responses ................................4 3.2. Different Administrative Management ........................5 3.3. Multiple Interfaces ........................................5 3.4. DNS Load Balancing .........................................5 3.5. DNS Search Path Issues .....................................5 3.6. Protocol Startup Sequence ..................................6 3.7. DHCP Option Variations .....................................6 3.8. Security Issues ............................................6 4. Potential Solutions .............................................7 4.1. Separate DHCP Servers ......................................7 4.2. Single DHCPv6 Server .......................................8 4.3. Optimising for Failure with Lists of Addresses .............9 4.4. Administrative and Other Areas ............................10 5. Summary ........................................................10 6. Security Considerations ........................................12 7. Acknowledgements ...............................................12 8. Informative References .........................................12
1. Introduction The original specification of the Dynamic Host Configuration Protocol (DHCP) was made with only IPv4 in mind. That specification has been subsequently revised, up to the latest version of DHCP . With the arrival of IPv6, a new DHCP specification for IPv6 has been designed and published as DHCPv6 . These protocols allow nodes to communicate via IPv4 or IPv6 (respectively) to retrieve configuration settings for operation in a managed environment. While an IPv6 node may acquire address-related configuration settings via IPv6 stateless address autoconfiguration , such a node may wish to use stateless DHCPv6  for other administratively configured options, such as DNS or NTP. In early IPv6 deployments, a dual-stack mode of operation is typically used. There will thus be nodes that require both IPv4 and IPv6 configuration settings. This document discusses issues with obtaining such settings in a dual-stack environment. There is a general multihoming issue to be solved for DHCP. A host might simultaneously be connected to multiple networks managed by multiple parties. Also, IPv4 and IPv6 might be managed by separate parties. While these issues are touched on in this document, here we focus on the specific issues for operating DHCP in a mixed (typically dual-stack) IPv4 and IPv6 environment within a single administrative domain. In this document, we refer to a "DHCP server" as a server implementing the original DHCP , and a "DHCPv6 server" as a server implementing DHCPv6  or its stateless subset . 2. Configuration Scenarios For a node in an IPv4-only or IPv6-only environment, the choice of DHCP server is a straightforward one; a DHCP server for IPv4, or a DHCPv6 server for IPv6. In a dual-stack environment a node in a managed environment will need to obtain both IPv4 and IPv6 configuration settings, such as the following: o IPv4 address o IPv6 address o NTP server
o DNS server o NIS server o DNS search path While the format of address settings will be IP specific, the node may equally well acquire IPv4 or IPv6 addresses for some settings, such as for DNS or NTP, if those services are available via IPv4 or IPv6 transport. Currently, a DHCP server returns IPv4 data, while a DHCPv6 server returns IPv6 data. It is worth noting that in an IPv4 environment, with a DHCP server, the choice of whether to use DHCP is made by the node. In an IPv6 environment, the use of the managed and other bits in the Router Advertisement can offer a hint to the node whether or not to use full DHCPv6 or its stateless variant. It is perhaps not clear whether a dual-stack node should do DHCP for IPv4 if Managed and OtherConfig flags in the Router Advertisement are both off; it seems most appropriate that the decision to use DHCP for IPv4 or not should be as if the host were IPv4-only. 3. Dual-Stack Issues In this section, we list issues that have been raised to date, related to dual-stack DHCP operation. It has been noted from comments that the first four, and possibly five, subsections here may also be viewed as multihoming issues. 3.1. Handling Multiple Responses The general question is how to handle configuration information that may be gathered from multiple sources. Where those sources are DHCP and DHCPv6 servers (which may be two physical nodes or two servers running on the same node) the client node needs to know whether to use the most recent data, or whether to perform some merger or union of the responses by certain rules. A method for merging lists of addresses, for options that carry such information, may also be required. A node may choose to ask a DHCPv6 server and only use a DHCP server if no response is received. Merging is possible, but is likely to be complex. There could be some priority, so that if both DHCP and DHCPv6 servers offer a value, only one is used. Or the node could choose to store and use both, in some order of its choosing. Merging issues are further discussed later in this document.
A node may also obtain information from other sources, such as a manual configuration file (for example, /etc/resolv.conf for DNS data on many UNIX systems). A node configured manually to use an IPv6 DNS server may lose that configuration if it is in a dual-stack environment and uses DHCP to obtain IPv4 settings; the new IPv4 settings from the DHCP response may then overwrite the manual IPv6 DNS setting. 3.2. Different Administrative Management In some deployments, the IPv4 and IPv6 services may not be administered by the same organisation or people, such as in a community wireless environment. This poses problems for consistency of data offered by either DHCP version. There may also be different connectivity for the protocols, and the client may gain advantage from knowing which 'administrative domain' is supplying which information. A client may need to use different received information depending on which connectivity is being used. In the example of the community wireless environment, the question of which connectivity is 'better' is a separate issue. 3.3. Multiple Interfaces A node may have multiple interfaces and run IPv4 and IPv6 on different interfaces. A question then is whether the settings are per interface or per node. Per-interface settings can be complex because a client node needs to know which interface system settings, like NTP server, came from. And it may not be apparent which setting should be used if, for example, an NTP server option is received on multiple interfaces, potentially over different protocols. 3.4. DNS Load Balancing In some cases it is preferable to list DNS server information in an ordered way per node for load balancing, giving different responses to different clients. Responses from different DHCP and DHCPv6 servers may make such configuration problematic, if the knowledge of the load balancing is not available to both servers. 3.5. DNS Search Path Issues The DNS search path may vary for administrative reasons. For example, a site under the domain example.com may choose to place an early IPv6 deployment under the subdomain ipv6.example.com, until it is confident of offering a full dual-stack service under its main
domain. The subtlety here is that the DNS search path then affects the choice of protocol used, such as IPv6 for nodes in ipv6.example.com. 3.6. Protocol Startup Sequence In the dual-stack environment, one needs to consider what happens if, for example, the IPv6 interface (transport) is started after DHCPv4 was used to configure the client. Should the client then simply discard the current IPv4 information, or merge it with a subsequent IPv6 response? It may also be possible that one protocol is shut down or started while the system is running. There are similarities here to issues when DHCP renewals have information that may appear that previously was not available (or no longer carry information that has been removed). 3.7. DHCP Option Variations Some options in DHCP are not available in DHCPv6 and vice versa. Some IP-version limitations naturally apply; for example, only IPv6 addresses can be in an IPv6 NTP option. The DHCP and DHCPv6 option numbers may be different. Some sites may choose to use IPv4-mapped addresses in DHCPv6-based options. The merits and drawbacks of such an approach need discussion. A site administrator may wish to configure all their dual-stack nodes with (say) two NTP servers, one of which has an IPv4 address, the other an IPv6 address. In this case, it may be desirable for an NTP option to carry a list of addresses, where some may be IPv4 and some may be IPv6. In general one could consider having DHCPv6 options that can carry a mix of IPv4 and IPv6 addresses. 3.8. Security Issues This document does not introduce any new security issues per se. A detailed analysis of DHCP and DHCPv6 security is out of scope for this document. While there is a specification for authentication for DHCP messages , the standard seems to have very few, if any, implementations. Thus DHCP and DHCPv6 servers are still liable to be spoofed. Adding an additional protocol may give an extra avenue for attack, should an attacker perhaps spoof a DHCPv6 server but not a DHCP server.
4. Potential Solutions Here we discuss the two broad solution strategies proposed within the IETF dhc WG. The first is to run separate DHCP and DHCPv6 servers (with the client merging information received from both where necessary, or perhaps choosing to query a particular version first). The second is to run only a DHCPv6 server and relay IPv4 configuration information within (new) IPv4 configuration options. 4.1. Separate DHCP Servers One solution is to run separate DHCP and DHCPv6 servers. These may or may not be run on the same physical node. The information served from the DHCP servers could be generated from a single database instance for consistency. One might have a single server instance supporting both DHCPv4 and DHCPv6 protocols. In this approach, some best practice guidance is required for how multiple responses are handled or merged. Administrators have the onus to maintain consistency (for example, scripts may generate common DHCP and DHCPv6 configuration files). In some cases, inconsistencies may not matter. In a simple case, an NTP server will give the same time whether accessed by IPv4 or IPv6. Even if different recursive DNS servers are offered via DHCP or DHCPv6, then those name servers should provide the same response to a given query. In cases where sites may be operating a 'two-faced DNS', this will still hold true if the node is on the same topological point on the network from an IPv4 or IPv6 perspective. The order of DNS servers in a node's configuration is not important, unless DNS load balancing is required. In other cases, inconsistencies may be an issue; for example, where lists of values are returned, an algorithm is needed for list merger (e.g., "alternate, DHCPv6 first"). Or there may be incompatible configuration values where, for example, DHCPv6 supplies domain names (such the SMTP or POP servers) whereas DHCPv4 provides only IPv4 addresses. In the case of separate servers, there are some options, like DNS search path, that aren't used in a specific IP protocol context. The multiple server approach will have some simplifications. The DHCPv4 and DHCPv6 servers may provide the same value for a particular parameter, in which case there is no conflict. In some cases, the value may be different, but the effect should be the same (such as an
NTP server). The crux of the issue is to identify where differences may occur and where these differences will have an impact on node behaviour. One possible solution is to have per-host preferences, or an ordered list of preferences, for example, "use manually configured", "prefer DHCPv4", or "prefer DHCPv6", assuming the host can act based upon which protocol is used. It is then up to the site administrator to ensure that values returned from either DHCP are consistent (a principle that extends if other methods are used, such as NIS or Service Location Protocol (SLP)). 4.2. Single DHCPv6 Server There is an argument for not having to configure and operate both DHCP and DHCPv6 servers in a dual-stack site environment. The use of both servers may also lead to some redundancy in the information served. Thus, one solution may be to modify DHCPv6 to be able to return IPv4 information. This solution is hinted at in the DHCPv6  specification: "If there is sufficient interest and demand, integration can be specified in a document that extends DHCPv6 to carry IPv4 addresses and configuration information." This solution may allow DHCP for IPv4 to be completely replaced by DHCPv6 with additional IPv4 information options, for dual-stack nodes. A general argument is that which DHCP protocol is used (whether it's over IPv4 or IPv6) shouldn't affect what kind of addresses you can get configured with it, and that simplicity and predictability come from using a single server over a single transport. IPv4-capable hosts will likely remain for at least 10 years, probably much longer; do we want dual-stack hosts (which will become the norm) to do both DHCPv4 and DHCPv6 forever while dual-stack? If you need both servers to configure interfaces with addresses, and get other configurations, then you rely on two separate protocols to work (servers and relays, etc.) in order for the host to behave correctly. This approach may require the listing of a mix of IPv4 and IPv6 addresses for an option. This could then be considered when new IPv6 options are introduced. There could be just two options needed, one new option for the address delegation, and one for doing encapsulation. Also, there are a number of paradigms in DHCPv6 that we miss in DHCPv4. An example is movement away from using MAC addresses for per-host address assignment and instead using DHCP Unique Identifier (DUIDs) or Identity Association Identifiers (IAIDs). As stated in Section 9 of RFC3315, DHCPv6 servers use DUIDs to identify clients for the selection of configuration parameters and in the association
of IAs with clients. DHCPv6 clients use DUIDs to identify a server in messages where a server needs to be identified. However, in this particular example, the new DHCPv6 functionality has recently been retrofitted to IPv4 via a specification for DUIDs for DHCPv4 . However, there are a number of potential problems with this approach: o IPv4-only nodes would not have any DHCP service available to them; such an approach is only possible in a fully dual-stack environment. o The client node may then be IPv6-only and receive IPv4 configuration settings that it does not want or be able to handle meaningfully. o The DHCPv4 servers need to be configured anyway to support IPv4- only hosts, so there is still duplication of information. o What happens if there are DHCPv6 servers that don't return IPv4 information? Does this mean the client can't run IPv4 (since it won't do DHCPv4)? o If IPv4 information is served from a DHCPv6 server as well as an IPv4 DHCP server, IPv4 address space will need to be allocated to both servers, fragmenting the potentially precious IPv4 global address resource for the site. 4.3. Optimising for Failure with Lists of Addresses There is a generic issue with any option that includes a list of addresses of servers (such as DNS server addresses). The list is offered to cater for resilience, such as whether the listed server itself fails or connectivity to the server fails. If the client does not know the cause of failure, its optimal strategy is to try a different server, via a different protocol. The problem today is that the IPv4 list is returned via DHCPv4, and the IPv6 list via DHCPv6; the client really has no way to "try a different server", since that information is lost by the protocol, even though it may be known by the server. Just putting merging heuristics in the client cannot provide the best behaviour, since information is lost. By comparison, if IPv4-mapped addresses were included in the DHCPv6 option along with IPv6 addresses, the DHCP server can give an intelligent order that takes into account which addresses are of the same DNS/whatever server. IPv6-only clients have to know to discard the IPv4-mapped addresses in this solution, and it's much easier to solve this in the combined- DHCP-server case than in the two-server case.
One can argue that this is only an optimisation, and in many cases the list has only two elements, so the "next" choice is forced. However, this particular issue highlights the subtleties of merging responses from separate servers. 4.4. Administrative and Other Areas There are also administrative issues or best practice that could be promoted. For example, it may be recommended that sites do not split their DNS name space for IPv6-specific testbeds. It may be worth considering whether separate manual configuration files should be kept for IPv4 and IPv6 settings, such as separate /etc/resolv.conf files for DNS settings on UNIX systems. However, this seems a complex solution. The problem should be better solved by other, more generalised methods. It may be important at times to be able to distinguish DHCP client and server identities. DHCPv6 introduces the idea of a DHCP Unique Identifier (DUID). The DUID concept has also been retrofitted to DHCPv4 , and thus it may form the basis of part of the solution space for the problem at hand. Some differences in DHCP and DHCPv6 may not be reconciled, but may not need to be, such as different ways to assign addresses by DUID in DHCPv6, or the lack of a comparable option in both DHCP versions. 5. Summary There are a number of issues in the operation of DHCP and DHCPv6 servers for nodes in dual-stack environments that should be clarified. While some differences in the protocols may not be reconciled, there may not be a need to do so. However, with DHCPv6 deployment growing, there is an operational requirement to determine best practice for DHCP server provision in dual-stack environments, which may or may not imply additional protocol requirements. The principal choice is whether separate DHCP and DHCPv6 services should be maintained by a site, or whether DHCPv6 should be extended to carry IPv4 configuration settings for dual-stack nodes. It can certainly be argued that until a site is completely dual- stack, an IPv4 DHCP service will always be required (for example, while there are still legacy printers, IP webcams, or other devices that still configure via DHCPv4), and a single IPv6 transport DHCP server offering configuration information for both protocols will then not be sufficient. In that case, IPv4 DHCP is required, and thus there
is a good rationale for focusing effort on how to combine the information received from separate IPv4 DHCP and (stateless) DHCPv6 servers. In theory, it should be relatively straightforward to write a configuration manager that would accept a single configuration specification from the service manager and distribute the correct (and consistent) configurations to the DHCPv4 and DHCPv6 servers (whether on the same host or not). In this case, maintaining coordinated configurations in two servers is an interface issue, not a protocol issue. The question then is whether the client has all the information it needs to make reasonable choices. We are aware of one implementation of separate DHCPv4 and DHCPv6 clients that is using a preference option for assisting client-side merging of the received information. Another issue for discussion is whether a combined DHCP service only available over IPv6 transport is a desirable longer-term goal for networks containing only dual-stack or IPv6-only nodes (or IPv4-only nodes where DHCPv4 is not needed). The transition to the long-term position may easily take more than 10 years. Upon reflection on the above observations, the dhc WG reached a strong consensus to adopt the two-server approach (separate DHCP and DHCPv6 servers), rather than have a combined single server returning IPv4 information over IPv6. The two servers may be co-located on a single node and may have consistent configuration information generated from a single asset database. It should be noted that deployment experience of DHCPv6 is still in its infancy; thus, a full understanding of the issues may only develop over time, but we feel we have reached the best consensus given the current status. Future work is now required to determine best practice for merging information from multiple servers, including merger of lists of addresses where options carry such information. As a footnote, we note that this work has overlap with multihoming and multi-interface configuration issues. It is also interwoven with the Detecting Network Attachment area, for example, where a node may move from an IPv4-only network to a dual-stack network, or vice versa. Both aspects may be best abstracted for discussion and progression in the respective IETF multi6, shim6, and dna WGs, in parallel with the two-server progression in the dhc WG.
6. Security Considerations There are no security considerations in this problem statement per se, as it does not propose a new protocol. 7. Acknowledgements The authors thank the following people for input to this document: Bernie Volz, AK Vijayabhaskar, Ted Lemon, Ralph Droms, Robert Elz, Changming Liu, Margaret Wasserman, Dave Thaler, Mark Hollinger, and Greg Daley. The document may not necessarily fully reflect the views of each of these individuals. The authors would also like to thank colleagues on the 6NET project for contributions to this document. 8. Informative References  Droms, R., "Dynamic Host Configuration Protocol", RFC 2131, March 1997.  Thomson, S. and T. Narten, "IPv6 Stateless Address Autoconfiguration", RFC 2462, December 1998.  Droms, R. and W. Arbaugh, "Authentication for DHCP Messages", RFC 3118, June 2001.  Droms, R., Bound, J., Volz, B., Lemon, T., Perkins, C., and M. Carney, "Dynamic Host Configuration Protocol for IPv6 (DHCPv6)", RFC 3315, July 2003.  Droms, R., "Stateless Dynamic Host Configuration Protocol (DHCP) Service for IPv6", RFC 3736, April 2004.  Lemon, T. and B. Sommerfeld, "Node-specific Client Identifiers for Dynamic Host Configuration Protocol Version Four (DHCPv4)", RFC 4361, February 2006.
Authors' Addresses Tim Chown University of Southampton School of Electronics and Computer Science Southampton, Hampshire SO17 1BJ United Kingdom EMail: firstname.lastname@example.org Stig Venaas UNINETT Trondheim NO 7465 Norway EMail: email@example.com Christian Strauf Clausthal University of Technology Erzstr. 51 Clausthal-Zellerfeld D-38678 Germany EMail: firstname.lastname@example.org
Full Copyright Statement Copyright (C) The Internet Society (2006). This document is subject to the rights, licenses and restrictions contained in BCP 78, and except as set forth therein, the authors retain all their rights. This document and the information contained herein are provided on an "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY AND THE INTERNET ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Intellectual Property The IETF takes no position regarding the validity or scope of any Intellectual Property Rights or other rights that might be claimed to pertain to the implementation or use of the technology described in this document or the extent to which any license under such rights might or might not be available; nor does it represent that it has made any independent effort to identify any such rights. Information on the procedures with respect to rights in RFC documents can be found in BCP 78 and BCP 79. Copies of IPR disclosures made to the IETF Secretariat and any assurances of licenses to be made available, or the result of an attempt made to obtain a general license or permission for the use of such proprietary rights by implementers or users of this specification can be obtained from the IETF on-line IPR repository at http://www.ietf.org/ipr. The IETF invites any interested party to bring to its attention any copyrights, patents or patent applications, or other proprietary rights that may cover technology that may be required to implement this standard. Please address the information to the IETF at email@example.com. Acknowledgement Funding for the RFC Editor function is provided by the IETF Administrative Support Activity (IASA).