Locating Session Initiation Protocol (SIP) Servers in a Dual-Stack IP Network

The Session Initiation Protocol (SIP, ) and the additional documents that extended it provide support for both IPv4 and IPv6. However, this support does not fully extend to the highly hybridized environments that are characteristic of the transitional migratory phase from IPv4 to IPv6 networks. During this phase, many server and client implementations run on dual-stack hosts. In such environments, a dual-stack host will likely suffer greater connection delay, and by extension an inferior user experience, than an IPv4-only host. The need to remedy this diminished performance of dual-stack hosts led to the development of the Happy Eyeballs algorithm, which has since been implemented in many protocols and applications. This document updates the DNS lookup procedures of RFC 3263 in preparation for the specification of the application of Happy Eyeballs to SIP. Happy Eyeballs will provide enhanced performance, and consequently user experience, in highly hybridized dual-stack SIP networks. The procedures described herein are such that a dual-stack client should look up both A and AAAA records in DNS and then select the best way to set up a network flow. The details of how the latter is done is considered out of scope for this document. See the Happy Eyeballs algorithm and implementation and design considerations in RFC 6555 for more information about issues with setting up dual-stack network flows. of this document clarifies the interaction of with and .

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 RFC 2119 . RFC 3261 defines additional terms used in this document that are specific to the SIP domain such as "proxy", "registrar", "redirect server", "user agent server" or "UAS", "user agent client" or "UAC", "back-to-back user agent" or "B2BUA", "dialog", "transaction", and "server transaction". This document uses the term "SIP server" that is defined to include the following SIP entities: user agent server, registrar, redirect server, a SIP proxy in the role of user agent server, and a B2BUA in the role of a user agent server. While this document focuses on the dual-stack situation described in RFC 6555 and other documents, concerning the migration from an IPv4-only network to a network supporting both IPv4 and IPv6, the techniques described can be used in other situations. Possible situations include when a device has multiple interfaces with distinct addressing characteristics and when additional IP address families are created in the future. This document uses the general term "dual-stack" to include all situations where the client has access to multiple communication methods that have distinct addressing characteristics. The term "address records" means the DNS records which translate a domain name into addresses within the address family(ies) that the entity supports (as A records provide IPv4 addresses and AAAA records provide IPv6 addresses), regardless of whether the address family was defined before or after this document was approved.

This specification introduces two normative DNS lookup procedures. These are designed to improve the performance of dual-stack clients in IPv4/IPv6 networks.

Once the transport protocol has been determined, the procedure for discovering an IP address if the TARGET is not a numeric IP address but the port is explicitly stated in the URI, is detailed in Section 4.2 of RFC 3263 . The piece relevant to this discussion is: If the TARGET was not a numeric IP address, but a port is present in the URI, the client performs an A or AAAA record lookup of the domain name. The result will be a list of IP addresses, each of which can be contacted at the specific port from the URI and transport protocol determined previously. Section 4.2 of RFC 3263 also goes on to describe the procedure for discovering an IP address if the TARGET is not a numeric IP address, and no port is present in the URI. The piece relevant to to this discussion is: If no SRV records were found, the client performs an A or AAAA record lookup of the domain name. The result will be a list of IP addresses, each of which can be contacted using the transport protocol determined previously, at the default port for that transport. Processing then proceeds as described above for an explicit port once the A or AAAA records have been looked up. Happy Eyeballs documents that looking up the "A or AAAA record" is not an effective practice for dual-stack clients and that it can add significant connection delay and greatly degrade user experience. Therefore, this document makes the following normative addendum to the DNS lookup procedures of Section 4.2 of RFC 3263 for IPv4/IPv6 hybrid SIP networks and recommends it as a best practice for such dual-stack networks: The dual-stack client SHOULD look up address records for all address families that it supports for the domain name and add the resulting addresses to the list of IP addresses to be contacted. A client MUST be prepared for the existence of DNS resource records containing addresses in families that it does not support; if such records may be returned by the client's DNS queries, such records MUST be ignored as unusable and the supported addresses used as specified herein.

The Happy Eyeballs algorithm is particularly effective for dual-stack HTTP client applications that have significant performance differences between their IPv4 and IPv6 network paths. This is because the client can initiate two TCP connections to the server, one using IPv4 and one using IPv6, and then use the connection which completes first. This works properly because the client can test each route by initiating a TCP connection, but simply opening a TCP connection to an HTTP server does not change the server's state; the client will send the HTTP request on only one connection. Unfortunately, in common SIP situations, it is not possible to "race" simultaneous request attempts using two address families. If the SIP requests are transmitted as single UDP packets, sending two copies of the request to two different addresses risks having two copies of the request propagating through the SIP network at the same time. The difference between SIP and HTTP is that in SIP the sender cannot test a route in a non-state-changing way. (If two copies of the same request arrive at the destination client, the client MUST reject the second of them with a 482 "Merged Request" response. But this rule is not sufficient to prevent user-visible differences in behavior. A proxy that is upstream of the second request to arrive at the client may (almost immediately!) serially fork the second request to further destinations (e.g., the voicemail service for the destination client).) In this common scenario it is often necessary for a dual-stack client to indicate a preference for either IPv4 or IPv6. A service may use DNS SRV records to indicate such a preference for an address family. This way, a server with a high-latency and/or low-capacity IPv4 tunnel may indicate a preference for being contacted using IPv6. A server that wishes to do this can use the lowest SRV priority to publish hostnames that only resolve in IPv6 and the next priority with host names that resolve in both address families. Note that hostnames that have addresses in only one address family are discouraged by . Such special-purpose hostnames SHOULD be used only as described in this section, as targets of SRV records for an aggregate host name, where the aggregate host name ultimately resolves to addresses in all families supported by the client.

Section 5 of specifies that the addresses from the address records for a single target DNS name for a server's DNS name must be contacted in the order specified by the source and destination address selection algorithms defined in . The set of addresses provided to a single invocation of the destination address selection algorithm MUST be the address records for the target DNS name in a single SRV record (or, if there are no SRV records, the DNS name in the URI or derived via NAPTR) -- the destination address selection algorithm MUST NOT reorder addresses derived from different SRV records. Typically, desination address selection is done by using the (relatively new) getaddrinfo() function to translate the target DNS name into a list of IPv4 and/or IPv6 addresses in the order in which they are to be contacted, as that function implements . Thus, if SRV lookup on the server's DNS name is successful, the major ordering of the complete list of destination addresses is determined by the priority and weight fields of the SRV records (as specified in ) and the (minor) ordering among the destinations derived from the "target" field of a single SRV record is determined by . For example, consider a server with DNS name example.com, with TCP transport specified. The relevant SRV records for example.com are: _sip._tcp.example.com. 300 IN SRV 10 1 5060 sip-1.example.com. _sip._tcp.example.com. 300 IN SRV 20 1 5060 sip-2.example.com. The processing of results in this ordered list of target domain names: sip-1.example.com sip-2.example.com The address records for sip-1.example.com, as ordered by , are sip-1.example.com. 300 IN AAAA 2001:0db8:58:c02::face sip-1.example.com. 300 IN AAAA 2001:0db8:c:a06::2:cafe sip-1.example.com. 300 IN AAAA 2001:0db8:44:204::d1ce sip-1.example.com. 300 IN A 192.0.2.45 sip-1.example.com. 300 IN A 203.0.113.109 sip-1.example.com. 300 IN A 198.51.100.24 and the address records for sip-2.example.com, as ordered by , are: sip-2.example.com. 300 IN AAAA 2001:0db8:58:c02::dead sip-2.example.com. 300 IN AAAA 2001:0db8:c:a06::2:beef sip-2.example.com. 300 IN AAAA 2001:0db8:44:204::c0de sip-2.example.com. 300 IN A 192.0.2.75 sip-2.example.com. 300 IN A 203.0.113.38 sip-2.example.com. 300 IN A 198.51.100.140 Thus, the complete list of destination addresses has this ordering: 2001:0db8:58:c02::face 2001:0db8:c:a06::2:cafe 2001:0db8:44:204::d1ce 192.0.2.45 203.0.113.109 198.51.100.24 2001:0db8:58:c02::dead 2001:0db8:c:a06::2:beef 2001:0db8:44:204::c0de 192.0.2.75 203.0.113.38 198.51.100.140 In particular, the destination addresses derived from sip-1.example.com and those derived from sip-2.example.com are not interleaved; does not operate on the complete list. This would be true even if the two SRV records had the same priority and were (randomly) ordered based on their weights, as the address records of two target DNS names are never interleaved.

This document introduces two new normative procedures to the existing DNS procedures used to locate SIP servers. A client may contact additional target addresses for a URI beyond those prescribed in , and/or it may contact target addresses in a different order than prescribed in . Neither of these changes introduce any new security considerations because it has always been assumed that a client desiring to send to a URI may contact any of its targets that are listed in DNS. The specific security vulnerabilities, attacks and threat models of the various protocols discussed in this document (SIP, DNS, SRV records, Happy Eyeballs requirements and algorithm, etc.) are well documented in their respective specifications.

This document does not require any actions by IANA.

The authors would like to acknowledge the support and contribution of the SIP Forum IPv6 Working Group. This document is based on a lot of tests and discussions at SIPit events, organized by the SIP Forum. This document has benefited from the expertise and review feedback of many participants of the IETF DISPATCH and SIPCORE WG mailing lists as well as those on the SIP Forum IPv6 Task Group mailing list. The authors wish to specifically call out the efforts and express their gratitude for the detailed and thoughtful comments and corrections of Dan Wing, Brett Tate, Rifaat Shekh-Yusef, Carl Klatsky, Mary Barnes, Keith Drage, Cullen Jennings, Simon Perreault, Paul Kyzivat, Adam Roach, Richard Barnes, Ben Campbell, Stefan Winter, Spencer Dawkins, and Suresh Krishnan. Adam Roach devised the example in .

[Note to RFC Editor: Please remove this entire section upon publication as an RFC.]

Remove the reference to RFC 3484, since that RFC has been superseded, and the reference was only the statement that 3484 had been superseded by RFC 6724. Added explanation why "racing" simultaneous copies of a SIP requests causes incorrect behavior. Acknowledged Spencer Dawkins for this. In , made explcit the ordered list of target domain names that result from processing the SRV records. Acknowledged Suresh Krishnan for suggesting this. Updated the Terminology section to remove the definitions of "IPv4-only", etc. (which weren't being used) and add a definition of "dual-stack" that includes all multiple-stack situations.

Update per Ben Campbell's AD evaluation. Update Vijay Gurbani's affiliation. Update per Stefan Winter's OPS-DIR review.

Acknowledged Adam Roach for providing the example in . Correct references to vs. references to .

Simplified the acknowledgments. Improve wording and punctuation. Rewrote based on critiques on the Sipcore list. Included an example by Adam Roach. Replaced "RR's" with "records" per suggestion by Jean Mahoney.

Changed the "updates" specification to add RFC 3263 and remove RFC 6157. Added Simon Perreault to the acknowledgments. Minor wording changes.

Described the relationship to RFC 3263 as "update", since the existing wording in 3263 is not what we want. Arguably, the new wording is what was intended in 3263, but the existing wording either does not say that or says it in a way that is easily misunderstood. Described the relationship to RFC 6157 as "clarification", since the described interaction between 3263 and 6157 appears to be the only reasonable interpretation. Revised wording, punctuation, and capitalization in various places. Clarified that this draft does not document Happy Eyeballs for SIP, but is preparatory for it. Attempted to use "update" for text that is definitively a change to the preexisting text and "clarify" for text that is a more clear statement of the (presumed) intention of the preexisting text. Removed normative words from section 1, the introduction. Copied definition of "address records" from RFC 2782 (SRV records) to allow the specifications to expand automatically to include any new address families. Relocated the text requiring a client to ignore addresses that it discovers in address families it does not support from section 4.2 (which describes why the situation arises) to section 4.1 (which describes how clients look up RRs). Clarified the interaction with RFC 6157 (source and destination address selection in IPv6) to specify what must have been intended: The major sort of the destinations is the ordering determined by priority/weight in the SRV records; the addresses derived from a single SRV record's target are minorly sorted based on RFC 6157. Removed editor's name from the acknowledgments list.