Cisco Systems, Inc.

170 West Tasman Drive San Jose 95134 California USA repenno@cisco.com

Cisco Systems

Cessna Business Park Bangalore India 560103 tasaxena@cisco.com

France Telecom

Rennes 35000 France mohamed.boucadair@orange.com

Cisco Systems

7100-8 Kit Creek Road Research Triangle Park 27709 North Carolina USA ssenthil@cisco.com

Transport Area Behavior Engineering for Hindrance Avoidance NAT64 DNS64 NAT-PT ALG (Application Layer Gateway) NAT traversal IPv4-IPv6 interconnection IPv4-IPv6 translation Due to specific problems, Network Address Translation - Protocol Translation (NAT-PT) was deprecated by the IETF as a mechanism to perform IPv6-IPv4 translation. Since then, new efforts have been undertaken within IETF to standardize alternative mechanisms to perform IPv6-IPv4 translation. This document analyzes to what extent the new stateful translation mechanisms avoid the problems that caused the IETF to deprecate NAT-PT.

This document uses stateful 64 (or 64 for short) to refer to the mechanisms defined in the following documents: IP/ICMP Translation Algorithm Stateful NAT64: Network Address and Protocol Translation from IPv6 Clients to IPv4 Servers DNS64: DNS Extensions for Network Address Translation from IPv6 Clients to IPv4 Servers IPv6 Addressing of IPv4/IPv6 Translators Framework for IPv4/IPv6 Translation

Stateful 64 is widely seen as a major interconnection technique designed to enable communications between IPv6-only and IPv4-only networks. One of the building blocks of the stateful 64 is decoupling the DNS functionality from the protocol translation itself. This approach is pragmatic in the sense that there is no dependency on DNS implementation for the successful NAT handling. As long as there is a function (e.g., DNS64 or other means) that can construct an IPv6-embedded IPv4 address with a pre-configured IPv6 prefix, an IPv4 address and a suffix (refer to ), NAT64 will work just fine. The focus of the stateful 64 is on the deployment and not the implementation details. As long as a NAT64 implementation conforms to the expected behavior, as desired in the deployment scenario, the details are not very important as mentioned in this excerpt from : A NAT64 MAY perform the steps in a different order, or MAY perform different steps, but the externally visible outcome MUST be the same as the one described in this document.

This document provides an analysis of how the proposed set of documents that specify stateful IPv6-only to IPv4-only translation and replace Network Address Translation - Protocol Translation (NAT-PT) address the issues raised in . As a reminder, it is worth mentioning the analysis is limited in the sense that hosts from IPv6 networks can initiate a communication to IPv4 network/Internet, but not vice versa. This corresponds to Scenarios 1 and 5 described in . Hence, the scenario of servers moving to IPv6 while clients remaining IPv4 remains unaddressed. Of course, IPv6-to-IPv4 communications can also be supported if static or explicit bindings (e.g., ) are configured on the stateful NAT64. Stateful 64, just like any other technique under development, has some positives and some drawbacks. The ups and downs of the proposal must be clearly understood while going forward with its future development. The scope of this document does not include stateless translation.

Of the set of problems pointed out in , the stateful 64 addresses some of them, whereas it leaves others unaddressed. Some issues mentioned in were solved by , , and . At the time when NAT-PT was published, these recommendations were not in place but they are orthogonal to the translation algorithm per se; therefore, they could be implemented with NAT-PT. On the other hand, NAT64 explicitly mentions that these recommendations need to be followed and thus should be seen as a complete specification. It is also worth pointing out that the scope of the stateful 64 is reduced when compared to NAT-PT. Following is a point-by-point analysis of the problems. This document classifies the issues listed in into three categories: Problems impossible to solve. Problems that can be solved. Problems solved.

Problems discussed in that are impossible to solve: Inability to redirect traffic for protocols that lack de-multiplexing capabilities or are not built on top of specific transport-layer protocols for transport address translations (Section 2.2 of ). Analysis: This issue is not specific to 64 but to all NAT-based solutions. Loss of information due to incompatible semantics between IPv4 and IPv6 versions of headers and protocols (Section 2.4 of). Analysis: This issue is not specific to 64 but is due to the design of IPv4 and IPv6. Need for the NAT64-capable device to act as proxy for correspondent node when IPv6 node is mobile, with consequent restrictions on mobility (Section 2.7 of). Analysis: This is not specific to NAT64 but to all NAT flavors. Refer to for an early analysis on mobility complications encountered when NAT64 is involved.

Problems discussed in that can be solved: Disruption of all protocols that embed IP addresses (and/or ports) in packet payloads or apply integrity mechanisms using IP addresses (and ports) (Section 2.1 of). Analysis: In the case of FTP, this problem can be mitigated in several ways (e.g., use a FTP64 Application Layer Gateway (ALG) or in the FTP client (e.g., )). In the case of SIP , no specific issue is induced by 64; the same techniques for NAT traversal can be used when a NAT64 is involved in the path (e.g., Interactive Connectivity Establishment (ICE) , maintain SIP-related NAT bindings as per Section 3.4 of , media latching , embedded SIP ALGs, etc.). provides more discussion on how to establish SIP sessions between IPv4 and IPv6 SIP user agents. The functioning of other protocols is left for future study. Note that the traversal of NAT64 by application embedding IP address literal is not specific to NAT64 but generic to all NAT-based solutions. Interaction with Stream Control Transmission Protocol (SCTP) and multihoming (Section 2.6 of). Analysis: Only TCP and UDP transport protocols are within the scope of NAT64 . SCTP is out of scope of this document. Inability to handle multicast traffic (Section 2.8 of). Analysis: This problem is not addressed by the current 64 specifications. Scalability concerns together with introduction of a single point of failure and a security attack nexus (Section 3.2 of). Analysis: This is not specific to NAT64 but to all stateful NAT flavors. The presence of a single point of failure is deployment-specific; some service providers may deploy state synchronization means while others may only rely on a distributed NAT64 model. Restricted validity of translated DNS records: a translated record may be forwarded to an application that cannot use it (Section 4.2 of). Analysis: If a node on the IPv4 side forwards the address of the other endpoint to a node that cannot reach the NAT box or is not covered under the endpoint-independent constraint of NAT, then the new node will not be able to initiate a successful session. Actually, this is not a limitation of 64 (or even NAT-PT) but a deployment context where IPv4 addresses managed by the NAT64 are not globally reachable. The same limitation can be encountered when referrals (even without any NAT in the path) include reachability information with limited reachability scope (see for more discussion about issues related to reachability scope). IPsec traffic using AH (Authentication Header) in both transport and tunnel modes cannot be carried through NAT-PT without terminating the security associations on the NAT-PT, due to the inclusion of IP header fields in the scope of AH's cryptographic integrity protection (Section 2.1 of). In addition, IPsec traffic using ESP (Encapsulating Security Payload) in transport mode generally uses UDP encapsulation for NAT traversal (including NAT-PT traversal) in order to avoid the problems described in (Section 2.1 of). Analysis: This is not specific to NAT64 but to all NAT flavors. Address selection issues when either the internal or external hosts implement both IPv4 and IPv6 (Section 4.1 of). Analysis: This is out of scope of 64 since Scenarios 1 and 5 of assume IPv6-only hosts. Therefore, this issue is not resolved and mitigation techniques outside the 64 need to be used (e.g., ). These techniques may allow one to offload NAT64 resources and prefer native communications that do not involve address family translation. Avoiding NAT devices in the path is encouraged for mobile nodes in order to save power consumption due to keepalive messages that are required to maintain NAT states ("always-on" services). An in-depth discussion can be found in .

Problems identified in that have been solved: Constraints on network topology (as it relates to DNS-ALG; see Section 3.1 of ). Analysis: The severity of this issue has been mitigated by the separation of the DNS from the NAT functionality. Nevertheless, a minimal coordination may be required to ensure that the NAT64 to be crossed (the one to which the IPv4-Converted IPv6 address returned to a requesting host) must be in the path and has also sufficient resources to handle received traffic. Need for additional state and/or packet reconstruction in dealing with packet fragmentation. Otherwise, implement no support for fragments (Section 2.5 of). Analysis: This issue is not specific to 64 but to all NAT-based solutions. specifies how to handle fragmentation; appropriate recommendations to avoid fragmentation-related DoS (Denial-of-Service) attacks are proposed (e.g., limit resources to be dedicated to out-of-order fragments). Inappropriate translation of responses to A queries from IPv6 nodes (Section 4.3 of). Analysis: DNS64 does not alter A queries. Address selection issues and resource consumption in a DNS-ALG with multi-addressed nodes (Section 4.4 of). Analysis: Since no DNS-ALG is required to be co-located with NAT64, there is no need to maintain temporary states in anticipation of connections. Note that explicit bindings (see Section 3 of ) are required to allow for communications initiated from an IPv4-only client to an IPv6-only server. Limitations on DNS security capabilities when using a DNS-ALG (Section 2.5 of). Analysis: A DNSSEC validating stub resolver behind a DNS64 in server mode is not supported. Therefore, if a host wants to do its own DNSSEC validation, and it wants to use a NAT64, the host has to also perform its own DNS64 synthesis. Refer to Section 3 of for more details. Creation of a DoS threat relating to exhaustion of memory and address/port pool resources on the translator (Section 3.4 of ). Analysis: This specific DoS concern on Page 6 of is under a DNS-ALG heading in that document, and refers to NAT-PT's creation of NAT mapping state when a DNS query occurred. With the new IPv6-IPv4 translation mechanisms, DNS queries do not create any mapping state in the NAT64. To mitigate the exhaustion of port pool issue (Section 3.4 of ), 64 must enforce a port limit similar to the one defined in . Thus, this concern can be fully eliminated in 64. Requirement for applications to use keepalive mechanisms to work around connectivity issues caused by premature timeout for session table and Binding Information Base entries (Section 2.3 of ). Analysis: Since NAT64 follows some of the , , and requirements, there is a high lower bound for the lifetime of sessions. In NAT-PT, this was unknown and applications needed to assume the worst case. For instance, in NAT64, the lifetime for a TCP session is approximately two hours, so not much keepalive signaling overhead is needed. Application clients (e.g., VPN clients) are not aware of the timer configured in the NAT device. For unmanaged services, a conservative approach would be adopted by applications that issue frequent keepalive messages to be sure that an active mapping is still maintained by any involved NAT64 device even if the NAT64 complies with , , and . Note that keepalive messages may be issued by applications to ensure that an active entry is maintained by a firewall, with or without a NAT in the path, which is located in the boundaries of a local domain. Lack of address mapping persistence: Some applications require address retention between sessions. The user traffic will be disrupted if a different mapping is used. The use of the DNS-ALG to create address mappings with limited lifetimes means that applications must start using the address shortly after the mapping is created, as well as keep it alive once they start using it (Section 3.3 of). Analysis: In the following, address persistence is used to refer to the support of "IP address pooling" behavior of "Paired" . In the context of 64, the external IPv4 address (representing the IPv6 host in the IPv4 network) is assigned by the NAT64 machinery and not the DNS64 function. Therefore, address persistence can be easily ensured by the NAT64 function (which complies with NAT recommendations and ). Address persistence should be guaranteed for both dynamic and static bindings. In the IPv6 side of the NAT64, the same IPv6 address is used to represent an IPv4 host; no issue about address persistence is raised in an IPv6 network.

The above analysis of the solutions provided by the stateful 64 shows that the majority of the problems that are not directly related to the decoupling of NAT and DNS remain unaddressed. Some of these problems are not specific to 64 but are generic to all NAT-based solutions. This points to several shortcomings of stateful 64 that must be addressed if the future network deployments have to move reliably towards 64 as a solution to IPv6-IPv4 interconnection. Some of the issues, as pointed out in , have possible solutions. However these solutions will require significant updates to the stateful 64, increasing its complexity. The following table summarizes the conclusions based on the analysis of stateful 64. Issue NAT-PT Specific Exists in NAT64 DNS ALG Specific Generic NAT Can be solved? Protocols embedding addresses No Yes No Yes Yes Protocols without demux capability No Yes No Yes No Binding state decay No Yes No Yes Yes Loss of information No Yes No No No Fragmentation No No No Yes Yes SCTP and Multihoming interaction No Yes No Yes Yes Proxy correspondent node for MIPv6 No Yes No No No Multicast No Yes No Yes Yes IPSec tunnel mode No Yes No Yes Yes Topology constraints with DNS-ALG Yes No Yes No Yes Scale and Single point of failure No Yes No Yes Yes Lack of address persistence No Yes No Yes Yes DoS attacks No Yes No Yes Yes Address selection issues with Dual stack hosts Yes No Yes No Yes Non-global validity of Translated RR records Yes No Yes No Yes Incorrect translation of A responses Yes No Yes No Yes DNS-ALG and Multi- addressed nodes No Yes No Yes Yes DNSSEC limitations No Yes No Yes Yes

This document does not specify any new protocol or architecture. It only analyzes how BEHAVE WG 64 documents mitigate concerns raised in and which ones are still unaddressed.

Many thanks to M. Bagnulo, D. Wing, X. Li, D. Anipko, and S. Moonesamy for their review and comments. D. Black provided the IPsec text.

The Port Control Protocol allows an IPv6 or IPv4 host to control how incoming IPv6 or IPv4 packets are translated and forwarded by a network address translator (NAT) or simple firewall, and also allows a host to optimize its outgoing NAT keepalive messages. Some networks are expected to support IPv4-only, dual-stack, and IPv6-only hosts at the same time. Such networks also want to IPv6/ IPv4 translation for the IPv6-only host so it can access servers on the IPv4 Internet. On such a network, the synthesized AAAA responses from a DNS64 can cause traffic to be translated. This document describes a solution to avoid that translation when the application uses DNS. The File transfer protocol(FTP) has a long histroy,, but still being widely used. The first concept of FTP was described RFC 114, and then was specified in RFC 354. FTP can work in IPv4 environment and then was extended to IPv6. RFC 2428 defines IPv6 extensions of FTP. In the IPv6-IPv4 translation scenario, considerations should be applied to FTP client, server and translation box to ensure FTP protocol work properly. This document discusses the details for FTP to work in IPv4-IPv6 transitiion scenario. This document gives recommendation regarding how IPv6 FTP client should behave in 6to4 scenario. RFC 6724 defines default address selection mechanisms for IPv6 that allow nodes to select an appropriate address when faced with multiple source and/or destination addresses to choose between. The RFC 6724 allowed for the future definition of methods to administratively configure the address selection policy information. This document defines a new DHCPv6 option for such configuration, allowing a site administrator to distribute address selection policy overriding the default address selection parameters and policy table, and thus control the address selection behavior of nodes in their site. This document defines common requirements for Carrier-Grade NAT (CGN). It updates RFC 4787. This memo discusses potential NAT64 technology repercussions for mobile nodes using Mobile IPv6. An ambiguity is identified related to the use of DNS during bootstrapping, which is likely to inhibit proper signaling between mobile node and home agent. Middleboxes are defined as any intermediary box performing functions apart from normal, standard functions of an IP router on the data path between a source host and destination host. Two such functions are network address translation and firewalling. When Application Layer Gateways, such as SIP entities, interact with NATs and firewalls, as described in the MIDCOM architecture, then problems may occur in the transport of media traffic when signaling protocol interaction takes place along the media path, as it is the case for recent key exchange proposals (such as DTLS-SRTP). This document highlights problems that may arise. Unfortunately, it is difficult for the end points to detect or predict problematic behavior and to determine whether the media path is reliably available for packet exchange. This document aims to summarize the various sources and effects of NAT and firewall control, the reasons that they exist, and possible means of improving their behavior to allow protocols that rely upon signaling along the media path to operate effectively. The purpose of a referral is to enable a given entity in a multiparty application to pass information to another party. This memo specifies a Generic Referral Object (GRO) to be used in the context of referrals. The proposed object is compact and is application- independent. Both IPv4 and IPv6 schemes are supported, as well as upper layer identifiers. Additional information to characterise an enclosed reference is also described. To allow proper interpretation of referrals, a new notion of scope identifiers is introduced.