Internet Draft P. Srisuresh Document: draft-ford-behave-top-03.txt Consultant Expires: April 23, 2007 B. Ford M.I.T. October 23, 2006 Complications from Network Address Translator Deployment Topologies Status of this Memo By submitting this Internet-Draft, each author represents that any applicable patent or other IPR claims of which he or she is aware have been or will be disclosed, and any of which he or she becomes aware will be disclosed, in accordance with Section 6 of BCP 79. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." The list of current Internet-Drafts can be accessed at http://www.ietf.org/1id-abstracts.html The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html Abstract This document identifies two deployment scenarios that have arisen from the unconventional network topologies formed using Network Address Translator devices (NATs). First, the simplicity of administering networks through the combination of NAT and DHCP has increasingly lead to the deployment of multi-level inter-connected private networks involving overlapping IP address spaces. Second, the proliferation of private networks in enterprises, hotels and conferences, and the wide spread use of Virtual Private Networks (VPNs) to access enterprise intranet from remote locations has increasingly lead to overlapping IP address space between remote and corporate networks. The document does not dismiss these unconventional scenarios as invalid, but recognizes them as real and Srisuresh & Ford [Page 1] Internet-Draft Complications from NAT Deployments October 2006 offers recommendations to ensure these real scenarios can funtion. Table of Contents 1. Introduction and Scope........................................ 2. Multi-Level NAT Network Topologies ........................... 2.1 Operational Details of the Multi-Level NAT Network ....... 2.1.1. Client/Server Communication ....................... 2.1.2. Peer-to-Peer Communication ........................ 2.2. Anomalies of the Multi-Level NAT Network ................ 2.2.1. Plug-and-Play NAT Devices ......................... 2.2.2. Unconventional Addressing on NAT Devices .......... 2.2.3. Multi-Level NAT Translations ...................... 2.2.4. Mistaken End Host Identity ........................ 2.3. Summary of Recommendations .............................. 3. Remote Access VPN Network Topologies ......................... 3.1. Operational Details of the Remote Access VPN Network .... 3.2. Anomalies of the Remote Access VPN Network .............. 3.2.1. Remote Router and DHCP Server Address Conflict ... 3.2.2. Simultaneous Connectivity Conflict ............... 3.2.3. VIP Address Conflict ............................. 3.2.4. Mistaken End Host Identity ....................... 3.3. Summary of Recommendations .............................. 4. Security Considerations ...................................... 5. Acknowledgements ............................................. 6. Normative References ......................................... 7. Informational References ..................................... 1. Introduction and Scope The Internet was originally designed to use a single, global 32-bit IP address space to uniquely identify hosts on the network, allowing applications on one host to address and initiate communications with applications on any other host regardless of the respective hosts' topological locations or administrative domains. For a variety of pragmatic reasons, however, the Internet has gradually drifted away from strict conformance to this ideal of a single flat global address space, and towards a hierarchy of smaller "private" address spaces [RFC1918] clustered around a large central "public" address space. The most important pragmatic causes of this unintended evolution of the Internet's architecture appear to be the following. 1. Depletion of the 32-bit IPv4 address space due to the exploding total number of hosts on the Internet. Although IPv6 promises to solve this problem, the uptake of IPv6 has in practice been slower than expected. 2. Perceived Security and Privacy: Traditional NAT devices provide a Srisuresh & Ford [Page 2] Internet-Draft Complications from NAT Deployments October 2006 filtering function that permits session flows to cross the NAT in just one direction, from private hosts to public network hosts. This filtering function is widely perceived as a security benefit. In addition, the NAT's translation of a host's original IP addresses and port number in private network into an unrelated, external IP address and port number is perceived by some as a privacy benefit. 3. Ease-of-use: NAT vendors often combine the NAT function with a DHCP server function in the same device, which creates a compelling, effectively "plug-and-play" method of setting up small Internet-attached personal networks that is often much easier in practice for unsophisticated consumers than configuring an IP subnet. The many popular and inexpensive consumer NAT devices on the market are usually configured "out of the box" to obtain a single "public" IP address from an ISP or "upstream" network via DHCP, and the NAT device in turn acts as both a DHCP server and default router for any "downstream" hosts (and even other NATs) that the user plugs into it. Consumer NATs in this way effectively create and manage private home networks automatically without requiring any knowledge of network protocols or management on the part of the user. Auto-configuration of private hosts makes NAT devices a compelling solution in this common scenario. The term NAT used throughout the document refers to the traditional NAT, as defined in [NAT-TERM] and specified in [NAT-TRAD]. [NAT-PROT] identifies various complications with application protocols due to NAT devices. This document acts as an adjunct to [NAT-PROT]. The scope of the document is restricted to the two scenarios identified in sections 2 and 3, as arising out of unconventional NAT deployment and private address space overlap. Even though the scenarios appear unconventional, they are real and and not uncommon to find. For each scenario, the document describes the seeming anomalies and offers recommendations on how best to make the topologies work. Section 2 describes the problem of private address space overlap due to multi-level NAT topology, the anomalies with the topology and recommendations to address the anomalies. Section 3 describes the problem of private address space overlap with remote access Virtual Private Network (VPN) connection, the anomalies with address overlap and recommendations to address the anomalies. Section 4 describes the security considerations in these scenarios. 2 Multi-Level NAT Network Topologies Srisuresh & Ford [Page 3] Internet-Draft Complications from NAT Deployments October 2006 Due to the pragmatic considerations discussed in the previous section and perhaps others, NATs are increasingly, and often unintentionally, used to create hierarchically interconnected clusters of private networks as illustrated in figure 1 below. The creation of multi-level hierarchies is often unintentional, since each level of NAT is typically deployed by a separate administrative entity such as an ISP, a corporation, or a home user. Public Internet (Public IP addresses) ----+---------------+---------------+---------------+---- | | | | | | | | 66.39.3.7 18.181.0.31 138.76.29.7 155.99.25.1 +-------+ Host A Host B +-------------+ | NAT-1 | (Alice) (Jim) | NAT-2 | | (Bob) | | (CheapoISP) | +-------+ +-------------+ 10.1.1.1 10.1.1.1 | | | | Private Network 1 Private Network 2 (private IP addresses) (private IP addresses) ----+--------+---- ----+-----------------------+---- | | | | | | | | | | 10.1.1.10 10.1.1.11 10.1.1.10 10.1.1.11 10.1.1.12 Host C Host D +-------+ Host E +-------+ | NAT-3 | (Mary) | NAT-4 | | (Ann) | | (Lex) | +-------+ +-------+ 10.1.1.1 10.1.1.1 | | | | Private Network 3 | Private Network 4 (private IP addresses)| (private IP addresses) ----+-----------+---+ ----+-----------+---- | | | | | | | | 10.1.1.10 10.1.1.11 10.1.1.10 10.1.1.11 Host F Host G Host H Host I Figure 1. Multi-level NAT topology with Overlapping Address Space In the above scenario, Bob, Alice, Jim, and CheapoISP have each obtained a "genuine", globally routable IP address from an upstream Srisuresh & Ford [Page 4] Internet-Draft Complications from NAT Deployments October 2006 service provider. Alice and Jim have chosen to attach only a single machine at each of these public IP addresses, preserving the originally intended architecture of the Internet and making their hosts, A and B, globally addressable throughout the Internet. Bob, in contrast, has purchased and attached a typical consumer NAT box. Bob's NAT obtains its external IP address (66.39.3.7) from Bob's ISP via DHCP, and automatically creates a private 10.1.1.x network for Bob's hosts C and D, acting as the DHCP server and default router for this private network. Bob probably does not even know anything about IP addresses; he merely knows that plugging the NAT into the Internet as instructed by the ISP, and then plugging his hosts into the NAT as the NAT's manual indicates, seems to work and gives all of his hosts access to Internet. CheapoISP, an inexpensive service provider, has allocated only one or a few globally routable IP addresses, and uses NAT to share these public IP addresses among its many customers. Such an arrangement is becoming increasingly common, especially in rapidly-developing countries where the exploding number of Internet-attached hosts greatly outstrips the ability of ISPs to obtain globally unique IP addresses for them. CheapoISP has chosen the popular 10.1.1.x address for its private network, since this is one of the three well-known private IP address blocks allocated in [RFC1918] specifically for this purpose. Of the three incentives listed in section 1 for NAT deployment, the last two still apply even to customers of ISPs that use NAT, resulting in multi-level NAT topologies as illustrated in the right side of the above diagram. Even three-level NAT topologies are known to exist. CheapoISP's customers Ann, Mary, and Lex have each obtained a single IP address on CheapoISP's network (Private Network 2), via DHCP. Mary attaches only a single host at this point, but Ann and Lex each independently purchase and deploy consumer NATs in the same way that Bob did above. As it turns out, these consumer NATs also happen to use 10.1.1.x addresses for the private networks they create, since these are the configuration defaults hard-coded into the NATs by their vendors. Ann and Lex probably know nothing about IP addresses, and in particular they are probably unaware that the IP address spaces of their own private networks overlap not only with each other but also with the private IP address space used by their immediately upstream network. Nevertheless, despite this direct overlap, all of the "multi-level NATted hosts" - F, G, H, and I in this case - all nominally function and are able to initiate connections to any public server on the public Internet that has a globally routable IP address. Connections made from these hosts to the main Internet are merely translated twice. Once by the consumer NAT (NAT-3 or NAT-44) into the IP Srisuresh & Ford [Page 5] Internet-Draft Complications from NAT Deployments October 2006 address space of CheapoISP's Private Network 2, and then again by CheapoISP's NAT-2 into the public Internet's global IP address space. 2.1 Operational Details of the Multi-Level NAT Network In the "de facto" Internet address architecture that has resulted from the above pragmatic and economic incentives, only the nodes on the public Internet have globally unique IP addresses assigned by the official IP address registries. IP addresses on different private networks are typically managed independently - either manually by the administrator of the private network itself, or automatically by the NAT through which the private network is connected to its "upstream" service provider. By convention, nodes on private networks are usually assigned IP addresses in one of the private address space ranges specifically allocated to this purpose in RFC 1918, ensuring that private IP addresses are easily distinguishable and do not conflict with the public IP addresses officially assigned to globally routable Internet hosts. However, when "plug-and-play" NATs are used to create hierarchically interconnected clusters of private networks, a given private IP address can be and often is reused across many different private networks. In figure 1 above, for example, private networks 1, 2, 3, and 4 all have a node with IP address 10.1.1.10. 2.1.1. Client/Server Communication When a host on a private network initiates a client/server-style communication session with a server on the public Internet, via the server's public IP address, the NAT intercepts the packets comprising that session (usually as a consequence of being the default router for the private network), and modifies the packets' IP and TCP/UDP headers so as to make the session appear externally as if it was initiated by the NAT itself. For example, if host C above initiates a connection to host A at IP address 18.181.0.31, NAT-1 modifies the packets comprising the session so as to appear on the public Internet as if the session originated from NAT-1. Similarly, if host F on private network 3 initiates a connection to host A, NAT-3 modifies the outgoing packet so the packet appears on private network 2 as if it had originated from NAT-3 at IP address 10.1.1.10. When the modified packet traverses NAT-2 on private network 2, NAT-2 further modifies the outgoing packet so as to appear on the public Internet as if it had originated from NAT-2 at public IP address 155.99.25.1. The NATs in Srisuresh & Ford [Page 6] Internet-Draft Complications from NAT Deployments October 2006 effect serve as proxies that give their private "downstream" client nodes a temporary presence on "upstream" networks to support individual communication sessions. In summary, all hosts on the private networks 1, 2, 3, and 4 in figure 1 above are able to establish a client/server-style communication sessions with servers on the public Internet. 2.1.2. Peer-to-Peer Communication While this network organization functions in practice for client/server-style communication, when the client is behind one or more levels of NAT and the server is on the public Internet, the lack of globally routable addresses for hosts on private networks makes direct peer-to-peer communication between those hosts difficult. For example, two private hosts F and H on the network shown above might "meet" and learn of each other through a well-known server on the public Internet, such as Host A, and desire to establish direct communication between G and H without requiring A to forward each packet. If G and H merely learn each other's (private) IP addresses from a registry kept by A, their attempts to connect to each other will fail because G and H reside on different private networks. Worse, if their connection attempts are not properly authenticated, they may appear to succeed but end up talking to the wrong host. For example, G may end up talking to Host F, the host on private network 3 that happens to have the same private IP address as Host H. Host H might similarly end up unintentionally connecting to Host I. In summary, peer-to-peer communication between hosts on disjoint private networks 1, 2, 3, and 4 in figure 1 above is a challenge without the assistance of a well known server on the public Internet. However, with assistance from a node in the public Internet, all hosts on the private networks 1, 2, 3, and 4 in figure 1 above are able to establish a peer-to-peer style communication sessions amongst themselves as well as with hosts on the public Internet. 2.2. Anomalies of the Multi-Level NAT Network Even though conventional wisdom would suggest that the network described above is seriously broken, in practice it still works in many ways. We break up figure 1 into two sub figures to better illustrate the anomalies. Figure 1.1 is the left half of figure 1 and reflects the conventional single NAT deployment that is widely prevalent in many last-mile locations. The deployment in figure 1.1 is popularly viewed as a pragmatic solution to work around the Srisuresh & Ford [Page 7] Internet-Draft Complications from NAT Deployments October 2006 depletion of IPv4 address space and is not considered broken. Figure 1.2 is the right half of figure-1 and is representative of the anomalies we are about to discuss. Public Internet (Public IP addresses) ----+---------------+---------------+----------- | | | | | | 66.39.3.7 18.181.0.31 138.76.29.7 +-------+ Host A Host B | NAT-1 | (Alice) (Jim) | (Bob) | +-------+ 10.1.1.1 | | Private Network 1 (private IP addresses) ----+--------+---- | | | | 10.1.1.10 10.1.1.11 Host C Host D Figure 1.1. Conventional Single-level NAT Network topology Srisuresh & Ford [Page 8] Internet-Draft Complications from NAT Deployments October 2006 Public Internet (Public IP addresses) ---+---------------+---------------+---- | | | | | | 18.181.0.31 138.76.29.7 155.99.25.1 Host A Host B +-------------+ (Alice) (Jim) | NAT-2 | | (CheapoISP) | +-------------+ 10.1.1.1 | | Private Network 2 (private IP addresses) ----+---------------+-------------+--+------- | | | | | | 10.1.1.10 10.1.1.11 10.1.1.12 +-------+ Host E +-------+ | NAT-3 | (Mary) | NAT-4 | | (Ann) | | (Lex) | +-------+ +-------+ 10.1.1.1 10.1.1.1 | | | | Private Network 3 Private Network 4 (private IP addresses) (private IP addresses) ----+-----------+------ ----+-----------+---- | | | | | | | | 10.1.1.10 10.1.1.11 10.1.1.10 10.1.1.11 Host F Host G Host H Host I Figure 1.2. Unconventional Multi-level NAT Network topology 2.2.1. Plug-and-Play NAT Devices Consumer NAT devices are predominantly "plug-and-play" NAT devices, and assume minimal user intervention during device setup. The "plug-and-play" NAT devices provide DHCP service on one interface and NAT function on another interface. Vendors of the consumer NAT devices make assumptions about how their consumers configure and hook up their PCs to the device. When consumers do not adhere to the vendor assumptions, the consumers end up with a broken network. Srisuresh & Ford [Page 9] Internet-Draft Complications from NAT Deployments October 2006 A "plug-and-play" NAT device provides DHCP service on the LAN attached to the private interface, and assumes that all private hosts at the consumer site have DHCP client enabled and are connected to the single LAN. Consumers should be informed of the assumption that all private hosts must be on a single LAN, with no router in between. A "Plug-and-Play" NAT device also assumes that there is no other NAT device or DHCP server device on the same LAN at the customer premises. When there are multiple "Plug-and-play" NAT devices on the same LAN, each NAT device will offer DHCP service on the same LAN, and likely from the same address pool. This could result in multiple end nodes on the same LAN ending up with identical IP addresses. That will break network connectivity to end hosts. Consumers should be cautioned against using more than one "plug-and-play" NAT device on the same LAN. Recommendation-1. Consumers should be informed that all private hosts behind a "Plug-and-play" NAT must be on a single LAN subnet, and that there should be no more than one "Plug-and-play" NAT device on the same LAN. 2.2.2. Unconventional Addressing on NAT Devices Let us consider the unconventional addressing with NAT-3 and NAT-4. NAT-3 and NAT-4 are apparently multi-homed on the same subnet through both their interfaces. NAT-3 is on subnet 10.1.1/24 through its external interface facing NAT-2, as well as through its private interface facing clients Host-F and Host-G. Likewise, NAT-4 also has two interfaces on the same subnet 10.1.1/24. In a traditional network, when a node has multiple interfaces with IP addresses on the same subnet, it is natural to assume that all interfaces with addresses on the same subnet must be on a single connected LAN (bridged LAN or a single physical LAN). Clearly, that is not the case here. Even though both NAT-3 and NAT-4 have two interfaces on the same subnet 10.1.1/24, the NAT devices view the two interfaces as being on two disjoint subnets and routing realms. The "plug-and-play" NAT devices are really not multi-homed on the same subnet as in a traditional sense. In a traditional network, both NAT-3 and NAT-4 in figure 1.2 should be incapable of communicating reliably as a transport endpoint with other nodes on their adjacent networks (ex: private networks 2 and 3 in the case of NAT-3 and private Networks 2 and 4 in the case of NAT-4). This is because applications on either of the NAT devices cannot know to differentiate packets from hosts on either of the Srisuresh & Ford [Page 10] Internet-Draft Complications from NAT Deployments October 2006 subnets bearing the same IP address. If NAT-3 attempts to resolve the IP address of a neighboring host in the conventional manner by broadcasting an ARP request on all of its physical interfaces bearing the same subnet, it may get a different response on each of its physical interfaces. Even though both NAT-3 and NAT-4 have hosts bearing the same IP address on the adjacent networks, the NAT devices do communicate effectively as end points. Many of the "plug-and-play" NAT devices offer a limited number of services on them. For example, many of the NAT devices respond to pings from hosts on either of the interfaces. Even though a NAT device is often not actively managed, many of the NAT devices are equipped to be managed from the private interface. This unconventional communication with NAT devices is achievable because NAT devices view the two interfaces as being on two disjoint routing domains and distinguish between sessions with hosts on either interface (private or public). Consumer oriented "Plug-and-Play" NAT devices must and all NATs should be able to handle multi-level NAT topologies such as the one described in figure 1.2, in which a private IP address space on one side of the NAT potentially conflicts with the private IP address space on the other side. Essentially NAT must be able to keep the two IP address spaces separate in its internal data structures, and base all packet processing decisions on the "side" or "port" from which the packet arrived and not just on the basis of the IP addresses it contains. This recommendation reiterates REQ-7 of [BEH-UDP]. Recommendation-2. As specified in REQ-7 of [BEH-UDP], NAT devices should support IP address space overlap between the address space on its private interface and the address space on its public interface. Essentially, a NAT device should keep the two IP address spaces separate and base all packet processing decisions on the "side" or "port" from which the packet arrived and not just on the basis of the IP addresses. 2.2.3. Multi-Level NAT Translations Use of a single NAT to connect private hosts to the public Internet as in figure 1.1 is a fairly common practice. Many consumer NATs are deployed this way. However, use of multi-level NAT translations as in figure 1.2 is not a common practice and is not well understood. Let us consider the conventional single NAT translation in figure 1.1. Because the public and private IP address ranges are Srisuresh & Ford [Page 11] Internet-Draft Complications from NAT Deployments October 2006 numerically disjoint, nodes on private networks can make use of both public and private IP addresses when initiating network communication sessions. Nodes on a private network can use private IP addresses to refer to other nodes on the same private network, and public IP addresses to refer to nodes on the public Internet. For example, host C in figure 1.1 is on private network 1 and can directly address hosts A, B and D using their assigned IP addresses. This is in spite of the fact that hosts A and B are on the public Internet and host D alone is on the private network. Next, let us consider the unconventional multi-level NAT topology in figure 1.2. In this scenario, private hosts are able to connect to hosts on the public Internet. But, private hosts are not able to connect with all other private hosts. For example, host F in figure 1.2 can directly address hosts A, B, and G using their assigned IP addresses, but F has no way to address any of the other hosts in the diagram. Host F in particular cannot address host E by its assigned IP address, even though host E is located on the immediately "upstream" private network through which F is connected to the Internet. Host E has the same IP address as host G. Yet, this addressing is "legitimate" in the NAT world because the two hosts are on different private networks. It would seem that the topology in figure 1.2 with multiple NAT translations is broken because private hosts are not able to address each other directly. However, the network is not broken. Nodes on any private network have no direct method of addressing nodes on other private networks. The private networks 1, 2, 3 and 4 are all disjoint. Hosts on private network 1 are unable to directly address nodes on private networks 2, 3 or 4 and vice versa. Multiple NAT translations was not the cause of this. As described in sections 2.1.1 and 2.1.2, client-server and peer-to-peer communication can and should be possible even with multi-level NAT topology deployment. A host on any private network must be able to communicate with any other host, no matter which private network the host is attached to or where the private network is located. Host F should be able to communicate with host E and carry out both client-server communication and peer-to-peer communication, and vice versa. Host F and host E form a hairpin session through NAT-2 to communicate with each other. Each host uses the public endpoint assigned by the Internet facing NAT (NAT-2) to address its peer. NAT devices should support hairpin translation ([P2P-STATE]) for session flows that originate from a host on one attached private network and targeted to a host on another private network also attached to the same NAT device. Hairpin translation is explained in detail in section 4.4 of [P2P-STATE]. Srisuresh & Ford [Page 12] Internet-Draft Complications from NAT Deployments October 2006 Ideally, ISPs should not use NAT devices to connect their customers, so the customers do not get caught up in a multi-level NAT scenario and hairpin session flows. NAT devices must support hairpin translations for all protocol sessions the NAT device supports. Hairpin translation support is a requirement for peer node connectivity in multi-level NAT topologies. This is reiterated in BEHAVE recommendations for NAT devices for UDP, TCP and ICMP protocols ([BEH_UDP], [BEH-TCP], [BEH-ICMP]). Recommendation-3. As specified in BEHAVE documents for IP protocols ([BEH-UDP], [BEH-TCP], [BEH-ICMP]), NAT devices need to support hairpin translation. 2.2.4. Mistaken End Host Identity There can be a potential security threat due to mistaken identity in figure 1.2. Suppose, the CheapoISP in figure 1.2 used host E as the ISP mail server. Host E is assigned an RFC 1918 address of 10.1.1.11. This address can potentially overlap with addresses on private networks 3 and 4. So, if a customer of CheapoISP had a mail server installed on his/her private network, bearing an IP address exactly the same as host E, this could pose a severe security threat to customer's mail messages. Potentially, the customer mail messages could be hijacked by the ISP's mail server. Ideally, ISPs should not use NAT devices to provide connectivity to their customers. If they do, any servers on the ISP's private network that need to be accessible to the ISP's customers (e.g., mail servers) should have global IP addresses, to ensure accessibility to customers who deploy NAT devices themselves. Recommendation-4. Ideally, ISPs should not use NAT devices to provide connectivity to their customers. If they do, any servers on the ISP's private network that need to be accessible to the ISP's customers (e.g., mail servers) should have global IP addresses, to ensure customer IP addresses don't conflict with IP addresses of the ISP's servers. 2.3. Summary of Recommendations The following is a summary of recommendations identified in section 2.2 to support the unconventional multi-level NAT topologies, such as the one identified in figure 1. The recommendations are addressed to NAT vendors, ISPs and consumers. Note, the recommendations listed are emanating merely from the perspective of a specific topology considered in the document. For this reason, the recommendations Srisuresh & Ford [Page 13] Internet-Draft Complications from NAT Deployments October 2006 should not be considered complete for NAT vendors, ISPs or consumers. Specifically, the recommendations for NAT vendors is limited. Readers should refer BEHAVE protocol documents ([BEH-UDP], [BEH-TCP] and [BEH-ICMP]) for a comprehensive list of requirements for NAT vendors. It may be noted that the recommendations provided for NAT vendors in this document (i.e., recommendations 2 & 3) are well in line with the recommendations in BEHAVE documents. Recommendation-1. Consumers should be informed that all private hosts behind a "Plug-and-play" NAT must be on a single LAN subnet, and that there should be no more than one "Plug-and-play" NAT device on the same LAN. Recommendation-2. As specified in REQ-7 of [BEH-UDP], NAT devices should support IP address space overlap between the address space on its private interface and the address space on its public interface. Essentially, a NAT device should keep the two IP address spaces separate and base all packet processing decisions on the "side" or "port" from which the packet arrived and not just on the basis of the IP addresses. Recommendation-3. As specified in BEHAVE documents for IP protocols ([BEH-UDP], [BEH-TCP], [BEH-ICMP]), NAT devices need to support hairpin translation. Recommendation-4. Ideally, ISPs should not use NAT devices to provide connectivity to their customers. If they do, any servers on the ISP's private network that need to be accessible to the ISP's customers (e.g., mail servers) should have global IP addresses, to ensure customer IP addresses don't conflict with IP addresses of the ISP's servers. 3. Remote Access VPN Network Topologies Enterprises use remote access VPN to allow secure access to the employees working outside the enterprise premises. While outside the enterprise premises, an employee may be located in his/her home office, hotel, conference or a partner's office. In all cases, it is desirable for the employee at the remote site to have unhindered access to his/her corporate network and the applications running on the corporate networks. This is so the employee can get his/her work done seamlessly without jeopardizing the integrity and confidentiality of the corporate network and the applications running on the network. IPsec, L2TP and SSL are some of the well known secure VPN technologies used by the remote access vendors. Besides Srisuresh & Ford [Page 14] Internet-Draft Complications from NAT Deployments October 2006 authenticating employees for granting access, remote access VPN servers often enforce different forms of security (e.g. IPsec, SSL) to protect the integrity and confidentiality of the run-time traffic between the VPN client and the VPN server. Many enterprises deploy their internal networks using RFC-1918 private address space and use NAT devices to connect to the public Internet. Further, many of the applications in the corporate network refer to information (such as URLs) and services using private addresses in the corporate network. Applications such as NFS rely on simple source IP address based filtering to restrict access to corporate users. These are some reasons why the remote access VPN servers are configured with a block of IP addresses from the corporate private network to assign to remote access clients. VPN clients use the virtual IP (VIP) address assigned to them (by the corporate VPN server) to access applications inside the corporate. Consider the remote access VPN scenario in figure 2 below. (Corporate Private network 10.0.0.0/8) ---------------+---------------------- | 10.1.1.10 +---------+-------+ | Enterprise Site | | Remote Access | | VPN Server | +--------+--------+ 129.32.34.18 | {---------+------} { } { } { Public Internet } { (Public IP addresses) } { } { } {---------+------} | 155.99.25.1 +--------+--------+ | Remote Site | | "Plug-and-Play" | | NAT router | +--------+--------+ 10.1.1.1 | Srisuresh & Ford [Page 15] Internet-Draft Complications from NAT Deployments October 2006 Remote Site Private Network | -----+-----------+-----------+-------------+----------- | | | | 10.1.1.10 10.1.1.11 10.1.1.12 10.1.1.13 Host A Host B +--------+ Host C | VPN | | Client | | on a PC| +--------+ Figure 2. Remote Access VPN with Overlapping Address Space In the above scenario, say an employee of the corporate is at a remote location and attempts to access the corporate network using the VPN client. The corporate network is laid out using RFC-1918 address pool of 10.0.0.0/8 and say the VPN server is configured with an address block of 10.1.1.0/24 to assign virtual IP addresses to remote access VPN clients. Now, say the employee at the remote site is attached to a network on the remote site which also happens to be using a RFC-1918 address space based network and coincidentally overlaps the corporate network. In this scenario, it is conventionally problematic for the VPN client to connect to the server(s) and other hosts at the enterprise. Nevertheless, despite the direct address overlap, the remote access VPN connection between the VPN client at the remote site and the VPN server at the enterprise should remain connected and should be made to work. I.e., the NAT device at the remote site should not obstruct the VPN connection traversing it. And, the remote user should be able to connect to any host at the enterprise through the VPN from the remote desktop. The following subsections describe the operational details of the VPN, anomalies with the address overlap and recommendations on how best to address the situation. 3.1. Operational Details of Remote Access VPN Network As mentioned earlier, in the "de facto" Internet address architecture, only the nodes on the public Internet have globally unique IP addresses assigned by the official IP address registries. Many of the networks in the edges use private IP addresses from RFC 1918 and use NAT devices to connect their private networks to the public Internet. Many enterprises adapted the approach of using private IP addresses internally. Employees within the enterprise's Intranet private network are "trusted" and may connect Srisuresh & Ford [Page 16] Internet-Draft Complications from NAT Deployments October 2006 to any of the internal hosts with minimal administrative or policy enforcement overhead. When an employee leaves the enterprise premises, remote access VPN provides the same level of intranet connectivity to the remote user. The objective of this section is to provide an overview of the operational details of a remote access VPN application so the reader has an appreciation for the problem of remote address space overlap. This is not a tutorial or specification of remote access VPN products, per se. When an employee at a remote site launches his/her remote access VPN client, the VPN server at the corporate premises demands the VPN client to authenticate itself. When the authentication succeeds, the VPN server assigns a Virtual IP (VIP) address to the client for connecting with the corporate Intranet. From this point onwards, while the VPN is active, outgoing IP packets directed to the hosts in the corporate Intranet are tunneled through the VPN, in that the the VPN server serves as the next-hop and the VPN connection as the next-hop link for these packets. Within the corporate Intranet, the outbound IP packets appear as arriving from the VIP address. so, IP packets from the corporate hosts to the remote user are sent to the remote user's VIP address and the IP packets are tunneled inbound to the remote user's PC through the VPN tunnel. This works well so long as the subnets in the corporate network do not conflict with subnets at the remote site where the remote user's PC is located. However, when the corporate network is built using RFC 1918 private address space and the remote location where the VPN client is launched is also using an overlapping network from RFC 1918 address space, there can be addressing conflicts. The remote user's PC will have a conflict in accessing nodes on the corporate site and nodes at the remote site bearing the same IP address simultaneously. Consequently, the VPN client may be unable to have full access to the employee's corporate network and the local network at the remote site simultaneously. In spite of address overlap, remote access VPN clients should be able to successfully establish connections with Intranet hosts in the enterprise. 3.2. Anomalies of the Remote Access VPN network Even though conventional wisdom would suggest that the remote access VPN scenario with overlapping address space would be seriously broken, in practice it still works in many ways. Let us look at some anomalies where there might be a problem and identify solutions Srisuresh & Ford [Page 17] Internet-Draft Complications from NAT Deployments October 2006 through which the remote access VPN application could be made to work even under the problem situations. 3.2.1. Remote Router and DHCP Server Address Conflict Routing and DHCP service are bootstrap services essential for a remote host to establish a VPN connection. Without DHCP lease, the remote host can not communicate over the IP network. Without a router to connect to the Internet, the remote host is unable to access past the local subnet to connect to the VPN server at the enterprise. It is essential that neither of these bootstrap services be tampered at the remote host in order for the VPN connection to stay operational. Typically, a "Plug-and-play" NAT device at the remote site provides both routing and DHCP services from the same IP address. When there is address overlap between hosts at corporate Intranet and hosts at the remote site, the remote VPN user is often unaware of the address conflict. Address overlap could potentially cause the remote user to lose connectivity to the enterprise entirely or lose connectivity to an arbitrary block of hosts at the enterprise. Consider, for example, a scenario where the IP address of the DHCP server at the remote site matched the IP address of a host at the enterprise network. When the remote user's PC is ready to renew the lease of the locally assigned IP address, the remote user's VPN client would incorrectly identify the IP packet as being addressed to an enterprise host and tunnel the DHCP renewal packet over the VPN to the remote VPN server. The DHCP renewal requests simply do not reach the DHCP server at the remote site. As a result, the remote PC would eventually lose the lease on the IP address and the VPN connection to the enterprise would be broken. Consider another scenario where the IP address of the remote user's router overlapped with the IP address of a host in the enterprise network. If the remote user's PC were to send ping or some type of periodic keep-alive packets to the router (say, to test the liveness of the router), the packets are intercepted by the VPN client and simply redirected to the VPN tunnel. This type of unintended redirection has the twin effect of hijackng critical packets addresed to the router as well as the host in the enterprise network (bearing the same IP address as the remote router) being bombarded with unintended keep-alive packets. Loss of connectivity to the router can result in the VPN connection being broken. Clearly, it is not desirable for the corporate intranet to conflict Srisuresh & Ford [Page 18] Internet-Draft Complications from NAT Deployments October 2006 with the IP addresses of the router and DHCP server at the remote site. VPN servers should, at a minimum, disallow access to corporate hosts that are using an IP address that might match any of the following entities at the remote site - a) client's next-hop router IP address used to access the VPN server, and b) DHCP server providing address lease on the remote host network interface. By doing this, VPN client on the remote PC will not intercept IP packets whose target IP addresses are not in the authorized list of enterprise hosts. And, the VPN connection remains. This however has the downside that the VPN client loses connectivity to a potentially mission critical host at the corporate site. Recommendation-5. When deploying a remote access VPN client, if there is conflict of address space between corporate Intranet and the remote site, the VPN server server should disallow access from the VPN client to corporate hosts bearing the same IP address as the router or DHCP server at the remote site. 3.2.2. Simultaneous Connectivity Conflict Ideally speaking, it is not desirable for the corporate intranet to conflict with any of the hosts at the remote site. As a general practice, if simultaneous communication with end hosts at the remote location is important, it is advisable to disallow access to any corporate network resource that overlaps the client's subnet at the remote site. By doing this, the remote user is able to connect to all local hosts simultaneously while the VPN connection is active. For example, if the PC's external network interface is configured with 10.1.1.1/24, the VPN server may be configured to disallow access to the corporate network that overlaps this subnet at the remote site for the VPN client. Such a configuration on the VPN server is also termed sometimes as "Split VPN" configuration. When "Split VPN" is enabled, the remote user is able to carry out simultaneous communication with hosts at the remote site and the hosts at the corporate intranet, with the exception of the hosts that overlapped the remote subnet. If simultaneous connectivity to local hosts is not important, the VPN server may be configured to require the VPN client to direct all communication traffic from the remote user to the VPN server across the VPN. This essentially ensures that all communication from the remote user's PC traverses the VPN link and routed through the VPN server. No communication takes place with hosts on the remote site. This configuration on the VPN server is also termed sometimes as "Non-split VPN". When "Non-Split VPN" is enabled, all traffic from the remote user's PC is directed to the VPN server, with the exception of traffic directed to the local Srisuresh & Ford [Page 19] Internet-Draft Complications from NAT Deployments October 2006 router and DHCP server. Recommendation-6. If simultaneous communication with end hosts at remote site is important, enterprises should configure the VPN server in "Split VPN" mode and disallow access to corporate hosts that overlap the client's subnet at remote site. If simultaneous connectivity to hosts at remote site is not important, enterprises should configure the VPN server in "Non-split VPN" mode, so the VPN client directs to the VPN server all traffic from the remote user, with the exception of traffic to the router and DHCP server at the remote site. 3.2.3. VIP Address Conflict When the VIP address assigned to the VPN client at the remote site is in direct conflict with the IP address of the existing network interface, the VPN client might be unable to establish the VPN connection. Consider a scenario where the VIP address assigned by the VPN server directly matched the IP address of the networking interface at the remote site. When the VPN client on the remote host attempts to set the VIP address on a virtual adapter (specific to the remote access application), the VIP address configuration will simply fail due to conflict with the IP address of the existing network interface. The configuration failure in turn will result in the remote access VPN tunnel not being established. Clearly, it is not advisable to have the VIP address overlap the IP address of the remote user's existing network interface. As a general rule, it is not advisable for the VIP address to overlap any IP address in the remote user's local subnet, as the VPN client on the remote PC might be forced to respond to ARP requests on the remote site and the VPN client might not process the handling of ARP requests gracefully. We recommend that VPN vendors offer provision to detect conflict of VIP address with remote site address space and switch between a minimum two VIP address pools on the VPN server. We also recommend enterprises deploying the VPN solution to use this vendor provision and configure the VPN server with a minimum of two distinct IP address pools. Alternately, the enterprises should deploy a minimum of two VPN servers with different address pools. By doing this, a VPN client that detected the conflict of VIP address with the local subnet is able to reconnect with the alternate VPN server using the alternate address pool that will not conflict. Srisuresh & Ford [Page 20] Internet-Draft Complications from NAT Deployments October 2006 Recommendation-7. VPN vendors should offer provision to detect conflict of VIP address with remote site address space and switch between a minimum of two VIP address pools with different subnets on the VPN server so the VIP address assigned is not in conflict with the address space at remote site. Recommendation-8. Enterprises deploying remote access VPN solution should adapt a strategy to avoid VIP address conflict with the subnet at remote site. Below are two suggestions. a) If the VPN device being deployed has provision to configure two address pools (as in recommendation above), configure the VPN server with a minimum of two distinct IP address pools. b) Deploy a minimum of two VPN servers with different address pools. By doing this, a VPN client that detected the conflict of VIP address with the subnet at remote site has the option to switch to alternate VPN server and eliminate VIP address conflict. 3.2.4. Mistaken End Host Identity When "Split VPN" configuration is set on the VPN server, there can be a potential security threat due to mistaken identity. Say, a certain service (ex: SMTP mail service) is configured on exactly the same IP address on both the corporate site and the remote site. The user could unknowingly be using the service on the remote site, thereby violating the integrity and confidentiality of the contents relating to that application. Potentially, remote user mail messages could be hijacked by the ISP's mail server. Enterprises deploying remote access VPN servers should allocate global IP addresses for the critical servers the remote VPN clients typically need to access. By doing this, even if most of the private corporate network uses RFC 1918 address space, this will ensure that the remote VPN clients can always access the critical servers regardless of the private address space used at the remote attachment point. Recommendation-9. When "Split VPN" is configured on the VPN server, enterprises deploying remote access VPN servers should allocate global IP addresses for the critical servers the remote VPN clients typically need to access. 3.3. Summary of Recommendations The following is a summary of recommendations identified in section 3.2 to support the address overlap in remote access VPN networks, such as the one identified in figure 2. The recommendations are Srisuresh & Ford [Page 21] Internet-Draft Complications from NAT Deployments October 2006 addressed to remote access VPN vendors, enterprises deploying the VPN servers and finally, the remote access VPN consumers. Following the recommendations will help ensure that a complete "network meltdown" is prevented. Recommendation-5. When deploying a remote access VPN client, if there is conflict of address space between corporate Intranet and the remote site, the VPN server server should disallow access from the VPN client to corporate hosts bearing the same IP address as the router or DHCP server at the remote site. Recommendation-6. If simultaneous communication with end hosts at remote site is important, enterprises should configure the VPN server in "Split VPN" mode and disallow access to corporate hosts that overlap the client's subnet at remote site. If simultaneous connectivity to hosts at remote site is not important, enterprises should configure the VPN server in "Non-split VPN" mode, so the VPN client directs to the VPN server all traffic from the remote user, with the exception of traffic to the router and DHCP server at the remote site. Recommendation-7. VPN vendors should offer provision to detect conflict of VIP address with remote site address space and switch between a minimum of two VIP address pools with different subnets on the VPN server so the VIP address assigned is not in conflict with the address space at remote site. Recommendation-8. Enterprises deploying remote access VPN solution should adapt a strategy to avoid VIP address conflict with the subnet at remote site. Below are two suggestions. a) If the VPN device being deployed has provision to configure two address pools (as in recommendation above), configure the VPN server with a minimum of two distinct IP address pools. b) Deploy a minimum of two VPN servers with different address pools. By doing this, a VPN client that detected the conflict of VIP address with the subnet at remote site has the option to switch to alternate VPN server and eliminate VIP address conflict. Recommendation-9. When "Split VPN" is configured on the VPN server, enterprises deploying remote access VPN servers should allocate global IP addresses for the critical servers the remote VPN clients typically need to access. 4. Security Considerations This document does not inherently create new security issues. Security issues known to DHCP servers and NAT devices are Srisuresh & Ford [Page 22] Internet-Draft Complications from NAT Deployments October 2006 applicable, but not within the scope of this document. Likewise, security issues specific to remote access VPN devices are also appliable to the remote access VPN topology, but not within the scope of this document. The security issues reviewed here only those relevant to the topologies described in sections 2 and 3, specifcally as they apply to private address space overlap in the topologies described. Mistaken end host identity is a security concern present in both topologies discussed. Mistaken end host identity, described in sections 2.2.4 and 3.2.4 for each of the topologies reviewed, essentially points the possibility of application services being hijacked by the wrong application server (ex: Mail server). Security violation due to mistaken end host identity arises principally due to critical servers being assigned RFC 1918 private addresses. The recommendation suggested for both scenarios is to assign globally unique pulic IP addresses for the critical servers. It is also recommended in section 2.1.2 that applications adapt end-to-end authentication and not depend on source IP address for authentication. Doing this will thwart connection hijacking and denial of service attacks. 5. Acknowledgements The authors wish to thank Dan Wing for reviewing the document in detail and making helpful suggestions in reorganizing the document format. 6. Normative References [BEH-ICMP] Srisuresh, P., Ford, B., Sivakumar, S., and Guha, S., "NAT Behavioral Requirements for ICMP Protocol", draft-ietf-behave-nat-icmp-01.txt (Work In Progress), October 2006. [BEH-TCP] Guha, S., Biswas, K., Ford, B., Sivakumar, S., and Srisuresh, P., "NAT Behavioral Requirements for TCP", draft-ietf-behave-nat-tcp-02.txt (Work In Progress), October 2006. [BEH-UDP] Audet, F. and Jennings, C., "NAT Behavioral Requirements for Unicast UDP", draft-ietf-behave-nat-udp-08.txt (Work In Progress), October 2006. [NAT-TERM] P. Srisuresh and M. Holdrege, "IP Network Address Translator Srisuresh & Ford [Page 23] Internet-Draft Complications from NAT Deployments October 2006 (NAT) Terminology and Considerations", RFC 2663, August 1999. [NAT-TRAD] P. Srisuresh and K. Egevang, "Traditional IP Network Address Translator (Traditional NAT)", RFC 3022, January 2001. [RFC1918] Rekhter, Y., Moskowitz, B., Karrenberg, D., de Groot,G. and Lear, E., "Address Allocation for Private Internets", BCP 5, RFC 1918, February 1996. 7. Informational References [NAT-PROT] Holdrege, M., and Srisuresh, P., "Protocol Complications with the IP Network Address Translator", RFC 3027, January 2001. [P2P-STATE] Srisuresh, P., Ford, B., and Kegel, D., "State of Peer-to- Peer(P2P) Communication Across Network Address Translators (NATs)", draft-ietf-behave-p2p-state-00.txt, October 2006, Work in Progress. Authors' Addresses: Pyda Srisuresh Consultant 20072 Pacifica Dr. Cupertino, CA 95014 U.S.A. Phone: (408) 836-4773 E-mail: srisuresh@yahoo.com Bryan Ford Computer Science and Artificial Intelligence Laboratory Massachusetts Institute of Technology 77 Massachusetts Ave. Cambridge, MA 02139 U.S.A. Phone: (617) 253-5261 E-mail: baford@mit.edu Web: http://www.brynosaurus.com/ Intellectual Property Statement The IETF takes no position regarding the validity or scope of any Intellectual Property Rights or other rights that might be claimed to Srisuresh & Ford [Page 24] Internet-Draft Complications from NAT Deployments October 2006 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 ietf-ipr@ietf.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. Srisuresh & Ford [Page 25]