INTERNET DRAFT Jeffrey Lo Expires November 1999 NEC USA Michael Borella David Grabelsky 3Com Corp Realm Specific IP: A Framework Status of this Memo This document is an Internet-Draft and is in full conformance with all provisions of Section 10 of RFC2026. 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/ietf/1id-abstracts.txt The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html. Abstract This document examines the general framework of Realm Specific IP (RSIP). All RSIP solutions must solve the same set of problems, and all RSIP-related proposals to date are similar in many ways. We attempt to enumerate the similarities and differences of these proposals, and expand the scope of RSIP to include several other possible mechanisms. We do not advocate any one RSIP solution over the other; instead, we present these solutions in the hope to clarify RSIP issues and generate further discussion towards adoption of RSIP. 1. Introduction While NAT has become a popular mechanism of sharing IP addresses amongst a number of hosts, it suffers from a lack of flexibility. In particular, a NAT router must examine and change the network layer, Lo et. al. Expires November 1999 [Page 1] INTERNET DRAFT Realm Specific IP: Framework May 1999 and possibly the transport layer, headers of each packet to or from the NAT subnet(s) sharing its IP address(es). This causes NAT to break the end-to-end nature of Internet connectivity, and disrupts protocols requring end-to-end connectivity, such as the network security protocols which embody IPSEC. Furthermore, any application that transmits IP address or port content, such as FTP, will require a proxy module (application layer gateway) within the NAT router. Given these limitations of NAT, RSIP has emerged as an attempt to remedy them. RSIP is based on the concept of granting host from one realm (e.g., privately addressed realm) a presence in another realm (e.g., publicly addressed realm) by granting it resources from the second realm. While this document is limited to the discussion of IPv4 networks, RSIP is general and may be applied beyond the limitations of IPv4 networks and used as a means of IPv4 public address assignment to IPv6 subnets. In this document we discuss the approaches of several possible RSIP systems and address the issues that any RSIP solution must face. 2. Terminology Private Realm A routing realm that uses private IP addresses from the ranges (10/8, 172.16/12, 192/16) specified in [RFC1918], or addresses that are non-routable from the Internet. Public Realm A routing realm with unique network addresses assigned by the Internet Assigned Number Authority (IANA) or an equivalent address registry. RSIP Server A router situated on the boundary between a private realm and a public realm and owns one or more public IP addresses. An RSIP server is responsible for public parameter management and assignment to RSIP clients. An RSIP server may act as a normal NAT box for hosts within the private realm that are not RSIP enabled. RSIP Client A host within the private realm that assumes publicly unique parameters from an RSIP server through the use of RSIP. Lo et. al. Expires November 1999 [Page 2] INTERNET DRAFT Realm Specific IP: Framework May 1999 RSA-IP: Realm Specific Address IP An RSIP method in which each RSIP client is allocated a unique IP address from the public realm. Dicussed in detail in [NAT-TERM]. RSAP-IP: Realm Specific Address and Port IP An RSIP method in which each RSIP client is allocated an IP address (possibly shared) and some number of per-address unique ports from the public realm. Dicussed in detail in [NAT-TERM]. All other terminology found in this document is consistent with that of [NAT-TERM]. 3. Specification of Requirements The keywords "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this documents are to be interpreted as described in [RFC2119]. 4. Architecture In a typical scenario where RSIP is deployed, there are some number of hosts within one addressing realm connected to another addressing realm by the RSIP server. This model is diagrammatically represented as follows: RSIP Client RSIP Server Host Xa Na Nb Yb [X]------( Addr sp. A )----[N]-----( Addr sp. B )-------[Y] ( Network ) ( Network ) Hosts X and Y belong to different addressing realms A and B, respectively, and N is an RSIP server (which may also act as a NAT). N has two addresses: Na on address space A, and Nb on address space B. N may have a pool of addresses in address space B which it can assign to or lend to X and other hosts in addressing realm A. These addresses are not shown above, but they can be denoted as Nb1, Nb2, Nb3 and so on. The hosts within address space A are likely to use private addresses while the RSIP server is multi-homed with one or more private addresses in addition to it's public addresses. Using the public parameters assigned by the RSIP server, RSIP clients route (usually tunnel) data packets to the RSIP server within address space A. If tunneling is used, the RSIP server acts as the end point of such tunnels, stripping off the outer headers and routing the Lo et. al. Expires November 1999 [Page 3] INTERNET DRAFT Realm Specific IP: Framework May 1999 inner packets onto the public realm. An RSIP server maintains a mapping of the assigned public parameters as demultiplexing tuples for uniquely mapping them to RSIP client private addresses. When a packet from the public realm arrives at the RSIP server and it matches a given set of demultiplexing tuples, then the RSIP server will tunnel it to the appropriate RSIP client. 5. RSIP Fundamentals This section discusses the issues that all RSIP schemes must address. Note that these issues are not orthogonal; thus, by addressing one, in some cases another issue is also addressed sufficiently. 5.1. Negotiation and/or determination of Demultiplexing Fields Assume that an RSIP client within a private realm has transmitted a request to a public server within a public realm, and the server has sent a response packet that successfully arrived at the RSIP server. Based on a pre-arranged mapping, the RSIP server must be able to determine the private IP address of the packet's destination; i.e., the RSIP client. The only information that the RSIP server may use is what is already contained within the headers of the inbound data packet. We will refer to these header fields as the "demultiplexing fields" as they are used to spread the incoming streams of packets to multiple destinations within the private realm. Depending on the type of mapping used by the RSIP server, demultiplexing parameters could either be public IPv4 addresses, TCP/UDP ports, IPSEC Security Payload Indexes (SPIs), ISAKMP initiator cookies, some combination of the above, or some other field(s). Such demultiplexing of incoming traffic resembles a decision tree, which could be represented as follows: - A unique public IP address is mapped to each RSIP client. - If the same IP address is used for more than one RSIP client, then subsequent headers must have at least one field that will be assigned a unqieu value per client so that it is usable as a demultiplexing field. - If the subsequent header is TCP or UDP, then destination port number can be used. Otherwise, there must exist another field usable as a demultiplexing field. - If the TCP or UDP port number is the same for more than one RSIP client, the payload section of the packet must contain a demultiplexing field that is guaranteed to be different for each RSIP client. In general, it is desirable for all demultiplexing fields to occur Lo et. al. Expires November 1999 [Page 4] INTERNET DRAFT Realm Specific IP: Framework May 1999 in well-known fixed locations so that an RSIP server can mask out and examine the appropriate fields on incoming packets. In some cases, for example, where different RSIP methods (RSA-IP, RSAP-IP, etc.) are used by the same RSIP client using just one IP address, the decision tree approach would be beneficial. Demultiplexing of incoming streams of packet requires pre- assignment of the demultiplexing fields to RSIP clients. Hence there exists a requirement for a negotiation process that enables these parameters to be negotiated between RSIP server and RSIP clients. Such a negotiation process can be based on the following approaches. - As an extension of current host configuration or policy protocols such as DHCP, COPS, RADIUS, DIAMETER, or SOCKS. - During tunnel establishment, for example as an extension to L2TP parameter negotiation. - As an RSIP specific protocol such as described in [RSIP-PROTO]. 5.2. Determination of other RSIP parameters Apart from negotiation of demultiplexing fields, other information pertaining to the assignment of those fields may also need to be negotiated. Examples of such parameters are: A binding identifier may be assigned for each public parameter assignment. The binding identifier serves to uniquely identify the resource(s) that has been allocated by an RSIP server. It may also be used during lookup to efficiently index existing bindings. A time duration (lease) may be associated with each bind of public parameters to an RSIP client. RSIP clients may require that the RSIP server specify how it allocates address and port resources (referred to as the RSIP method). RSIP servers may only allocate a public IP address to each unique host, known as RSA-IP. Or, RSIP-servers may distribute a (potentially shared) public IP address and a unique port range per that IP address to each host, termed RSAP-IP. The negotiation and assignment mechanism SHOULD be extensible and facilitate vendor specific parameters. 5.3. Tunnel Use and Establishment Once the public demultiplexing fields have been allocated by the Lo et. al. Expires November 1999 [Page 5] INTERNET DRAFT Realm Specific IP: Framework May 1999 RSIP server, RSIP clients will be able to use them freely. However, RSIP implementations generally requires data packets to be tunnelled between the RSIP client and server within the private realm since public routing information is not advertised in the private realm. While it is possible to imagine an RSIP implementation that does not require tunneling, it seems that tunneling is a flexible method for solving address ambiguity problems. The type of tunnel may be IP-IP, GRE, IPSEC tunnel mode, L2TP, or another form of tunnel. There is a disadvantage to not using a tunnel between the RSIP client and the RSIP server. It is likely that an RSIP server will also act as a firewall or packet filter for the private network. In this case, it publically addressed packets are transmitted on the private network, the RSIP server may consider these packets tobe part of an attack. Tunnels may be established statically or dynamically between RSIP clients and servers. A static tunnel is established at host boot and remains in service until the host is no longer using the network. A dynamic tunnel is established at the beginning of a session or flow and exists only for the lifetime of the session. Both types of tunnels may allow for on-the-fly re-negotiation of demultiplexing fields and re-assignment of parameters to RSIP clients. If tunneling is used to route the publicly addressed packet within private realm, public parameter negotiation could be associated with tunnel establishment mechanisms. Alternatively, a negotiation protocol may enable the negotiation of tunnel type as well. 6. Miscellaneous Issues The resolution of a number of RSIP issues are still open. Although solutions may exist for these problems, they may have unattractive side effects. In this section we discuss several such issues. 6.1. Policy and Accounting All RSIP-clients SHOULD have a mechansims of authenticating themselves to RSIP-servers, in order to alleviate possible denial of service attacks in which a malicious RSIP client attempts to utilize the resources assigned to a different RSIP client. Any RSIP implementation SHOULD implement accounting of irregular event seen by the RSIP-server. Events such as denial of service attacks, illegal use of resources (traffic without bindings or after binding expirations) and public resource depletion SHOULD be logged. Lo et. al. Expires November 1999 [Page 6] INTERNET DRAFT Realm Specific IP: Framework May 1999 6.2. Contacting Internal Servers In order for an RSIP implementation to allow private hosts to run servers that can be contacted from the public network, these servers must be registered with the RSIP server. Registration of servers with unique and/or well known listen ports may be limited to one per private realm unless other means beyond port number are used for demultiplexing (e.g. multiple WWW domains may be disambiguated by looking into the HTTP headers). 6.3. Determining Locality of Destinations In general, an RSIP client must know, for a particular IP address, whether it should transmit the packet normally for local delivery, or tunnel the packet to the RSIP server. Since more than one subnet may be behind an RSIP server, looking at a local subnet mask will not always work. We'd rather not have to propagate routing tables to all RSIP clients. A simple alternative, proposed in [RSIP-PROTO], that will solve this problem is to require that the RSIP server knows all of the subnets that are on the private network. This information can be manually entered because it is not expected to change often. Then, if an IP address in question is not on a host's local subnet, the host can query the server with the address. The RSIP server will return a simple "yes or "no" answer - yes, this address is local, or no, it is not. As proposed in [RSIP-PROTO], a queried RSIP server may respond with the list of subnets supported. An RSIP client may cache this information. However, in large enterprise networks, an RSIP server may not be aware of all private subnets. Alternatively, RSIP-clients could send all packets for destinations without an explicit static route to the RSIP server. If they arrive at the RSIP server, it informs the host that it should instead tunnel the packet. The host then acquires the necessary public parameters and tunnels the packet, to the RSIP server. This approach may require further changes to the TCP/IP stack at the host, since, in the case of TCP traffic, a half-open TCP socket must be discarded. Likewise, the RSIP client could at first tunnel the packets to the RSIP server. If the server determines that the destination is local, it would inform the host of this fact and the host could then transmit the packet in the standard fashion. Regardless of the solution chosen, RSIP clients caching the "locality" of recently-contacted IP addresses may be beneficial. 6.4. Implementing RSIP Client Deallocation As currently defined in [RSIP-PROTO], an RSIP client MAY free Lo et. al. Expires November 1999 [Page 7] INTERNET DRAFT Realm Specific IP: Framework May 1999 resources that it has determined that it no longer requires. For example, on a large RSAP-IP subnet with a limited number of public IP addresses, locally-unique port numbers may become scarse. Thus, if RSIP clients are able to deallocate ports that they no longer need, RSIP will be more scalable. However, this functionality may require significant modifications to a vanilla TCP/IP stack in order to implement properly. The RSIP client must be able to determine which TCP or UDP sessions are using RSIP resources. If those resources are unused for a period of time, then the RSIP client may deallocate them. When an open socket's resources are deallocated, it will cause some associated applications to fail. An analogous case would be TCP and UDP sessions that must terminate when a PPP interface that they are using loses connectivity. On the other hand, this issue can be considered a resource allocation problem. It is not recommended that a large number (hundreds) of hosts share the same IP address, for performance purposes. Even if, say, 100 hosts each are allocated 100 ports, the total number of ports in use by RSIP would be still less than one-sixth the total port space for an IP address. If more hosts or more ports are needed, more IP addresses should be used. Thus, it is reasonable, that in many cases, RSIP clients will not have to deallocate ports for the lifetime of their activity. Similarly, it is non-trivial for an RSIP client to know when to allocate ports. It will have to detect activity on a socket, determine if the destination host is local or external, and then request the appropriate resources. In cases when the allocation requires multiple rounds, for example when more than one public resources are to be allocated and multiple assignment requests are issued or a request gets denied a number of times, delays may be introduced by the resource allocation process. 7. Cascaded RSIP It is possible for RSIP to allow for cascading of RSIP-servers. For example, consider an ISP that uses RSIP for address sharing amongst its customers. It might assign resources (e.g., IP addresses and ports) to a particular customer. This customer may further subdivide the port ranges and address(es) amongst individual end hosts. A reference architecture is depicted below. Lo et. al. Expires November 1999 [Page 8] INTERNET DRAFT Realm Specific IP: Framework May 1999 +-----------+ | | | RSIP | | server +---- 10.0.0.0/8 | B | | | +-----+-----+ | | 10.0.1.0/24 +-----------+ | (149.112.240.0/25) | | | 149.112.240.0/24| RSIP +--+ ----------------+ server | | A +--+ | | | +-----------+ | 10.0.2.0/24 | (149.112.240.128/25) | +-----+-----+ | | | RSIP | | server +---- 10.0.0.0/8 | C | | | +-----------+ RSIP-server A is in charge of the IP addresses of subnet 149.112.240.0/24. It distributes these addresses to RSIP-clients and RSIP-servers. In the given configuration, it distributes addresses 149.112.240.0 - 149.112.240.127 to RSIP-server B, and addresses 149.112.240.128 - 149.112.240.254 to RSIP-server C. Note that the subnet broadcast address, 149.112.240.255, must remain unclaimed, so that broadcast packets can be distributed to arbitrary hosts behind RSIP-server A. Also, the subnets between RSIP-server A and RSIP- servers B and C will use private addresses. Due to the tree-like fashion in which addresses will be cascaded, we will refer to RSIP-servers A as the 'parent' of RSIP-servers B and C, and RSIP-servers B and C as 'children' of RSIP-servers A. An arbitrary number of levels of children may exist under a parent RSIP- server. A parent RSIP-server will not necessarily be aware that the address(es) and port blocks that it distributes to a child RSIP- server will be further distributed. Thus, the RSIP-clients MUST tunnel their outgoing packets to the nearest RSIP-server. This server will then verify that the sending host has used the proper address and port block, and then tunnel the packet on to its parent Lo et. al. Expires November 1999 [Page 9] INTERNET DRAFT Realm Specific IP: Framework May 1999 RSIP-server. For example, in the context of the diagram above, host 10.0.0.1, behind RSIP-server C will use its assigned external IP address (say, 149.112.240.130) and tunnel its packets over the 10.0.0.0/8 subnet to RSIP-server C. RSIP-server C strips off the outer IP header. After verifying that the source public IP address and source port number is valid, RSIP-server C will tunnel the packets over the 10.0.2.0/8 subnet to RSIP-server A. RSIP-server A strips off the outer IP header. After verifying that the source public IP address and source port number is valid, RSIP-server A transmits the packet on the public network. While it may be more efficient in terms of computation to have a RSIP-client tunnel directly to the overall parent of an RSIP-server tree, this would introduce significant state and administrative difficulties. A RSIP-server that is a child MUST take into consideration the parameter assignment constraints that it inherits from its parent when it assigns parameters to its children. For example, if a child RSIP-server is given a lease time of 3600 seconds on an IP address, it MUST compare the current time to the lease time and the time that the lease was assigned to compute the maximum allowable lease time on the address if it is to assign the address to a RSIP-client or child RSIP-server. 8. To Do - RSIP impact on ALGs? - Pros and cons of RSIP riding on top of existing protocols such as DHCP, RADIUS, SOCKS, etc. 9. Changelog 00 to 01: - Synched up terminology with the latest NAT terminology draft. - Changed all instances of "global" to "public" - Modified section on "Architecture" - Added discussion of demultiplexing parameters tree to the "Negotiation and assignment of demultiplexing fields" section - Added discussion of subnet list query in "Determining Locality of Destination" section - Added "RSIP Client Deallocation" discussion section - Added more explanation in "Tunnel Use and Establishment" section 10. Acknowledgements Lo et. al. Expires November 1999 [Page 10] INTERNET DRAFT Realm Specific IP: Framework May 1999 The authors would like to thank Gabriel Montenegro, Pyda Srisuresh, Dan Nessett, Gary Jaszewski, and Rick Cobb for their input. 11. References [RSIP-PROTO] Michael Borella, David Grabelsky, Jeffrey Lo and Kuni Taniguchi, "Realm Specific IP: Protocol Specification," , work in progress, Apr. 1999. [RFC2119] S. Bradner, "Key words for use in RFCs to indicate requirement levels," RFC 2119, Mar. 1997. [RFC1918] Y. Rekhter, B. Moskowitz, D. Karrenberg, G. J. de Groot, and E. Lear, "Address Allocation for Private Internets," RFC 1918, Feb. 1996. [NAT-TERM] P. Srisuresh and Matt Holdrege, "IP Network Address Translator (NAT) Terminology and Considerations," , work in progress, Apr. 1999. 12. Authors' Addresses Jeffrey Lo NEC USA C&C Research Labs. 110 Rio Robles San Jose, CA 95134 (408) 943 3033 jlo@ccrl.sj.nec.com Michael Borella 3Com Corp. 1800 W. Central Rd. Mount Prospect IL 60056 (847) 842 6093 mike_borella@3com.com David Grabelsky 3Com Corp. 1800 W. Central Rd. Mount Prospect IL 60056 (847) 222 2483 david_grabelsky@3com.com Copyright (c) The Internet Society (1999). All Rights Reserved. Lo et. al. Expires November 1999 [Page 11] INTERNET DRAFT Realm Specific IP: Framework May 1999 This document and translations of it may be copied and furnished to others, and derivative works that comment on or otherwise explain it or assist in its implementation may be prepared, copied, published and distributed, in whole or in part, without restriction of any kind, provided that the above copyright notice and this paragraph are included on all such copies and derivative works. However, this document itself may not be modified in any way, such as by removing the copyright notice or references to the Internet Society or other Internet organizations, except as needed for the purpose of developing Internet standards in which case the procedures for copyrights defined in the Internet Standards process must be followed, or as required to translate it into languages other than English. The limited permissions granted above are perpetual and will not be revoked by the Internet Society or its successors or assigns. This document and the information contained herein is provided on an "AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING TASK FORCE DISCLAIMS 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. Lo et. al. Expires November 1999 [Page 12]