Network Working Group M. Stenberg Internet-Draft S. Paavolainen Expires: January 12, 2001 T. Ylonen T. Kivinen SSH Communications Security Corp July 14, 2000 IPsec NAT-Traversal draft-stenberg-ipsec-nat-traversal-00.txt 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. This Internet-Draft will expire on January 12, 2001. Copyright Notice Copyright (C) The Internet Society (2000). All Rights Reserved. Abstract IPsec architecture is based on the concept of keeping data secure while it is being transported across a network. Therefore there are problems when packet headers change while in transmit across the network, by virtue of NAT devices. This draft details the changes needed in order to make both initial IKE negotiation and subsequent authenticated/encrypted communications across IPsec AH/ESP SAs work despite the changes in the headers, and possible protocol transformations. Stenberg, et. al. Expires January 12, 2001 [Page 1] Internet-Draft IPsec NAT-Traversal July 2000 Table of Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 1.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . 3 2. Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 4 2.1 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . 4 2.2 IPsec cases . . . . . . . . . . . . . . . . . . . . . . . . 4 2.2.1 Host-to-host . . . . . . . . . . . . . . . . . . . . . . . . 5 2.2.2 Host-to-network . . . . . . . . . . . . . . . . . . . . . . 5 2.2.3 Network-to-network . . . . . . . . . . . . . . . . . . . . . 5 2.3 Issues stemming from NAT technology . . . . . . . . . . . . 5 2.4 Summary of issues . . . . . . . . . . . . . . . . . . . . . 5 3. Solution . . . . . . . . . . . . . . . . . . . . . . . . . . 7 3.1 IKE probe . . . . . . . . . . . . . . . . . . . . . . . . . 7 3.1.1 Determining of support . . . . . . . . . . . . . . . . . . . 7 3.1.2 NAT-Traversal need-probe . . . . . . . . . . . . . . . . . . 8 3.2 IPsec SA traffic encapsulation . . . . . . . . . . . . . . . 8 3.3 Heartbeat . . . . . . . . . . . . . . . . . . . . . . . . . 10 3.3.1 Heartbeat format . . . . . . . . . . . . . . . . . . . . . . 11 3.4 Built-in NAT . . . . . . . . . . . . . . . . . . . . . . . . 11 4. Known issues with the solution . . . . . . . . . . . . . . . 13 4.1 Conceptual issues . . . . . . . . . . . . . . . . . . . . . 13 4.2 Overhead . . . . . . . . . . . . . . . . . . . . . . . . . . 13 4.3 Security considerations . . . . . . . . . . . . . . . . . . 14 4.4 Intellectual property rights . . . . . . . . . . . . . . . . 14 References . . . . . . . . . . . . . . . . . . . . . . . . . 15 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . 15 Full Copyright Statement . . . . . . . . . . . . . . . . . . 16 Stenberg, et. al. Expires January 12, 2001 [Page 2] Internet-Draft IPsec NAT-Traversal July 2000 1. Introduction NAT devices have proliferated recently. The eventual wider adoption of IPv6 will also cause great deal of NAT activity, as IPv4 is here to stay for the foreseeable future. Thus, there will be need for bridging between IPv4 and IPv6 networks, and as long as there are NATs around, basic IPsec as defined by RFC[1] will not work. It is quite important that there is a defined standard for handling IPsec traffic in networks with NAT devices. Preferably, a standard will evolve to fit all possible cases that may arise. In Section 2, most of the different IPsec+NAT permutations are analyzed and finally, a list of issues is presented. Section 3 details the proposed solution to these issues. Finally, potential problems in the solution are noted in Section 4. 1.1 Definitions Following definitions will be used when discussing different types of NATs. o Static NAT: A NAT device that (for the traffic under discussion) uses only a single static NAT translation. o Dynamic NAT: A NAT device that (for the traffic under discussion) creates address mappings dynamically based on some configured static rules. o Host NAT: A NAT that changes only the src/dst in packets. o Host/port NAT: A NAT that changes src/dst and srcport/dstport in packets. o Protocol NAT: A NAT that changes the protocol of the packet; this usually involves a whole new header for the packet. Stenberg, et. al. Expires January 12, 2001 [Page 3] Internet-Draft IPsec NAT-Traversal July 2000 2. Analysis 2.1 Assumptions It can be safely assumed that IKE[2] works. IKE negotiations are handled with normal UDP traffic, and therefore it should work despite network address changes across the route. IP packet payloads are assumed to be left unmodified; changes to the UDP headers can occur, as long as nothing drops the packets before they reach the host. As normal IPsec traffic does not pass across host/port NATs (and may not pass across protocol NATs), a complete NAT-Traversal design should encapsulate IPsec SAs in UDP packets, which are (in most of the important respects) like IP packets, except that they can pass through all types of NATs. 2.2 IPsec cases Initially, most of the different IPsec+NAT combinations are listed here to make sure that all implications of NAT use are addressed. IPsec cases can be divided to three different categories (with possible NATs in various places along the route between hosts employing IPsec). o Host-to-host (tunnel or transport mode) o Host-to-network (tunnel mode) o Network-to-network (tunnel mode) In all cases, the IKE responder must be, at best, only behind a series of static host NATs; dynamic NATs do not work, for obvious reasons (the IKE initiator cannot contact such an address), and host/port NATs do not work because IKE is defined to be port-500-only. The IKE initiator can be behind any kind of NAT, although in cases where initiation of traffic from both directions should be allowed (primarily VPN-like cases), the same restrictions that apply to the responder apply also to the initiator. ISSUE0: Both hosts need to know that there is a NAT in the middle, but currently IKE/IPsec do not provide such methodology (beyond the fact that all IPsec SA packets, if they even arrive, will be dropped as invalid). ISSUE1: It is obvious that programs residing on an IKE responder that is behind a host NAT cannot know about the existence of the Stenberg, et. al. Expires January 12, 2001 [Page 4] Internet-Draft IPsec NAT-Traversal July 2000 host NAT, nor about the specific address mappings configured there. Thus, the IKE responder implementation should have advance knowledge about the address mappings. 2.2.1 Host-to-host Host-to-host traffic using tunnel or transport mode is the most basic case; it only becomes interesting if there is no shared address space between the parties (i.e., a VPN of sorts) and there are NATs in between. ISSUE2: If NATs are employed across the route, there may be addressing conflicts in tunnel mode (and there WILL be conflicts in transport mode). From the IKE responder point of view, the IKE initiators' addresses may conflict if they are in private networks (such as the IANA-assigned 10.0.0.0 subnet). 2.2.2 Host-to-network Only tunnel mode is applicable for host-to-network communication, and the only apparent problem is the potential lack of shared address space (i.e., a host without an address in the remote network that it is accessing). Therefore, there is potential for ISSUE2-type of problems. 2.2.3 Network-to-network Only tunnel mode is applicable in network-to-network communication, and ISSUE2 is potentially a problem as well, although accounting for network-to-network non-unique address mappings may be obscure. 2.3 Issues stemming from NAT technology ISSUE3: The dynamic NATs may change their address mapping suddenly (or they may be rebooted), making the remote host concept unworkable even as a unique (host, port) pair. 2.4 Summary of issues There are basically four problems that need addressing: 1. detection of network address translation during IKE negotiation (ISSUE0), 2. a way of sending packets across the network so that NAT effects can be countered, yet the security of the system will not be affected (UDP encapsulation; assumption), 3. keeping NAT mapping static - NAT devices with dynamic host/port Stenberg, et. al. Expires January 12, 2001 [Page 5] Internet-Draft IPsec NAT-Traversal July 2000 allocation configurations typically contain timeouts that will cause changes in addressing, if not circumvented by using a heartbeat to keep the specific mappings up (ISSUE3), and 4. the lack of unique IP addresses in the NAT world; it is possible for a server to have several clients with the same configured IP address, although they appear to the server to be from different hosts/ports (ISSUE2). ISSUE1 (host NAT case, where the IKE responder does not know what address to use) is trivial to solve, as seen in the end of Section 3.2. Stenberg, et. al. Expires January 12, 2001 [Page 6] Internet-Draft IPsec NAT-Traversal July 2000 3. Solution The solution that resolves all the issues mentioned in Section 2.4 can be divided into four different parts, which are detailed in this section: o IKE probe to detect NAT presence, o IPsec SA traffic encapsulation to counter NAT effects, o NAT translation keepalive heartbeat which maintains NAT mappings, and o built-in NAT (if needed) to make addresses unique. Incoming packet Outgoing packet / | | \ / | | \ NAT-T decap. | | Un-NAT dst | | | | IPsec IPsec IPsec IPsec | | NAT src NAT-T encap. Figure 1: IPsec processing with and without a NAT-Traversal process. 3.1 IKE probe There is a need for two different exchanges in the IKE Phase 1 negotiation; initially, determining whether or not both sides support NAT-Traversal. Then, if both sides do support it, there should be a probe sequence that results in knowledge about whether or not the network between hosts contains a NAT device. As there is a need for two exchanges, of two messages each, it is obvious that NAT-Traversal cannot be supported in P1 modes with less than four messages sent across. Therefore, Aggressive Mode cannot be used with this NAT-Traversal approach. IKE Main Mode exchange contains 6 messages and therefore the probing sequence can be done within it. 3.1.1 Determining of support The NAT-Traversal capability of the remote host is determined by an exchange of vendor strings; in Main Mode's four first messages, the vendor id for this specification of NAT-Traversal ("draft-stenberg-ipsec-nat-traversal-00") MUST be sent if supported Stenberg, et. al. Expires January 12, 2001 [Page 7] Internet-Draft IPsec NAT-Traversal July 2000 (and it MUST be received by remote side) for the NAT-Traversal probe to continue. 3.1.2 NAT-Traversal need-probe Once the NAT-Traversal support of both parties has been determined, in the last two encrypted messages of Main Mode, there are additional private payloads sent in both directions. Initially, in the 5th message of Main Mode, the initiator will add one private payload to the message. PAYLOAD_TYPE (from private range) is 211. The payload should contain the following: {perceived remote identity - IP address and port} {one or more local identities - local interface address+port numbers} Figure 2: Probe payload in Main Mode message #5 The probe payload is encoded as a series of Identification Payloads of [3], with the perceived remote identity as the first payload, and the local identities as the following payloads. Once the IKE responder receives the payload represented in Figure 2, the remote should check whether or not the remote identity, as perceived by the IKE initiator, matches one of the locally configured interface addresses (with proper port number). Also, the remote identity as perceived by the IKE responder should match one of the address+port pairs sent in the packet. If one (or two) of those tests fails, the responder knows that NAT-Traversal is needed. The decision about whether to use NAT-Traversal or not is left up to the responder, and the responder transmits the decision as a private payload of type 211 in the last message of Main Mode. The payload is just one byte long, and contains 0 when NAT-Traversal is not elected to be used, and 1 when NAT-Traversal is chosen. 3.2 IPsec SA traffic encapsulation Automatic use of NAT-Traversal encapsulation for IKE-negotiated IPsec SAs MUST NOT be done. Instead, NAT-Traversal MUST be used only when IKE negotiation has resulted in a decision to use NAT-Traversal, or when manually keyed IPsec SAs are configured to use it. Traffic that is not in AH or ESP format MUST NOT be encapsulated Stenberg, et. al. Expires January 12, 2001 [Page 8] Internet-Draft IPsec NAT-Traversal July 2000 using this scheme, as that provides a way to create DDoS attacks, and possibly some other security problems as well. Normal AH/ESP traffic does not pass through NATs unmodified; typically, the addresses may change (src/dst), and that makes the resulting AH/ESP packet unusable. Thus, there has to be enough redundant data to always be able to recreate a packet to its original form. Additionally, it should preferably follow the same NAT route as IKE packets, to make the implementation simpler. Therefore (as noted before), the traffic has to be encapsulated as UDP packets between two hosts (which implies that they follow same route even in host/port NATs) using the IKE port. The basic idea behind this NAT-Traversal data encapsulation format is that it should be a format that can be adapted to future needs; therefore, the only requirement for this initial version is that it contains a version number, and it is invalid for IKE purposes. An IPsec NAT-Traversal envelope for IPv4 packet encapsulation looks like this: 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 +---------------+---------------+---------------+---------------+ | NAT-T version | IP4 hlen | IP4 ToS | IP4 protocol | +---------------+---------------+---------------+---------------+ | IP4 ID | IP4 frag_offset | +---------------+---------------+---------------+---------------+ | IP4 src | +---------------+---------------+---------------+---------------+ | | +---------------+---------------+---------------+---------------+ | | 0 (IKE ver.) | +---------------+---------------+ Figure 3: A NAT-Traversal envelope for an IPv4 IPsec packet. The variables with the IP4 prefix are the original values that are normally sent to the network as the header for the IPsec ESP data. Stenberg, et. al. Expires January 12, 2001 [Page 9] Internet-Draft IPsec NAT-Traversal July 2000 The header leftovers, if any, are the difference between the new packet's IP4 header length and the original packet's IP4 header length. The IP4 dst is not stored, as the remote host should be able to know which address it is using to communicate with each host (in the host-to-host case, with the responder behind a host NAT, the only recourse is manual configuration data). The NAT-T version field specifies the encapsulation specified; defined types are as follows (and undefined types will be dropped silently): Name | Value ------------------------------- TYPE_IPV4_ENCAPSULATION | 0x01 TYPE_HEARTBEAT | 0x02 The IPv4 encapsulation is defined in this section, and the heartbeat is specified in Section 3.3. Encapsulation occurs after IPsec processing, and it copies all variables as-is from the original packet. Decapsulation occurs before IPsec processing, and it copies values from the envelope to the real packet and discards the envelope. In some cases, it may involve changing of dst to be the host NAT address (if the remote side negotiated with the host NAT address, not the real configured address, and thus the host:spi pair is that of the host NAT address:spi). 3.3 Heartbeat Disclaimer: the IKE SA heartbeat should probably be used whenever one becomes a standard. Until then, the NAT-Traversal will have its own heartbeat that is entirely separate from the IKE SA and is used between two hosts. The sole purpose of the heartbeat is to keep the NATs in the network route between hosts from removing the mapping from their dynamic configuration (if any). Therefore, the actual contents of the heartbeat can be more or less ignored (unless they stop arriving), and thus encrypting them would serve no useful purpose. Heartbeats MUST be sent as long as there is at least one IKE-probed IPsec SA in existence between two hosts that employ NAT-Traversal to communicate with each other. The IKE initiator MUST send a heartbeat packet every HEARTBEAT_INTERVAL (=10) seconds. The IKE responder SHOULD reply to it. There MUST NOT be replies to address+port pairs with no IPsec SAs up, or when there are too many heartbeat packets going on (i.e., Stenberg, et. al. Expires January 12, 2001 [Page 10] Internet-Draft IPsec NAT-Traversal July 2000 there should be one reply [at most] in HEARTBEAT_INTERVAL/2 seconds). HEARTBEAT_INTERVAL MAY be higher, but as it is not negotiated, both sides must be configured for a higher HEARTBEAT_INTERVAL independently. The heartbeat sequence in practice works as follows, using the heartbeat format defined in Section 3.3.1: Initiator Responder -> Heartbeat (flag = FLAG_HEARTBEAT_PING) <- Heartbeat (flag = FLAG_HEARTBEAT_REPLY) 3.3.1 Heartbeat format The heartbeat packet format is very simple; it uses the same kind of pseudo-IKE encapsulation as the previously defined IP4 envelope, but with less fields in use. 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 +---------------+---------------+---------------+---------------+ | NAT-T version | HB flag | | +---------------+---------------+---------------+---------------+ | | +---------------+---------------+---------------+---------------+ | | +---------------+---------------+---------------+---------------+ | | +---------------+---------------+---------------+---------------+ | | 0 (IKE ver.) | +---------------+---------------+ Figure 5: Heartbeat format. Name | Value ---------------------------- FLAG_HEARTBEAT_PING | 0x01 FLAG_HEARTBEAT_REPLY | 0x02 Other values SHOULD be silently ignored. 3.4 Built-in NAT Built-in host NAT implementation within the IPsec stack is necessary in some tunnel cases and all transport cases. To stay consistent with [2], which specifies that both tunnel and transport mode MUST be supported, we define that there MUST be a built-in host NAT implementation for NAT-Traversal use. The built-in NAT is needed in some cases where ISSUE2 surfaces (see Stenberg, et. al. Expires January 12, 2001 [Page 11] Internet-Draft IPsec NAT-Traversal July 2000 Section 2.2 for details) to make the remote host(s) unique. Typically, the host mapping should be from (perceived_remote_host, perceived_remote_port) to some internal A- or B-class network. Whenever the remote side successfully initiates IPsec SA employing NAT-Traversal, there should be an internal NAT definition for the (remote_host, remote_port) if one is required according to the local configuration (or if transport mode is used, in which case internal host NAT SHOULD always be employed). Whenever IPsec processing for an incoming packet is done, the internal host NAT should be done to the src. Whenever an outgoing packet headed towards an internal NAT address enters the IPsec, the internal NAT address should be changed to the address that was used for negotiating the IPsec SA. In tunnel mode, it is possible that whole networks may need masking. In the NAT-Traversal+IPsec case, a separate NAT box would not be able to know about the (perceived_remote_host, perceived_remote_port) pair which provides uniqueness to the tunneled IP addresses. Therefore, there is a need for NAT within the IPsec implementation. This MAY be supported, but no details about implementation details will be provided here. Stenberg, et. al. Expires January 12, 2001 [Page 12] Internet-Draft IPsec NAT-Traversal July 2000 4. Known issues with the solution 4.1 Conceptual issues Most of the solution is solid. An IPv6 and IPv4-IPv6 interoperability specification will be added to the next draft version. However, the non-unique hosts may cause problems, as there is a potential problem of (host-port-proto-spi) not being unique any more. The problem does not surface in the incoming traffic, but it may occur in the outgoing case. There are (at least) a couple of different solutions to the problem: o Tying the remote-host,remote-port of NAT-T IPsec SA decapsulation and the (host-port-proto-spi). o Refusal of duplicate IPsec SA SPIs during IKE P2QM negotiation. 4.2 Overhead Overall, this solution is almost the most minimal one possible that covers most of the eventual possibilities and does not become overly complex. Different types of overhead caused by this draft are noted here, as well as possible ways of decreasing/removing the overheads involved. Processing time and memory overhead are ignored as negligible (some more processing for each packet, potentially logarithmic searches for free addresses, minimal extra data for each IPsec SA). o IKE P1 negotiation extra payloads: Moderately small, typically less than 200 bytes. Does not appear to be reducible. o Each IPv4-based IPsec SA packet will contain extra overhead of 8 (UDP header) + 17 (NAT-T header) = 25 bytes. This might be lowered slightly by dropping the IKE version field. Additionally, more intelligent probing about what fields are changed across the route would decrease the overhead as well, although at the cost of increasing complexity. Some overhead is inevitable. o Heartbeat overhead of 20 (IP header) + 8 (UDP header) + 17 (NAT-T header) = 45 bytes * 2 every HEARTBEAT_INTERVAL. 9 bytes/second may seem excessive, but as long as a general-purpose solution is desired, it cannot be bypassed unless the encapsulation changes from IKE-compliance (see previous entry). Heartbeat delay negotiation for slower-reaction dynamic NAT routes (and disabling when there are only static NATs) might be in order. This will be investigated further in future versions of this draft. Stenberg, et. al. Expires January 12, 2001 [Page 13] Internet-Draft IPsec NAT-Traversal July 2000 4.3 Security considerations Whenever changes to some fundamental parts of a security protocol are proposed, the examination of security implications cannot be skipped. Therefore, here are some observations regarding what is affected, and whether or not the effect matters. This section will be expanded further in future versions of this draft. o IKE probe reveals NAT-Traversal support to everyone. This should be a non-issue. o IPsec encapsulation which contains source address before NAT is a security leak of sorts, if internal network characteristics are desired to be kept hidden. o Obviously, the value of authentication mechanisms based on IP addresses gets near zero once NATs are in the picture. That is not necessarily a bad thing; for any real security, other authentication measures than IP addresses should be used in any case. o Some DoS implications exist; a single malicious user can possibly allocate up to (number-of-hosts-available) * 65535 (=number-of-ports-on-host) internal host IP addresses at the same time - and cause that many negotiations as well (this is 65535 times as much DoS potential as traditional IKE). As the IP addresses are allocated only after authentication is successful, the culprit is known, however, and therefore this can be considered a slight risk at best. o The encapsulation scheme prevents some control of IP-level headers. Therefore, there is some potential for forging of some fields of the ESP transport mode IP header. Because the decapsulated destination+spi MUST remain unchanged, there are no apparent security risks, unless the remote end's IP-level handling has some exploitable bugs (and even then only in implementation approach B of Appendix A of [1]). AH+ESP should be employed if IP-level header integrity is desired, as usual. Therefore, this is a non-issue; although firewall administration loses some control over IP headers that are passed through, use of flawed IP protocol implementations is in itself a bad idea. 4.4 Intellectual property rights SSH Communications Security Corp has patent applications which may cover parts of this technology. If this technology starts to progress on the IETF standards track, SSH is willing to seek a licensing solution that allows widespread use of this technology. Stenberg, et. al. Expires January 12, 2001 [Page 14] Internet-Draft IPsec NAT-Traversal July 2000 References [1] Kent, S. and R. Atkinson, "Security Architecture for the Internet Protocol", RFC 2401, November 1998. [2] Harkins, D. and D. Carrel, "The Internet Key Exchange (IKE)", RFC 2409, November 1998. [3] Piper, D., "The Internet IP Security Domain of Interpretation for ISAKMP", RFC 2407, November 1998. Authors' Addresses Markus Stenberg SSH Communications Security Corp Fredrikinkatu 42 FIN-00100 Helsinki Finland EMail: mstenber@ssh.com Santeri Paavolainen SSH Communications Security Corp Fredrikinkatu 42 FIN-00100 Helsinki Finland EMail: santtu@ssh.com Tatu Ylonen SSH Communications Security Corp Fredrikinkatu 42 FIN-00100 Helsinki Finland EMail: ylo@ssh.com Tero Kivinen SSH Communications Security Corp Fredrikinkatu 42 FIN-00100 Helsinki Finland EMail: kivinen@ssh.com Stenberg, et. al. Expires January 12, 2001 [Page 15] Internet-Draft IPsec NAT-Traversal July 2000 Full Copyright Statement Copyright (C) The Internet Society (2000). All Rights Reserved. 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. Acknowledgement Funding for the RFC editor function is currently provided by the Internet Society. Stenberg, et. al. Expires January 12, 2001 [Page 16]