NSIS Working Group M. Stiemerling Internet-Draft NEC Expires: April 17, 2004 H. Tschofenig Siemens M. Martin NEC C. Aoun Nortel Networks October 18, 2003 A NAT/Firewall NSIS Signaling Layer Protocol (NSLP) draft-ietf-nsis-nslp-natfw-00 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 April 17, 2004. Copyright Notice Copyright (C) The Internet Society (2003). All Rights Reserved. Abstract This draft describes scenarios, problems and solutions for path-coupled Network Address Translator and Firewall signaling. This is one of the two NSIS Signaling Layer Protocols (NSLPs) the working group will address during its work. Stiemerling, et al. Expires April 17, 2004 [Page 1] Internet-Draft A NAT/FW NSIS NSLP October 2003 Table of Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . 4 2. Terminology and Abbreviations . . . . . . . . . . . . . . 5 3. Framework and Scenarios . . . . . . . . . . . . . . . . . 7 3.1 Scope . . . . . . . . . . . . . . . . . . . . . . . . . . 7 3.2 What problem should be solved? . . . . . . . . . . . . . . 7 3.3 Basic NSIS Usage for NAT/FW traversal . . . . . . . . . . 8 3.4 Scenarios for Protocol Functionality . . . . . . . . . . . 9 3.4.1 Firewall traversal . . . . . . . . . . . . . . . . . . . . 9 3.4.2 NAT with two private networks . . . . . . . . . . . . . . 10 3.4.3 NAT with private network on sender side . . . . . . . . . 11 3.4.4 NAT with private network on receiver side . . . . . . . . 11 3.4.5 Both end hosts in same private network behind NATs . . . . 12 3.4.5.1 Both end hosts in same private network behind same NAT . . 13 3.4.6 IPv4/v6 NAT with two private networks . . . . . . . . . . 13 3.5 Trust Relationship and Authorization . . . . . . . . . . . 14 3.5.1 Peer-to-Peer Trust Relationship . . . . . . . . . . . . . 14 3.5.2 Intra-Domain Trust Relationship . . . . . . . . . . . . . 15 3.5.3 End-to-Middle Trust Relationship . . . . . . . . . . . . . 16 4. Problems and Challenges . . . . . . . . . . . . . . . . . 18 4.1 Missing Network-to-Network Trust Relationship . . . . . . 18 4.2 End-to-end significance . . . . . . . . . . . . . . . . . 19 4.3 Relationship with routing . . . . . . . . . . . . . . . . 19 4.4 Dynamic state installation and maintenance . . . . . . . . 20 4.5 Affected Parts of the Network . . . . . . . . . . . . . . 20 4.6 NSIS backward compatibility with NSIS unaware NAT and Firewalls . . . . . . . . . . . . . . . . . . . . . . . . 20 4.7 Authentication and Authorization . . . . . . . . . . . . . 21 4.8 Directional Properties . . . . . . . . . . . . . . . . . . 21 4.9 Routing Asymmetry . . . . . . . . . . . . . . . . . . . . 22 4.10 Addressing . . . . . . . . . . . . . . . . . . . . . . . . 22 4.11 NTLP/NSLP NAT Support . . . . . . . . . . . . . . . . . . 23 4.12 Route changes . . . . . . . . . . . . . . . . . . . . . . 23 4.13 Combining Middlebox and QoS signaling . . . . . . . . . . 23 4.14 Difference between sender- and receiver-initiated signaling . . . . . . . . . . . . . . . . . . . . . . . . 24 4.15 Inability to know the scenario . . . . . . . . . . . . . . 24 5. NSIS NAT Handling Solution . . . . . . . . . . . . . . . . 25 5.1 Problem Description . . . . . . . . . . . . . . . . . . . 25 5.2 Solution Overview . . . . . . . . . . . . . . . . . . . . 28 5.2.1 Destination IP address Selection . . . . . . . . . . . . . 30 6. Protocol Description . . . . . . . . . . . . . . . . . . . 32 Stiemerling, et al. Expires April 17, 2004 [Page 2] Internet-Draft A NAT/FW NSIS NSLP October 2003 6.1 Basic protocol overview . . . . . . . . . . . . . . . . . 32 6.2 Message format and types . . . . . . . . . . . . . . . . . 32 6.3 Message flow . . . . . . . . . . . . . . . . . . . . . . . 35 6.4 Middle box configuration . . . . . . . . . . . . . . . . . 36 6.5 Message handling . . . . . . . . . . . . . . . . . . . . . 37 6.5.1 Detailed behaviour . . . . . . . . . . . . . . . . . . . . 38 6.5.1.1 Reserve external address message . . . . . . . . . . . . . 38 6.5.1.2 Return external address message . . . . . . . . . . . . . 39 6.5.1.3 Create message . . . . . . . . . . . . . . . . . . . . . . 39 6.5.1.4 Delete session message . . . . . . . . . . . . . . . . . . 41 6.5.1.5 Prolong session message . . . . . . . . . . . . . . . . . 41 6.5.1.6 Path Succeded message . . . . . . . . . . . . . . . . . . 41 7. Solution examples . . . . . . . . . . . . . . . . . . . . 43 7.1 Firewall traversal . . . . . . . . . . . . . . . . . . . . 43 7.2 NAT with private network on sender side . . . . . . . . . 43 7.3 NAT with private network on receiver side . . . . . . . . 45 7.4 Both end hosts are in same private network behind NATs . . 49 7.5 IPv4/v6 NAT with two private networks . . . . . . . . . . 52 7.6 Full example for NAT/FW with two private networks . . . . 52 8. NSIS NAT and Firewall transitions issues . . . . . . . . . 59 9. Security Considerations . . . . . . . . . . . . . . . . . 60 10. Open Issues . . . . . . . . . . . . . . . . . . . . . . . 62 11. Contributors . . . . . . . . . . . . . . . . . . . . . . . 63 Normative References . . . . . . . . . . . . . . . . . . . 64 Informative References . . . . . . . . . . . . . . . . . . 65 Authors' Addresses . . . . . . . . . . . . . . . . . . . . 66 A. Interworking of SIP with NSIS NATFW NSLP . . . . . . . . . 68 A.1 The Session Initiation Protocol . . . . . . . . . . . . . 68 A.2 Conclusions . . . . . . . . . . . . . . . . . . . . . . . 73 B. Ad-Hoc networks . . . . . . . . . . . . . . . . . . . . . 74 C. Interworking of Security Mechanisms and NSIS NATFW NSLP . 75 D. Solution approaches in case of missing authorization . . . 76 D.1 Solution Approach: Local authorization from both end points . . . . . . . . . . . . . . . . . . . . . . . . . . 76 D.2 Solution Approach: Access Network-Only Signaling . . . . . 77 D.3 Solution Approach: Authorization Tokens . . . . . . . . . 77 Stiemerling, et al. Expires April 17, 2004 [Page 3] Internet-Draft A NAT/FW NSIS NSLP October 2003 E. Acknowledgements . . . . . . . . . . . . . . . . . . . . . 80 Intellectual Property and Copyright Statements . . . . . . 81 Stiemerling, et al. Expires April 17, 2004 [Page 4] Internet-Draft A NAT/FW NSIS NSLP October 2003 1. Introduction Even though the NSIS WG (Next Steps in Signaling) has QoS signaling as a primary application, other types of applications should be possible. In this draft, we address the scenario, framework, problems, and issues when signaling Network Address Translators (NAT) and/or Firewalls to allow packet traversal. One of the requirements in NSIS [1] is that the NTLP signaling protocol must be independent of the service requested. In some cases the NSIS protocol suite could be used to request end-to-end or edge-to-edge QoS from IP networks; in other cases the service might be allow packet traversal from one host to the other through all the NATs and firewalls deployed on the data path. See also [17] and [18] for proposals to use RSVP or CASP for NAT and Firewall traversal. Stiemerling, et al. Expires April 17, 2004 [Page 5] Internet-Draft A NAT/FW NSIS NSLP October 2003 2. Terminology and Abbreviations This document uses terms defined in [22]. In addition the following terms are used: o NSIS NAT Forwarding State: The term "NSIS NAT Forwarding State" in this context refers to a state used to forward the NSIS signaling message beyond the targeted destination address; that state is typically used when the NSIS Responder address is not known o Sender-/Receiver Initiated Signaling Sender-initiated: NAT bindings and firewall rules are created immediately when the "path" message hits the NSIS nodes. With "path" message we refer to the signaling message traveling from the data sender towards the data receiver. Receiver-initiated: NAT bindings and firewall rules are created when the "resv" message returns from the other end. With "resv" message we refer to a signaling message on the reverse path, this means from the receiver to the sender (i.e. backwards routed). Note that these definitions have nothing to do with number of roundtrips, who performs authorization etc. o Firewalls vs. Security Gateway: As discussed in Section 3.1 different types of firewalls exist. This document focuses on firewalls, which perform packet filtering, and possibly application level filtering and does not address IPsec based security gateways. o NSIS Initiator (NI): the signaling entity which makes the resource request, usually as a result of user application request. o NSIS Responder (NR): the signaling entity that acts as the endpoint for the signaling and can optionally interact with applications as well. o NSIS Forwarder (NF): the signaling entity between an NI and NR which propagates NSIS signaling further through the network. o Receiver (DR or R): the node in the network which is receiving the data packets in a flow. o Sender (DS or S): the node in the network which is sending the data packets in a flow. Stiemerling, et al. Expires April 17, 2004 [Page 6] Internet-Draft A NAT/FW NSIS NSLP October 2003 o NATFW NSLP session: an artifact that groups the installed states for a given data flow. o Middlebox: from [21]: "A middlebox is defined as any intermediate device performing functions other than the normal, standard functions of an IP router on the datagram path between a source host and a destination host". The term middlebox in context of this document and in NSIS refers to firewalls and NATs only. Other types of middlebox are currently outside the scope. Stiemerling, et al. Expires April 17, 2004 [Page 7] Internet-Draft A NAT/FW NSIS NSLP October 2003 3. Framework and Scenarios 3.1 Scope The term firewall and middlebox in general raises different expectations about the functionality provided by such a device. Different groups have worked on the problem of securing access to a network by using different procedures and protocols. From an abstract point of view two different mechanisms for restricting access to a network can be differentiated: o Packet Filters o Cryptographically protected data traffic Within this document we assume that packet filters are installed at devices along the path. These packet filters may consist of a 5 tuple (src/dst ip address, transport protocol, src/dst port). Some devices entitled as firewalls only accept traffic after cryptographic verification (i.e. IPsec protected data traffic). Particularly for network access scenarios either link layer or network layer data protection is common. Hence we do not address these types of devices (referred as security gateways) since per-flow signaling is rather uncommon in this environment. For a discussion of network access authentication and associated scenarios the reader is referred to the PANA working group (see [16]). In mobility scenarios an often experienced problem is the traversal of a security gateway at the edge of the corporate network. Network administrators often rely on the policy that only authenticated data traffic is allowed to enter the network. A problem statement for the traversal of these security gateways in the context of Mobile IP can be found at [15]). 3.2 What problem should be solved? The goal of NSIS FW/NAT signaling therefore focuses on packet filter installation combined with associated actions to be enforced on packets matching the filter expression. In the case of NATs and Firewalls the associated actions are packet forwarding and address/ port translation in packet headers. Discovering security gateways, which was also mentioned as an application for NSIS signaling, for the purpose of executing an IKE to create an IPsec SA, is already solved without requiring NSIS. Installing packet filters provides some security but has some weaknesses, which heavily depend on the type of packet filter Stiemerling, et al. Expires April 17, 2004 [Page 8] Internet-Draft A NAT/FW NSIS NSLP October 2003 installed. A packet filter cannot prevent an adversary to inject traffic (due to the IP spoofing capabilities). This type of attack might not be particular helpful if the packet filter is a standard 5 tuple which is very restrictive. If packet filter installation, however, allows specifying a rule, which restricts only the source IP address, then IP spoofing allows transmitting traffic to an arbitrary address. NSIS aims to provide path-coupled signaling and therefore an adversary is somewhat restricted in the location from which attacks can be performed. Some trust is therefore assumed from nodes and networks along the path. 3.3 Basic NSIS Usage for NAT/FW traversal The basic high-level picture of NSIS usage is that of the endhosts signaling to establish packet filters and NAT bindings on a data path, which allows the said data to travel from the endhost to the endpoint unobstructed. Therefore it is necessary that each firewall and each NAT involved in the signaling communication runs an NSIS daemon. There might be several NATs and FWs in various possible combinations on a path between two hosts. The reader is referred to Section 3.4 where different scenarios are presented. Stiemerling, et al. Expires April 17, 2004 [Page 9] Internet-Draft A NAT/FW NSIS NSLP October 2003 Application Application Server (0, 1, or more) Application +----+ +----+ +----+ | +------------------------+ +------------------------+ | +-+--+ +----+ +-+--+ | | | NSIS Agents | +-+--+ +----+ +-----+ +-+--+ | +--------+ +----------------------------+ +-----+ | +-+--+ +-+--+ +--+--+ +-+--+ | | ------ | | | | //// \\\\\ | | +-+--+ +-+--+ |/ | +-+--+ +-+--+ | | | | | Internet | | | | | | +--------+ +-----+ +----+ +-----+ | +----+ +----+ |\ | +----+ +----+ \\\\ ///// sender NAT/FW (1+) ------ NATFW (1+) receiver Figure 1: Generic View on NSIS in a NAT / Firewall case 3.4 Scenarios for Protocol Functionality This section introduces several scenarios for middleboxes in the Internet. These middleboxes are firewalls or different flavours of NATs [5], like NAPT. Combinations of Firewalls and NATs in the same device are possible as well. Each section introduces a different scenario for a different set of middleboxes and their ordering within the topology. 3.4.1 Firewall traversal The following scenario shows two end hosts behind a firewall but connected via the public Internet. The application can somehow trigger firewall traversal (e.g. via an API call) at the NSIS agent at the local host. The NSIS agent then signals this request to the next NSIS aware node and therefore to the receiver. Each firewall in the middle must provide traversal service in order to permit the NSIS message to reach the other end host. The difference between this scenario and the following is that firewalls are on the path, but no NATs. This has specific implication concerning the used destination address for path-coupled signaling message sent by the NSIS Initiator to an NR hosted behind a NAT. Stiemerling, et al. Expires April 17, 2004 [Page 10] Internet-Draft A NAT/FW NSIS NSLP October 2003 +----+ //----\\ +----+ S -----| FW |---| |------| FW |--- R +----+ \\----// +----+ private public private FW: Firewall S: Data Sender R: Data Receiver Figure 2: Firewall Traversal Scenario 3.4.2 NAT with two private networks This scenario deals with NATs on both ends of the network. Therefore, each application instance is behind a NAT and is connected to the public Internet (see Figure 3 ). The case where more than one MB on each side ("only" two are shown in the figure) is present must be taken into account. This aspect introduces more topology problems. +----+ +----+ //----\\ +----+ +----+ S --| MB |-----| MB |---| |---| MB |-----| MB |--- R +----+ +----+ \\----// +----+ +----+ private public private MB: Middlebox S: Data Sender R: Data Receiver Figure 3: NAT with two private networks Scenario Data traffic from the sender to the receiver has to traverse all four middleboxes on the path and all four middleboxes must be configured properly to allow subsequent data packets to flow. The sender has to know the IP address of the receiver in advance, i.e. before any NSIS message can be sent. Or more general the NSIS Initiator must know the IP addresses of the NSIS Responder, otherwise he cannot send a single NSIS signaling message towards the responder. Note that this IP address is not the private IP address of the responder. Instead a NAT binding (including a public IP address) has to be obtained from Stiemerling, et al. Expires April 17, 2004 [Page 11] Internet-Draft A NAT/FW NSIS NSLP October 2003 the NAT which subsequently allows packets hitting the NAT to be forwarded to the receiver within the private address realm. This in general requires further support from an application layer protocol for the purpose of discovering and exchanging information. The receiver might have a number of ways to learn its public IP address and port number and might need to signal this information to the sender using the application level signaling protocol. 3.4.3 NAT with private network on sender side This scenario shows an application instance at the sending node which is behind one ore more FW/NATs. The receiver is located in the public internet. +----+ +----+ //----\\ S --| MB |-----| MB |---| |--- R +----+ +----+ \\----// private public MB: Middlebox S: Data Sender R: Data Receiver Figure 4: NAT with private network on sender scenario The traffic from the sender to the receiver has to traverse only middleboxes on the sender's side. The receiver has a public IP address and therefore the procedure is simple. The sender sends its signaling message directly to the receiver whereby it is intercepted by the middleboxes along the path. Note that the data sender does not necessarily know whether the receiver is behind a NAT or not, hence, it is the receiving side that has to detect the whether it is behind a NAT or not. As described in Section 5 NSIS can also provide help for this procedure. 3.4.4 NAT with private network on receiver side The application instance receiving data is behind one or more NATs. Stiemerling, et al. Expires April 17, 2004 [Page 12] Internet-Draft A NAT/FW NSIS NSLP October 2003 //----\\ +----+ +----+ S ---| |---| MB |-----| MB |--- R \\----// +----+ +----+ public private MB: Middlebox S: Data Sender R: Data Receiver Figure 5: NAT with private network on receiver Scenario First, the sender must determine the public IP address of the receiver. One possibility is that an application level protocol is used. In this case, the receiver must first find out its public IP addresses at the middlebox on its side. This information about IP address and port numbers could be signalled somehow to the actual sender directly or indirectly via a third party. In the scenario, this means the receiver has to determine its public IP address (NAT binding) and register this address with the third party. The sender can start NSIS signaling after he has received information about the receiver's address and port number. Note: The solution design will determine where to discontinue forwarding the signaling messages. 3.4.5 Both end hosts in same private network behind NATs This is a special case, where the main problem is to detect that both nodes are within the same network behind a NAT. This scenario primarily addresses performance aspects. Sender and receiver are both within a private address realm and potentially have overlapped addresses. Figure 6 shows the ordering of NATs. This is a common configuration in several networks, particularly after the merging of companies that have overlapped addresses. Stiemerling, et al. Expires April 17, 2004 [Page 13] Internet-Draft A NAT/FW NSIS NSLP October 2003 public +----+ +----+ //----\\ S --| MB |--+--| MB |---| | +----+ | +----+ \\----// | | +----+ +--| MB |------------ R +----+ private MB: Middlebox S: Data Sender R: Data Receiver Figure 6: NAT to public, receiver in same private network Scenario The middleboxes are twice-NATs, i.e. they map the IP addresses and port numbers on both sides, private and public interface. From a protocol point of view, this means that the protocol must be robust enough to at least not break with this scenario. In the worst case, both sender and receiver obtain a public IP address at the NAT and the communication path is not optimal anymore. 3.4.5.1 Both end hosts in same private network behind same NAT A slight variation of the previous scenario consists in having both the DS and the DR behind the same NAT. In case the data sender and data receiver can not communicate which address is the most suitable among the several available (local scoped or globally scoped) addresses to be used for the NSIS messages, the protocol should be robust enough to handle the worst case, where the messages are sent through the NAT. 3.4.6 IPv4/v6 NAT with two private networks This scenario combines the usage case mentioned in Section 3.4.2 with the IPv4 to IPv6 transition scenario, i.e. using Network Address and Protocol Translators (NAT-PT). The difference to the other scenarios lies in the use of IPv6 - IPv4 address translation, which happens in both directions. Additionally, the base NTLP must take care of this case for its own functionality of forwarding messages between NSIS peers. Stiemerling, et al. Expires April 17, 2004 [Page 14] Internet-Draft A NAT/FW NSIS NSLP October 2003 +----+ +----+ //----\\ +----+ //----\\ +----+ +----+ S --| MB |--| MB |--| |--| MB |-| |--| MB |--| MB |-- R +----+ +----+ \\----// +----+ \\----// +----+ +----+ private public public private IPv4 IPv6 MB: Middlebox S: Data Sender R: Data Receiver Figure 7: IPv4/v6 NAT with two private networks 3.5 Trust Relationship and Authorization Trust relationships and authorization are very important for the protocol machinery. Trust and authorization closely related to each other in the sense that a certain degree of trust is required to authorize a particular action. If the action is "create/delete packet filters" then authorization is very important due to the nature of a firewall. It is not particularly surprising that differences exist between authorization in a QoS signaling environment and firewall signaling. As elaborated in [13] the establishment of a financial relationship is very important for QoS, signaling whereas for firewall signaling is not directly of interest. In the subsequent sections different trust relationships will be described which appear in firewall signaling environments. Peer-to-peer trust relationships are those, which are used in QoS signaling today and seem to be the simplest. However, there are reasons to believe that this is not the only type of trust relationship found in today's networks. 3.5.1 Peer-to-Peer Trust Relationship Starting with the simplest scenario it is assumed that neighboring nodes trust each other. The required security association to authenticate and to protect a signaling message is either available (manual configuration) or dynamically established with the help of an authentication and key exchange protocol. If nodes are located closely together it is assumed that security association establishment is easier than establishing it between far distant node. It is, however, difficult to describe this relationship generally due to the different usage scenarios and environments. Stiemerling, et al. Expires April 17, 2004 [Page 15] Internet-Draft A NAT/FW NSIS NSLP October 2003 Authorization heavily depends on the participating entities but for this scenario it is assumed that neighboring entities trust each other (at least for the purpose of packet filter creation and deletion). Note that Figure 8 does not illustrate the trust relationship between the end host and the access network, which is dynamically established as part of the network access authentication procedure as stated in Section 1. +------------------------+ +-------------------------+ | | | | | Network A | | Network B | | | | | | +---------+ +---------+ | | +-///-+ Middle- +---///////----+ Middle- +-///-+ | | | | box 1 | Trust | box 2 | | | | | +---------+ Relationship +---------+ | | | | | | | | | | | | | | | | | | | | | | Trust | | Trust | | | | Relationship | | Relationship | | | | | | | | | | | | | | | | | | | | | +--+---+ | | +--+---+ | | | Host | | | | Host | | | | A | | | | B | | | +------+ | | +------+ | +------------------------+ +-------------------------+ Figure 8: Peer-to-Peer Trust Relationship 3.5.2 Intra-Domain Trust Relationship In larger corporations often more than one firewall is used to protect different departments. In many cases the entire enterprise is controlled by a security department, which gives instructions to the department administrators. In such a scenario a peer-to-peer trust-relationship might be prevalent. Sometimes however it might be necessary to preserve authentication and authorization information within the network. As a possible solution a centralized approach could be used whereby an interaction between the individual middleboxes and a central entity (for example a policy decision point - PDP) takes place. As an alternative individual firewalls could exchange the authorization decision to another firewalls within the Stiemerling, et al. Expires April 17, 2004 [Page 16] Internet-Draft A NAT/FW NSIS NSLP October 2003 same trust domain. Individual middleboxes within an administrative domain should exploit their trust relationship instead of requesting authentication and authorization of the signaling initiator again and again. Thereby complex protocol interaction is avoided. This provides both a performance improvement without a security disadvantage since a single administrative domain can be seen as a single entity. Figure 9 illustrates a network structure which uses a centralized entity. +-----------------------------------------------------------+ | | | Network A | | | | | | +---------+ +---------+ | +----///--------+ Middle- +------///------++ Middle- +--- | | | box 2 | | box 2 | | | +----+----+ +----+----+ | | | | | | +----+----+ | | | | | Middle- +--------+ +---------+ | | | | box 1 | | | | | | +----+----+ | | | | | | | | | | | - | | | | | - | +----+-----+ | | | | | | Policy | | | | +--+---+ +-----------+ Decision +----------+ | | | Host | | Point | | | | A | +----------+ | | +------+ | +-----------------------------------------------------------+ Figure 9: Intra-domain Trust Relationship 3.5.3 End-to-Middle Trust Relationship In some scenarios a simple peer-to-peer trust relationship between participating nodes is not sufficient. Network B might require additional authorization of the signaling message initiator. If authentication and authorization information is not attached to the initial signaling message then the signaling message arriving at Middlebox 2 would cause an error message to be created, which indicates the additional authorization requirement. In many cases the signaling message initiator is already aware of the additionally Stiemerling, et al. Expires April 17, 2004 [Page 17] Internet-Draft A NAT/FW NSIS NSLP October 2003 required authorization before the signaling message exchange is executed. Replay protection is a requirement for authentication to the non-neighboring firewall which might be difficult to accomplish without adding additional roundtrips to the signaling protocol (e.g. by adding a challenge/response type of message exchange). Figure 10 shows the slightly more complex trust relationships in this scenario. +----------------------+ +--------------------------+ | | | | | Network A | | Network B | | | | | | | Trust | | | | Relationship | | | +---------+ +---------+ | | +-///-+ Middle- +---///////----+ Middle- +-///-+ | | | | box 1 | +-------+ box 2 | | | | | +---------+ | +---------+ | | | | | | | | | | |Trust | | | | | | |Relationship | | | | | | | | | | Trust | | | | | | | Relationship| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | +--+---+ | | | +--+---+ | | | Host +----///----+------+ | | Host | | | | A | |Trust | | B | | | +------+ |Relationship | +------+ | +----------------------+ +--------------------------+ Figure 10: End-to-Middle Trust Relationship Stiemerling, et al. Expires April 17, 2004 [Page 18] Internet-Draft A NAT/FW NSIS NSLP October 2003 4. Problems and Challenges This section describes a number of problems which have to be addressed for NSIS NAT/Firewall. These are also of relevance for other NSLP protocols. 4.1 Missing Network-to-Network Trust Relationship Peer-to-peer trust relationship, as shown in Figure 8, is a very convenient assumption that allows simplified signaling message processing. However, it might not always be applicable, especially between two arbitrary access networks (over a core network where signaling messages are not interpreted) does possibly not exist because of the large number of networks and the unwillingness of administrators to have other network operators to create holes in their firewalls without proper authorization. Hence in the following scenario we assume a somewhat different message processing and show three possible approaches to tackle the problem. None of these three approaches is without drawbacks or constraining assumptions. Stiemerling, et al. Expires April 17, 2004 [Page 19] Internet-Draft A NAT/FW NSIS NSLP October 2003 +----------------------+ +--------------------------+ | | | | | Network A | | Network B | | | | | | +---------+ Missing +---------+ | | +-///-+ Middle- | Trust | Middle- +-///-+ | | | | box 1 | Relation- | box 2 | | | | | +---------+ ship +---------+ | | | | | or | | | | | | Authorization| | | | | | | | | | | Trust | | Trust | | | | Relationship | | Relationship | | | | | | | | | | | | | | | | | | | | | +--+---+ | | +--+---+ | | | Host | | | | Host | | | | A | | | | B | | | +------+ | | +------+ | +----------------------+ +--------------------------+ Figure 11: Missing Network-to-Network Trust Relationship Figure 11 illustrates a problem whereby an external node is not allowed to manipulate (create, delete, query, etc.) packet filters at a firewall. Opening pinholes is only allowed for internal nodes or with a certain authorization permission. Hence the solution alternatives in Section 5 focus on establishing the necessary trust with cooperation of internal nodes. 4.2 End-to-end significance In the case of NAT/firewalls traversal, the NSIS signaling messages need to be sent all they way from the DS and DR or viceversa. This is so because a middle box does not know wether the remaining path to the destination is clear of pottentially obstructing middleboxes or not. 4.3 Relationship with routing The data path is following the "normal" routes. The NAT/FW devices along the data path are those providing the service. In this case the service is something like "open a pinhole" or even more general "allow for connectivity between two communication partners". The benefit of using path-coupled signaling is that the NSIS NATFW NSLP does not need to determine what middleboxes or in what order the data Stiemerling, et al. Expires April 17, 2004 [Page 20] Internet-Draft A NAT/FW NSIS NSLP October 2003 flow will go through. Creating NAT bindings modifies routing of data packets between end points. This is unlike other NSIS NSLPs, which do not interfere with routing - instead they only follow the path of the data packets. 4.4 Dynamic state installation and maintenance For NAT/Firewall traversal, the lifetime of a NAT binding or a packet filter must be determined is often mantained through periodic refresh.For short-lived flows, having unpredictable filtersi, signaling for pinholes and NAT bindings is preferable as opposed to statically configured NAT bindings and pinholes requested for long duration in time. The capability to specify a lifetime for a NAT binding provides some advantages to what exists today where unknown NAT binding lifetimes can lead to unexpected protocol actions. For static state other mechanisms than an NSIS signaling protocol might be preferable; such mechanisms would include a management protocols such as SNMP or CLI. 4.5 Affected Parts of the Network NATs and Firewalls tend to be located at the edge of the network, whereby other signaling applications affect all nodes along the path. One typical example is QoS signaling where all networks along the path must provide QoS in order to achieve true end-to-end QoS. In the NAT/Firewall case, only some of the domains/nodes are affected (typically access networks), whereas most parts of the networks and nodes are unaffected (e.g. the core network). This fact raises some questions. Should an NSIS NTLP node intercept every signaling message independently of the upper layer signaling application or should it be possible to make the discovery procedure more intelligent to skip nodes. These questions are also related to the question whether NSIS NAT/FW should be combined with other NSIS signaling applications. 4.6 NSIS backward compatibility with NSIS unaware NAT and Firewalls Backward compatibility is key for NSIS deployments, as such the NSIS protocol suite should be sufficiently robust to allow traversal of none NSIS aware routers (Qos gates, Firewalls, NATs, etc ) NSIS NATFW NSLP's backward compatibility issues is different than the NSIS QoS NSLP backward compatibility issues, where an NSIS unaware Qos gate will simply forward the Qos NSLP message. An NSIS unaware firewall rejects NSIS messages, since firewalls typically implement Stiemerling, et al. Expires April 17, 2004 [Page 21] Internet-Draft A NAT/FW NSIS NSLP October 2003 the policy "default = deny". The NSIS backward compatibility support on none NSIS aware firewall would typically consist of configuring a static policy rule that allows the forwarding of the NSIS protocol messages (either protocol type if raw transport mode is used or transport port number in case a transport protocol is used). For NATs backward compatibility is more problematic since signaling messages are forwarded (at least in one direction), but with a changed IP address and changed port numbers. The content of the NSIS signaling message is, however, unchanged. This can lead to unexpected results, both due to embedded unchanged local scoped addresses and none NSIS aware firewalls configured with specific policy rules allowing forwarding of the NSIS protocol (case of transport protocols are used for the NTLP). NSIS unaware NATs must be detected to maintain a well known deterministic mode of operation for all the involved NSIS entities. Such a "legacy" NAT detection procedure can be done during the NSIS discover procedure itself. Based on experience it was discovered that routers unaware of the Router Alert IP option [RFC 2113] discarded packets,this is certainly a problem for NSIS signaling. 4.7 Authentication and Authorization Since a firewall has security functionality, strong authentication and authorization means MUST be provided. For NATs security is not a major concern, but might play a role in the perceived security measure of some administrators. For NAT sometimes address depletion is considered as a threat. 4.8 Directional Properties There two directional properties that need to be addressed by the NATFW NSLP: o Directionality of the data o Directionality of NSLP signaling Both properties are relevant to NATFW NSLP aware NATs and Firewalls. With regards to NSLP signaling directionality: As stated in the previous sections, the authentication and authorization of NSLP signaling messages received from hosts within the same trust domain (typically from hosts located within the security perimeter delimited Stiemerling, et al. Expires April 17, 2004 [Page 22] Internet-Draft A NAT/FW NSIS NSLP October 2003 by firewalls) is normally simpler than received messages sent by hosts located in different trust domains. The way NSIS signaling messages enters the NSIS agent of a firewall (see Figure 2) might be important, because different policies might apply for authentication and admission control. Hosts deployed within the secured network perimeter delimited by a Firewall, are protected from hosts deployed outside the secured network perimeter, hence by nature the firewall has more restrictions on flows triggered from hosts deployed outside the security perimeter. As opposed to traditional NAT [6] where the NAT bind was implicitly created when an outbound packet was sent; the NSIS NATFW NSLP aware NATs will be explicitly requested to create NAT binds. Hence data packet directionality is no longer significant for the bind creation, in the case of signaled NAT bind creations on NSIS aware NATs. However depending on the used NAT implementation, data directionality may still apply to implicitly maintain the NAT bind (since NAT binds consume memory and addressing resources); as such some NAT implementation might still require that outbound packet ONLY maintain (implicitly and not explicitly) binds. 4.9 Routing Asymmetry Routing asymmetry [14] is a general problem for path-coupled signaling, especially when installed states on NSIS forwarders are related to bi-directional flows. Path state, on an NSIS forwarder, including the next NSIS hop (for packets sent from the NR to NI), is used to handle routing asymmetry for NSIS messages but not for data flows (i.e. no route pinning for data flows). Similarly to path-coupled QoS signaling, firewall signaling also has to be aware of the routing asymmetry when bi-directional flows relevant states need to be installed on NSIS aware nodes, although the routing asymmetry might not be significant within the local networks where firewalls are typically located. For signaling NAT bindings this issue comes with a different flavor since an established NAT binding changes the path of the data packets. Hence a data receiver might still be able to send NSIS signaling messages to create a NAT binding, although they travel the previously "wrong" path. 4.10 Addressing Stiemerling, et al. Expires April 17, 2004 [Page 23] Internet-Draft A NAT/FW NSIS NSLP October 2003 A more general problem of NATs is the addressing of the end-point. NSIS signaling message have to be addressed to the other end host to follow data packets subsequently sent. Therefore a public IP address of the receiver has to be known prior to sending an nsis message. When NSIS signaling messages contain IP addresses of the sender and the receiver in the signaling message payloads, then an NSIS agent must modify them. This is one of the cases, where a NSIS aware NATs is also helpful for other types of signaling applications e.g. QoS signaling. 4.11 NTLP/NSLP NAT Support It must be possible for NSIS NATs along the path to change NTLP and/ or NSLP message payloads , which carry IP address and port information. This functionality includes the support of providing mid-session and mid-path modification of these payloads. As a consequence these payloads must not be reordered, integrity protected and/or encrypted in a non peer-to-peer fashion (e.g. end-to-middle, end-to-end protection). Ideally these mutable payloads must be marked (e.g. a protected flag) to assist NATs in their effort of adjusting these payloads. 4.12 Route changes The effect of route changes are more severe than in other signaling applications since a firewall pinhole and NAT binding is needed before further communication can takes place. This is true for both NSIS signaling and for subsequent data traffic. If a route changes and NSIS signaling messages do not configure NSIS NATs and firewalls along the new path then the communication is temporarily interrupted. This is naturally a big problem for networks where routes frequently change e.g. ad-hoc networks or in case of fast mobility. In these cases state refresh messages have to provide a mechanism for fast reaction. 4.13 Combining Middlebox and QoS signaling In many cases, middlebox and QoS signaling has to be combined at least logically. Hence it was suggested to combine them into a single signaling message or to tie them together with the help of some sort of data connection identifier, later on refered as Session ID. This, however, has some disadvantages such as: - NAT/FW NSLP signaling affects a much small number of NSIS nodes along the path (for example compared to the QoS signaling). - NAT/FW signaling might show different signaling patterns (e.g. required end-to-middle communication). Stiemerling, et al. Expires April 17, 2004 [Page 24] Internet-Draft A NAT/FW NSIS NSLP October 2003 - The refresh interval is likely to be different. - The number of error cases increase as different signaling applications are combined into a single message. The combination of error cases has to be considered. 4.14 Difference between sender- and receiver-initiated signaling For NAT/FW signaling there seems to be little difference between sender- and receiver- initiated signaling messages. Some other characteristics of QoS signaling protocols are not applicable (e.g. the adspec object) to the NAT/FW context. It seems that a full roundtrip is always required if the protocol aims to be generic enough. 4.15 Inability to know the scenario In Section 3.4 a number of different scenarios are presented. In some scenario NSIS signaling is fairly easy whereas in others it is quite complex. Additionally different trust relationships exist between networks along the path, which might require interaction with the end host or a different signaling behavior. However, the user (or the NSIS agent initially) typically does not know which scenario is currently applicable. To make things worse, the scenario might actually change with moving networks, adhoc networks or with mobility in general. Hence NSIS signaling must assume the worst case and cannot put responsibility to the user to know which scenario is currently applicable. As a result, it might be necessary to perform a "discovery" periodically such that the NSIS agent at the end host has enough information to decide which scenario is currently applicable. This additional messaging, which might not be necessary in all cases, eats performance, bandwidth and adds complexity. Additional information by the user can provide information to assist this "discovery" process but cannot replace it. Some protocols already aim to provide a solution for an end host to learn something about the topology such as STUN [11]. To some extend these protocols can help NSIS NAT/FW signaling. Stiemerling, et al. Expires April 17, 2004 [Page 25] Internet-Draft A NAT/FW NSIS NSLP October 2003 5. NSIS NAT Handling Solution This section describes a mechanism for allowing NSIS signaling messages to travel end-to-end in the presence of NATs at the receiving side. This requires to establish state information at the NSIS-aware NAT device. Note: The discussed mechanism only creates state relevant for NSIS message handling. It does not create NAT bindings for data traffic. 5.1 Problem Description NSIS signaling messages follow the data path from the data sender to the data receiver. To provide this property of being path-coupled the discovery process sends signaling messages along the same route as taken by subsequent data packets. The NSIS messages are directed to a particular destination IP address and hence the destination address needs to be known in advance before NSIS signaling can start. Stiemerling, et al. Expires April 17, 2004 [Page 26] Internet-Draft A NAT/FW NSIS NSLP October 2003 +-------------+ AS-Data Receiver Communication +-------->| Application |<-----------------------------+ | | Server | | | +-------------+ | | IP(R-NAT_B) | | NSIS Signaling Message +-------+--+ | +------------------------------------------>| NAT/NAPT | | | | B | | | +-------+--+ | | | AS-Data| | | Receiver| | +----------+ | Comm.| | | NAT/NAPT | | | | | A | | | | +----------+ | | | | | | | | | | | | | v | IP(R) v +--------+ +---------+ | Data | | Data | | Sender | | Receiver| +--------+ +---------+ Figure 12: The Data Receiver behind NAT problem Figure 12 describes a typical message communication in a peer-to-peer networking environment whereby the two end points learn of each others existence with the help of a third party (referred as Application Server). The communication with the application server and the two end points (data sender and data receivers) serves a number of functions. As one of the most important functions it enables the two end hosts to learn the IP address of each other. Without the proposed mechanism it would not be possible to establish a NAT binding end-to-end in all scenarios. Some sort of communication between the end hosts and a third party is typically necessary (independently of NSIS). NSIS signaling messages cannot be used to communicate application level relevant end point identifiers (in the generic case at least) as a replacement for the communication with the application server. If the data receiver is behind a NAT then an NSIS signaling message will be addressed to the IP address allocated at the NAT (if there Stiemerling, et al. Expires April 17, 2004 [Page 27] Internet-Draft A NAT/FW NSIS NSLP October 2003 was one allocated). If no corresponding NSIS NAT Forwarding State at NAT/NAPT B exists (binding IP(R-NAT B) <-> IP(R)) then the signaling message will terminate at the NAT device (most likely without proper response message). The signaling message transmitted by the data sender cannot install the NAT binding or NSIS NAT Forwarding State "on-the-fly" since this would assume that the data sender knows the topology at the data receiver side (i.e. the number and the arrangement of the NAT and the private IP address(es) of the data receiver). The primary goal of path-coupled middlebox communication was not to force end hosts to have this type of topology knowledge. A number of solutions exist to allow nodes behind a NAT to establish a NAT binding to allow the receiver to receive IP packets. These solutions can at best be labeled as hacks (see [NATP2P]) and they have their drawbacks: o They assume a certain behavior of NAT boxes. o They work in some environments whereas in others they do not properly function. o They only allow NAT bindings for UDP traffic to be established. o They often fail. Some other solutions assume that both nodes are registered in the DNS directory (see [19]). The requirements for an NSIS solution are two-fold: 1. NSIS signaling messages must be able to travel end-to-end (between data sender and data receiver) - if desired. This is important for a number of NSIS NSLPs 2. NSIS relies on a generic solution which works in all scenarios (see section 5 of [22]). Since the NSIS signaling messages are intercepted at each NSIS device the NAT solution depends on the properties of the NTLP. In particular, multiplexing capability is important. Two possible options are feasible: 1. Multiplexing with the help of transport layer information (i.e. port information) 2. Multiplexing at the NSIS application layer (e.g. based on session identifier) Stiemerling, et al. Expires April 17, 2004 [Page 28] Internet-Draft A NAT/FW NSIS NSLP October 2003 We describe the second approach although we believe that alternatives are possible. Enough information has to be available to convert IP address information of an incoming signaling message to different IP addresses of an outgoing NSIS message. Finally the signaling message must reach the data receiver. It seems that the session identifier can be used to associate state information of the two independent signaling exchanges. The two exchanges (as described in Section 5.2) are: 1. Signaling exchange from the data receiver (NR) to the NAT(s) 2. End-to-end NSIS signaling message exchange from the NI to the NR. If the session identifier is used for this purpose then it is necessary to communicate the session id from the data receiver (NR) to the NI. Communicating the IP address information instead (as an alternative solution approach) is easier since this functionality is already provided by SIP whereas securely exchanging (e.g. confidentiality protected) the Session Identifier is not available. 5.2 Solution Overview The data receiver starts to signal an NSIS Create-NAT-Binding message into the "wrong direction". By "wrong" we refer to the usual behavior of path-coupled signaling where the data sender starts signaling in order to tackle with routing asymmetry. The data receiver would typically return signaling messages to the data sender in the reverse direction by utilizing state created at nodes along the path (i.e. to reverse route signaling messages). The concept of path-coupled or path-decoupled signaling is, however, no relevant for this special type of signaling communication. In case of establishing NAT bindings (and NSIS NAT Forwarding State) the direction does not matter since routing is modified. Subsequent NSIS messages (and also data traffic) will travel through the same NAT boxes. The proposed solution requires two NSIS signaling messages: 1. Create-NAT-Binding Request 2. Create-NAT-Binding Acknowledgement The semantic of the two messages will be described in detail in this section. Stiemerling, et al. Expires April 17, 2004 [Page 29] Internet-Draft A NAT/FW NSIS NSLP October 2003 The data receiver sends a Create-NAT-Binding NSIS signaling message into the local network (before the data sender starts NSIS signaling). In Section Section 5.2.1 we will discuss where to address this signaling message (i.e. which destination IP address to use). The signaling message creates NSIS NAT Forwarding State at intermediate NSIS NAT node(s). Furthermore it has to be ensured that the edge NAT device is discovered as part of this process. By edge NAT we refer to the NAT device which is reachable from outside and has a globally routable IP address. The end host cannot be assumed to know this device - instead the NAT box itself is assumed to know that it has such a capability. Forwarding of the Create-NAT-Binding NSIS message beyond this entity is not necessary, and should be prohibited as it provides information on internal hosts capabilities. Create-NAT-Binding Request +-------+ +-------+ +-------+ +---------+ | NAT X |<---| NAT Y |<---| NAT Z |<---| Data | | |--->| |--->| |--->| Receiver| +-------+ +-------+ +-------+ +---------+ Create-NAT-Binding Response ========================================> Data Traffic Direction Figure 13: Create-NAT-Binding NSIS Signaling Message The goal of this signaling message exchange is: o to create one (or more) NAT binding(s) o to allow the data receiver to learn its global routable IP address (for communication with NSIS) o not to require the data receiver to learn topology information. Figure 13 shows a number of NAT devices at the data receivers network side. NSIS NAT Forwarding State is established at these network elements. The Create-NAT-Binding Request message triggers the state creation and the discovery. The message carries information where the sender expects incoming NSIS signaling messages. Stiemerling, et al. Expires April 17, 2004 [Page 30] Internet-Draft A NAT/FW NSIS NSLP October 2003 The Create-NAT-Binding Response message confirms the state creation and allows to return information about the NATs and the topology to the end host (for informational purposes). As a result the end host will learn the public IP address which can be used by the data sender to address NSIS signaling messages. 5.2.1 Destination IP address Selection The Create-NAT-Binding Request message has to be addressed to a specific destination IP address. Since there is no natural candidate a few alternatives might be considered. The discussed options refer to entities of Figure 12 Possible options are: 1. Public IP address of the data sender 2. Public IP address of the data receiver (allocated at NAT B) 3. IP address at the Application Server Actually, there is no "correct" answer to this question and from a theoretical point of view it does not really matter as long as Host A learns an IP address where he has to send the NSIS signaling message. >From a performance point of view there is, however, a difference since it would be desirable to create an "optimal" routing path. 1. Public IP address of the data sender: * Assumption: + The data receiver already learned the IP address of the data sender (e.g. via a third party). * Problems: + The data sender might also be behind a NAT. In this case the public IP address of the data receiver is the IP address allocated at this NAT. + Due to routing asymmetry it might be possible that the routes taken by a) the data sender and the application server b) the data sender and NAT B might be different. As a consequence it might be necessary to advertise a new (and different) external IP address with SIP after using NSIS to establish a NAT binding. Stiemerling, et al. Expires April 17, 2004 [Page 31] Internet-Draft A NAT/FW NSIS NSLP October 2003 2. Public IP address of the data receiver (allocated at NAT B): * Assumption: + The data receiver already learned his externally visible IP address (e.g. based on the third party communication). * Problems: + Communication with a third party is required. 3. IP address at the Application Server: * Assumption: + An application server (or a different third party) is available. * Problems: + If the NSIS signaling message is not terminated at the NAT of the local network then an NSIS unaware application server might discard the message. + Routing might not be optimal since the route between a) the data receiver and the application server b) the data receiver and the data sender might be different. Stiemerling, et al. Expires April 17, 2004 [Page 32] Internet-Draft A NAT/FW NSIS NSLP October 2003 6. Protocol Description 6.1 Basic protocol overview In the NSIS protocol, hosts try to communicate with each other in spite of firewalls or NAT's by opening pinholes or making NAT bindings in them. For things to work, we need to signal the proper boxes, and that means we will have to send the signals on the data path, and in the proper direction. So, if the Data Sender (DS) wants to send data to the Data Receiver (DR), it is DS that will have to signal towards DR. Problems arise when DR has a private IP address and is not publicly reachable. In that case, DR will have to make due for it, and get a public IP address it can use by using specific NSIS signaling. DS will then signal towards the public address DR got. Given this fact, we have two modes of operation: o First, DR making itself available by signaling on the reverse path (DR towards DS). This way, it reserves publicly reachable addresses and ports, which it communicates to DR through means out of the scope of this document, but probably involving a third party and an application level protocol. o And second, DS signaling directly to DR, and creating NAT bindings and installing FW rules. Nothe that the first mode will usually make reservations only, which will be "activated" by the signaling from DS towards DR. The first mode is detailed in the Section 5 The protocol is meant to work on a soft-state basis. This means, that whatever state is installed or reserved on a middle box, will expire, and thus be uninstalled/forgotten after a certain timeout. To prevent this the involved boxes will have to specifically request a session prolongation. An explicit NATFW NSLP state deletion message is also provided by the protocol. Middle boxes should report back in case of error, so that appropriate measures and debugging can be performed. 6.2 Message format and types At the moment of writing this document, no decision has been reached on the details of the NTLP, (NSIS Transport Layer Protocol), hence the current NATFW NSLP protocol specification uses raw IP transport with an arbitrary protocol number. The NSIS Protocol Data Unit is constituted of a header, and a series of objects. Stiemerling, et al. Expires April 17, 2004 [Page 33] Internet-Draft A NAT/FW NSIS NSLP October 2003 The NSIS header contains: o header_len: length of the nsisp header in bytes o version: protocol version number. o obj_count: number of objects that follow after the nsis header. It is followed by a series of objects. Each object is composed of an object header and the object data. The object header format is common to all the object types, and has the following format: o obj_len: total length of the object o obj_h_len: length of the object header o obj_type: type of NATFW NSLP object. Identifies the data that follows. The object data is attached next to the object header, and varies depending on the required action to be taken. For the moment four object types are defined as follows, if required others could be defined later on. Create session (a temporary pinhole or a nat-binding): o sid: Session ID. Randomly generated by the endhosts. o src_addr: where the data will come from. If it is DS sending data to DR, the source address is either DS or the closest NAT in the route from DS to the middlebox that gets the packet; That is, the address where each middlebox will see the packet come from. o dst_addr: whre the data is headed. If it is DS sending data to DR, the destination address is either DR or the public address DR reserved itself. o src_port: the transport port the data will come from o dst_port: the transport port the data will go to. o proto: protocol number. Note: you might want to leave the source address or port set to ANY, to accept any source address port. This makes the pinhole not so pin like, but might be necessary at the integration with certain NAT/FW types. Whether this loose pinhole is authorized or not by the middle box, is a policy decision based on the middle box configuration. Stiemerling, et al. Expires April 17, 2004 [Page 34] Internet-Draft A NAT/FW NSIS NSLP October 2003 Prolong session: o sid: the session we intend to prolong. Delete session: o sid: the session we want to delete. Reserve external addr: o sid: session for the reserve external address process. o tgt_addr: address we want our external address to point to. If A reserves an external address in a NAT, it will want that external address to point to A. o tgt_port: port the external addres:port pair should point to. o src_addr: The address the data will come from. We might want to set this to 0.0.0.0, to make a loose pinhole reservation, if we don't the src_addr yet. o src_port: The port the data will come from. Can also be set to zero if unknown. Note that no state, be it a firewall rule or a nat binding, is installed as a result of this message. The state is only remembered, and might be later installed by a create message. Return external addr: o sid: the session this packet is replyng about. o ext_addr: the external address that the nat has just created for the endhost. o ext_port: the external port. Path succeeded: o sid: the session ID for which a path was succesfully installed Error: o sid: the session id of the object that generated the error o error_code: what the error was. Stiemerling, et al. Expires April 17, 2004 [Page 35] Internet-Draft A NAT/FW NSIS NSLP October 2003 6.3 Message flow Of the shown message types, 4 of them are sent by the end hosts to achieve some service (create, reserve, prolong, and delete) and 3 are sent back as a response (return external address, path succeded and error). The usual message flow for the reservation mode of operation, to achieve an external address, is: DS Public Internet NAT Private addres DR | | space | | | | | | Reserve | | |<---------------------------| | | | | | Return ext addr / Error | | |--------------------------->| | | | | | | In the case of the creation mode, when rules are actually installed, the message flow is as follows: Stiemerling, et al. Expires April 17, 2004 [Page 36] Internet-Draft A NAT/FW NSIS NSLP October 2003 DS Public Internet NAT Private addres DR | | space | | Create | | |----------------------------->| | | | | | Error (if necessary) | | |<-----------------------------| Create | | |--------------------------->| | | | | | Path Succeeded/Error | | Path Succeeded/Error |<---------------------------| |<-----------------------------| | | | | | | | When it comes to prolonging or deleting sessions: DS Public Internet NAT Private addres DR | | space | | Prolong/Delete | | |----------------------------->| | | | | | Error (if necessary) | | |<-----------------------------| Prolong/Delete | | |--------------------------->| | | | | | Error (if necessary) | | Error (if necessary) |<---------------------------| |<-----------------------------| | | | | | | | 6.4 Middle box configuration The way a message is handled in each box will depend on its configuration. Currently defined ocal NSIS NAT/FW configuration parameters are: o nat_capabilities: am I a NAT? o nat_external_addr: if I'm a nat, what is the address closest to Stiemerling, et al. Expires April 17, 2004 [Page 37] Internet-Draft A NAT/FW NSIS NSLP October 2003 the public internet? o edge_nat: true, false or auto. Tells us how much of an edge_nat we are. See explanation below. The edge_nat parameter requires some more explanation, though. When a host reserves an external address, as seen in Section 5, it makes itself reachable by the other end. This does not always mean a public internet address. In Section 7.4 we see that a private address can also be reachable by another box in the same address space. Still, things work well if we get a public ip address, only the route is unnecesarily long (see an example in Section 7.4). For this reason, we can set up the boxes to always or never behave as an edge nat, or else set it to auto. The auto setting will compare the destination ip address of the reserve address objects, and see if they belong to the same address space as the nat external address. If it is so, the nat behaves as an edge_nat for this connection, otherwise, it doesn't. Note that the address spaces are set in RFC3330 [2]. 6.5 Message handling When a box receives an NSIS message, it might take an action based on the message type, the nature of the box, its configuration and its security policies. As a summary, here's the behaviour of the boxes, depending on message type and config parameters: NAT FW NAT+FW DS DR reserve 5 - 5 + + ret_ext_addr - - - + 8 create 1 2 3 + 4 prolong 6 6 6 + 4 delete 7 7 7 + 4 path_succeed - - - 8 + 1: install the rule, rewriting either the source or destination address depending on whether the packet comes from the external_address or not. Always forward. 2: Install the rule, always forward. 3: 1+2. The order depends on whether it comes from the outside address (NAT, then FW) or the inside one (FW, then NAT) Stiemerling, et al. Expires April 17, 2004 [Page 38] Internet-Draft A NAT/FW NSIS NSLP October 2003 4: If it fits one of its requests, send a path_succeeded packet back. Drop the packet. 5: Make a reservation. If edge_nat is set, send back the external_addr and don't forward. Otherwise, forward and don't send anything back. 6: Prolong the session. Always forward. 7: Terminate the session. Always forward. 8: hand it over to superior layers, and drop it. -: ignore and forward. +: ignore and drop. Note that the order is important, when it comes to NAT+FW boxes, because the filter rules have to be set up according to the packet they will see. Source NAT is done at the end so it doesn't disturb routing decisons, this means the filter sees the original packets. Destination NAT, on the other hand, is done at the beginning, so it can be routed properly, and so the filter sees the modified packets. Note also that for each action, the host might demand a certain degree of authorization, and thus refuse to take the action, sending an error message back instead. The details of the message handling in each of the boxes follows. 6.5.1 Detailed behaviour 6.5.1.1 Reserve external address message o NAT Box: When a NAT box gets a Reserve external addressm message, it checks whether it arrived on the public address, or the private one. If it arrived in the public one, an error message of the type: "Requested an external address from the outside" is sent back. If it arrived on the private side, an entry is made in the internal reservation list with the packet information. If the box is an edge nat (either by configuring it to true, or just for that connection if it is set to auto), it drops the message, and replies with a return external address message containing the allocated address port pair. If it is not an edge NAT, it forwards the packet on. Stiemerling, et al. Expires April 17, 2004 [Page 39] Internet-Draft A NAT/FW NSIS NSLP October 2003 o Firewall Box: Reserve messages are silently ignored in Firewall boxes. They are simply forwarded on. o NAT+FW Box: When a box that integrates both a NAT and a Firewall gets a reserve message, it will hand it to its NAT part. Its firewall part will simple ignore it. o Data Sender: The message should never get here. It should be ignored and dropped. o Data Receiver: The message should never get here. It should be ignored and dropped. 6.5.1.2 Return external address message o NAT Box, Firewall Box and NAT+Firewall Box: When one of these boxes gets a Return external address message, it must simply ignore it and let it traverse. o Data Sender: The message should never get here. It should be ignored and dropped. o Data Receiver: A return external address message in the Data receiver, has reached its destination. It must be dropped, and it's information handed to superior layers. 6.5.1.3 Create message o NAT Box: When a NAT box gets a create message, it first checks if it arrived on the public address or not. Stiemerling, et al. Expires April 17, 2004 [Page 40] Internet-Draft A NAT/FW NSIS NSLP October 2003 If it came from the public side, it means an external box will try to send data. It then looks for a reservation in its reservation list, that matches the dst_addr and dst_port of the packet. If it doesn't find it, it returns an error message of the type "No reservation found". If it finds it, it fills in the reservation with the data from the packet, and installs the given rule. It then changes the dst_addr and dst_port fields of the create packet and forwards it to the tgt_addr of the reservation. If it came from the private side, it installs the nat rule with the information in the packet. It then changes the src_addr and src_port of the create message to its own external address and port. o Firwall Box: When a firewall box gets a create message, it simply installs the rule specified in the message and forwards the packet. o NAT+FW Box: When a box that integrates both a NAT and a Firewall gets a create message, it first checks whether it arrived on the public address or not. If it arrived on the public side, the NAT part of the box takes care of the packet first, as said in the NAT Box case. Afterwards, the modified packet is handed to the firewall part, where it is handled as in the Firewall Box case. If it arrived on the private side, the message is handed to the firewall part first, and then to the NAT one. o Data Sender: The message should never get here. It should be ignored and dropped. o Data Receiver: If the data receiver gets a create message, it means all the boxes on the way accepted it, and so the signaling succeded. All it does is drop the packet, and send back a Path Succeded message to the ip packet source address. Stiemerling, et al. Expires April 17, 2004 [Page 41] Internet-Draft A NAT/FW NSIS NSLP October 2003 6.5.1.4 Delete session message o NAT Box, Firewall Box and NAT+Firewall Box: When one of these boxes gets a Delete session message, it should erase the session refered in the message. o Data Sender: As in the create session message, this packet should never get to the DS, so it is to be ignored and dropped. o Data Receiver: As in the create session message, this is the final destination of the message. If it got here, everything is fine. No further action should be taken on the packet, which means it is dropped at the endhost. 6.5.1.5 Prolong session message o NAT Box, Firewall Box and NAT+Firewall Box: When one of these boxes gets a Prolong session message, the expiration time of the session should be changed to the time of reception plus the configured session lifetime. o Data Sender: As in the create session message, this packet is sent from the DS to the DR, and should never arrive at the DS. Again, it should be ignored and dropped. o Data Receiver: The same behaviour as in the case of a Delete session message on the DR should be applied. 6.5.1.6 Path Succeded message o NAT Box, Firewall Box and NAT+Firewall Box: When one of these boxes gets a Path succeded message, it should simply ignore it and allow it to traverse. o Data Sender: Stiemerling, et al. Expires April 17, 2004 [Page 42] The message ends here. It must be dropped, and its contents sent to upper software layers. o Data Receiver: The message should never get here. It should be ignored and dropped. Stiemerling, et al. Expires April 17, 2004 [Page 43] Internet-Draft A NAT/FW NSIS NSLP October 2003 7. Solution examples 7.1 Firewall traversal DS wants to send ata traffic to DR through tight firewalls, as seen in Figure 18. To do that, it will have to signal using NSIS, on the data path. +-----+ +-----+ //----\\ +-----+ +-----+ DS --| FW1 |-----| FW2 |---| |---| FW3 |-----| FW4 |--- DR +-----+ +-----+ \\----// +-----+ +-----+ private public private Data Flow ===============================> Figure 18: Firewall Traversal Scenario Therefore, DS initiates signaling to DR by sending a create object to the ip address od DR. Note that DS already knows its source address and port (say, 1111), and the destination address of DR. The destination port (let's say 9999) has been send to DS by DR via application layer messages, possibily, but not necessarily involving a third party. The message looks like: o dst_addr = DR o dst_port = 9999 o src_addr = DS o src_port = 1111 This message is received by FW1, which installs the state that reads: "Any packet coming from DS:1111 headed for DR:9999 will be allowed traversal" FW2, FW3 and FW4 do exactly the same, and forward the packet to each other, until it finally reaches DR. At this point, the data path is open, and DR sends back a Path succeeded message to DS, which can now start sending traffic. 7.2 NAT with private network on sender side Stiemerling, et al. Expires April 17, 2004 [Page 44] Internet-Draft A NAT/FW NSIS NSLP October 2003 In the example in Figure 19, DS is in a private network and wants to send data to DR, out in the public internet. To do so, DS will have to initiate NSIS signaling towards DR +------+ +------+ //----\\ DS --| NAT1 |-----| NAT2 |---| |--- DR +------+ +------+ \\----// private public Figure 19: NAT with private network on sender scenario Now, this is a deceivingly easy example. Apparently, the normal NAT functionality will take care of sending the data from DS out into the public internet, and route back the replies from DR. This is indeed true, but doesn't give NSIS control on what the source address or port is, as it is usually asigned dinamycally by the NAT. Moreover, the NSLP would have no information on this hops, and could not install proper pinholes, as it would set DS as the source address, and not that of the last NAT. DS builds a create packet with the information he has, which is the same as that in Section 7.1. It looks like this: o dst_addr = DR o dst_port = 9999 o src_addr = DS o src_port = 1111 NAT1 is the first to get the packet; It is not coming from its configured "nat external address", and so, it knows it will have to rewrite the information on the source, and not that of the destination. NAT1 then picks a free port (incidentally 1011) and installs a nat rule that reads: "Whatever packet comes from DS:111, heading for DR:9999 will be rewriten so that the source address looks like NAT1:1011". It then rewrites the packet it received as follows: o dst_addr = DR o dst_port = 9999 Stiemerling, et al. Expires April 17, 2004 [Page 45] Internet-Draft A NAT/FW NSIS NSLP October 2003 o src_addr = NAT1 o src_port = 1011 And forwards the packet. NAT2 gets it now, and does exactly the same. Port 2022 is chosen, and the rule: "Whatever packet comes from NAT1:1011, heading for DR:9999 will be rewriten so that the source address looks like NAT2:2022" is installed. The packet gets modified as follows: o dst_addr = DR o dst_port = 9999 o src_addr = NAT2 o src_port = 2022 And is forwarded. It eventually reaches DR, who sends back a path succeded message. Data flow from DS:1111 to DR:9999 is now possible. 7.3 NAT with private network on receiver side In this example, DS wants to send data to DR over the network in Figure 20: //----\\ +------+ +------+ DS ---| |---| NAT1 |-----| NAT2 |--- DR \\----// +------+ +------+ public private Figure 20: NAT with private network on receiver Scenario The problem, of course, is that DR is not publicly reachable. Because of that, DR will have to signal on the data path, in the opposite direction (DR->DS) to get itself a public address it can use. This method is described in Section 5 To get an external address, DR sends a packet to DS. It could actually send it to anything in the public internet, as it would force it to traverse what NATs are on its way. In the case of multihomed environments, though, more than one NAT to the outside is possible, so the better we "aim" the more the chances we go out the Stiemerling, et al. Expires April 17, 2004 [Page 46] Internet-Draft A NAT/FW NSIS NSLP October 2003 right NATs and get more optimal routes. The said packet is an NSIS reserve_addr object which looks like this: o tgt_addr = DR o tgt_port = 9999 o src_addr = 0.0.0.0 o src_port = 0 Notice that this is a really loose pinhole, since any src_addr and port is allowed. NAT2 gets the packet and looks for a free port (say, 2022, for clarity's sake). It then adds an entry to its reservation list. The entry looks like this: o src_addr = 0.0.0.0 o src_port = 0 o dst_addr = NAT2 o dst_port = 2022 o tgt_addr = DR o tgt_port = 9999 This means simply that packets coming from any source, destinated to the public address we just reserved, shoud be targeted to the internal box DR, on port 9999 It then rewrites the packet so that it looks like: o tgt_addr = NAT2 o tgt_port = 2022 o src_addr = 0.0.0.0 o src_port = 0 Because it is not an edge nat, it forwards the modified packet and does not sent a return_external_addr object to DR. Note that no nat binding is installed so far in NAT2, although the state is now Stiemerling, et al. Expires April 17, 2004 [Page 47] Internet-Draft A NAT/FW NSIS NSLP October 2003 reserved. NAT1 now gets the packet, picks free port 1011 and adds the following entry to its reservation list: o src_addr = 0.0.0.0 o src_port = 0 o dst_addr = NAT1 o dst_port = 1011 o tgt_addr = NAT2 o tgt_port = 2022 As it turns out, NAT1 IS an edge_nat, so it doesn't forward the packet. Instead, it replies to DR sending back a return external address packet on the same connection, so it finds its way back through the NATs: o ext_addr = NAT2 o ext_port = 2022 By using some application layer protocol, and possibly, although not necesarily, using a third party box, DR sends it's freshly allocated external address and port to DS. DS now knows who to signal, so it sends a create message: o dst_addr = NAT1 o dst_port = 1011 o src_addr = DS o src_port = 1111 When it reaches NAT1, it does so through NAT1 external address. It realizes it is being asked to forward the traffic from some outside box towards the inside. It then looks up its reservation list, looking for a session that has the external address and port NAT1:1011 assigned. It finds this: o src_addr = 0.0.0.0 Stiemerling, et al. Expires April 17, 2004 [Page 48] Internet-Draft A NAT/FW NSIS NSLP October 2003 o src_port = 0 o dst_addr = NAT1 o dst_port = 1011 o tgt_addr = NAT2 o tgt_port = 2022 Using the information in the create object, it then fills in this structure to: o src_addr = DS o src_port = 1111 o dst_addr = NAT1 o dst_port = 1011 o tgt_addr = NAT2 o tgt_port = 2022 This IS a tight pinhole. NAT1 installs the rules now, which say: "Whatever packet comes from DS:1111 heading for NAT1:1011, should have its destination address changed to NAT2:2022, and be forwarded". The packet is also rewritten into this: o src_addr = DS o src_port = 1111 o dst_addr = NAT2 o dst_port = 2022 And is forwarded to NAT2. Upon arrival, a similar process issues. NAT2 finds its reservation entry: o src_addr = 0.0.0.0 o src_port = 0 o dst_addr = NAT2 o dst_port = 2022 Stiemerling, et al. Expires April 17, 2004 [Page 49] Internet-Draft A NAT/FW NSIS NSLP October 2003 o tgt_addr = DR o tgt_port = 9999 Fills it in accordingly: o src_addr = DS o src_port = 1111 o dst_addr = NAT2 o dst_port = 2022 o tgt_addr = DR o tgt_port = 9999 Rewrites the packet: o src_addr = DS o src_port = 1111 o dst_addr = DR o dst_port = 2222 And forwards it to DR. Once there, DR acknowledges it by sending back a path succeded message in reply, back to DS. The path is now open and data transmission from DS:1111->DR:9999 can comence. 7.4 Both end hosts are in same private network behind NATs In this example (see Figure 21), DS, in a private address space, wants to send data to DR, in another private address space. The point marked "%" is yet another private address space. Notice that since NAT1 and NAT3 have addresses in the same address space, NAT3 might want to consider itself an edge nat. We will consider both situations. Stiemerling, et al. Expires April 17, 2004 [Page 50] Internet-Draft A NAT/FW NSIS NSLP October 2003 public +------+ % +------+ //----\\ DS --| NAT1 |--+--| NAT2 |---| | +------+ | +------+ \\----// | | +------+ +--| NAT3 |------------ DR +------+ private Figure 21: NAT to public, receiver in same private network Scenario We will first assume that NAT3 has the edge_nat option set to false. In this case, the connection is a combination of Section 7.3 and Section 7.2. Firstly DR will signal against on the data path, against the data flow, with a reserve external address object. NAT3 will reserve the address and forward the packet on to NAT2, who IS an edge nat in all cases. NAT2 will reply with the external address, and the connection goes on just as in Section 7.2, except for the fact the topology becomes: Stiemerling, et al. Expires April 17, 2004 [Page 51] Internet-Draft A NAT/FW NSIS NSLP October 2003 public +------+ +------+ DS --| NAT1 |-----o------o---+ +------+ | | | | NAT2 |---+ | | | +--o------o---+ | +------+ | | +------+ +--| NAT3 |------------ DR +------+ private Figure 22: New topology due to the non optimal edge nat parameter decision This is not optimal, but the connection does succeed, and data flow can commence. Let us now solve the case in which NAT3 has edge_nat set to auto. Back in Figure 21, NAT3 will decide it IS an edge_nat if the destination we pick up for the reserve address packet is in the address space marked as "%", and will NOT consider itself an edge_nat if we point it anywhere else. This is an optimality issue such as the one pointed out in Section 7.3. Well so, if it doesn't consider itself an edge nat, we already saw what the topological equivalent is, and how it proceeds. If it IS an edge nat, the topological equivalent would be: Stiemerling, et al. Expires April 17, 2004 [Page 52] Internet-Draft A NAT/FW NSIS NSLP October 2003 +------+ DS --| NAT1 |--+ +------+ | | | +------+ +--| NAT3 |------------ DR +------+ private Figure 23: A good edge nat decision brings an optimal route And we would proceed in the same way, only on a more optimal route. 7.5 IPv4/v6 NAT with two private networks TBD 7.6 Full example for NAT/FW with two private networks The NAT's have the nat_capabilities variable set to true. NAT+FW3 and NAT+FW5 have the edge_nat variable set to true. The rest of boxses have it set to false. Let's now suppose that DR wants to get a data stream from DS in Figure 24. For that, we need some way for B to get messages from A, be it through some third party application server or some publicly reachable proxy, perhaps made public through a nat binding. Stiemerling, et al. Expires April 17, 2004 [Page 53] Internet-Draft A NAT/FW NSIS NSLP October 2003 +-----+ +--------| FW4 |--------+ | +-----+ | +---------+ +---------+ | NAT+FW3 | | NAT+FW5 | +---------+ +---------+ | | +---------+ +---------+ | NAT3 | | NAT6 | +---------+ +---------+ | | +---------+ +---------+ | FW1 | | FW7 | +---------+ +---------+ | | +---------+ +---------+ | DS | | DR | +---------+ +---------+ Data Flow ==========================> Figure 24: Example network topology DR wants a data stream from DS, which means that the direction of the data is DS->DR. A will have to make itself publicly reachable by signaling its NATs and firewalls as necessary. This is a step by step guide to the whole process. In steps 1 to 4, DR makes itself publicly reachable. From 5 and on, DS is signaling on the data path towards DR. 1. DR wants to get data from DS, so it sends a reserve_addr object to a target in the public internet. The closest this target is, the more the chances that the outcoming route is optimal, but any will work. The reserve_addr obj looks like this: o tgt_addr = DR o tgt_port = 888 o src_addr = 0.0.0.0 o src_port = 0 Notice that this is a really loose pinhole, since any src_addr and port is allowed. Stiemerling, et al. Expires April 17, 2004 [Page 54] Internet-Draft A NAT/FW NSIS NSLP October 2003 2. FW7 gets the packet, ignores its contents and forwards it. Firewalls always ignore reserve_addr objects. 3. NAT6 gets the packet, and looks for a free port (say, 666, for clarity's sake). It then adds an entry to its reservation list. The entry looks like this: o src_addr = 0.0.0.0 o src_port = 0 o dst_addr = NAT6 o dst_port = 666 o tgt_addr = DR o tgt_port = 888 It then rewrites the packet so that it looks like: o tgt_addr = NAT6 o tgt_port = 666 o src_addr = 0.0.0.0 o src_port = 0 Because it is not an edge nat (edge_nat=false), it does not sent a return_external_addr object to DR, but rather forwards the modified packet. Note that no nat binding is installed so far in NAT6, although the state is now reserved. 4. NAT+FW5 receives the packet. The firewall part gets the object, but, being as it is an address reservation only object, it ignores it. The NAT part gets it next. Because it is a NAT, it binds a free port, which is thus reserved. An entry to the reservation list is added: o src_addr = 0.0.0.0 o src_port = 0 o dst_addr = NAT+FW5 o dst_port = 555 Stiemerling, et al. Expires April 17, 2004 [Page 55] Internet-Draft A NAT/FW NSIS NSLP October 2003 o tgt_addr = NAT6 o tgt_port = 666 Because it is an edge_nat, it sends a return_external_addr packet with address NAT+FW5 and port 555 back to DR. It does so by simply sending it back to the source IP addr in the IP header of the packet. In this case, it is NAT6. The standard capabilities of NAT6 will send it back to DR, since we are always working on the same connection. Because it is an edge_nat and this is a reserve_external_addr packet, it does not forward the packet. At this stage, the end host DR has learned what its (reserved) external address is, even if it can not be used. It is now publicly reachable, and path-coupled nsis signaling in direction DS->DR can start. 5. Firstly, DR tells DS about it's freshly reserved outside address through some higher layer protocol, using the third-party box. 6. DS now initiates signaling to DR by sending a create object to the brand new public address of DR. It looks like: o dst_addr = NAT+FW5 o dst_port = 555 o src_addr = DS o src_port = 111 7. The firewall FW1 gets it, and installs the requested pinhole. (Note this IS a tight pinhole with well defined source and destination). It then forwards the packet. 8. NAT2 gets the packet. Because it is NOT coming from it's external address, it realizes it is being asked to forward DS's future data packets, and so, it will have to rewrite it's source address. To do so, NAT2 picks a random free port (which turns out to be 222), and installs a nat rule that says: "Whatever packet comes from DS:111, heading for NAT+FW5:555 will be rewriten so that the source address looks like NAT2:222". That is usually known as Source NAT. The nsis create request is then rewriten to look like: o dst_addr = NAT+FW5 o dst_port = 555 Stiemerling, et al. Expires April 17, 2004 [Page 56] Internet-Draft A NAT/FW NSIS NSLP October 2003 o src_addr = NAT2 o src_port = 222 Because it is not an edge nat, it simply forwards the modified packet. 9. NAT+FW3 gets the packet next. Because it is NOT coming from the extarnal_addr of the NAT+FW, The firewall part gets it first, and installs the filter rule that says: "Allow traversal of packets going from NAT2:222 towards NAT+FW5:555". It then hands it to the NAT part. The NAT part gets it then. It is not coming from its external addr, and so, it does as NAT2, binding a port (333) and installing a rule that says: "Whatever packet comes from NAT2:222, heading for NAT+FW5:555, will be rewriten so that the source address looks like NAT+FW3:333". It will then rewrite the create object to: o dst_addr = NAT+FW5 o dst_port = 555 o src_addr = NAT+FW3 o src_port = 333 Note that the box won't send a packet back to DS informing it of its external address, because DS will never need that. 10. FW4 gets the create object, and installs the rule "Allow traversal of packets going from NAT+FW3:333 towards NAT+FW5:555" It then forwards the object. 11. NAT+FW5 gets the create object. It arrived at its external address, so it realizes it doesn't have to change the source address of the future data packets of DS, but rather its destination. It also means that the NAT part will have to handle it first. It then tries to find out where it has to re-destinate it to, by looking up its reservation tables. It finds the previous reservation, by matching it with ther dst_addr and dst_port of the create object: o src_addr = 0.0.0.0 o src_port = 0 o dst_addr = NAT+FW5 Stiemerling, et al. Expires April 17, 2004 [Page 57] Internet-Draft A NAT/FW NSIS NSLP October 2003 o dst_port = 555 o tgt_addr = NAT6 o tgt_port = 666 And proceeds to fill it in with the information of the create object (src_addr and src_port): o src_addr = NAT+FW3 o src_port = 333 o dst_addr = NAT+FW5 o dst_port = 555 o tgt_addr = NAT6 o tgt_port = 666 It then installs a NAT rule with that information. It reads: "Whatever packet comes from NAT+FW3:333, heading for NAT+FW5:555 will be rewritten, so that its destination address looks like NAT6:666". The reservation is erased and the rule starts working. The NAT binding becomes thus usable. The object is modified, so that it now looks like: o dst_addr = NAT+FW3 o dst_port = 333 o src_addr = NAT6 o src_port = 666 The FW part now gets the object, and installs the rule: "Allow traversal of whatever packet that comes from NAT+FW3:333 heading for NAT6:666". The packet is then forwarded. 12. NAT6 gets the packet. As it comes from the external adddress, it does as NAT+FW5, looking up the reservation list and filling it in with: o src_addr = NAT+FW3 o src_port = 333 Stiemerling, et al. Expires April 17, 2004 [Page 58] Internet-Draft A NAT/FW NSIS NSLP October 2003 o dst_addr = NAT6 o dst_port = 666 o tgt_addr = DR o tgt_port = 888 It then installs the rule: "Whatever packet comes from NAT+FW3:333, heading for NAT6:666 will be rewritten, so that its destination address looks like DR:888". The rule reservation is erased, and the NAT binding becomes active. The object is rewritten as: o src_addr = NAT+FW3 o src_port = 333 o dst_addr = DR o dst_port = 888 The object is thus forwarded. 13. FW7 gets the packet now, and installs the rule: "Allow traversal of whatever packet that comes from NAT+FW3:333 heading for DR:888". It forwards the packet. 14: DR gets (finally) the packet. It realizes it is a create object headed for him, to the port which he expected, and so it sees everything went well. A reply to the packet is send, and the NAT's on the way, knowing the already established connection, will route it to DS. The packet is a path_succesful message, which simply means "Everything's fine, send data whenever you want". Stiemerling, et al. Expires April 17, 2004 [Page 59] Internet-Draft A NAT/FW NSIS NSLP October 2003 8. NSIS NAT and Firewall transitions issues NSIS NAT and Firewall transition issues are premature and will be addressed in a separate draft (see [8]). An update of this section will be based on consensus. Stiemerling, et al. Expires April 17, 2004 [Page 60] Internet-Draft A NAT/FW NSIS NSLP October 2003 9. Security Considerations Security is of major concern particularly in case of firewall traversal. Generic threats for NSIS signaling have been discussed in [3] and are applicable here as well. It is necessary to provide proper signaling message protection and proper authorization. Note that the NAT is likely to be co-located with a firewall and might therefore require packet filters to be changed in order to allow the signaling message to process and to traverse. This section aims to raise some items for further discussion and illustrates the problems the authors faced when creating a security solution for the NAT/ Firewall NSLP. Without doubts there is a dependency on the security provided by the NTLP. At the time of writing this version of the document no NTLP draft is available (or accessible for the authors). Section Section 4 and Section 3.5 motivates some trust relationship and authorization scenarios. These scenarios deserve a discussion since some of them (particularly one with a missing network-to-network trust relationship) is different to what is know from QoS signaling. To address some of these trust relationships and authorization issues requires security mechanisms between non-neighboring nodes at the NSLP layer. For the group of authors it seems that peer-to-peer and end-to-middle security needs to be provided. An NSLP security mechanism between neighboring NSLP peers might be necessary if security mechanisms at the NTLP do not provide adequate protection mechanisms. This issue is, however, still in discussion. As a design goal it seems to be favorable to reuse existing mechanisms to the best extend possible. In most cases it is necessary to carry the objects for end-to-middle as NSLP payloads since the precesence of NATs might prevent direct communication. Three security mechanisms have to be considered in more detail in a future version of this document: CMS [9] and Identity Representation for RSVP [10]. The authors believe that CMS more suitable (since it provides much more functionality). The details deserve further discussion and implementation experience. With regard to signal between two end hosts even though the receiver is behind a NAT this proposal suggests creating state by the data receiver first. This allows NSIS signaling messages to traverse a NAT at the receiver side (due to the established state at this NAT box) and simplifies security handling.To achieve this behavior it is required to install NSIS NTLP and NSLP state. Furthermore, it is envisioned to associate the two signaling parts (one part from the data sender to the NAT and the other part from the NAT to the data receiver) with the help of the Session Identifier. As such, the discussion in [10] is relevant for this document. Another interesting property of this protocol proposal is to prevent Denial of Service attacks against NAT boxes whereby an adversary Stiemerling, et al. Expires April 17, 2004 [Page 61] Internet-Draft A NAT/FW NSIS NSLP October 2003 allocates NAT bindings with the help of data packets. Since these data packets do not provide any type of authentication and are not authorized any adversary is able to mount such an attack. This attack has been mentioned at several places in the literature already and is particularly harmful if no NAPT functionality is used (i.e. if a new NAT binding consumes one IP address of a pool of IP addresses). Using the protocol described in this document additional security can be achieved and more fairness can be provided. Stiemerling, et al. Expires April 17, 2004 [Page 62] Internet-Draft A NAT/FW NSIS NSLP October 2003 10. Open Issues At least the following issues require further discussion: o Message format: The exact message format is still to be determined, both in regards of bit level details and on parameters, such as the need for an object header length, since, until now, that is a constant. o Error codes: error codes have to be defined still. Among others, we will need: missing authorization, out of resources, unable to understand the packet, or maximum resources for that individual already allocated. o Middle box default policies: allow for the configuration of the default policies of the box. For a NAT+Firewall box, for instance, the firewall default policy might be "accept", and so, no packet filters would have to be installed on that regard (we would still need the nat bindings, though). Stiemerling, et al. Expires April 17, 2004 [Page 63] Internet-Draft A NAT/FW NSIS NSLP October 2003 11. Contributors A number of individuals have contributed to this draft. Since it was not possible to list them all in the authors section, it was decided to split it and move Marcus Brunner and Henning Schulzrinne into the contributors section. Separating into two groups was done without treating any one of them better (or worse) than others. Stiemerling, et al. Expires April 17, 2004 [Page 64] Internet-Draft A NAT/FW NSIS NSLP October 2003 Normative References [1] Brunner et al., M., "Requirements for Signaling Protocols", DRAFT draft-ietf-nsis-req-07.txt, March 2003. [2] IANA, "Special-Use IPv4 Addresses", RFC 3330, September 2002. [3] Tschofenig, H. and D. Kroeselberg, "Security Threats for NSIS", DRAFT draft-ietf-nsis-threats-01.txt, January 2003. [4] Srisuresh, P., Kuthan, J., Rosenberg, J., Molitor, A. and A. Rayhan, "Middlebox communication architecture and framework", RFC 3303, August 2002. Stiemerling, et al. Expires April 17, 2004 [Page 65] Internet-Draft A NAT/FW NSIS NSLP October 2003 Informative References [5] Srisuresh, P. and M. Holdrege, "IP Network Address Translator (NAT) Terminology and Considerations, RFC 2663", August 1999. [6] Srisuresh, P. and E. Egevang, "Traditional IP Network Address Translator (Traditional NAT), RFC 3022", January 2001. [7] Tschofenig, H., Schulzrinne, H., Hancock, R., McDonald, A. and X. Fu, "Security Implications of the Session Identifier", June 2003. [8] Aoun and others...., C., "NAT/Firewall NSLP migration, routing, NTLP requirements and off-path Considerations", October 2003. [9] Housley, R., "Cryptographic Message Syntax (CMS)", RFC 3369, August 2002. [10] Yadav, S., Yavatkar, R., Pabbati, R., Ford, P., Moore, T., Herzog, S. and R. Hess, "Identity Representation for RSVP", RFC 3182, October 2001. [11] Rosenberg, J., Weinberger, J., Huitema, C. and R. Mahy, "STUN - Simple Traversal of User Datagram Protocol (UDP) Through Network Address Translators (NATs)", RFC 3489, March 2003. [12] Manner, J., Suihko, T., Kojo, M., Liljeberg, M. and K. Raatikainen, "Localized RSVP", DRAFT draft-manner-lrsvp-00.txt, November 2002. [13] Tschofenig, H., Buechli, M., Van den Bosch, S. and H. Schulzrinne, "NSIS Authentication, Authorization and Accounting Issues", draft-tschofenig-nsis-aaa-issues-01 (work in progress), March 2003. [14] Amini, L. and H. Schulzrinne, "Observations from router-level internet traces", DIMACS Workshop on Internet and WWW Measurement, Mapping and Modelin Jersey) , Februar 2002. [15] Adrangi, F. and H. Levkowetz, "Problem Statement: Mobile IPv4 Traversal of VPN Gateways", draft-ietf-mobileip-vpn-problem-statement-req-02.txt (work in progress), April 2003. [16] Ohba, Y., Das, S., Patil, P., Soliman, H. and A. Yegin, "Problem Space and Usage Scenarios for PANA", draft-ietf-pana-usage-scenarios-06 (work in progress), April 2003. Stiemerling, et al. Expires April 17, 2004 [Page 66] Internet-Draft A NAT/FW NSIS NSLP October 2003 [17] Shore, M., "The TIST (Topology-Insensitive Service Traversal) Protocol", DRAFT draft-shore-tist-prot-00.txt, May 2002. [18] Tschofenig, H., Schulzrinne, H. and C. Aoun, "A Firewall/NAT Traversal Client for CASP", DRAFT draft-tschofenig-nsis-casp-midcom-01.txt, March 2003. [19] Srisuresh, P., Tsirtsis, G., Akkiraju, P. and A. Heffernan, "DNS extensions to Network Address Translators (DNS_ALG)", RFC 2694, September 1999. [20] Rosenberg, J., Schulzrinne, H., Camarillo, G., Johnston, A., Peterson, J., Sparks, R., Handley, M. and E. Schooler, "SIP: Session Initiation Protocol", RFC 3261, June 2002. [21] Carpenter, B. and S. Brim, "Middleboxes: Taxonomy and Issues", RFC 3234, February 2002. [22] Brunner, M., Stiemerling, M., Martin, M., Tschofenig, H. and H. Schulzrinne, "NSIS NAT/FW NSLP: Problem Statement and Framework", DRAFT draft-brunner-nsis-midcom-ps-00.txt, June 2003. Authors' Addresses Martin Stiemerling Network Laboratories, NEC Europe Ltd. Kurfuersten-Anlage 36 Heidelberg 69115 Germany Phone: +49 (0) 6221 905 11 13 EMail: stiemerling@ccrle.nec.de URI: Hannes Tschoefenig Siemens AG Otto-Hahn-Ring 6 Munich 81739 Germany Phone: EMail: Hannes.Tschofenig@siemens.com URI: Stiemerling, et al. Expires April 17, 2004 [Page 67] Internet-Draft A NAT/FW NSIS NSLP October 2003 Miquel Martin Network Laboratories, NEC Europe Ltd. Kurfuersten-Anlage 36 Heidelberg 69115 Germany Phone: +49 (0) 6221 905 11 16 EMail: lopez@ccrle.nec.de URI: Cedric Aoun Nortel Networks France EMail: cedric.aoun@nortelnetworks.com Stiemerling, et al. Expires April 17, 2004 [Page 68] Internet-Draft A NAT/FW NSIS NSLP October 2003 Appendix A. Interworking of SIP with NSIS NATFW NSLP This document aims at pinpointing the problems of using SIP in nowadays networks, focusing on the problems derived of NAT's, Firewalls and multi-path communications. It is intended to fit in a scenario description that shows the necessity of NSIS, as well as depicting it's requirements. However, note that there are a number of other solutions available. For example the IETF Midcom working group is working on [4]. A.1 The Session Initiation Protocol [20] describes the Session Initiation Protocol, an application-layer control protocol that can establish, modify, and terminate multimedia sessions. This often involves several flows for video and voice, which are transported over new connections. These use of dynamically allocated ports which results in protocol complexity which can not be handled by nowadays NAT's and Firewalls. Session initiation when one or both of the users is behind a NAT is also not possible, given the impossibility to address a private IP over the internet. Moreover, network deployments often allow for different paths per connection and direction, making the setup of the middle boxes even more complicated. The following figure depicts a typical SIP connection: Stiemerling, et al. Expires April 17, 2004 [Page 69] Internet-Draft A NAT/FW NSIS NSLP October 2003 Ernie(192.0.2.1) Bert(192.0.2.2) | | | 1# SIP INVITE | +--------------------------------------->| | | 2# SIP Ringing | |<---------------------------------------+ | | | 3# SIP OK | <-- Call accepted |<---------------------------------------+ | | | 4# SIP ACK | +--------------------------------------->| | | | 5# DATA | |=======================================>| |<=======================================| | | 1# SIP Invite (192.0.2.1:? -> 192.0.2.2:SIP): I Listen on 192.0.2.1:1000 Ernie invites Bert to the conference, and informs it's awaiting media data on port 1000. 2# SIP Ringing (192.0.2.2:SIP -> 192.0.2.1:?): Ringing Bert's phone The ringing simply inplies that there's something sip aware on Berts side, and that it's ringing his phone 3# SIP OK (192.0.2.2:SIP -> 192.0.2.1:?): Call accepted, I listen on 192.0.2.2:2000 This OK means that the Bert took the phone off hook, and thus accepted the call. It also informs Ernie that Bert is awaiting his media data at port 2000 4# SIP ACK (192.0.2.1:? -> 192.0.2.2:SIP): All is fine, start transmitting. ACK means the ports are accepted and the call can start in the slected data ports on both sides. 5# DATA (192.0.2.1:? -> 192.0.2.2:2000 and 192.0.2.2:? -> 192.0.2.1:1000): Voice,image, video.. This is the actual data being transmited. In the above example, SIP is used successfully to establish a communication, which includes negotiating the data ports for the actual transmission. Unfortunatelly, this scheme will not work for more complex setups. Let's now consider one firewall in the data path, be it on Ernie's or Bert's network, or elsewhere in the middle. We assume that the Stiemerling, et al. Expires April 17, 2004 [Page 70] Internet-Draft A NAT/FW NSIS NSLP October 2003 firewall is allowing traffic directed to the SIP port. As to the rest of the ports, a typical setup involves outgoing connections being allowed, and incoming connections being dropped, except for those already established. That is, we allow packets to go out and their replies to come in, but disable all other traffic. In this case, the connection is as follows, for the case of a firewall on Ernie's network: Ernie(192.0.2.1) FW Bert(192.0.2.2) | | | | 1# SIP INVITE | | +--------------------------------------->| | | | | | 2# SIP Ringing | |<---------------------------------------+ | | | | | 3# SIP OK | <-- Call accepted |<---------------------------------------+ | | | | 4# SIP ACK | | +--------------------------------------->| | | | | 5# DATA | | |=======================================>| | |<=======================| | | | Notice how the SIP messages #1 and #4 traverse the firewall, because they are outbound, and how 2# and 3# traverse it too, because they are replies to the connection established at 1#. Notice now how 5# can go outwards, but Bert can not go through the firewall to reach Ernie's port 1000. The reason is the connection is a new one, and the firewall won't allow it through. Bert will now get media from Ernie, but Ernie is never going to get anything from Bert. The call is thus considered unsuccessful. The reason is that the application level port negotiation is never acknowledge by the network-transport layer firewall, which doesn't know what to expect. We would still face the same problem if the connection used a SIP Proxy, for it would only translate names into IP addresses. Let us now assume that we indeed have an application layer firewall, Stiemerling, et al. Expires April 17, 2004 [Page 71] Internet-Draft A NAT/FW NSIS NSLP October 2003 be it by design, or because we load some sort of SIP module to it. The previous case would now work, since the firewall can now understand the packets going through it and open the necessary ports. Still, we cannot assume that SIP signalization packets and the actual data follow the same path. The following figure shows a likely setup. FW+ stands for one or more firewalls: SIP Signalization Path +-----+ /---------------->-----------| FW+ |-------\ | +-----+ | +------+ +------+ +-----+ |Ernie |----|Router| |Bert | +------+ +------+ +-----+ | Data Path +-----+ | \---------------->-----------| FW+ |-------/ +-----+ The SIP packets with the information about the listening ports now travels on the SIP Signalization path, and so the firewalls on that path can read them. The Data, though, is traveling through the Data path, and the firewalls in that path never get to see the Invite and Ok packets. They are thus unable to open the ports. Two issues are arisen here: first, we need on-path signalization unless we already know the path our packets will take; a highly unlikely situation in today's internet. Second, if we patch the firewalls to understand SIP, we will provide any caller with a hole-puncher for the firewall, since SIP is not provisioned with proper authentication mechanism. It is now clear that tight firewalls prevent SIP from successfully working. There is still another obstacle: NATs. NATs provide for a link between two different address spaces, typically connecting a private range network to a public range one. As a consequence, connections going from the inside (usually the private range) are translated using the NAT's public interface address, and the replies are routed back. The public side of the network can only see the NATs public interface, and know nothing of the private network inside. This means computers outside the NAT won't be able to address computers inside the NAT. Let us analyze the SIP example when Ernie is behind a NAT. The following figure depicts a typical session: Stiemerling, et al. Expires April 17, 2004 [Page 72] Internet-Draft A NAT/FW NSIS NSLP October 2003 Ernie(10.0.0.2) (10.0.0.1) NAT (192.0.2.1) Bert(192.0.2.2) | | | | 1# SIP INVITE | | +--------------------------\ | | |----------------->| | | | | | 2# SIP Ringing | | /------------------+ |<-------------------------| | | | | | | 3# SIP OK | <-- Call accepted | /------------------+ |<-------------------------| | | | | | 4# SIP ACK | | +--------------------------\ | | |----------------->| | | | | 5# DATA | | |==========================\ | | |=================>| | | ?<=============| | | | The communication is analogous to the one in the previous examples, except for the fact the NAT is rewriting the source address of the packets as they traverse it. For instance, packet 1# is going from 10.0.0.2:? towards 192.0.2.2:SIP. The NAT box intercepts the message and puts 192.0.2.1:? as the source address and port, with ? being a dynamically picked port, which might be different from the original one 1# used. On the way back, Bert is replyinc to the source of the IP packet, that is, 192.0.2.1, and so, when 2# reaches 192.0.2.1, the NAT know it is a reply from 1#, because it established a NAT binding, and this replaces the destination address, 192.0.2.1:? with 10.0.0.2:? and forwards the packet inside the NAT. As a result, Ernie never knows there is a NAT in his communication path, since he sends and receives packets from 192.0.2.2 normally. This means that the INVITE packet will tell Bert to send data back to 10.0.0.2, a private IP. Once the signalization is finished, and the actual DATA transmission starts, Bert tries to connect to 10.0.0.2, a private IP address, from the internet; The routers don't know how to route this, and the packet is eventually dropped. Stiemerling, et al. Expires April 17, 2004 [Page 73] Internet-Draft A NAT/FW NSIS NSLP October 2003 One possible solution would be for Ernie to know the NAT exists, and already indicate that it listens on 192.0.2.1, and not 10.0.0.2. That, still would not work, since the NAT binding is not performed at the NAT box. A.2 Conclusions The above examples display the inability to use standard SIP through tight firewalls or NATs, and points at the necessity of a secure on-path protocol to negotiate firewall pinholes and NAT bindings. Stiemerling, et al. Expires April 17, 2004 [Page 74] Internet-Draft A NAT/FW NSIS NSLP October 2003 Appendix B. Ad-Hoc networks Some forms of ad-hoc networks exist where trust in the network is not justified. Figure Figure 29 mainly illustrates the problems of malicious NSIS entities graphically: +------------------------------------------+ +--------// | Adhoc | | ISP | Network | | | regular data | | | traffic by +---------+ | | | node A |Malicious| | +-+--------+ | +-------------->+ Node +-----+///-->+ Firewall +-// | ^ | 3 |===========>| 1 | | | +---------+ injected +-+--------+ | | data traffic | | | | | | | | | | +---+-----+ +---------+ | | | + Node | | Node | | | | | 1 | | 2 | | | | +---------+ +---------+ | | | ^ | +--------// | | | +----------+-------------------------------+ | +--+---+ | Node | | A | +------+ Figure 29: Limits of packet filter security An ad-hoc network consists of a number of nodes between the end host (Node A) and the ISP to which Node A wants to get access. Although Node A uses an authentication and key exchange protocol to create a policy rule at the firewall 1 it is still possible for an untrusted node (in this case Node 3) to inject data traffic which will pass Firewall 1 since the data traffic is not authenticated. To prevent this type of threat two approaches are possible. First, a restrictive packet filter limits the capabilities of an adversary. Finally, there is always the option of using data traffic protection. Stiemerling, et al. Expires April 17, 2004 [Page 75] Internet-Draft A NAT/FW NSIS NSLP October 2003 Appendix C. Interworking of Security Mechanisms and NSIS NATFW NSLP TBD Stiemerling, et al. Expires April 17, 2004 [Page 76] Internet-Draft A NAT/FW NSIS NSLP October 2003 Appendix D. Solution approaches in case of missing authorization D.1 Solution Approach: Local authorization from both end points The first approach makes use of local authorization from both end points. If Host A sends a signaling message toward the destination to Middlebox 1 the message will perform the desired action in Network A. Middlebox 1 establishes some state information and forwards the signaling message towards Host B. Signaling message protection between the two access networks might be difficult. A missing trust relationship does not necessarily mean that no security association establishment is possible. The lacking trust disallows Middlebox 1 (or indirectly Host A where the signaling message was initiated) to create packet filters at Middlebox 2. We assume that the NSIS signaling message is allowed to pass the firewall then it finally reaches Host B. Due to the missing authorization no packet filter specific state is created. The filters will be installed later after receiving an authorization from Host B. When Host B returns a confirmation or acknowledgement then Middlebox 2 treats it as an authorization and finally triggers filter creation. The message is then forwarded to Middlebox 1, where filters are either already installed or require an additional confirmation. Finally the signaling message is forwarded to Host A, which can be assured that subsequent data traffic can be transmitted end-to-end from Host A to Host B. The same procedure has to be applied again to signal information for the other direction (Host B to Host A). The following behavior has to be assumed in order for this approach to be applicable: 1. Signaling messages must be allowed to pass firewalls along the path. 2. NSIS signaling must operate in the described manner which could be described as: Install where you have authorization - delay and forward where you have no authorization. This approach suffers from the following drawbacks: 1. Firewalls which block NSIS signaling from external networks or nodes prevent a successful operation. 2. A full roundtrip is required to signal packet filter information. The NSIS signaling message must therefore provide the capability to route signaling message in both direction which might either require state installation at nodes along the path (route pinning) or a stateless version via record-route. Some risk of DoS protection might exist. Stiemerling, et al. Expires April 17, 2004 [Page 77] Internet-Draft A NAT/FW NSIS NSLP October 2003 D.2 Solution Approach: Access Network-Only Signaling The next approach is based on signaling packet filter information by both hosts into the local access network only. An NSIS allows specifying such a behavior by indicating the signaling endpoint with the help of scoping (for example with domain name or a "local network only" flag). Scoping means that the signaling message although addressed to a particular destination IP address terminates somewhere along the path. If packet filters for both directions have to be installed then the signaling messages have to make packet filter installations up- and downstream along the data path. Similar to proposals in the area of QoS signaling some problems are likely to occur. One such problem is that downstream signaling in general causes problems because of asymmetric routes. In particular it is difficult to determine the firewall where the downstream data traffic will enter a network. The problem of triggering downstream reservations is for example described in [12] . Another problem for example is the placement of a firewall or NAT along the path other than in the access network. This would prevent a successful data exchange. The following behavior has to be assumed in order for this approach to be applicable: 1. It must be possible to trigger a signaling message exchange for a downstream signaling message exchange at the firewall where the data traffic enters the network. 2. No other firewalls or NATs are present along the path other than in the access network. This approach suffers from the following drawbacks: 1. To signal policy rules only within the access network (by both end-points) has a number of disadvantage and challenges (see for example [12] ). The complex message processing caused by this approach strongly argues against it although it might sound simple (and even might be simple in restricted environments). 2. Complex topologies might lead to ineffective policy rules (i.e. data traffic hits firewalls hits wrong firewalls). D.3 Solution Approach: Authorization Tokens The last approach is based on some exchanged authorization tokens which are created by an authorized entity (such as the PDP) or by a trusted third party. Both end hosts need to exchange these tokens Stiemerling, et al. Expires April 17, 2004 [Page 78] Internet-Draft A NAT/FW NSIS NSLP October 2003 with protocols such as SIP or HTTP since these protocols are likely to be allowed to bypass the firewall. The basic idea of this approach is to provide an end host, which requests access to the network, with credentials (referred as authorization tokens). These tokens have to possess some properties, namely: 1. They have to be restrictive by including lifetimes, source and destination identifiers, usage indication and more. 2. They have to provide basic replay protection to prevent unauthorized reuse. 3. The have be cryptographically protected to prevent manipulations. 4. There has to be a mechanism to dynamically create them for a specific reason and to distribute them to the end points. 5. It has to be possible to exchange tokens via a trusted third part in cases where no direct communication between the end hosts is possible (due to NAT). 6. The token can be created locally at the network or by a trusted third party. An example of a possible signaling communication could have the following structure: After exchanging the tokens between the two end hosts. Host A would include the received authorization token to the signaling message for Network B. When the signaling message arrives at Middlebox 2 then the token is verified by the token-creating entity. In order to prevent parties from reusing the token timestamps (e.g. token creation, token lifetime, etc.) have to be included. Adding IP address information about Host A would create difficulties in relationship with NATs. Information about Host B might be possible to include in order to limit attacks where a token is lost and reused by a different host for a different purpose. The goal is to restrict the usage of the token for a specific session. The content of the token only needs to be verified by the originator of the token since it only has to be verified locally. Since authorization needs to be linked to the authorized actions, which have to be performed on the packets matching the packet filter, the token may include the associated action or a reference to it. The following behavior has to be assumed in order for this approach to be applicable: 1. The exchange of authorization tokens between end-systems must be possible. These protocols must be allowed to pass the firewalls. 2. An end-system must be able to request such an authorization token Stiemerling, et al. Expires April 17, 2004 [Page 79] Internet-Draft A NAT/FW NSIS NSLP October 2003 at some entity in the local network or at a trusted third party. This approach suffers from the following drawback: 1. Possibly an additional protocol is required for an end host to request an authorization token from an entity in the local network. Stiemerling, et al. Expires April 17, 2004 [Page 80] Internet-Draft A NAT/FW NSIS NSLP October 2003 Appendix E. Acknowledgements We would like to acknowledge Vishal Sankhla and Joao Girao for their input to this draft. Stiemerling, et al. Expires April 17, 2004 [Page 81] Internet-Draft A NAT/FW NSIS NSLP October 2003 Intellectual Property Statement The IETF takes no position regarding the validity or scope of any intellectual property or other rights that might be claimed to 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; neither does it represent that it has made any effort to identify any such rights. Information on the IETF's procedures with respect to rights in standards-track and standards-related documentation can be found in BCP-11. Copies of claims of rights made available for publication 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 implementors or users of this specification can be obtained from the IETF Secretariat. The IETF invites any interested party to bring to its attention any copyrights, patents or patent applications, or other proprietary rights which may cover technology that may be required to practice this standard. Please address the information to the IETF Executive Director. Full Copyright Statement Copyright (C) The Internet Society (2003). 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 assignees. 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 Stiemerling, et al. Expires April 17, 2004 [Page 82] Internet-Draft A NAT/FW NSIS NSLP October 2003 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. Stiemerling, et al. Expires April 17, 2004 [Page 83]