Next Steps in Signaling H. Schulzrinne Internet-Draft Columbia U. Expires: April 19, 2004 R. Hancock Siemens/RMR October 20, 2003 GIMPS: General Internet Messaging Protocol for Signaling draft-ietf-nsis-ntlp-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. Copyright Notice Copyright (C) The Internet Society (2003). All Rights Reserved. Abstract This document specifies protocol stacks for the routing and transport of per-flow signaling messages along the path taken by that flow through the network. The solution uses existing transport and security protocols under a common messaging layer, the Generic Internet Messaging Protocol for Signaling (GIMPS), which provides a universal service for diverse signaling applications. GIMPS does not handle signaling application state itself, but manages its own internal state and the configuration of the underlying transport and security protocols to enable the transfer of messages in both directions along the flow path. The combination of GIMPS and the lower layer protocols provides a solution for the base protocol component of the "Next Steps in Signaling" framework. Schulzrinne & Hancock Expires April 19, 2004 [Page 1] Internet-Draft GIMPS October 2003 Table of Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 2. Requirements Notation and Terminology . . . . . . . . . . . 4 3. Design Methodology . . . . . . . . . . . . . . . . . . . . . 6 3.1 Overall Approach . . . . . . . . . . . . . . . . . . . . . . 6 3.2 Design Attributes . . . . . . . . . . . . . . . . . . . . . 8 4. Overview of Operation . . . . . . . . . . . . . . . . . . . 10 4.1 GIMPS State . . . . . . . . . . . . . . . . . . . . . . . . 10 4.2 Basic Message Processing . . . . . . . . . . . . . . . . . . 12 4.3 Routing State and Messaging Association Maintainance . . . . 15 5. Message Formats and Encapsulations . . . . . . . . . . . . . 19 5.1 Message Formats . . . . . . . . . . . . . . . . . . . . . . 19 5.2 Encapsulation in Datagram Mode . . . . . . . . . . . . . . . 20 5.3 Encapsulation Options in Connection Mode . . . . . . . . . . 20 6. Advanced Protocol Features . . . . . . . . . . . . . . . . . 28 6.1 Route Changes . . . . . . . . . . . . . . . . . . . . . . . 28 6.2 Policy-Based Forwarding . . . . . . . . . . . . . . . . . . 31 6.3 Route Recording . . . . . . . . . . . . . . . . . . . . . . 32 6.4 NAT Traversal . . . . . . . . . . . . . . . . . . . . . . . 32 6.5 Interaction with IP Tunnelling . . . . . . . . . . . . . . . 33 6.6 IPv4-IPv6 Transition and Interworking . . . . . . . . . . . 34 7. Security Considerations . . . . . . . . . . . . . . . . . . 36 7.1 Message Confidentiality and Integrity . . . . . . . . . . . 36 7.2 Peer Node Authentication . . . . . . . . . . . . . . . . . . 37 7.3 Routing State Integrity . . . . . . . . . . . . . . . . . . 37 7.4 Denial of Service Prevention . . . . . . . . . . . . . . . . 38 8. Open Issues . . . . . . . . . . . . . . . . . . . . . . . . 40 8.1 Protocol Naming . . . . . . . . . . . . . . . . . . . . . . 40 8.2 General IP Layer Issues . . . . . . . . . . . . . . . . . . 40 8.3 Encapsulation and Addressing for Datagram Mode . . . . . . . 40 8.4 Intermediate Node Bypass and Router Alert Values . . . . . . 42 8.5 Messaging Association Flexibility . . . . . . . . . . . . . 43 8.6 Messaging Association Setup Message Sequences . . . . . . . 43 8.7 Connection Mode Encapsulation . . . . . . . . . . . . . . . 45 8.8 GIMPS State Teardown . . . . . . . . . . . . . . . . . . . . 45 8.9 Datagram Mode Retries and Single Shot Message Support . . . 45 8.10 GIMPS Support for Message Scoping . . . . . . . . . . . . . 46 8.11 Mandatory or Optional Reverse Routing State . . . . . . . . 46 8.12 Additional Discovery Mechanisms . . . . . . . . . . . . . . 47 8.13 Protocol Design Details . . . . . . . . . . . . . . . . . . 47 Normative References . . . . . . . . . . . . . . . . . . . . 49 Informative References . . . . . . . . . . . . . . . . . . . 50 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . 51 A. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 52 B. Example Routing State Table . . . . . . . . . . . . . . . . 53 C. Connection Mode Messaging Association Configurations . . . . 54 Intellectual Property and Copyright Statements . . . . . . . 55 Schulzrinne & Hancock Expires April 19, 2004 [Page 2] Internet-Draft GIMPS October 2003 1. Introduction Signaling involves the manipulation of state held in network elements. 'Manipulation' could mean setting up, modifying and tearing down state; or it could simply mean the monitoring of state which is managed by other mechanisms. The case of path-coupled signaling involves network elements which are located on the path taken by a particular data flow, possibly including but not limited to the flow endpoints. Indeed, there are almost always more than two participants in a path-coupled-signaling session, although there is no need for every router on the path to participate. Path-coupled signaling thus excludes end-to-end higher-layer application signaling (except as a degenerate case); an example of such application signaling would be ISUP (telephony signaling for Signaling System #7) messages being transported by SCTP between two nodes. In the context of path-coupled signaling, examples of state management include network resource allocation (for "resource reservation"), firewall configuration, and state used in active networking; examples of state monitoring are the discovery of instantaneous path properties (such as available bandwidth, or cumulative queuing delay). Each of these different uses of path-coupled signaling is referred to as a signaling application. Every signaling application involves a set of state management rules, as well as protocol support to exchange messages along the data path. Several aspects of this protocol support are common to all or a large number of applications, and hence should be developed as a common standard. The framework given in [18] provides a rationale for a function split between the common and application specific protocols, and gives outline requirements for the former, the 'NSIS Transport Layer Protocol' (NTLP). This specification provides a concrete solution for the NTLP. It is based on the use of existing transport and security protocols under a common messaging layer, the Generic Internet Messaging Protocol for Signaling (GIMPS). Different signaling applications may make use of different services provided by GIMPS, but GIMPS does not handle signaling application state itself; in that crucial respect, it differs from application signaling protocols such as the control component of FTP, SIP and RTSP. Instead, GIMPS manages its own internal state and the configuration of the underlying transport and security protocols to ensure the transfer of signaling messages on behalf of signaling applications in both directions along the flow path. Schulzrinne & Hancock Expires April 19, 2004 [Page 3] Internet-Draft GIMPS October 2003 2. Requirements Notation and Terminology The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [2]. The terminology used in this specification is fully defined in this section. The basic entities relevant at the GIMPS level are shown in Figure 1. GIMPS (adjacent) peer nodes IP addresses = Signaling IP address Source/Destination Addresses IP address = Flow (depending on signaling direction) = Flow Source | | Destination Address | | Address V V +--------+ Data +------+ +------+ +--------+ | Flow |-----------|------|-------------|------|-------->| Flow | | Sender | Flow | | | | |Receiver| +--------+ |GIMPS |============>|GIMPS | +--------+ | Node |<============| Node | +------+ Signaling +------+ Flow >>>>>>>>>>>>>>>>> = Downstream direction <<<<<<<<<<<<<<<<< = Upstream direction Figure 1: Basic Terminology [Data] Flow: A set of packets following a unique path through the network, identified by some fixed combination of header fields. Only unicast, unidirectional flows are considered. Session: A single application layer flow of information for which some network control state information is to be manipulated or monitored. IP mobility may cause the mapping between sessions and flows to change, and IP multihoming may mean there is more than one flow for a given session. [Flow] Sender: The node in the network which is the source of the packets in a flow. Could be a host or a router (if the flow is actually an aggregate). [Flow] Receiver: The node in the network which is the sink for the packets in a flow. Schulzrinne & Hancock Expires April 19, 2004 [Page 4] Internet-Draft GIMPS October 2003 Downstream: In the same direction as the data flow. Upstream: In the opposite direction to the data flow. GIMPS Node: Any node along the data path supporting GIMPS (regardless of what signaling applications it supports). Adjacent peer: The next GIMPS node along the data path, in the upstream or downstream direction. Whether two nodes are adjacent or not is determined implicitly by the GIMPS peer discovery mechanisms; it is possible for adjacencies to 'skip over' intermediate nodes if they have no interest in the signaling messages being exchanged. Datagram mode: A mode of sending GIMPS messages between nodes without using any transport layer state or security protection. Upstream messages are sent UDP encapsulated directly to the signaling destination; downstream messages are sent towards the flow receiver with a router alert option. Connection mode: A mode of sending GIMPS messages directly between nodes using point to point transport protocols and security associations, i.e. messaging associations (see below). Messaging association: A single connection between two explicitly identified GIMPS adjacent peers, i.e. between a given signaling source and destination address. A messaging association uses a specific transport protocol and known ports, and may be run over specific network layer security associations, or use a transport layer security association internally. A messaging association is bidirectional. Schulzrinne & Hancock Expires April 19, 2004 [Page 5] Internet-Draft GIMPS October 2003 3. Design Methodology 3.1 Overall Approach The generic requirements identified in [18] for transport of path-coupled signaling messages are essentially two-fold: "Routing": Determine how to reach the adjacent signaling node along the data path (the GIMPS peer); "Transport": Deliver the signaling information to that peer. To meet the routing requirement, for downstream signaling the node can either use local state information (e.g. gathered during previous signaling exchanges) to determine the identity of the GIMPS peer explicitly, or it can just send the signaling towards the flow destination address and rely on the peer to intercept it. For upstream signaling, only the first technique is possible. Once the routing decision has been made, the node has to select a mechanism for transport of the message to the peer. GIMPS divides the transport problems into two categories, the easy and the difficult ones. It handles the easy cases within GIMPS itself, avoiding complexity and latency, while drawing on the services of well-understood reliable transport protocols for the harder cases. Here, with details discussed later, "easy" messages are those that are sized well below the lowest MTU along a path, are infrequent enough not to cause concerns about congestion and flow control, and do not need transport or network-layer security support. In addition, in many cases, signaling information needs to be delivered reliably between GIMPS peers. Some applications may implement their own reliability mechanism, but experience with RSVP has shown [12] that relying on soft-state refreshes itself may yield unsatisfactory performance if signaling messages are lost even occasionally. The provision of this type of reliability is also the responsibility of the underlying transport protocols. In [18] all of these routing and transport requirements are assigned to a single notional protocol, the 'NSIS Transport Layer Protocol' (NTLP). The strategy of splitting the transport problem leads to a layered structure for the NTLP, as a specialised GIMPS 'messaging' layer running over standard transport protocols, as shown in Figure 2. Schulzrinne & Hancock Expires April 19, 2004 [Page 6] Internet-Draft GIMPS October 2003 ^^ +-------------+ || | Signaling | || +------------|Application 2| || | Signaling +-------------+ NSIS |Application 1| | Signaling +-------------+ | Application | +-------------+ | Level | | Signaling | | || | |Application 3| | || | +-------------+ | VV | | | =======================|==========|========|======================= ^^ +----------------------------------------------+ || |+-----------------------+ +------------+ | || || GIMPS | | GIMPS | | || || Encapsulation |<<<>>>|Maintainance| | NSIS |+-----------------------+ +------------+ | Transport |GIMPS: messaging layer | Level +----------------------------------------------+ ("NTLP") | | | | || +----+ +----+ +----+ +----+ || |UDP | |TCP | |SCTP| |DCCP|.... VV +----+ +----+ +----+ +----+ ===========================|=======|=======|=======|=============== | | | | +----------------------------------------------+ | IP | +----------------------------------------------+ Figure 2: Protocol Stacks for Signaling Transport For efficiency, GIMPS offers two modes of transport operation: Datagram mode: for small, infrequent messages with modest delay constraints; Connection mode: for larger data objects or where fast setup in the face of packet loss is desirable, or where channel security is required. The datagram mode uses a lower-layer unreliable unsecured datagram transport mechanism, with UDP as the initial choice. The connection mode can use any stream or message-oriented transport protocol, including TCP and SCTP. It may employ specific network layer security associations (using IPsec), or an internal transport layer security association (using TLS). It is possible to combine these two modes along a chain of nodes, Schulzrinne & Hancock Expires April 19, 2004 [Page 7] Internet-Draft GIMPS October 2003 without coordination or manual configuration. This allows, for example, the use of datagram mode at the edges of the network and connection mode in the core of the network. Such combinations may make operation more efficient for mobile endpoints, while allowing multiplexing of signaling messages across shared security associations and transport connections between core routers. It must be understood that the routing and transport decisions made by GIMPS are not totally independent. If the message transfer has performance requirements that enforce the use of connection mode (e.g. because fragmentation is required), this can only be used between explicitly identified nodes. In such cases, the GIMPS node must carry out signaling in datagram mode to identify the peer node and set up the necessary transport connection - even for downstream signaling, the datagram mode option of sending the message in the direction of the flow receiver and relying on interception is not available. In general, the state associated with connection mode messaging to a particular peer (signaling destination address, protocol and port numbers, internal protocol configuration and state information) is referred to as a "messaging association". There may be any number of messaging associations between two GIMPS peers (although the usual case is 0 or 1), and they are set up and torn down by managment actions within the GIMPS layer itself. 3.2 Design Attributes Soft state: All parts of GIMPS state are subject to time-out ("soft-state"). Explicit removal is also logically possible (see Section 8.8. 'State' here includes the transport associations managed by GIMPS. Application-neutral: GIMPS is designed to support the largest range of signaling applications. While a number of such applications have been identified, it appears likely that new applications will emerge. (This was the case after the development of RSVP, for example.) Mobility support: End systems can change their network attachment point and network address during a session. GIMPS minimises the use of IP addresses as identifiers for non-topological information (e.g. authentication state). Efficient: Signaling often occurs before an application such as an IP telephone conversation can commence, so that any signaling delay becomes noticeable to the application. Signaling delays are incurred by the delay in finding signaling nodes along the path Schulzrinne & Hancock Expires April 19, 2004 [Page 8] Internet-Draft GIMPS October 2003 (peer discovery), in retransmitting lost signaling messages and in setting up security associations between nodes, among other factors. GIMPS attempts to minimise these delays by several mechanisms, such as the use of high performance transport protocols to circumvent message loss, and the re-use of messaging associations to avoid setup latency. If discovery is needed it is a lightweight process which only probes local topology, and GIMPS also allows it to be bypassed completely for downstream datagram mode messages. IP version neutral: GIMPS supports both IPv4 and IPv6: it can use either for transport, largely as a result of their support in the underlying transport protocols, and can signal for either type of flow. In addition, GIMPS is able to operate on dual-stack nodes (to bridge between v4 and v6 regions) and also to operate across v4/v6 boundaries and other addressing boundaries. Specific transition issues are considered in Section 6.6. Transport neutral: GIMPS can operate over any message or stream-oriented transport layer, including UDP, DCCP, TCP and SCTP. Messages sent over protocols that do not offer a native fragmentation service, such as UDP, are strictly limited in size and rate to avoid network congestion and loss-amplification problems. Proxy support: The end systems in a session may not be capable of handling either the signaling transport or the application and may instead rely on proxies to initiate and terminate signaling sessions. GIMPS decouples the operation of the messaging functions from the flow source and destination addresses, treating these primarily as data. Scaleable: As will be discussed in Section 4.3, up to one messaging association is generally kept for each next-hop node and thus transport state scales better than the number of sessions. (Many next-hops may not have transport state at all, if there are no messages on sessions visiting those nodes that warrant such treatment.) Transport associations are removed based on policy at each node, depending on trade-offs between fast peer-to-peer communication and state overhead. In short, transport state can be removed immediately after the last signaling session to a particular next-hop is removed, after some delay to wait for new sessions or only if resource demands warrant it. Schulzrinne & Hancock Expires April 19, 2004 [Page 9] Internet-Draft GIMPS October 2003 4. Overview of Operation This section describes the basic structure and operation of GIMPS. It is divided into three parts. Section 4.1 gives an overview of the per-flow and per-peer state that GIMPS maintains for the purpose of routing messages. Section 4.2 describes how messages are processed in the case where any necessary messaging associations and associated routing state already exist; this includes the simple scenario of pure datagram mode operation, where no messaging associations are necessary in the first place (equivalent to the transport functionality of base RSVP as defined in [8].) Section 4.3 describes how routing state is maintained and how messaging associations are initiated and terminated. 4.1 GIMPS State For each flow, the GIMPS layer can maintain message routing state to manage the processing of outgoing messages. This state is conceptually organised into a table with the following structure. The primary key (index) for the table is the combination of {flow routing info, session id, signaling application id}: Flow: the header N-tuple that determines the route taken by the flow through the network; in particular, including the flow destination address. Session: a cryptographically random and (probabilistically) globally unique identifier of the application layer session that is using the flow. For a given flow, different signaling applications may or may not use the same session identifier. Often there will only be one flow for a given session, but in mobility/multihoming scenarios there may be more than one and they may be differently routed. Signaling application: an IANA assigned identifier of the signaling application which is generating messages for this flow. The inclusion of this identifier allows the routing state to be different for different signaling applications (e.g. because of different adjacencies or different messaging requirements). For a given flow and signaling application there can only be a single row in the table. The state information for a given key is as follows: Schulzrinne & Hancock Expires April 19, 2004 [Page 10] Internet-Draft GIMPS October 2003 Upstream peer: the adjacent GIMPS peer closer to the flow source for this flow and signaling application. This could be an IP address (learned from previous signaling) or a pointer to a messaging association. It could also be null, if this node is the flow sender or this node is not storing reverse routing state, or a special value to indicate that this node is the last upstream node (but not the sender). Downstream peer: the adjacent GIMPS peer closer to the flow destination for this flow and signaling application. This could be a pointer to a messaging association, or it could be null, if this node is the flow receiver or this node is only sending downstream datagram mode messages for this flow and signaling application, or a special value to indicate that this node is the last downstream node (but not the receiver). Note that both the upstream and downstream peer state may be null, and the session identifier information is not required for message processing; in that case, no state information at all needs to be stored in the table. Both items of peer identification state have associated timers for how long the identification can be considered accurate; when these timers expire, the peer identification (IP address or messaging association pointer) is purged if it has not been refreshed. An example of a routing state table for a simple scenario is given in Appendix B. Note also that the information is described as a table of flows, but that there is no implied constraint on how the information is stored. For example, in a network using pure destination address routing (without load sharing or any form of policy-based forwarding), the downstream peer information might be possible to store in an aggregated form in the same manner as the IP forwarding table. In addition, many of the per-flow entries may point to the same per-peer state (e.g. the same messaging association) if the flows go through the same adjacent peer. The per-flow message routing state is not the only state stored by GIMPS. There is also the state implied in the messaging associations. Since we assume that these associations are typically per-peer rather than per-flow, they are stored separately. As well as the messaging association state itself, this table also stores per-association information including: o messages pending transmission while an association is being established; o an inactivity timer for how long the association has been idle. Schulzrinne & Hancock Expires April 19, 2004 [Page 11] Internet-Draft GIMPS October 2003 4.2 Basic Message Processing This section describes how signaling application messages are processed in the simple case where any necessary messaging associations and routing state are already in place. The description is divided into several parts: message reception, local processing and delivery to signaling applications, and message transmission. An overview is given in Figure 3. +---------------------------------------------------------+ | Signaling Application Processing | | | +---------------------------------------------------------+ ^ V ^ NSLP Payloads V ^ V +---------------------------------------------------------+ | GIMPS Processing | | | +--x-----------u--d---------------------d--u-----------x--+ x u d d u x x u d>>>>>>>>>>>>>>>>>>>>>d u x x u d Bypass at d u x +--x-----+ +--u--d--+ GIMPS level +--d--u--+ +-----x--+ | C-mode | | D-mode | | D-mode | | C-mode | |Handling| |Handling| |Handling| |Handling| +--x-----+ +--u--d--+ +--d--u--+ +-----x--+ x u d d u x x uuuuuu d>>>>>>>>>>>>>>>>>>>>>d uuuuuu x x u d Bypass at d u x +--x--u--+ +-----d--+ router +--d-----+ +--u--x--+ |IP Host | | RAO | alert level | RAO | |IP Host | |Handling| |Handling| |Handling| |Handling| +--x--u--+ +-----d--+ +--d-----+ +--u--x--+ x u d d u x +--x--u-----------d--+ +--d-----------u--x--+ | IP Layer | | IP Layer | | (Receive Side) | | (Transmit Side) | +--x--u-----------d--+ +--d-----------u--x--+ x u d d u x x u d d u x x u d d u x uuuuuuuuuuuuuu = upstream datagram mode messages dddddddddddddd = downstream datagram mode messages xxxxxxxxxxxxxx = connection mode messages RAO = Router Alert Option Schulzrinne & Hancock Expires April 19, 2004 [Page 12] Internet-Draft GIMPS October 2003 Figure 3: Message Paths through a GIMPS Node Note that the same messages are also used for internal GIMPS state maintenance operations. The state maintenance takes place as a result of processing specific GIMPS payloads in these messages. The processing of these payloads is the subject of Section 4.3. Message Reception: Messages can be received in connection or datagram mode, and from upstream or downstream peers. Reception in connection mode is simple: incoming packets undergo the security and transport treatment associated with the messaging association, and the messaging association provides complete messages to the GIMPS layer for further processing. Reception in datagram mode depends on the message direction. Upstream messages (from a downstream peer) will arrive UDP encapsulated and addressed directly to the receiving signaling node. Each datagram contains a single complete message which is passed to the GIMPS layer for further processing, just as in the connection mode case. Downstream datagram mode messages are UDP encapsulated and addressed to the flow receiver with an IP router alert option to cause interception. The signaling node will therefore 'see' all such messages, including those in which it is not interested. There are then two cases. If the node categorises the message as 'not interesting', it is passed on for message transmission without further processing (criteria and mechanisms for categorisation are discussed in Section 8.4). If the message is determined to be interesting, it is passed up to the GIMPS layer for further processing as in the other cases. Local Processing: Once a message has been received, by any method, it is processed locally within the GIMPS layer. The GIMPS processing to be done depends on the payloads carried; most of the GIMPS-internal payloads are associated with state maintenance and are covered in the next subsection. One GIMPS-internal payload which can be carried in any message and requires processing is the GIMPS hop count. This is decremented on input processing, and checked to be greater than zero on output processing. The primary purpose of the GIMPS hop count is to prevent message looping. The remainder of the GIMPS message consists of an NSLP payload. This is delivered locally to the signaling application identified at the GIMPS level; the format of the NSLP payload is not Schulzrinne & Hancock Expires April 19, 2004 [Page 13] Internet-Draft GIMPS October 2003 constrained by GIMPS, and the content is not interpreted. Signaling applications can generate their messages for transmission, either asynchronously, or in response to an input message. Messages may also be forwarded in the GIMPS layer if there is no appropriate local signaling application to process them. Regardless of the source, outgoing messages are passed downwards for message transmission processing. Message Transmission: When a message is available for transmission, GIMPS uses internal policy and the stored routing state to determine how to handle it. The following processing applies equally to locally generated messages and messages forwarded from within the GIMPS or signaling application levels. The main decision is whether the message must be sent in connection mode or datagram mode. Reasons for using the former could be: * NSLP requirements: for example, the signaling application has requested channel secured delivery, or reliable delivery; * protocol specification: for example, this document could specify that a message that requires fragmentation MUST be sent over a messaging association; * local GIMPS policy: for example, a node may prefer to send messages over a messaging association to benefit from congestion control. In principle, as well as determining that some messaging association must be used, GIMPS could select between a set of alternatives, e.g. for load sharing or because different messaging associations provide different transport or security attributes (see Section 8.5 for further discussion). If a messaging association is not required, the message is sent in datagram mode. The processing in this case depends on whether the message is directed upstream or downstream. * If the upstream peer IP address is available from the per-flow routing table, the message is UDP encapsulated and sent directly to that address. Otherwise, the message cannot be forwarded (i.e. this is an error condition). * In the downstream direction, messages can always be sent. They are simply UDP encapsulated and IP addressed to the flow receiver, with the appropriate router alert option. Schulzrinne & Hancock Expires April 19, 2004 [Page 14] Internet-Draft GIMPS October 2003 If the use of a messaging association is selected, the message is queued on the association (found from the upstream or downstream peer state table), and further output processing is carried out according to the details of the protocol stack used for the association. If no appropriate association exists, the message is queued while one is created (see next subsection). If no association can be created, this is again an error condition. 4.3 Routing State and Messaging Association Maintainance The main responsibility of the GIMPS layer is to manage the routing state and messaging associations which are used in the basic message processing described above. Routing state is installed and maintained by datagram mode messages containing specific GIMPS payloads. Messaging associations are dependent on the existence of routing state, but are actually set up by the normal procedures of the transport and security protocols that comprise the messaging association. Timers control routing state and messaging association refresh and expiration. The complete sequence of possible messages for state setup between adjacent peers is shown in Figure 4 and described in detail in the following text. Schulzrinne & Hancock Expires April 19, 2004 [Page 15] Internet-Draft GIMPS October 2003 +----------+ +----------+ | Querying | |Responding| | Node | | Node | +----------+ +----------+ GIMPS-query -------------------------> Router Alert Option Flow/Session/Sig App ID Querying Node Addressing [Query cookie, response request] [NSLP Payload] GIMPS-response <------------------------- Flow/Session/Sig App ID Query cookie Responder node addressing [Responder cookie, response request] [NSLP Payload] Messaging Association Setup <========================> Final handshake -------------------------> Flow/Session/Sig App ID Responder cookie Querying Node Addressing [NSLP Payload] Figure 4: Message Sequence at State Setup The initial message in any routing state maintenance operation is a downstream datagram mode message, sent from the querying node and intercepted at the responding node. This is encapsulated and addressed just as in the normal case; in particular, it has addressing and other identifiers appropriate for the flow and signaling application that state maintenance is being done for, and it is allowed to contain an NSLP payload. Processing at the querying and responding nodes is also essentially the same. However, the querying node includes additional payloads: its own address information, and optionally a 'discover-query' payload, which contains a response request flag and a query cookie. This message is informally referred to as a 'GIMPS-query', although it is just a normal message with a particular payload set. In the responding node, the GIMPS level processing of the discover-query payload triggers the generation of a 'GIMPS-response' Schulzrinne & Hancock Expires April 19, 2004 [Page 16] Internet-Draft GIMPS October 2003 message. This is also a normally encapsulated and addressed message with particular payloads, this time in the upstream direction. Again, it can contain an NSLP payload (possibly a response to the NSLP payload in the initial message). It includes its own addressing information and the query cookie, and optionally a 'discover-response' payload, which contains another response request flag and a responder cookie. Note that if a messaging association already exists towards the querying node, this can be used to deliver the GIMPS-response message; otherwise, datagram mode is used. The querying node installs the responder address as downstream peer state information after verifying the query cookie in the GIMPS-response. The responding node can install the querying address as upstream peer state information after the receipt of the initial GIMPS-query, or after a third message in the downstream direction containing the responder cookie. The detailed constraints on precisely when state information is installed are driven by security considerations on prevention of denial-of-service attacks and state poisoning attacks, which are discussed further in Section 7. Setup of messaging associations begins when both downstream peer addressing information is available and a new messaging association is actually needed. (In many cases, the GIMPS-response message above will identify a downstream peer for whom an appropriate messaging association already exists, in which case no further GIMPS management is needed.) Setup of the messaging association always starts from the upstream node, but once the querying node has sent any queued signaling application messages it can be used equally in both directions. Refresh and expiration of all types of state is controlled by timers. State in the routing table has a per-flow, per-direction timer, which expires after a routing state lifetime (RSL). It is the responsibility of the querying node to generate a GIMPS-query message, optionally with a discover-query payload, before this timer expires, if it believes that the flow is still active. Receipt of the message at the responding node will refresh upstream peer addressing state, and receipt of a GIMPS-response at the querying node will refresh any downstream peer addressing state if it exists. Note that nodes do not control the refresh of upstream peer state themselves, they are dependent on the upstream peer to manage this. Messaging associations can be managed by either end. Management consists of tearing down unneeded associations. Whether an association is needed is a local policy decision, which could take into account the cost of keeping the messaging association open, the level of past activity on the association, and the likelihood of future activity (e.g. if there are flows still in place which might Schulzrinne & Hancock Expires April 19, 2004 [Page 17] Internet-Draft GIMPS October 2003 generate messages that would use it). Because messaging associations can always be set up on demand, and messaging association status is not made directly visible outside the GIMPS layer, even if GIMPS tears down and later re-establishes a messaging association, signaling applications cannot distinguish this from the case where the association is kept permanently open. Schulzrinne & Hancock Expires April 19, 2004 [Page 18] Internet-Draft GIMPS October 2003 5. Message Formats and Encapsulations 5.1 Message Formats All GIMPS messages begin with a common header which includes a version number field. For messages sent over a messaging association, it may also include transport protocol specific format information as discussed in Section 5.3.1. The remainder of the message is is encoded in an RSVP-style format, i.e., consisting of type-length-value (TLV) objects. A later version of this specification will contain more details on rules for object encodings which enable protocol extensibility. These items are contained in each GIMPS message: Flow routing information: Information sufficient to define the route that the flow will take through the network. Minimally, this could just be the flow destination address; however, to account for policy based forwarding and other issues a more complete set of header fields should be used (see Section 6.2 and Section 6.4 for further discussion). This object also contains a flag to indicate whether the signaling message is being sent upstream or downstream with respect to the flow. Session identifier: The GIMPS session identifier is a long, cryptographically random identifier chosen by the initiator. The length is open, but 128 bits should be more than sufficient to make the probability of collisions orders of magnitude lower than other failure reasons. Signaling application identifier: This describes the signaling application, such as resource reservation or firewall control. GIMPS Hop counter: A hop counter prevents a message from looping indefinitely. (Since messages may get translated between different lower-layer transport protocols, the IP hop count cannot be relied upon.) The following items are optional: Lifetime: The lifetime of a routing state in the absence of refreshes, measured in seconds. Defaults to 30 seconds. Node addressing: Minimally, this is the IP address at which the GIMPS node originating the message can be reached; this will be used to fill in peer routing state. Additional information can be provided on node capabilities and policy on messaging association management, as well as currently available associations. The level Schulzrinne & Hancock Expires April 19, 2004 [Page 19] Internet-Draft GIMPS October 2003 of flexibility required in this field is discussed in Section 8.5. Cookies: A query-cookie is optional in a GIMPS-query message and if present must be echoed in a GIMPS-response; A response-cookie is optional in a GIMPS-response message, and if present must be echoed in the following downstream message. The optional cookies and response request flags make up the discover-query and discover-response payloads. Cookies are X-octets long and need to be designed so that a node can determine the validity of a cookie without keeping state. Message identifier: A four-octet message counter, used to associate GIMPS messages with their confirmations. The initial use for this is in state maintenance exchanges (GIMPS-query/response); possible use to provide a simple one-off reliable exchange for signaling application messages is considered in Section 8.9. NSLP Payload: The payload carries one or more NSLP objects. GIMPS does not interpret the payload content. 5.2 Encapsulation in Datagram Mode Encapsulation in datagram mode is simple. The complete set of GIMPS payloads for a single message is concatenated together with the common header, and placed in the data field of a UDP datagram. UDP checksums should be enabled. Upstream messages are addressed to the adjacent peer, downstream messages are addressed to the flow receiver and encapsulated with a Router Alert Option. Open issues about alternative encapsulations, addressing possibilities, and router alert option value field setting are discussed in Section 8.2, Section 8.3 and Section 8.4 respectively. The source UDP port is selected by the message sender. A destination UDP port should be allocated by IANA. Note that GIMPS may send messages addressed as {flow sender, flow receiver} which could make their way to the flow receiver even if that receiver were GIMPS-unaware. This should be rejected (with an ICMP message) rather than delivered to another application. Therefore, a well-known port would seem to be required. 5.3 Encapsulation Options in Connection Mode Encapsulation in connection mode is more complex, firstly because of the increased demands on transport functionality. This issue is treated in Section 5.3.1. In addition, a consequence of the restriction to the use of existing transport and security protocols is that messaging associations must run between fixed nodes. However, Schulzrinne & Hancock Expires April 19, 2004 [Page 20] Internet-Draft GIMPS October 2003 there are still several options for how the IP packets for the messaging association should be carried on the wire between these nodes. These alternatives have different functionality, flexibility and performance tradeoffs, and are considered in Section 5.3.2 - Section 5.3.4. 5.3.1 General Considerations It is a general requirement of the NTLP defined in [18] that it should be able to support bundling (of small messages), fragmentation (of large messages), and message boundary delineation. Not all transport protocols natively support all these features. SCTP [5] satisfies all requirements. (The bundling requirement is met implicitly by the use of Nagle-like algorithms inside the SCTP stack.) DCCP [6] is message based but does not provide bundling or fragmentation. These could both be provided by an additional (simple) encapsulation above the DCCP level. TCP provides both bundling and fragmentation, but not message boundaries. Again, this could be done with a simple length field embedded in the TCP data stream to mark each message. UDP could be augmented as in the DCCP case. (However, note that it should probably not be used for messages requiring fragmentation, because of UDP's lack of congestion control functionality.) If it is desired to support all these protocol options, it is probably easiest to handle this as the addition of transport protocol specific shim layers at the top of the messaging association, so that the GIMPS layer sees a uniform functionality transport service interface. The control information for these shim layers would be carried inside the GIMPS fixed header. 5.3.2 Raw Encapsulation The simplest encapsulation is to carry the GIMPS and signaling application payloads directly in the transport protocol, and to carry that protocol on the wire in the normal way. This situation is shown in Figure 5; it applies to both upstream and downstream messages. Schulzrinne & Hancock Expires April 19, 2004 [Page 21] Internet-Draft GIMPS October 2003 +---------------------------------------------+ | L2 Header | +---------------------------------------------+ | IP Header | ^ | Source address = signaling source | ^ | Destination address = signaling destination | . +---------------------------------------------+ . | L4 Header | . ^ | (Standard TCP/SCTP/DCCP/UDP header) | . ^ +---------------------------------------------+ . . | GIMPS Fixed Header | . . ^ | (Optional shim fields, depending on L4) | . . ^ Scope of +---------------------------------------------+ . . . security | GIMPS Payloads | . . . protection |(Sequence of TLV Elements, as in Section 6.1)| . . . (depending +---------------------------------------------+ . . . on channel | NSLP Payloads | . . . security | (Opaque to GIMPS) | V V V mechanism +---------------------------------------------+ V V V in use) Figure 5: Raw Messaging Association Encapsulation The advantages and disadvantages of this approach are as follows: +) Practically all transport and security protocols can be used. +) Use of existing protocol stack implementations is simple. +) There is very little protocol overhead. -) In order to do any processing on any of the GIMPS or signaling application payloads, the complete messaging association must be terminated at the processing node. This applies even if the processing to be done is very lightweight (e.g. an address translation) and/or the node does not support the signaling application in question. This property makes it hard to offer useful transport and security properties between signaling application peers which need them, particularly flow control/ reliable delivery and signaling application payload confidentiality and integrity. -) More generally, for a given signaling application on two nodes, there is very little room for flexibility in the NSIS processing at nodes in between. Nodes have to process everything or nothing, igoring the application completely. Schulzrinne & Hancock Expires April 19, 2004 [Page 22] Internet-Draft GIMPS October 2003 -) Signaling messages are only ever delivered between peers established in GIMPS-query/response exchanges. Any route change is not detected until another GIMPS-query/response procedure takes place; in the meantime, signaling messages are misdelivered. The raw encapsulation is most appropriate (has very few disadvantages) in an environment where routes are stable and the signaling nodes are highly homogeneous (all nodes have identical processing requirements). A scenario which illustrates the problems that arise in more heterogeneous environments is given in Appendix C. 5.3.3 Peer-Peer Tunnelled Encapsulation In order to allow intermediate nodes to do limited processing of some GIMPS payloads without interfering with the overall transport and security service being provided by the messaging association, it is necessary to carry some of the payloads outside the messaging association protocols. One method for doing this is shown in Figure 6; it also applies to both upstream and downstream messages. Schulzrinne & Hancock Expires April 19, 2004 [Page 23] Internet-Draft GIMPS October 2003 +---------------------------------------------+ | L2 Header | +---------------------------------------------+ | IP Header | | Source address = signaling source |<---+ | Destination address = signaling destination | | +---------------------------------------------+ | Repeated | Router Alert Option | | IP header +---------------------------------------------+ | | UDP Header | | | (Possibly using datagram mode ports) | | +---------------------------------------------+ | | Mutable GIMPS Header and Payloads | | | (Subset of full GIMPS payload set) | | +---------------------------------------------+ | | IP Header | | ^ | Source address = signaling source |<---+ ^ | Destination address = signaling destination | . +---------------------------------------------+ . | L4 Header | . | (Standard TCP/SCTP/DCCP/UDP header) | .Similar +---------------------------------------------+ .to 'raw' | GIMPS Fixed Header | . case | (Optional shim fields, depending on L4) | . +---------------------------------------------+ . | 'End to end' GIMPS Payloads | . |(Sequence of TLV Elements, as in Section 6.1)| . +---------------------------------------------+ . | NSLP Payloads | . | (Opaque to GIMPS) | V +---------------------------------------------+ V Figure 6: Peer-Peer Tunnelled Messaging Association Encapsulation The 'raw' encapsulation is carried within a tunnel; the outer tunnel encapsulation is similar to datagram mode (except with fixed addresses for both the upstream and downstream directions). Advantages and disadvantages are as follows: +) Any message-based transport and security protocol can be used. +) Intermediate nodes (between the signaling source and destination, which are assumed to be peers processing the full signaling application) can intercept the packet because of the router alert option and process the 'mutable' GIMPS payloads. This allows for lighter-weight intermediate nodes. To avoid disrupting the operation of the transport protocols, this processing should be restricted to operations that can be carried out on a 'fast Schulzrinne & Hancock Expires April 19, 2004 [Page 24] Internet-Draft GIMPS October 2003 signaling path', for example avoiding AAA operations. +) However, reliability and other transport functions continue to run between the 'outer' signaling application peers, and channel security also applies (except of course for the mutable fields) between them. +) In principle, signaling application payloads could also be carried in the 'mutable' area. -) Stream-based transport and security protocols would be hard to use in this way (i.e. ruling out TCP/TLS). -) Integration of existing protocol stacks implementations is more complex. When a message is sent or received at the messaging association endpoints, the mutable field contents have to be provided as ancilliary data somehow. -) The mutable area is vulnerable to spoofing and interception. (However, such attacks can only be launched by on-path nodes.) -) There is some additional protocol overhead, mainly the repeated IP header. (It might be possible to omit this.) -) This encapsulation is 'blind' to route changes as in the raw case. -) If policy-based forwarding is in use between the signaling source and destination, the intermediate nodes traversed may not be the correct ones for the flow path, because the outer IP header does not match the flow packet IP header. (This could be handled with the functionality of Section 6.3.) The application of this method to a heterogeneous signaling scenario is shown in Appendix C. This tunnelled encapsulation is conceptually similar to the use of a new IP option to carry the mutable fields. However, the use of a new IP option format would be problematic for IPv4; for IPv6, we would either have to allow the intermediate nodes to process destination options on seeing the router alert, or use a hop-by-hop option, with similar effects to the IPv4 case. Although the cost is a duplicate IP header, the UDP tunnelling method seems more attractive. 5.3.4 End-to-End Tunnelled Encapsulation In order to get round the need for parallel route change discovery, an alternative tunnel encapsulation can be considered for downstream signaling messages. One method for doing this is shown in Figure 7. (The corresponding upstream encapsulation would be the point-to-point Schulzrinne & Hancock Expires April 19, 2004 [Page 25] Internet-Draft GIMPS October 2003 tunnel mode.) +---------------------------------------------+ | L2 Header | +---------------------------------------------+ | IP Header | | Source address = flow sender |<---+ | Destination address = flow receiver | | +---------------------------------------------+ | Different | Router Alert Option | | IP header +---------------------------------------------+ | | UDP Header | | | (Possibly using datagram mode ports) | | +---------------------------------------------+ | | Mutable GIMPS Header and Payloads | | | (Subset of full GIMPS payload set) | | +---------------------------------------------+ | | IP Header | | ^ | Source address = signaling source |<---+ ^ | Destination address = signaling destination | . +---------------------------------------------+ . | L4 Header | . | (Standard TCP/SCTP/DCCP/UDP header) | .Similar +---------------------------------------------+ .to 'raw' | GIMPS Fixed Header | . case | (Optional shim fields, depending on L4) | . +---------------------------------------------+ . | 'End to end' GIMPS Payloads | . |(Sequence of TLV Elements, as in Section 6.1)| . +---------------------------------------------+ . | NSLP Payloads | . | (Opaque to GIMPS) | V +---------------------------------------------+ V Figure 7: End-to-End Tunnelled Messaging Association Encapsulation The sole change is the use of flow sender and receiver addresses in the outer IP header. This has a number of significant implications compared to the previous case: +) Route changes are automatically detected by messages sent in the downstream direction. The receiving node will determine that the messaging association headers (in fact, the inner IP header) is incorrect and can notify the signaling source that a discovery exchange is needed. Schulzrinne & Hancock Expires April 19, 2004 [Page 26] Internet-Draft GIMPS October 2003 +) Intermediate nodes can still process the 'mutable' GIMPS payloads without breaking end-to-end security and transport properties of the messaging association. +) Again, signaling application payloads could also be carried in the 'mutable' area. +) Policy-based forwarding will be automatically handled in the downstream direction. -) Stream-based transport and security protocols would be hard to use in this way (i.e. ruling out TCP/TLS). In addition, because some messages will be 'lost' from the messaging association on a route change, the protocol must allow for partial reliability (e.g. PR-SCTP [7] rather than full SCTP, also allowing only IPsec and not TLS for channel security.) -) For the same reason, the congestion processing of the transport protocols becomes more complex, because there is an additional source of non-congestive message loss which the transport protocol itself is not aware of. -) Integration of existing protocol stacks implementations is as complex as in the peer-to-peer case, and there is the same additional protocol overhead. The attractiveness of this encapsulation depends on the perceived value of the automatic route change detection. If route change detection is primarily done by tracking events outside GIMPS itself, background rediscovery at a lower rate is no longer a significant overhead, and the advantages over peer-to-peer tunnelling are limited. Schulzrinne & Hancock Expires April 19, 2004 [Page 27] Internet-Draft GIMPS October 2003 6. Advanced Protocol Features 6.1 Route Changes When re-routing takes place in the network, GIMPS and signaling application state needs to be updated for all flows whose paths have changed. The updates to signaling application state may be signaling application dependent: for example, if the path characteristics have actually changed, simply moving state from the old to the new path is not sufficient. Therefore, GIMPS cannot carry out the complete path update processing. Its main responsibility is to detect the route change, update its own routing state consistently, and inform interested signaling applications at affected nodes. 6.1.1 Route Change Detection There are three primary mechanisms for a GIMPS node to detect that the downstream flow path has changed: 1. In trigger mode, a node finds out that the next hop has changed. This is the RSVP trigger mechanism where some form of notification mechanism from the routing table to the protocol handler is assumed. Clearly, in RSVP and GIMPS, this only works if the routing change is local, not if the routing change happens somewhere a few routing hops away. 2. An extended trigger, where the node checks the link-state routing table to discover that the path has changed. This makes certain assumptions on consistency of route computation (but you probably need to make those to avoid routing loops) and only works intra-domain and for OSPF and similar link-state protocols. Where available, this offers the most accurate and expeditious indication of route changes, but requires more access to the routing internals than a typical OS may provide. 3. In probing mode, each GIMPS node periodically repeats the discovery operation. This is similar to RSVP behavior, except that there is an extra degree of freedom since not every message needs to repeat the discovery, depending on the likely stability of routes. All indications are that leaving mobility aside that routes are stable for hours and days, so this may not be necessary on a 30-second interval. (Sender mobility is handled differently issue since the signaling message will visit "virgin" territory, rather than nodes with existing sessions.) For the first two mechanisms, once the change has been detected, the discovery-query/response sequence is then triggered at the detecting node (for the third case this sequence is part of the detection Schulzrinne & Hancock Expires April 19, 2004 [Page 28] Internet-Draft GIMPS October 2003 mechanism anyway). This propagates downstream until it rejoins the old path; the node where this happens may also have to carry out local repair actions. In addition, there are several semi-heuristic techniques to detect that the flow path might have changed, based on monitoring changes in flow or signaling packet behaviour at downstream nodes (these are discussed in [18]). These are mainly implementation options at the detecting node; exploiting them might require an additional GIMPS payload to be sent in the upstream direction with the semantics 'the route for this flow may have changed', hence triggering discovery again. 6.1.2 Local Repair Storms Even where a node has detected a route change in the downstream direction, there are still two possible cases: o the detecting node may be the true crossover router, i.e. the point in the network where old and new paths diverge, or o the path change may actually have taken place upstream of the detecting node (so that this node is no longer on the path at all) In some circumstances, it is hard to distinguish these cases; an example is shown in Figure 8. Here, after the link failure, node A is the true crossover, but nodes B1/C1/D1 may all also initiate local repair operations. A later version of this document will consider how to control this. Schulzrinne & Hancock Expires April 19, 2004 [Page 29] Internet-Draft GIMPS October 2003 xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x +--+ +--+ +--+ x Initial x .|B1|_.......|C1|_.......|D1| x Configuration x . +--+. .+--+. .+--+\. x x . . . . . . x >>xxxxxx . . . . . . xxxxxx>> +-+ . .. .. . +-+ .....|A|/ .. .. .|E|_.... +-+ . . . . . . +-+ . . . . . . . . . . . . . +--+ +--+ +--+ . .|B2|_.......|C2|_.......|D2|/ +--+ +--+ +--+ +--+ +--+ +--+ Configuration .|B1|........|C1|........|D1| after failure . +--+ .+--+ +--+ of D1-E link . \. . \. ./ . . . . . +-+ . .. .. +-+ .....|A|. .. .. .|E|_.... +-+\. . . . . . +-+ >>xxxxxx . . . . . . xxxxxx>> x . . . . . . x x . +--+ +--+ +--+ . x x .|B2|_.......|C2|_.......|D2|/ x x +--+ +--+ +--+ x xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx ........... = physical link topology >>xxxxxxx>> = flow direction _.......... = indicates outgoing link for flow xxxxxx given by local forwarding table Figure 8: A Re-Routing Event 6.1.3 Propagating Route Change Notifications A second problem that may occur with route change detection is that the detecting GIMPS node may not implement all the signaling applications that need to be informed. Therefore, the GIMPS node needs to be able to send a notification back along the unchanged path Schulzrinne & Hancock Expires April 19, 2004 [Page 30] Internet-Draft GIMPS October 2003 to trigger the nearest signaling application aware node to take action. If multiple signaling applications are in use, it will be hard to define when to stop propagating this notification. If the intermediate node bypass capabilities described in Section 4.3, Section 5.3 and Section 8.4 are fully used, one consequence is that this separation between the nodes which do route change detection and signaling application processing no longer occurs, so this problem in its raw form no longer arises. Conversely, the ability of GIMPS to detect route changes by purely local monitoring of forwarding tables is more limited. (This is probably an appropriate limitation of GIMPS functionality. If we need a protocol for distributing notifications about local changes in forwarding table state, a flow signaling protocol is probably not the right starting point.) 6.2 Policy-Based Forwarding Signaling messages almost by definition need to contain address and port information to identify the flow they are signaling for. We can divide this information into two categories: Flow Routing Information: This is the information needed to determine how a flow is routed within the network. Minimally it is the flow destination address, but to handle load sharing, QoS routing, and other forms of policy based forwarding it can be extended to include the full IPv4 5-tuple or IPv6 3-tuple, and possibly traffic class information as well. It is carried as an object in each GIMPS message (see Section 5.1). Additional Packet Classification Information: This is any further higher layer information needed to select a subset of packets for special treatment by the signaling application. The need for this is highly signaling application specific, and so this information is invisible to GIMPS (if indeed it exists). The correct pinning of signaling messages to the data path depends on how well the downstream messages in datagram mode can be made to be routed correctly. Two strategies are used: The messages themselves have the flow destination address, and possibly source address and traffic class (see Section 8.3 for further discussion). In many cases, this will cause the messages to be routed correctly even by GIMPS unaware nodes. A GIMPS aware node carrying out policy based forwarding on higher layer identifiers (in particular, the protocol and port numbers for IPv4) should take into account the entire flow routing Schulzrinne & Hancock Expires April 19, 2004 [Page 31] Internet-Draft GIMPS October 2003 information object in selecting the outgoing interface rather than relying on the IP layer. 6.3 Route Recording The basic procedures of Section 4.3 describe how messages between adjacent GIMPS peers are used to fill in address information for use in subsequent message exchanges. When GIMPS messages only go between adjacent peers, this is all that is needed. However, the more flexible encapsulations of Section 5.3.3 and Section 5.3.4 allow signaling messages to be sent via intermediate GIMPS nodes which do some limited processing of GIMPS or signaling application payloads. It is necessary to ensure that the correct sequence of intermediate nodes is traversed, in the downstream direction and possibly also in the upstream direction. (For example, if the intermediate node is a NAT, payload translations must be made for both directions of signaling message.) However, this may not happen automatically, in the downstream direction because of policy based forwarding, and in the upstream direction because of asymmetric routing. One solution for this problem is to generalise the notion of the upstream/downstream peer address into a sequence of addresses, i.e. a recorded route. This route can be built up in GIMPS-query messages: nodes which need to be kept on the signaling path but which do not wish to maintain per-flow forwards or reverse routing state can append their outgoing address to a GIMPS 'route record' payload, a generalisation of the 'peer addressing information' object of Section 5.1. This is then stored at the receiver in place of the upstream peer address. The GIMPS-response message sent back can be source routed using this, and can gather another route record of the upstream path which replaces the downstream peer address at the querying node. The address manipulations at intermediate nodes are very similar to source routing in IPv4 or IPv6, and it might be possible to use these underlying protocol features instead of a specific GIMPS function. 6.4 NAT Traversal A already noted, GIMPS messages must carry packet addressing and higher layer information as payload data in order to define the flow signalled for. (This applies to all GIMPS messages, regardless of how they are encapsulated or which direction they are travelling in.) At an addressing boundary the data flow packets will have their headers translated; if the signaling payloads are not likewise translated, the signaling messages will refer to incorrect (and probably Schulzrinne & Hancock Expires April 19, 2004 [Page 32] Internet-Draft GIMPS October 2003 meaningless) flows after passing through the boundary. The simplest solution to this problem is to require that a NAT is GIMPS aware, and to allow it to update/rewrite the flow routing information payload described in Section 6.2. (This is making the implicit assumption that NATs only rewrite the header fields included in this payload, and not higher layer identifiers.) Provided this is done consistently with the data flow header translation, signaling messages will be valid each side of the boundary, without requiring the NAT to be signaling application aware. For this reason, the flow routing information is a candidate to be carried in the 'mutable' area if either tunnelled encapsulation for connection mode messages is used (Section 5.3.3 and Section 5.3.4), since then the NAT does not even have to terminate the transport association protocols. It does raise security issues about unauthorised modifications to this payload. The case of traversing a GIMPS unaware NAT is for further study. There is a dual problem of whether the GIMPS peers either side of the boundary can work out how to address each other, and whether they can work out what translation to apply to the flow routing information from what is done to the signaling packet headers. The fundamental problem is that GIMPS mssages contain 3 or 4 interdependent addresses which all have to be consistently translated, and existing generic NAT traversal techniques such as STUN [17] can process only two. 6.5 Interaction with IP Tunnelling The interaction between GIMPS and IP tunnelling is very simple. An IP packet carrying a GIMPS message is treated exactly the same as any other packet with the same source and destination addresses: in other words, it is given the tunnel encapsulation and forwarded with the other data packets. Tunnelled packets will not have the router alert option or the standard GIMPS protocol encapsulation (e.g. port numbers). They will not be identifiable as GIMPS messages until they leave the tunnel again. If signaling is needed for the tunnel itself, this has to be initiated as a separate signaling session by one of the tunnel endpoints - that is, the tunnel counts as a new flow. Because the relationship between signaling for the 'microflow' and signaling for the tunnel as a whole will depend on the signaling application in question, we are assuming that it is a signaling application responsibility to be aware of the fact that tunnelling is taking place and to carry out additional signaling if necessary; in other words, one tunnel endpoint must be signaling application aware. In some cases, it is the tunnel exit point (i.e. the node where Schulzrinne & Hancock Expires April 19, 2004 [Page 33] Internet-Draft GIMPS October 2003 tunnelled data and downstream signaling packets leave the tunnel) that will wish to carry out the tunnel signaling, but this node will not have knowledge or control of how the tunnel entry point is carrying out the data flow encapsulation. This information could be carried as additional data (an additional GIMPS payload) in the tunnelled signaling packets if the tunnel entry point was at least GIMPS aware. This payload would be the GIMPS equivalent of the RSVP SESSION_ASSOC object of [9]. Whether this functionality should really be part of GIMPS and if so how the payload should be handled will be considered in a later version. 6.6 IPv4-IPv6 Transition and Interworking GIMPS itself is essentially IP version neutral (version dependencies are isolated in the formats of certain GIMPS payloads, and GIMPS also depends on the version independence of the protocols that underlie messaging associations). In mixed environments, GIMPS operation will be influenced by the IP transition mechanisms in use. This section provides a high level overview of how GIMPS is affected, considering only the currently predominant mechanisms. Dual Stack: (This applies both to the basic approach described in [11] as well as the dual-stack aspects of more complete architectures such as [21].) In mixed environments, GIMPS should use the same IP version as the flow it is signaling for; hosts which are dual stack for applications and routers which are dual stack for forwarding should have GIMPS implementations which can support both IP versions. In theory, for some connection mode encapsulation options, a single messaging association could carry signaling messages for flows of both IP versions, but the saving seems of limited value. The IP version used in datagram mode is closely tied to the IP version used by the data flow, so it is intrinsically impossible for a IPv4-only or IPv6-only GIMPS node to support signaling for flows using the other IP version. Applications with a choice of IP versions might select a version for which GIMPS support was available in the network, which could be established by running parallel discovery procedures. In theory, a GIMPS message related to a flow of one IP version could flag support for the other; however, given that IPv4 and IPv6 could easily be separately routed, the correct GIMPS peer for a given flow might well depend on IP version anyway, making this flagged information irrelevant. Schulzrinne & Hancock Expires April 19, 2004 [Page 34] Internet-Draft GIMPS October 2003 Packet Translation: (Applicable to SIIT [4] and NAT-PT [10].) Some transition mechanisms allow IPv4 and IPv6 nodes to communicate by placing packet translators between them. From the GIMPS perspective, this should be treated essentially the same way as any other NAT operation (e.g. between 'public' and 'private' addresses) as described in Section 6.4. In other words, the translating node needs to be GIMPS aware; it will run GIMPS with IPv4 on some interfaces and with IPv6 on others, and will have to translate the flow routing information payload between IPv4 and IPv6 formats for flows which cross between the two. The translation rules for the fields in the payload (including e.g. traffic class and flow label) are as defined in [4]. Tunnelling: (Applicable to 6to4 [13] and a whole host of other tunnelling schemes.) Many transition mechanisms handle the problem of how an end to end IPv6 (or IPv4) flow can be carried over intermediate IPv4 (or IPv6) regions by tunnelling; the methods tend to focus on minimising the tunnel administration overhead. From the GIMPS perspective, the treatment should be as similar as possible to any other IP tunnelling mechanism, as described in Section 6.5. In particular, the end to end flow signaling will pass transparently through the tunnel, and signaling for the tunnel itself will have to be managed by the tunnel endpoints. However, additional considerations may arise because of special features of the tunnel management procedures. For example, [14] is based on using an anycast address as the destination tunnel endpoint. It might be unwise to carry out signaling for the tunnel to such an address, and the GIMPS implementation there would not be able to use it as a source address for its own signaling messages (e.g. GIMPS-responses). Further analysis will be contained in a future version of this specification. Schulzrinne & Hancock Expires April 19, 2004 [Page 35] Internet-Draft GIMPS October 2003 7. Security Considerations The security requirement for the GIMPS layer is to protect the signaling plane against identified security threats. For the signaling problem as a whole, these threats have been outlined in [19]; the NSIS framework [18] assigns a subset of the responsibility to the NTLP. The main issues to be handled can be summarised as: Message Protection: Signaling message content should be protected against interception, modification etc. while in transit. This applies both to GIMPS payloads, and GIMPS should also provide such protection as a service to signaling applications between adjacent peers. State Integrity Protection: It is important that signaling messages are delivered to the right nodes (and not delivered to the wrong ones). GIMPS uses internal state to decide how to route signaling messages, and this state needs to be protected against corruption. Prevention of Denial of Service Attacks: GIMPS nodes and the network have finite resources (state storage, processing power, bandwidth). The protocol should try to minimise exhaustion attacks against these resources and not allow GIMPS nodes to be used to launch attacks on other network elements. The main missing issue is handling authorisation for executing signaling operations (e.g. allocating resources). This is assumed to be done in each signaling application. In many cases, GIMPS relies on the security mechanisms available in messaging associations to handle these issues, rather than introducing new security measures. Obviously, this requires the interaction of these mechanisms with the rest of the GIMPS protocol to be understood and verified. 7.1 Message Confidentiality and Integrity GIMPS can use messaging association functionality, such as TLS or IPsec, to ensure message confidentiality and integrity. In many cases, confidentiality of GIMPS information itself is not likely to be a prime concern, in particular since messages are often sent to parties which are unknown ahead of time. Signaling applications may have their own mechanism for securing content as necessary; however, they may find it convenient to rely on protection provided by messaging associations, particularly if this is provided efficiently and if it runs unbroken between signaling application peers. Schulzrinne & Hancock Expires April 19, 2004 [Page 36] Internet-Draft GIMPS October 2003 7.2 Peer Node Authentication Cryptographic protection (of confidentiality or integrity) typically requires session keys, which can established during authentication exchanges based on shared secrets or public key techniques. Authentication and key agreement is possible using the protocols associated with the messaging association being secured (TLS incorporates this functionality directly, IKE provides it for IPsec). GIMPS nodes rely on these protocols to authenticate the identity of the next hop, and GIMPS has no authentication capability of its own. However, with discovery, there are few effective ways to know what is the legitimate next or previous hop as opposed to an impostor. In other words, authentication here only provides assurance that a node is 'who' it is (i.e. the legitimate owner of identity in some namespace), not 'what' it is (i.e. the correct peer to carry out signaling with for a particular flow). Authentication provides only limited protection, in that a known peer is unlikely to lie about its role. Additional methods of protection against this type of attack are considered in Section 7.3 below. 7.3 Routing State Integrity The internal state in a node (see Section 4.1), specifically the upstream and downstream peer identification, is used to route messages along the data flow path. If this state is corrupted, signaling messages may be misdirected. The routing state table is the local GIMPS view of what routes are being taken by flows through the network. Since routes are only weakly secured (e.g. there is usually no cryptographic binding of a flow to a route), and there is no other authoritative information about flow routes than the current state of the network itself. Therefore, consistency between GIMPS and network routing state has to be ensured by directly interacting with the routing mechanisms to ensure that the upstream and downstream signaling peers are the appropriate ones for any given flow. A good overview of security issues and techniques in this sort of context is provided in [20]. Downstream peer identification is installed and refreshed only on receiving a GIMPS-reponse message. This must echo the cookie from a previous GIMPS-query message, which will have been sent downstream along the flow path (in datagram mode, i.e. end-to-end addressed). Hence, only the true next peer or an on-path attacker will be able to generate such a message, provided freshness of the cookie can be checked at the querying node. Upstream peer identification can be installed directly on receiving a Schulzrinne & Hancock Expires April 19, 2004 [Page 37] Internet-Draft GIMPS October 2003 GIMPS-query message containing addressing information for the upstream peer. However, any node in the network could generate such a message (indeed, almost any node in the network could be the genuine upstream peer for a given flow). To protect against this, two strategies are possible: Filtering: the receiving node may be able to reject signaling messages which claim to be for flows with flow source addresses which would be ruled out by ingress filtering. An extension of this technique would be for the receiving node to monitor the data plane and to check explicitly that the flow packets are arriving over the same interface and if possible from the same link layer neighbour as the datagram mode signaling packets. (If they are not, it is likely that at last one of the signaling or flow packets is being spoofed.) Signaling applications should only install state on the route taken by the signaling itself. Authentication: the receiving node may refuse to install upstream state until it has handshaked by some means with the upstream peer. This handshaking could be as simple as requesting the upstream peer to echo the response cookie in the discover-response payload of a GIMPS-response message (to discourage nodes impersonating upstream peers from using forged source addresses); or, it could be full peer authentication within the messaging association, the reasoning being that an authenticated peer can be trusted not to pretend to be on path when it is not. The second technique also plays a role in denial of service prevention, see below. In practice, a combination of both techniques may be appropriate. 7.4 Denial of Service Prevention GIMPS is designed so that each connectionless discovery message only generates at most one response, so that a GIMPS node cannot become the source of a denial of service amplification attack. However, GIMPS can still be subjected to denial-of-service attacks where an attacker using forged source addresses forces a node to establish state without return routability, causing a problem similar to TCP SYN flood attacks. There are two types of state attacks and one computational resource attack. In the first state attack, an attacker floods a node with messages that the node has to store until it can determine the next hop. If the destination address is chosen so that there is no GIMPS-capable next hop, the node would accumulate messages for several seconds until the discovery retransmission attempt times out. The second type of state-based attack causes GIMPS state to be established by bogus messages. A related Schulzrinne & Hancock Expires April 19, 2004 [Page 38] Internet-Draft GIMPS October 2003 computational/network-resource attack uses unverified messages to cause a node to make AAA queries or attempt to cryptographically verify a digital signature. (RSVP is vulnerable to this type of attack.) There are at least three defenses against these attacks: 1. The responding node does not establish a session or discover its next hop on receiving the GIMPS-query message, but can wait for a setup message on a reliable channel. If the reliable channel exists, the additional delay is a one one-way delay and the total is no more than the minimal theoretically possible delay of a three-way handshake, i.e., 1.5 node-to-node round-trip times. The delay gets significantly larger if a new connection needs to be established first. 2. The response to the initial discovery message contains a cookie. The previous hop repeats the discovery with the cookie included. State is only established for messages that contain a valid cookie. The setup delay is also 1.5 round-trip times. (This mechanism is similar to that in SCTP [5].) 3. If there is a chance that the next-hop node shares a secret with the previous hop, the sender could include a hash of the session ID and the sender's secret. The receiver can then verify that the message was likely sent by the purported source. This does not scale well, but may work if most nodes tend to communicate with a small peer clique of nodes. (In that case, however, they might as well establish more-or-less permanent transport sessions with each other.) These techniques are complementary; we chose a combination of the first and second method. Once a node has decided to establish routing state, there may still be transport and security state to be established between peers. This state setup is also vulnerable to additional denial of service attacks. GIMPS relies on the lower layer protocols that make up messaging associations to mitigate such attacks. The current description assumes that the upstream node is always the one wishing to establish a messaging association, so it is typically the downstream node that is protected. Extensions are considered in Section 8.6; these would require further analysis. Schulzrinne & Hancock Expires April 19, 2004 [Page 39] Internet-Draft GIMPS October 2003 8. Open Issues 8.1 Protocol Naming Alternate names: GIST: Generic Internet Signaling Transport GIMPS: Generic Internet Messaging Protocol for Signaling LUMPS: Lightweight Universal Messaging for Path associated Signaling There is a danger of some ambiguity as to whether the protocol name refers to the complete transport stack below the signaling applications, or only to the additional protocol functionality above the standard transport protocols (UDP, TCP etc.) The NSIS framework uses the term NTLP for the first, but this specification uses the GIST/variants names for the second (see Figure 2 in Section 3.1). In other words, this specification proposes to meet the requirements for NTLP functionality by layering GIMPS/... over existing standard transport protocols. It isn't clear if additional terminological surgery is needed to make this clearer. 8.2 General IP Layer Issues Some NSIS messages have to be addressed end-to-end but intercepted at intermediate routers, and this imposes some special constraints on how they can be encapsulated. RSVPv1 [8] primarily uses raw IP with specific protocol number (46); a UDP encapsulation is also possible for hosts unable to perform raw network i/o. RSVP aggregation [15] uses an additional protocol number (134) to bypass certain interior nodes. The critical requirements for the encapsulation at this level are that routers should be able to identify signaling packets for processing, and that they should not mis-identify packets for 'normal' end-to-end user data flows as signaling packets. The current assumption is that UDP encapsulation can be used for such messages, by allocating appropriate (new) value codes for the router alert option (RAO) [1][3] to identify NSIS messages. Specific open issues about how to allocate such values are discussed in Section 8.4. An alternative approach would be to use raw IP with the RSVP protocol numbers and a new RSVP version number. 8.3 Encapsulation and Addressing for Datagram Mode The discussion in Section 4 essentially assumes that datagram mode messages are UDP encapsulated. This leaves open the question of whether other encapsulations are possible, and exactly how these messages should be addressed. Schulzrinne & Hancock Expires April 19, 2004 [Page 40] Internet-Draft GIMPS October 2003 As well as UDP/IP (and raw IP as discussed and temporarily ruled out in Section 8.2), DCCP/IP and UDP/IPsec could also be considered as 'datagram' encapsulations. However, they still require explicit addressing between GIMPS peer nodes and some per-peer state to be set up and maintained. Therefore, it seems more appropriate to consider these encapsulation options as possible messaging association types, for use where there is a need for congestion control or security protection but without reliability. This would leave UDP/IP as the single encapsulation allowed for all datagram mode messages. Addressing for upstream datagram mode messages is simple: the IP source address is the signaling source address, and the IP destination address is the signaling destination address (compare Figure 1). For downstream datagram mode messages, the IP destination address will be the flow destination address, but the IP source address could be either of the flow source address or signaling source address. Some of the relative merits of these options are as follows: o Using the flow source address makes it more likely that the message will be correctly routed through any intermediate NSIS-unaware region which is doing load sharing or policy routing on the {source, destination} address pair. If the signaling source address is used, the message will be intercepted at some node closer to the flow destination, but it may not be the same as the next node for the data flow packets. o Conversely, using the signaling source address means that ICMP error messages (specifically, unreachable port or address) will be correctly delivered to the message originator, rather than being sent back to the flow source. Without seeing these messages, it is very difficult for the querying node to recognise that it is the last NSIS node on the path. In addition, using the signaling source address may make it possible to exchange messages through GIMPS unaware NATs (although it isn't clear how useful the resulting messages will be, see Section 6.4). It is not clear which of these situations it is more important to handle correctly and hence which source addressing option to use. (RSVP uses the flow source address, although this is primarily for multicast routing reasons.) A conservative approach would be to allow both, possibly even in parallel (although this might open up the protocol to amplification attacks). As well as the addressing, downstream datagram mode messages could be given the same traffic class (and, in the case of IPv6, flow label) as data flow messages. Doing this would increase the faithfulness of the routing of such messages. The traffic class may not be strictly Schulzrinne & Hancock Expires April 19, 2004 [Page 41] Internet-Draft GIMPS October 2003 appropriate for signaling; however, in many cases the bulk of the signaling messages will be sent in connection mode (for which a different traffic class can be freely chosen). 8.4 Intermediate Node Bypass and Router Alert Values We assume that the primary mechanism for intercepting messages is the use of the RAO. The RAO contains a 16 bit value field, within which 35 values have currently been assigned by IANA. It is open how to assign values for use by GIMPS messages to optimise protocol processing, i.e. to minimise the amount of slow-path processing that nodes have to carry out for messages they are not actually interested in the content of. There are two basic reasons why a GIMPS node might wish to ignore a message: o because it is for a signaling application that the node does not process; o because even though the signaling application is present, the node is only processing signaling messages at the aggregate level and not for individual flows (compare [15]). Once a message is ignored and forwarded onwards towards the next node, the current node has essentially dropped out of the signaling path for this flow and signaling application. Its address will not be discovered and used for upstream datagram mode messages or any connection mode messages. Some or all of this information could be encoded in the RAO value field, which would then allow these messages to be filtered on the fast path. (Even if the information is encoded deeper into the message, such as in the GIMPS fixed header, it may still be possible to process it on the fast path, and the semantics of the node dropping out of the signaling path are the same however the filtering is done. However, using the RAO seems the most efficient method if this is possible.) It is not clear whether the signaling application and aggregation level should be directly identified in the RAO value, or whether IANA should allocate special values for 'popular' applications or groups of applications or network configurations. The former is most flexible but is liable to be expensive in terms of value allocations. There is a special consideration in the case of the aggregation level. In this case, whether a message should be processed depends on the network region it is in. There are then two basic possibilities: Schulzrinne & Hancock Expires April 19, 2004 [Page 42] Internet-Draft GIMPS October 2003 1. All routers have essentially the same algorithm for which messages they process, i.e. all messages at aggregation level 0. However, messages have their aggregation level incremented on entry to an aggregation region and decremented on exit. 2. Routers process messages only above a certain aggregation level and ignore all others. The aggregation level of a message is never changed; signaling messages for end to end flows have level 0, but signaling messages for aggregates are generated with a higher level. The first technique requires aggregating/deaggregating routers to be configured as being at a particular aggregation level, and also requires consistent message rewriting at these boundaries. The second technique eliminates the rewriting, but requires interior routers to be configured also. It is not clear what the right trade-off between these options is. 8.5 Messaging Association Flexibility The language of Section 4 is mainly based on the use of 0 or 1 messaging associations between any pair of GIMPS nodes. However, logically it would be quite possible to have more than one association, for example: o to allow different reliability characteristics; o to provide different levels of security protection or to have security separation between different signaling streams; o even simply to have load split between different connections according to priority (so there could be two associations with identical protocol stacks). It is possible to imagine essentially infinite flexibility in these options, both in terms of how many possibilities are allowed and how nodes signal their capabilities and preferences, without much changing the overall GIMPS structure. (The GIMPS-query and GIMPS-response messages described in section Section 4.3 can be used to exchange this information.) What is not clear is how much flexibility is actually needed. 8.6 Messaging Association Setup Message Sequences The discussion of Section 4.3 assumes a simple fixed message sequence, which we can picture as follows: Schulzrinne & Hancock Expires April 19, 2004 [Page 43] Internet-Draft GIMPS October 2003 +---------------------------------+---------------------------------+ | Direction | Message | +---------------------------------+---------------------------------+ | ---> | GIMPS-query message | | | | | <--- | GIMPS-response message | | | | | ===> | Querying node initiates | | | messaging association setup | | | messages | | | | | <--> | Signaling messages exchanged | +---------------------------------+---------------------------------+ There are several variants which could be considered at this level, for example whether the messaging association could be set up by the responding node: +---------------------------------+---------------------------------+ | Direction | Message | +---------------------------------+---------------------------------+ | ---> | GIMPS-query message | | | | | <=== | Responding node initiates | | | messaging association setup | | | messages | | | | | <--> | Signaling messages exchanged | +---------------------------------+---------------------------------+ This saves a message but may be vulnerable to additional denial of service attacks. Another open area is how the responding node propagates the signaling message downstream. It could initiate the downstream discovery process as soon as it received the initial GIMPS-query message, or it could wait until the first signaling application message has been received (which might not be until a messaging association has been established). A similar timing question applies to when it should initiate its own downstream messaging associations. It is possible that all these options are simply a matter for implementation freedom, although leaving them open will make mobility and re-routing behaviour rather harder to analyse, and again there are denial of service implications for some approaches (see Section 7.4). A final open area is how to manage the protocol exchanges that take place in setting up the messaging association itself. It is probably an implementation matter to consider whether to carry out, for Schulzrinne & Hancock Expires April 19, 2004 [Page 44] Internet-Draft GIMPS October 2003 example, the SCTP 4-way handshake only after IKE exchanges (for IPsec SA initialisation) have completed, or whether these can be done partly in parallel. A more radical step is to carry the initial request and response messages of both exchanges as payloads in the GIMPS-query/response exchange, with the request message initially formatted by the querying node with an unspecified destination IP address. This would require modifications to the protocol implementations (especially at the querying node) similar to what is needed for NAT traversal; it would have to be evaluated whether this was worth the one or two round trip times that are saved. 8.7 Connection Mode Encapsulation Section 5.3.2 - Section 5.3.4 present 3 options for encapsulation in connection mode. Allowing all three would be a significant complexity cost, especially given the interaction between encapsulation mode and feasible protocols. If one option has to be chosen, point-to-point tunnelling currently seems most attractive. There might be efficiency saving in being able to use raw mode in the core of the networks, for example on single-hop interdomain links where these intermediate node problems do not arise. The absence of intermediate nodes could be determined during the GIMPS-query/ response exchange, and then raw mode encapsulation chosen only when possible. 8.8 GIMPS State Teardown The description of Section 4.3 provides for GIMPS state (per-flow routing state and per-peer messaging association state) to be removed on timer expiry; routing state can also be replaced (updated). This is the fundamental technique. An additional possibility would be to have explicit removal, i.e. a protocol mechanism to tear down GIMPS state immediately for a particular flow. On recieving such a message, a GIMPS node would clear routing entries and possibly take down messaging associations that were no longer used. Even if one peer indicates that routing state is no longer required, the receiving GIMPS node would have to ensure that no other peers (e.g. supporting different signaling applications) might generate messages of their own still needing the state. In addition, it is not clear how useful it is to remove GIMPS state promptly, since maintaining it only requires table storage without retention of any actual network resources. 8.9 Datagram Mode Retries and Single Shot Message Support The GIMPS-query and GIMPS-response messages may suffer from message Schulzrinne & Hancock Expires April 19, 2004 [Page 45] Internet-Draft GIMPS October 2003 loss (e.g. due to congestion or corruption). Because a successful handshake is necessary before a messaging association can even be initiated, GIMPS must provide its own recovery method for these cases. A working assumption is that the querying node can repeat the GIMPS-query with an exponential backoff until a response is received or some retry threshold is reached. The same technique could be used by the GIMPS layer to provide a 'low-cost' reliable message transfer service, restricted to short messages, without incurring the cost of setting up a messaging association. (If a messaging association exists, it will often be cheaper to discover and use that.) Providing such a service would require some minor extensions to the basic GIMPS protocol. It isn't clear if this additional option fills an important gap in the spectrum between datagram and connection mode message transfer. 8.10 GIMPS Support for Message Scoping Many signaling applications are interested in sending messages over a specific region of the network. Message scoping of this nature seems to be hard to achieve in a topologically robust way, because such region boundaries are not well defined in the network layer. It may be that the GIMPS layer can assist such scoping, by detecting and counting different types of nodes in the signaling plane. The simplest solution would be to count GIMPS nodes supporting the relevant signaling application - this is already effectively done by the GIMPS hop count. A more sophisticated approach would be to track the crossing of aggregation region boundaries, as introduced in Section 8.4. Whether this is plausible depends on the willingness of operators to configure such boundary information in their routers. 8.11 Mandatory or Optional Reverse Routing State Reverse routing state (i.e. the upstream peer addressing information in the routing state table) is installed per-flow when receiving a downstream datagram mode message containing an addressing information payload for the signaling source. Technically, the presence of this payload (and hence the installation of the state) is optional. This allows for very lightweight sending of multi-hop downstream signaling messages (even all the way from flow sender to flow receiver) because no state needs to be installed or managed by GIMPS at the intermediate nodes. However, this rules out the possibility of any downstream node sending signaling responses (including error messages) directly upstream; they have to be sent via the flow endpoints, leading to additional processing there, as well as more complex security considerations. Schulzrinne & Hancock Expires April 19, 2004 [Page 46] Internet-Draft GIMPS October 2003 It is possible that the requirement for lightweight intermediate nodes can be better matched by using one of the tunnelled encapsulations described in Section 5.3. This would allow for a restricted subset of processing at the intermediate nodes, while still allowing the use of bidirectional communication between the 'outer' GIMPS peers, including reverse routing state at the downstream one. This would eliminate the complexity of considering reverse routing state maintenance as optional. 8.12 Additional Discovery Mechanisms The routing state maintenance procedures described in Section 4.3 are strongly focussed on the problem of discovering, implicitly or explicitly, the neighbouring peers on the flow path - which is the necessary functionality for path-coupled signaling. As well as the GIMPS-query/response discovery mechanism, other techniques may sometimes also be possible. For example, in many environments, a host has a single access router, i.e. the downstream peer (for outgoing flows) and the upstream peer (for incoming ones) are known a priori. More generally, a link state routing protocol database can be analysed to determine downstream peers in more complex topologies, and maybe upstream ones if strict ingress filtering is in effect. More radically, much of the GIMPS protocol is unchanged if we consider off-path signaling nodes, although there are significant differences in some of the security analysis (Section 7.3). However, none of these possibilities are considered further in this specification. 8.13 Protocol Design Details Clearly, not all details of GIMPS operation have been defined so far. This section provides a list of slightly non-trivial areas where more detail is need, where these have not been mentioned elsewhere in the text. o Datagram mode still requires (primitive) transport functions for backoff on retransmission and rate limiting in general. o Processing of the GIMPS hop count and IP TTL needs to be clarified, especially for messages which are being bypassed without going through full GIMPS processing. o Receiver initiated signaling applications need to have reverse path state set up in the network. Should this be done by GIMPS carrying out the discovery for the specific signaling application (which requires the flow sender to know what signaling applications are going to be used), or should the discovery Schulzrinne & Hancock Expires April 19, 2004 [Page 47] Internet-Draft GIMPS October 2003 attempt to find every GIMPS node and the signaling applications they support? Schulzrinne & Hancock Expires April 19, 2004 [Page 48] Internet-Draft GIMPS October 2003 Normative References [1] Katz, D., "IP Router Alert Option", RFC 2113, February 1997. [2] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. [3] Partridge, C. and A. Jackson, "IPv6 Router Alert Option", RFC 2711, October 1999. [4] Nordmark, E., "Stateless IP/ICMP Translation Algorithm (SIIT)", RFC 2765, February 2000. [5] Stewart, R., Xie, Q., Morneault, K., Sharp, C., Schwarzbauer, H., Taylor, T., Rytina, I., Kalla, M., Zhang, L. and V. Paxson, "Stream Control Transmission Protocol", RFC 2960, October 2000. [6] Kohler, E., "Datagram Congestion Control Protocol (DCCP)", draft-ietf-dccp-spec-04 (work in progress), July 2003. [7] Stewart, R., "SCTP Partial Reliability Extension", draft-ietf-tsvwg-prsctp-01 (work in progress), August 2003. Schulzrinne & Hancock Expires April 19, 2004 [Page 49] Internet-Draft GIMPS October 2003 Informative References [8] Braden, B., Zhang, L., Berson, S., Herzog, S. and S. Jamin, "Resource ReSerVation Protocol (RSVP) -- Version 1 Functional Specification", RFC 2205, September 1997. [9] Terzis, A., Krawczyk, J., Wroclawski, J. and L. Zhang, "RSVP Operation Over IP Tunnels", RFC 2746, January 2000. [10] Tsirtsis, G. and P. Srisuresh, "Network Address Translation - Protocol Translation (NAT-PT)", RFC 2766, February 2000. [11] Gilligan, R. and E. Nordmark, "Transition Mechanisms for IPv6 Hosts and Routers", RFC 2893, August 2000. [12] Berger, L., Gan, D., Swallow, G., Pan, P., Tommasi, F. and S. Molendini, "RSVP Refresh Overhead Reduction Extensions", RFC 2961, April 2001. [13] Carpenter, B. and K. Moore, "Connection of IPv6 Domains via IPv4 Clouds", RFC 3056, February 2001. [14] Huitema, C., "An Anycast Prefix for 6to4 Relay Routers", RFC 3068, June 2001. [15] Baker, F., Iturralde, C., Le Faucheur, F. and B. Davie, "Aggregation of RSVP for IPv4 and IPv6 Reservations", RFC 3175, September 2001. [16] 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. [17] 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. [18] Hancock, R., "Next Steps in Signaling: Framework", draft-ietf-nsis-fw-04 (work in progress), September 2003. [19] Tschofenig, H. and D. Kroeselberg, "Security Threats for NSIS", draft-ietf-nsis-threats-02 (work in progress), July 2003. [20] Nikander, P., "Mobile IP version 6 Route Optimization Security Design Background", draft-nikander-mobileip-v6-ro-sec-01 (work in progress), July 2003. [21] Bound, J., "Dual Stack Transition Mechanism", Schulzrinne & Hancock Expires April 19, 2004 [Page 50] Internet-Draft GIMPS October 2003 draft-bound-dstm-exp-00 (work in progress), August 2003. Authors' Addresses Henning Schulzrinne Columbia University Department of Computer Science 450 Computer Science Building New York, NY 10027 US Phone: +1 212 939 7042 EMail: hgs+nsis@cs.columbia.edu URI: http://www.cs.columbia.edu Robert Hancock Siemens/Roke Manor Research Old Salisbury Lane Romsey, Hampshire SO51 0ZN UK EMail: robert.hancock@roke.co.uk URI: http://www.roke.co.uk Schulzrinne & Hancock Expires April 19, 2004 [Page 51] Internet-Draft GIMPS October 2003 Appendix A. Acknowledgements This document is based on the discussions within the IETF NSIS working group. It has been informed by prior work and formal and informal inputs from: Bob Braden, Xiaoming Fu, Ruediger Geib, Eleanor Hepworth, Georgios Karagiannis, John Loughney, Jukka Manner, Andrew McDonald, Charles Shen, Melinda Shore, Mike Thomas, Hannes Tschofenig, Sven van den Bosch, and Lars Westberg. We look forward to inputs and comments from many more in the future. Schulzrinne & Hancock Expires April 19, 2004 [Page 52] Internet-Draft GIMPS October 2003 Appendix B. Example Routing State Table Figure 9 shows a signaling scenario for a single flow being managed by two signaling applications. The flow sender and receiver and one router support both, two other routers support one each. A B C D E +------+ +-----+ +-----+ +-----+ +--------+ | Flow | +-+ +-+ |NSLP1| |NSLP1| | | | Flow | |Sender|====|R|====|R|====|NSLP2|====| |====|NSLP2|====|Receiver| | | +-+ +-+ |GIMPS| |GIMPS| |GIMPS| | | +------+ +-----+ +-----+ +-----+ +--------+ ------------------------------>> Flow Direction Figure 9: A Signaling Scenario +------------+------------+-------------+-------------+-------------+ | Flow | Session ID | Signaling | Upstream | Downstream | | Routing | | Application | Peer | Peer | | Informatio | | | | | | n | | | | | +------------+------------+-------------+-------------+-------------+ | {IP-#A, | 0xABCDEF | NSLP1 | IP-#A | (null) | | IP-#E, | | | | | | protocol & | | | | | | ports} | | | | | | | | | | | | {IP-#A, | 0x123456 | NSLP2 | IP-#A | Pointer to | | IP-#E, | | | | B-D | | protocol & | | | | messaging | | ports} | | | | association | +------------+------------+-------------+-------------+-------------+ The table shows the routing state at Node B for the single flow from A to E. The upstream state is just the same address for each application. For the downstream case, NSLP1 only requires datagram mode messages and so no explicit routing state towards C is needed. NSLP2 requires a messaging association for its messages, and node C has elected to bypass the GIMPS-query/response exchange, so the downstream peer state for NSLP2 is a pointer to a messaging association that runs directly from B to D. Note that E is not visible in the state table (except implicitly in the address in the flow routing information); routing state is stored only for adjacent peers. Schulzrinne & Hancock Expires April 19, 2004 [Page 53] Internet-Draft GIMPS October 2003 Appendix C. Connection Mode Messaging Association Configurations Figure 10 shows a segment of a path between flow sender and receiver (not shown). Nodes A and F are GIMPS nodes which also fully process NSLP1. They need to carry out relatively complex negotiations, maybe exchanging large messages or requiring a secured channel. There are several other nodes on the path between them: o Node B is GIMPS aware but only processes NSLP2; o Node C is a GIMPS aware NAT (see Section 6.4) but processes no signaling applications; o Node D is a router; o Node E processes a subset of NSLP1 (e.g. insertion of local resource availability status for use by Node F). A B C D E F +-----+ +-----+ +-----+ +-----+ +-----+ |NSLP1| | | | | +-+ |NSLP1| |NSLP1| >>>>>| |>>>>|NSLP2|>>>>| NAT |>>>>|R|>>>>|-lite|>>>>| |>>>>> |GIMPS| |GIMPS| |GIMPS| +-+ |GIMPS| |GIMPS| +-----+ +-----+ +-----+ +-----+ +-----+ Figure 10: Heterogeneous Signaling The possible messaging association arrangements depend on the connection mode encapsulation: Raw (as in Section 5.3.2): Messaging associations are needed from A-C, C-E and E-F (node D is invisible to GIMPS and node B can ignore discovery for NSLP1). Achieving properties such as reliability or flow control or channel security between A and F depends on the quality of implementation and trust in the administration of nodes C and E. Tunnelled (as in Section 5.3.3 and Section 5.3.4): There is a single messaging association from A-F directly. C can do NAT processing if the flow routing information is placed in the mutable area, and E can manipulate NSLP1 objects in the mutable area also. Schulzrinne & Hancock Expires April 19, 2004 [Page 54] Internet-Draft GIMPS 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 Schulzrinne & Hancock Expires April 19, 2004 [Page 55] Internet-Draft GIMPS October 2003 HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Acknowledgment Funding for the RFC Editor function is currently provided by the Internet Society. Schulzrinne & Hancock Expires April 19, 2004 [Page 56]