MPLS & RSVP Working Groups Daniel Awduche Internet Draft UUNET Technologies, Inc. Expiration Date: February 1999 Der-Hwa Gan Juniper Networks, Inc. Tony Li Juniper Networks, Inc. George Swallow Cisco Systems, Inc. Vijay Srinivasan Torrent Networks, Inc. August 1998 Extensions to RSVP for Traffic Engineering draft-swallow-mpls-rsvp-trafeng-00.txt Status of this Memo This document is an Internet-Draft. 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." To view the entire list of current Internet-Drafts, please check the "1id-abstracts.txt" listing contained in the Internet-Drafts Shadow Directories on ftp.is.co.za (Africa), ftp.nordu.net (Northern Europe), ftp.nis.garr.it (Southern Europe), munnari.oz.au (Pacific Rim), ftp.ietf.org (US East Coast), or ftp.isi.edu (US West Coast). Abstract This document describes the use of RSVP, including all the necessary extensions, to support traffic engineering with MPLS as specified in [6]. Swallow, editor [Page 1]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 We propose several additional objects that extend RSVP, allowing the establishment of explicitly routed label switched paths (LSPs), using RSVP as a signaling protocol. The result is the instantiation of label-switched sessions which can be automatically routed away from network failures, congestion, and bottlenecks. Swallow, editor [Page 2]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 Contents 1 Introduction ........................................... 4 2 Overview of operation .................................. 5 2.1 Service Classes ........................................ 6 2.2 Reservation styles ..................................... 6 2.2.1 Fixed Filter (FF) style ................................ 7 2.2.2 Wildcard Filter (WF) style ............................. 7 2.2.3 Shared Explicit (SE) style ............................. 8 2.3 LSP Tunnels ............................................ 8 2.4 Rerouting LSP Tunnels .................................. 9 3 RSVP Message Formats ................................... 10 3.1 Path message ........................................... 10 3.2 Resv message ........................................... 11 4 Objects ................................................ 11 4.1 Label Object ........................................... 11 4.1.1 Handling Label Objects in Resv messages ................ 12 4.1.2 Non-support of the Label Object ........................ 13 4.2 Label Request Object ................................... 13 4.2.1 Handling of LABEL_REQUEST .............................. 14 4.2.2 Non-support of the Label Request Object ................ 14 4.3 Explicit Route Object .................................. 15 4.3.1 Subobjects ............................................. 15 4.3.2 Applicability .......................................... 16 4.3.3 Semantics of the Explicit Route Object ................. 16 4.3.4 Strict and Loose subobjects ............................ 17 4.3.5 Loops .................................................. 18 4.3.6 Subobject semantics .................................... 18 4.3.7 Processing of the Explicit Route Object ................ 20 4.3.8 Non-support of the Explicit Route Object ............... 21 4.4 Record Route Object .................................... 22 4.4.1 Subobjects ............................................. 22 4.4.2 Applicability .......................................... 24 4.4.3 Handling RRO ........................................... 25 4.4.4 Loop Detection ......................................... 26 4.4.5 Non-support of RRO ..................................... 26 4.5 Error subcodes for ERO and RRO ......................... 27 4.6 Session, Sender Template, and Filter Spec Objects ...... 27 4.6.1 Session Object ......................................... 27 4.6.2 Sender Template Object ................................. 28 4.6.3 Filter Specification Object ............................ 29 4.6.4 Reroute procedure ...................................... 29 4.7 Session Attribute Object ............................... 30 5 RSVP Aggregate Message ................................. 33 5.1 Aggregate Header ....................................... 33 5.2 Message Formats ........................................ 35 Swallow, editor [Page 3]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 5.3 Sending RSVP Aggregate Messages ........................ 35 5.4 Receiving RSVP Aggregate Messages ...................... 36 5.5 Forwarding RSVP Aggregate Messages ..................... 37 5.6 Aggregate-capable bit .................................. 37 6 Tear Confirm ........................................... 37 7 Security Considerations ................................ 39 8 Acknowledgments ........................................ 39 9 References ............................................. 39 1. Introduction For hosts and routers that support both RSVP [1] and Multi-Protocol Label Switching [2], it is possible to associate labels with RSVP flows [4]. The result is that a router can identify the appropriate reservation state for a packet based on its label value, thus greatly simplifying packet classification. This design also improves network performance because the same label lookup identifies forwarding information of the packet. Using RSVP to establish label switched paths (LSPs) clearly enables the allocation of resources to an LSP. For example, you can allocate bandwidth to an LSP using standard RSVP reservations and Integrated Services service classes [7]. While this is useful, reservations are not required. An LSP can also be established to carry best-effort traffic without a resource reservation. It is possible to add explicit routing capability on top of label- switched RSVP flows [3] [5] by adding a simple EXPLICIT_ROUTE object to RSVP. By using this object, the paths taken by label-switched RSVP flows can be predetermined, independent of conventional IP routing. The hops in the path can be manually configured, or computed automatically based on the QoS requirements of the flow and the current network load. The purpose of this document is to organize all the objects from [3], [4], and [5] into a single document that fully describes all the procedures and packet formats so that interoperable implementations are possible. A few new objects are also suggested for enhancing management and diagnostics of LSPs. All objects described are optional, and this document describes what happens when an object is not supported by a node. Finally, an RSVP aggregate message is proposed to help alleviate one of the RSVP scaling issues: how to efficiently handle large number of Swallow, editor [Page 4]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 RSVP messages that are periodically transmitted between neighbors. The document concentrates on unicast LSPs. Explicitly routed multicast LSPs are left for further study. 2. Overview of operation When an RSVP flow originates in or crosses an MPLS domain, the flow may be label switched. To initiate label switching, the first MPLS node inserts a LABEL_REQUEST object into the Path message. The LABEL_REQUEST object indicates that a label binding for this path is requested, and also provides an indication of the network layer protocol that is to be carried over this path. The reason for this is that the network layer protocol sent down an LSP cannot be assumed to be IPv4, and cannot be deduced from the L2 header, which simply identifies the higher layer protocol as being MPLS. If the sender node has prior knowledge of an alternative route that has better likelihood of meeting the flow's QoS requirement or that makes more efficient use of network resources, the node can decide to reroute some of its sessions. To do this, the node adds an EXPLICIT_ROUTE object to the Path message. If, during a session, the sender node finds a better route, the session can be rerouted on the fly by simply changing the EXPLICIT_ROUTE object. If there are problems with an EXPLICIT_ROUTE object, either because it causes a routing loop or some intermediate routers do not support it, the sender node is notified. If the RECORD_ROUTE object is added to Path messages, the sender node can receive information about the exact routing path and can prompt for notifications from the network if the routing path changes for any reason. Finally, a SESSION_ATTRIBUTE object can be added to Path messages for aiding in session identification and diagnostics. Additional control information, such as preemption, priority, and fast-reroute, is also included in this object. When the EXPLICIT_ROUTE object (ERO) is present, the Path message is forwarded towards its destination along a path specified by the ERO. Each node along the path records the ERO in its path state block. Nodes may also modify the ERO before forwarding the Path message, in which case the modified ERO should be stored in the path state block. The LABEL_REQUEST object requests intermediate routers and receiving Swallow, editor [Page 5]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 nodes to provide a label binding for the session. If a node is incapable of providing a label binding it sends a PathErr message with an "unknown object class" error. If the LABEL_REQUEST object is not supported end to end, the sender node will be notified by the first node which lacks the support. The destination node includes a LABEL object in its response Resv message. The LABEL object is inserted in the filter spec list immediately following the filter spec to which it pertains. When the LABEL object propagates upstream to the sender node, a label-switched path is already set up for use. The Resv message is sent back towards the sender. A node that receives a Resv message containing a label uses that label for outgoing traffic on this path. It also allocates a new label and places that label in the corresponding LABEL object of the Resv message before sending it upstream. This is the label that this node will use for incoming traffic on this path. This label now serves as shorthand for the Filter Spec. 2.1. Service Classes This document does not restrict the type of Integrated Service requested on a reservations. However, an implementation should always be ready to accept the Controlled-Load service [7]. An LSP may not need a bandwidth reservation or a QoS guarantee. Such LSPs can be used to deliver best-effort traffic, even if RSVP is used for setting up LSPs. When no resources need to be allocated to the LSP, the Sender_TSpec in the Path message can specify a token bucket rate of zero and a token bucket size of zero. The corresponding FLOWSPEC (in the Resv message) should carry a zero rate and size as well. LSPs with no bandwidth reservation are not subject to Admission Control and do not require traffic policing. 2.2. Reservation styles The receiver node can select from among a set of possible reservation styles for each session, and each RSVP session must have a particular style. Senders have no influence on the choice of reservation style. The receiver can choose different reservation styles for different LSPs. An RSVP session is identified by a unique (destination address, protocol, destination port) tuple. An RSVP session can create one or more LSPs, depending on the Swallow, editor [Page 6]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 reservation style chosen. Some reservation styles, such as FF, dedicate a particular reservation to an individual sender node. Other reservation styles, such as WF and SE, can share a reservation among several sender nodes. The following sections discuss the different reservation styles and their advantages and disadvantages. 2.2.1. Fixed Filter (FF) style The Fixed Filter (FF) reservation style creates a distinct reservation for traffic from each sender that is not shared by other senders. This style is common for applications in which traffic from each sender is likely to be concurrent and independent. The total amount of reserved bandwidth on a link for sessions using FF is the sum of the reservations for the individual senders. Because each sender has its own reservation, a unique label and a separate label-switched-path is assigned to each sender. This results in a point-to-point LSP between every sender/receiver pair. Because the network state overhead is proportional to the number of LSPs, having more LSPs means that more network resources are consumed. 2.2.2. Wildcard Filter (WF) style With the Wildcard Filter (WF) reservation style, a single shared reservation is used for all senders. The total reservation on a link remains the same regardless of the number of senders. This style is useful in applications in which not all senders send traffic at the same time. A phone conference, for example, is an application where not all speakers talk at the same time. A single label-switched-path is created for all senders, because all senders to the session are covered by the reservation. On links that senders share, a single label is allocated. If there is only one sender, the LSP looks like normal point-to-point connection. When multiple senders are present, a multipoint-to-point LSP (a reversed tree) is created. This has the advantage of minimizing the number of LSPs (and the memory and CPU resources used for each LSP), allowing the network to scale better. Because of the merging rules, EXPLICIT_ROUTE objects cannot be used with WF reservations. Hence, the use of the WF style should be discouraged in the presence of ERO. Swallow, editor [Page 7]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 2.2.3. Shared Explicit (SE) style Unlike the WF style, where any sender is allowed to share the reservation, the Shared Explicit (SE) style allows a receiver to explicitly specify the senders to be included. There is a single reservation on a link for all the senders listed. Only listed senders can join the reservation. Because each sender is explicitly listed in the Resv message, you can assign separate labels to each sender, therefore creating separate LSPs for each sender. [4] describes the reason why separate LSPs are needed. Having separate LSPs for each sender also eliminates the incompatibility with the EXPLICIT_ROUTE object. Path messages from different senders can carry their own ERO, and the paths taken by the senders can converge and diverge at any point. Unlike the FF style, all SE LSPs share the single reservation. Unlike the WF style, a separate LSP is created for each sender. 2.3. LSP Tunnels When LSPs are used to carry flows, it becomes possible to be more flexible in the definition of a flow. The first node in an LSP can use any of a variety of means to determine which packets will be assigned a particular label. Once that label is assigned, the label becomes the definition of the flow. We refer to such an LSP as an LSP Tunnel due to the opaque nature of the flow. In support of this, a new SESSION object, LSP_TUNNEL_IPv4 is defined. The semantics of this object are that the flow is defined solely on the basis of packets arriving from the PHOP with the particular label value(s) assigned by this node to senders to the session. In fact, the IPv4 appearing in the object name only denotes that the destination address is an IPv4 address. An application of particular interest is traffic engineering. By establishing ER-LSPs a node at the edge of an MPLS domain can control the path which traffic from this node will take through that domain. These capabilities can be used to optimize the utilization of network resources and enhance traffic oriented performance characteristics. Swallow, editor [Page 8]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 2.4. Rerouting LSP Tunnels One of the requirements for Traffic Engineering is the ability to have an LSP tunnel re-routed upon a failure of a resource along its current path. A further requirement is the ability to have the LSP tunnel return to its original path when the failed resource is restored. It is also desirable not to disrupt traffic while rerouting is in progress. The adaptive rerouting requirement calls for establishing a new LSP while keeping the old LSP intact. On links that old and new LSPs share, one wishes to (1) not release resources from the old LSP that one wants to use for the new LSP, and (2) not double-count reservations, because this might cause Admission Control to deny the new LSP. The combination of the LSP_TUNNEL_IPv4 SESSION object and the SE reservation style naturally achieves smooth transitions. The LSP_TUNNEL_IPv4 SESSION object is used to narrow the scope of the RSVP session to the particular tunnel in question. To uniquely identify a tunnel we use the combination of the destination IP address, a Tunnel ID, and the sender's IP address which is placed in the Extended Tunnel ID field. During the reroute operation, the source needs to be able to appear as two different sources to RSVP. This is achieved by the use of a "LSP ID", which is carried in the SENDER_TEMPLATE and FILTER_SPEC objects. Since the semantics of these objects is changed, a new C- Type is assigned. To effect a reroute, the source node picks a new LSP ID and forms a new SENDER_TEMPLATE. It creates a new ERO to define the new path. The node sends a new Path Message using the original SESSION object and the new SENDER_TEMPLATE and ERO. It continues to use the old LSP and refresh the old Path message. On links which are not in common, the new Path message is treated as any new LSP tunnel setup. On links held in common, the shared SESSION object and SE style allow the LSP to be established sharing the same resources. Once the sender receives a Resv message for the new LSP, it is free to begin using it and to tear down the old LSP. Also new C-Types are assigned for the SESSION, SENDER_TEMPLATE, and FILTER_SPEC objects. Detailed descriptions of the new objects are given in later sections. All new objects are optional with respect to RSVP. An implementation Swallow, editor [Page 9]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 3. RSVP Message Formats Five new objects are defined in this document: Object name Applicable RSVP messages --------------- ------------------------ LABEL_REQUEST Path LABEL Resv EXPLICIT_ROUTE Path RECORD_ROUTE Path, Resv SESSION_ATTRIBUTE Path can choose to support some but not other objects. However, the LABEL_REQUEST and LABEL objects are mandatory with respect to this document. The LABEL and RECORD_ROUTE objects, are sender specific. They must immediately follow either the SENDER_TEMPLATE in Path messages, or the FILTER_SPEC in Resv messages. The placement of EXPLICIT_ROUTE, LABEL_REQUEST, and SESSION_ATTRIBUTE objects is simply a suggestion. While it is recommended that an implementation follow this format, the ordering of these objects is not important, so an implementation must be prepared to accept objects in any order. 3.1. Path message The format of the Path message is as follows: <Path Message> ::= <Common Header> [ <INTEGRITY> ] <SESSION> <RSVP_HOP> <TIME_VALUES> [ <EXPLICIT_ROUTE> ] <LABEL_REQUEST> [ <SESSION_ATTRIBUTE> ] [ <POLICY_DATA> ... ] [ <sender descriptor> ] <sender descriptor> ::= <SENDER_TEMPLATE> [ <SENDER_TSPEC> ] [ <ADSPEC> ] [ <RECORD_ROUTE> ] Swallow, editor [Page 10]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 3.2. Resv message The format of the Resv message is as follows: <Resv Message> ::= <Common Header> [ <INTEGRITY> ] <SESSION> <RSVP_HOP> <TIME_VALUES> [ <RESV_CONFIRM> ] [ <SCOPE> ] [ <POLICY_DATA> ... ] <STYLE> <flow descriptor list> <WF flow descriptor> ::= <FLOWSPEC> <LABEL> [ <RECORD_ROUTE> ] <FF flow descriptor list> ::= <FLOWSPEC> <FILTER_SPEC> <LABEL> [ <RECORD_ROUTE> ] | <FF flow descriptor list> <FF flow descriptor> <FF flow descriptor> ::= [ <FLOWSPEC> ] <FILTER_SPEC> <LABEL> [ <RECORD_ROUTE> ] <SE flow descriptor> ::= <FLOWSPEC> <SE filter spec list> <SE filter spec list> ::= <SE filter spec> | <SE filter spec list> <SE filter spec> <SE filter spec> ::= <FILTER_SPEC> <LABEL> [ <RECORD_ROUTE> ] Note: LABEL and RECORD_ROUTE (if present), are bound to the preceding FILTER_SPEC. No more than one LABEL and/or RECORD_ROUTE may follow each FILTER_SPEC. 4. Objects 4.1. Label Object Labels may be carried in Resv messages. When a label is to be associated with a single sender, it must immediately follow the FILTER_SPEC for that sender in the Resv message. The LABEL object was first documented in [4]. The LABEL object has the following format: The contents of a LABEL object are a stack of labels, where each label is encoded right aligned in 4 octets. The top of the stack is in the right 4 octets of the object contents. A LABEL object that Swallow, editor [Page 11]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 LABEL class = 16, C_Type = 1 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Length (bytes) | Class-Num | C-Type | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | // (Object contents) // | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | (top label) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ contains no labels is illegal. Each label is an unsigned integer in the range 0 through 1048575. The decision of whether to create a label stack with more than one label, when to push a new label, and when to pop the label stack are to be specified in a separate document. For implementations that do not support a label stack, only the top label is examined. The rest of the label stack should be passed through unchanged. Such implementations are required to generate a label stack of depth 1 when initiating the first LABEL. 4.1.1. Handling Label Objects in Resv messages For unicast sessions, only Resv messages contain the LABEL object. If a router does not wish to support MPLS for the session, the router can ignore the received LABEL objects and continue processing the rest of Resv message. The router uses the top label carried in the LABEL object as the outgoing label associated with the session (if WF) or sender (if FF or SE). The router allocates a new label and binds it to the incoming interface of this session/sender. This is the same interface that the router uses to forward Resv messages to the previous hops. To construct a new LABEL object, the router replaces the top label (from the received Resv message) with the locally allocated new label. The router then sends the new LABEL object as part of the Resv message to the previous hop. The LABEL object should be kept in the Reservation State Block. It is then used in the next Resv refresh event for formatting the Resv message. A router can decide to send a Resv message before its refresh timers Swallow, editor [Page 12]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 expire if the contents of the LABEL object have changed. A received Resv message without a LABEL object indicates that the next-hop router does not wish to support MPLS for this session/sender. A label can be withdrawn without removing the reservation by sending a Resv with no LABEL object. The receiving router should stop sending label switched packets toward the next-hop router. The RSVP session itself is not affected. If, however, the session is of type LSP_TUNNEL_IPv4, then the label withdrawal procedure must not be used and a ResvTear sent instead. 4.1.2. Non-support of the Label Object An RSVP router that does not recognize the LABEL object sends a ResvErr with the error code "Unknown object class" toward the receiver. This causes the reservation to fail. The receiver should notify management that a LSP cannot be established, and possibly take action to continue the reservation without the LABEL object. RSVP is designed to cope gracefully with non-RSVP routers anywhere between senders and receivers. However, non-RSVP routers cannot receive label-switched packets conveyed in PATH or RESV messages. This means that if a router has a neighbor who is not RSVP capable, the router must not advertise the LABEL object when sending messages that pass through the non-RSVP router. [1] describes how routers can determine the presence of non-RSVP routers. 4.2. Label Request Object The LABEL_REQUEST object was first documented in [3]. A LABEL_REQUEST object has the following format: class = 19, C_Type = 1 (need to get an official class num from the IANA) 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Length (bytes) | Class-Num | C-Type | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Reserved | L3PID | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ The L3PID is an identifier of the layer 3 protocol using this path. Standard Ethertype values are used. Swallow, editor [Page 13]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 4.2.1. Handling of LABEL_REQUEST The sender creates a Path message with a LABEL_REQUEST object. The LABEL_REQUEST object indicates that a label binding for this path is requested and provides an indication of the network layer protocol that is to be carried over this path. This permits non-IP network layer protocols to be sent down an LSP. The information can also be useful in assigning the actual label on a link, because some reserved labels are protocol specific. See [8]. The LABEL_REQUEST should be stored in the Path State Block, so that refreshes of the Path messages will also contain the LABEL_REQUEST object. When the Path message reaches its receiver, the presence of the LABEL_REQUEST object triggers the receiver to allocate a label and to place the label in the LABEL object for the corresponding Resv message. A receiver that accepts a LABEL_REQUEST object, must include a LABEL object in Resv messages. A node that accepts a LABEL_REQUEST object must be ready to accept and correctly process a LABEL object in the corresponding Resv messages. A node that recognizes a LABEL_REQUEST object, but that is unable to support it (possibly because of a failure to allocate labels), should send a PathErr with the error code "Routing problem" and the subcode "MPLS label allocation failure." If a node cannot support the protocol L3PID, it should send a PathErr with the error code "Routing problem" and the subcode "Unsupported L3PID." This causes the RSVP session to fail. 4.2.2. Non-support of the Label Request Object An RSVP router that does not recognize the LABEL_REQUEST object sends a PathErr with the error code "Unknown object class" toward the sender. This causes the path setup to fail. The sender should notify management that a LSP cannot be established and possibly take action to continue the reservation without the LABEL_REQUEST. RSVP is designed to cope gracefully with non-RSVP routers anywhere between the sender and the receiver. However, non-RSVP routers cannot receive label-switched packets. This means that if a router has a neighbor that is not RSVP capable, the router must not advertise LABEL_REQUEST objects when sending messages that pass through the non-RSVP routers. The router should send a PathErr back to the sender, with the error code "Routing problem" and the subcode "MPLS being negotiated, but a non-RSVP capable router stands in the path." [1] describes how routers can determine the presence of non-RSVP Swallow, editor [Page 14]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 routers. 4.3. Explicit Route Object As stated earlier, explicit routes are to be specified through a new EXPLICIT_ROUTE object in RSVP. RSVP Path messages carry this object. The EXPLICIT_ROUTE object has the following format: 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Length (bytes) | Class-Num | C-Type | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | // (Object contents) // | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Class-Num The Class-Num for an EXPLICIT_ROUTE object is 193 (need to get an official one from the IANA with the high order two bits set to 11) C-Type The C-Type for an EXPLICIT_ROUTE object is 1 (need to get an official one from the IANA) If a Path message contains multiple EXPLICIT_ROUTE objects, only the first object is meaningful. Subsequent EXPLICIT_ROUTE objects may be ignored and should not be propagated. 4.3.1. Subobjects The contents of an EXPLICIT_ROUTE object are a series of variable- length data items called subobjects. Each subobject has the form: 0 1 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-------------//----------------+ |L| Type | Length | (Subobject contents) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-------------//----------------+ Swallow, editor [Page 15]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 L The L bit is an attribute of the subobject. The L bit is set if the subobject represents a loose hop in the explicit route. If the bit is not set, the subobject represents a strict hop in the explicit route. Type The Type indicates the type of contents of the subobject. Currently defined values are: 0 Reserved 1 IPv4 prefix 2 IPv6 prefix 32 Autonomous system number 64 MPLS label switched path termination Length The Length contains the total length of the subobject in bytes, including the L, Type and Length fields. The Length must always be a multiple of 4, and at least 4. 4.3.2. Applicability The EXPLICIT_ROUTE object is intended to be used only for unicast situations. Applications of explicit routing to multicast are a topic for further research. The EXPLICIT_ROUTE object is to be used only when all routers along the explicit route support RSVP and the EXPLICIT_ROUTE object. The mechanisms for determining, a priori, that such support is present are beyond the scope of this document. 4.3.3. Semantics of the Explicit Route Object An explicit route is a particular path in the network topology. Typically, the explicit route is computed by a node, with the intent of directing traffic down that path. An explicit route is described as a list of groups of nodes along the explicit route. Certain operations to be performed along the path can also be encoded in the EXPLICIT_ROUTE object. In addition to the ability to identify specific nodes along the path, Swallow, editor [Page 16]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 an explicit route can identify a group of nodes that must be traversed along the path. This capability allows the routing system a significant amount of local flexibility in fulfilling a request for an explicit route. In turn, this allows the generator of the explicit route to have imperfect information about the details of the path. The explicit route is encoded as a series of subobjects contained in an EXPLICIT_ROUTE object. Each subobject may identify a group of nodes in the explicit route or may be an operation to be performed along the path. An explicit route is then a path including all of the identified groups of nodes, with the specified operations occurring along the path. To simplify the discussion, we call each group of nodes an abstract node. Thus, we can also say that an explicit route is a path including all of the abstract nodes, with the specified operations occurring along that path. As an example, consider an explicit route that consists solely of autonomous system number subobjects. Each subobject corresponds to an autonomous system in the network topology. Each autonomous system is an abstract node. In this case, the explicit route is a path including each of the specified autonomous systems. There may be multiple hops within each autonomous system. 4.3.4. Strict and Loose subobjects The L bit in the subobject is a one-bit attribute. If the L bit is set, then the value of the attribute is `loose.' Otherwise, the value of the attribute is `strict.' For brevity, we say that if the value of the subobject attribute is `loose' then it is a `loose subobject.' Otherwise, it's a `strict subobject.' Further, we say that the abstract node of a strict or loose subobject is a strict or a loose node, respectively. Loose and strict nodes are always interpreted relative to their prior abstract nodes. The path between a strict node and its prior node MUST include only network nodes from the strict node and its prior abstract node. The path between a loose node and its prior node MAY include other network nodes that are not part of the strict node or its prior abstract node. The L bit has no meaning in operation subobjects. Swallow, editor [Page 17]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 4.3.5. Loops While the EXPLICIT_ROUTE object is of finite length, the existence of loose nodes implies that it is possible to construct forwarding loops during transients in the underlying routing protocol. This can be detected by the originator of the explicit route through the use of another opaque route object called the RECORD_ROUTE object. The RECORD_ROUTE object is used to collect detailed path information and is useful for loop detection as well as diagnostic purposes. 4.3.6. Subobject semantics 4.3.6.1. Subobject 1: The IPv4 prefix The contents of an IPv4 prefix subobject are a 4-octet IPv4 address, a 1-octet prefix length, and a 1-octet pad. The abstract node represented by this subobject is the set of nodes that have an IP address which lies within this prefix. Note that a prefix length of 32 indicates a single IPv4 node. The length of the IPv4 prefix subobject is 8 octets. The contents of the 1 octet of padding must be zero on transmission and must not be checked on receipt. 4.3.6.2. Subobject 2: The IPv6 address 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Type | Length | IPV6 address (16 bytes) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | IPV6 address (continued) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | IPV6 address (continued) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | IPV6 address (continued) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | IPV6 address (continued) | Mask | Padding | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Type 0x82 IPv6 address Length Swallow, editor [Page 18]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 The Length contains the total length of the subobject in bytes, including the Type and Length fields. The Length is always 20. IPv6 address A 128-bit unicast host address. Mask 128 Padding Zero on transmission. Ignored on receipt. 4.3.6.3. Subobject 32: The autonomous system number The contents of an autonomous system (AS) number subobject are a 2- octet autonomous system number. The abstract node represented by this subobject is the set of nodes belonging to the autonomous system. The length of the AS number subobject is 4 octets. 4.3.6.4. Subobject 64: MPLS label switched path termination The contents of an MPLS label switched path termination subobject are 2 octets of padding. The subobject is an operation subobject. This object is only meaningful if there is a LABEL_REQUEST object in the Path message. If a LABEL_REQUEST object is present in the Path message, this Path message is being used to establish a Label Switched Path. In this case, this subobject indicates that the prior abstract node should remove one level of label from all packets following this Label Switched Path. The length of the MPLS label termination subobject is 4 octets. Swallow, editor [Page 19]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 4.3.7. Processing of the Explicit Route Object 4.3.7.1. Selection of the next hop A Path message containing an EXPLICIT_ROUTE object must determine the next hop for this path. Selection of this next hop may involve a selection from a set of possible alternatives. The mechanism for making a selection from this set is implementation dependent and is outside of the scope of this specification. Selection of particular paths is also outside of the scope of this specification, but it is assumed that each node will make a best effort attempt to determine a loop-free path. Note that such best efforts may be overridden by local policy. To determine the next hop for the path, a node performs the following steps: 1) The node receiving the RSVP message must first evaluate the first subobject. If the node is not part of the abstract node described by the first subobject, it has received the message in error and should return a "Bad initial subobject" error. If the first subobject is an operation subobject, the message is in error and the system should return a "Bad EXPLICIT_ROUTE object" error. If there is no first subobject, the message is also in error and the system should return a "Bad EXPLICIT_ROUTE object" error. 2) If there is no second subobject, this indicates the end of the explicit route. The EXPLICIT_ROUTE object should be removed from the Path message. This node may or may not be the end of the path. Processing continues with section 4.3.2, where a new EXPLICIT_ROUTE object may be added to the Path message. 3) Next, the node evaluates the second subobject. If the subobject is an operation subobject, the node records the subobject, deletes it from the EXPLICIT_ROUTE object and continues processing with step 2, above. Note that this changes the third subobject into the second subobject in subsequent processing. The precise operations to be performed by this node must be defined by the operation subobject. 4) If the node is also a part of the abstract node described by the second subobject, then the node deletes the first subobject and continues processing with step 2, above. Note that this makes the second subobject into the first subobject of the next iteration. 5) The node determines whether it is topologically adjacent to the abstract node described by the second subobject. If so, the node selects a particular next hop which is a member of the abstract node. The node then deletes the first subobject and continues processing Swallow, editor [Page 20]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 with section 4.3.2. 6) Otherwise, the node selects a next hop within the abstract node of the first subobject that is along the path to the abstract node of the second subobject. If no such path exists then there are two cases: 6a) If the second subobject is a strict subobject, there is an error and the node should return a "Bad strict node" error. 6b) Otherwise, if the second subobject is a loose subobject, the node selects any next hop that is along the path to the next abstract node. If no path exists, there is an error, and the node should return a "Bad loose node" error. 7) Finally, the node replaces the first subobject with any subobject that denotes an abstract node containing the next hop. This is necessary so that when the explicit route is received by the next hop, it will be accepted. 4.3.7.2. Adding subobjects to the Explicit Route Object After selecting a next hop, the node may alter the explicit route in the following ways. If, as part of executing the algorithm in section 4.3.1, the EXPLICIT_ROUTE object is removed, the node may add a new EXPLICIT_ROUTE object. Otherwise, if the node is a member of the abstract node for the first subobject, a series of subobjects may be inserted before the first subobject or may replace the first subobject. Each subobject in this series must denote an abstract node that is a subset of the current abstract node. Alternately, if the first subobject is a loose subobject, an arbitrary series of subobjects may be inserted prior to the first subobject. 4.3.8. Non-support of the Explicit Route Object An RSVP router that does not recognize the EXPLICIT_ROUTE object sends a PathErr with the error code "Unknown object class" toward the sender. This causes the path setup to fail. The sender should notify management that a LSP cannot be established and possibly take action to continue the reservation without the EXPLICIT_ROUTE or via Swallow, editor [Page 21]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 a different explicit route. 4.4. Record Route Object The format of the RECORD_ROUTE object (RRO) is described below. 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Length (bytes) | Class-Num | C-Type | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | // (Subobjects) // | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Class-Num The Class-Num for a RECORD_ROUTE object is 194 (need to get an official one from the IANA with the high order two bits set to 11) C-Type The C-Type for a RECORD_ROUTE object is 1 (need to get an official one from the IANA) The RRO can show up in both RSVP Path and Resv messages. The presence of the RRO in Path messages is semantically unrelated to the presence of RRO in Resv message. The presence of RRO in one message type does not necessarily require RRO in other message types. If a message contains multiple RROs, only the first RRO is meaningful. Subsequent RROs can be ignored and should not be propagated. 4.4.1. Subobjects The contents of a RECORD_ROUTE object are a series of variable-length data items called subobjects. Each subobject has its own Length field, the Length contains the total length of the subobject in bytes, including the Type and Length fields. The length must always be a multiple of 4, and at least 4. Swallow, editor [Page 22]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 Subobjects are organized as a last-in-first-out stack. The first subobject relative to the beginning of RRO is considered the top. The last subobject is considered bottom. When a new subobject is added, it is always added to the top. An empty RRO with no subobjects is considered illegal. Two kinds of subobjects are currently defined. 4.4.1.1. Subobject 1: The IPv4 address 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Type | Length | IPV4 address (4 bytes) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | IPV4 address (continued) | Mask | Padding | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Type 0x81 IPv4 address IPv4 address A 32-bit unicast, host address. Any network-reachable interface address is allowed here. Illegal addresses, such as loopback addresses, should not be used. Length The Length contains the total length of the subobject in bytes, including the Type and Length fields. The Length is always 8. Mask 32 Padding Zero on transmission. Ignored on receipt. Swallow, editor [Page 23]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 4.4.1.2. Subobject 2: The IPv6 address 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Type | Length | IPV6 address (16 bytes) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | IPV6 address (continued) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | IPV6 address (continued) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | IPV6 address (continued) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | IPV6 address (continued) | Mask | Padding | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Type 0x82 IPv6 address Length The Length contains the total length of the subobject in bytes, including the Type and Length fields. The Length is always 20. IPv6 address A 128-bit unicast host address. Mask 128 Padding Zero on transmission. Ignored on receipt. 4.4.2. Applicability In Path messages, the RRO can be used for both unicast and multicast RSVP sessions. In Resv messages, only the procedure for use in unicast sessions is defined here. There are three possible uses of RRO in RSVP. First, it can function as a loop detection mechanism to discover L3 routing loops. The exact procedure for doing so is described in later sections of this Swallow, editor [Page 24]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 document. Second, RRO collects up-to-date detailed path information hop-by-hop about RSVP sessions, providing valuable information to the sender or receiver. Any path change (due to network topology changes) is quickly reported. Third, RRO syntax is designed so that, with minor changes, the whole object can be used as input to the EXPLICIT_ROUTE object. This is useful if the sender receives RRO from the receiver in a Resv message, applies it to EXPLICIT_ROUTE object in the next Path message in order to "pin down session path". 4.4.3. Handling RRO Typically, a node initiates an RSVP session by adding the RRO to the Path message. The initial RRO contains only one subobject - the sender's IP addresses. When a Path message containing a RRO is received by an intermediate router, the router stores a copy of it in the Path State Block. The RRO is then used in the next Path refresh event for formating Path messages. When a new Path message is to be sent, the router adds a new subobject to the RRO and appends the resulting RRO to the Path message before transmission. The newly added subobject must be this router's IP address. The address to be added should be the interface address of the outgoing Path messages. If there are multiple addresses to choose from, the decision is local matter. However, it is recommended that the same address be chosen consistently. If the newly added subobject causes the RRO to be too big to fit in a Path message, the Path message shall be dropped and a PathErr message should be sent back to the sender. An RSVP router can decide to send Path messages before its refresh time if the RRO in the next Path message is different from the previous one. This can happen if the contents of the RRO received from the previous hop router changes or if this RRO is newly added to (or deleted from) the Path message. A received Path message without an RRO indicates that the sender node no longer needs route recording. Subsequent Path messages shall not contain an RRO. Likewise, RSVP session receiver nodes initiate the RRO process by adding an RRO to Resv messages. The processing mirrors that of the Swallow, editor [Page 25]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 Path messages. The only difference is that the RRO in a Resv message records the path information in the reverse direction. 4.4.4. Loop Detection As part of processing an incoming RRO, the intermediate router looks into all subobjects contained within the RRO. If the router determines that it is already in the list, a forwarding loop exists. An RSVP session is loop-free if receiver nodes receive Path messages with no routing loops detected in the contained RRO or sender nodes receive Resv messages with no looping detected. There are two broad classifications of forwarding loops. The first class is the transient loop, that occurs as a normal part of operations as L3 routing tries to converge on a consistent forwarding path for all destinations. The second class of forwarding loop is the permanent loop, that normally results from network mis- configuration. The action performed on receipt depends on the message type in which the RRO is received. For Path messages containing a forwarding loop, the router builds and send a "Routing problem" PathErr message, with the subcode "loop detected," and drops the Path message. Until the loop is eliminated, this session is not suitable for forwarding user data packets. Eliminating the loop is beyond the scope of this document. For Resv messages containing a forwarding loop, the router simply drops the message. Resv messages should not loop if Path messages do not loop. 4.4.5. Non-support of RRO An RSVP router that does not recognize RRO forwards it unchanged. This has no impact on the reservation. The presence of non-RSVP routers anywhere between senders and receivers has no impact on the object either. The worst result is that RRO does not reflect the full path information. Swallow, editor [Page 26]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 4.5. Error subcodes for ERO and RRO In the processing described above, certain errors must be reported as part of a "Routing Problem" PathErr message. The value of the "Routing Problem" error code is 24 (TBD). The following defines the subcodes for the routing problem PathErr message: Value Error: 1 Bad EXPLICIT_ROUTE object 2 Bad strict node 3 Bad loose node 4 Bad initial subobject 5 No route available toward destination 6 RRO syntax error detected 7 RRO indicated routing loops 8 MPLS being negotiated, but a non-RSVP-capable router stands in the path 9 MPLS label allocation failure 10 Unsupported L3PID 4.6. Session, Sender Template, and Filter Spec Objects New C-Types are defined for the SESSION, SENDER_TEMPLATE and FILTER_SPEC objects. The LSP_TUNNEL_IPv4 objects have the following format: 4.6.1. Session Object IPv4 tunnel end point address IPv4 address of the destination node for the tunnel. Tunnel ID Swallow, editor [Page 27]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 Class = SESSION, C-Type = LSP_TUNNEL_IPv4 (7) 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Length (bytes) | Class-Num | C-Type | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | IPv4 tunnel end point address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Must be zero | Tunnel ID | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Extended Tunnel ID | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ A 16-bit identifier used in the SESSION that remains constant over the life of the tunnel. Extended Tunnel ID A 32-bit identifier used in the SESSION that remains constant over the life of the tunnel. Normally set to all zeros. Source nodes which wish to narrow the scope of a SESSION to the source destination pair may place their IPv4 address here as a globally unique identifier. 4.6.2. Sender Template Object Class = SENDER_TEMPLATE, C-Type = LSP_TUNNEL_IPv4 (7) 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Length (bytes) | Class-Num | C-Type | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | IPv4 tunnel sender address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Must be zero | LSP ID | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ IPv4 tunnel sender address IPv4 address for a sender node LSP ID a 16-bit identifier used in the SENDER_TEMPLATE and the FILTER_SPEC that can be changed to allow a sender to share resources with itself. Swallow, editor [Page 28]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 4.6.3. Filter Specification Object Class = FILTER SPECIFICATION, C-Type = LSP_TUNNEL_IPv4 (7) 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Length (bytes) | Class-Num | C-Type | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | IPv4 tunnel sender address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Must be zero | LSP ID | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ IPv4 tunnel sender address IPv4 address for a sender node LSP ID a 16-bit identifier used in the SENDER_TEMPLATE and the FILTER_SPEC that can be changed to allow a sender to share resources with itself. 4.6.4. Reroute procedure A tunnel which is capable of maintaining resource (without double counting) while it is being rerouted or attempting to increase its bandwidth is setup as follows. In the initial Path message, the source node forms a SESSION object, picking a Tunnel_ID and placing its IPv4 address in the Extended_Tunnel_ID. It forms a SENDER_TEMPLATE picking a Tunnel_Path_ID. Tunnel setup continues with normal processing. The destination node sends a Resv message with the STYLE to Shared Explicit. [Note I think we should add a flag to the SESSION_ATTRIBUTE for the source to indicate that it wishes the SE style.] When a source node with an established path that wants to change the path it forms a new Path message as follows. The existing SESSION object is used, in particular the Tunnel_ID and Extended_Tunnel_ID are unchanged. It picks a new Tunnel_Path_ID to form a new SENDER_TEMPLATE. It creates an EXPLICIT_ROUTE object with the new route. The new Path message is sent. The source node refreshes both Swallow, editor [Page 29]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 the old and new path messages The destination node responds with a Resv message with an SE flow descriptor formated as: <FLOW_SPEC><old_FILTER_SPEC><old_LABEL_OBJECT><new_FILTER_SPEC> <new_LABEL_OBJECT> (Note that if the PHOPs are different, then two messages are sent each with the appropriate FILTER_SPEC and LABEL_OBJECT.) When the Source node receives the Resv Message(s) it may begin using the new route. It should send a PathTear message for the old route. 4.7. Session Attribute Object The format of the SESSION_ATTRIBUTE object is described below. 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Length (bytes) | Class-Num | C-Type | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Setup Prio | Reserv. Prio | Flags | Name Length | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | // Session Name (NULL padded display string) // | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Class-Num The Class-Num indicates that the object is 207. (TBD) C-Type The C-Type is 7. Flags 0x01 = Fast-reroute This flag permits transit routers to precompute and pre-establish detour paths for this session. Upon fault detection on a immediate downstream link or node, transit routers reroute traffic onto the detour path for fast fail-over. Swallow, editor [Page 30]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 0x02 = Merging permitted This flag permits transit routers to merge this session with other RSVP sessions for the purpose of reducing resource overhead on downstream transit routers, thereby providing better network scalability. Setup Priority the range of 0 to 7. 0 is the highest priority. The Setup Priority is used in deciding whether this session should displace another session. Reservation Priority in the range of 0 to 7. 0 is the highest priority. Reservation Priority is used in deciding whether this session should be displaced by another session. Name Length The length of the display string before padding, in bytes. Session Name A null padded string of characters. The support of setup and reservation priorities is optional. A node can recognize this information but be unable to perform the requested operation. The node should pass the information downstream unchanged. Preemption is implemented by two priorities. The Setup Priority is the priority for taking resources. The Reservation Priority is the priority for holding a resource. The Setup Priority should never be higher than the Reservation Priority. The Reservation Priority is the priority at which resources assigned to this request will be reserved. When a new reservation is considered for admission, the bandwidth requested is compared with the bandwidth available at the priority specified in the Setup Priority. The bandwidth available at a particular priority is the unused bandwidth plus the bandwidth reserved at all priorities lower than the Setup Priority. If the bandwidth is not available a PathErr message is returned with a Error Code of 01, Admission Control failure, and an Error Value of 0x0002. The first 0 means globally defined subcode and not informational. The 002 means "requested bandwidth unavailable". If the requested bandwidth is less than the unused bandwidth, Swallow, editor [Page 31]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 processing is complete. If the requested bandwidth is available, but is in use by lower priority sessions, then lower priority sessions (beginning with the lowest priority) are pre-empted to free the necessary bandwidth. When pre-emption is supported, each pre-empted reservation triggers a TC_Preempt() upcall to local clients, passing a subcode indicating the reason. A ResvErr and/or PathErr with the code "Policy Control failure" should be sent toward the downstream receivers and upstream senders. [Editor: we need to define a subcode; if we stay with SESSION_ATTRIBUTE (not POLICY) we should also define an error code] The support of fast-reroute is optional. A node can recognize this information but be unable to perform the requested operation. The node should pass the information downstream unchanged. The support of merging is optional. A node can recognize this information but be unable to perform the requested operation. The node should pass the information downstream unchanged. If a Path message contains multiple SESSION_ATTRIBUTE objects, only the first SESSION_ATTRIBUTE object is meaningful. Subsequent SESSION_ATTRIBUTE objects can be ignored and not forwarded. The contents of the Session Name field are a string, typically displayable characters. The Length must always be a multiple of 4 and must be at least 8. For an object length that is not multiple of 4, the object is padded with trailing NULL characters. The Name Length field contains the actual string length. All RSVP routers, whether they support this object or not, shall forward the object unmodified. The presence of non-RSVP routers anywhere between senders and receivers has no impact on the object. Note that the granting of one reservation may result in the preemption of other reservations. We will also need an error code to indicate that a reservation has been preempted. I suggest we do that with both a PathTear and a ResvTear with a Error Code of 02, Policy Control failure, and a Error Value of 0x8002, where the subcode 002 means "Reservation Preempted". Swallow, editor [Page 32]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 5. RSVP Aggregate Message The resource requirement (processing and memory) for running RSVP on a router increases proportionally with the number of sessions. Supporting a large number of sessions on may present scaling problems. This section describes an approach to help alleviate one of the scaling issues. Path and Resv messages must be periodically refreshed. The approach here simply reduces the volume of messages which must be periodically sent and received. Another way of addressing the refresh volume problem is to increase the refresh timer R. Increasing the value of R provides linear improvement on transmission overhead, but at the cost of increasing refresh timeout. With the proposed aggregate message (see below), network administrators can reduce R for faster detection of connectivity problems while enjoying an order of magnitude less overhead. If topology failures occur, every node adjacent to the failure might wish to notify all affected sender and receiver nodes. These notification messages are either tear or error messages. Depending on how many sessions are affected and how fast every node is willing to react, these messages represent a flood that ripples out from the original failure point. Aggregate messages provide the mechanism to reduce message flooding and network overload. They also enhance the efficiency and reliability in delivering of RSVP tear or error messages. An RSVP aggregate message consists of an aggregate header followed by a body consisting of a variable number of standard RSVP messages. The following subsections define the formats of the aggregate header and the rules for including standard RSVP messages as part of message. 5.1. Aggregate Header 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Vers | Flags | Msg type | RSVP checksum | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Send_TTL | (Reserved) | RSVP length | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ The format of the aggregate header is identical to the format of the Swallow, editor [Page 33]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 RSVP common header [1]. The fields in the header are as follows: Vers: 4 bits Protocol version number. This is version 1. Flags: 4 bits 0x01: Aggregate capable If set, indicates to RSVP neighbors that this node is willing and capable of receiving aggregate messages. This bit is meaningful only between adjacent RSVP neighbors. 0x02-0x08: Reserved Msg type: 8 bits 12 = Aggregate RSVP checksum: 16 bits The one's complement of the one's complement sum of the entire message, with the checksum field replaced by zero for the purpose of computing the checksum. An all-zero value means that no checksum was transmitted. Because individual submessages carry their own checksum as well as INTEGRITY object for authentication, this field is recommended to be left as zero. Send_TTL: 8 bits The IP TTL value with which the message was sent. This is used by RSVP to detect a non-RSVP hop by comparing the IP TTL that an Aggregate message sent to the TTL in the received message. RSVP length: 16 bits The total length of this RSVP aggregate message in bytes, including the aggregate header and the submessages that follow. Swallow, editor [Page 34]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 5.2. Message Formats An RSVP aggregate message must contain at least one submessage. A submessage is one of the RSVP Path, PathTear, PathErr, Resv, ResvTear, ResvErr, or ResvConf messages. Empty RSVP aggregate message should not be sent. It is illegal to include another RSVP aggregate message a as submessage. 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Vers | Flags | 12 | RSVP checksum | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Send_TTL | (Reserved) | RSVP length | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | // First submessage // | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | // More submessage... // | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 5.3. Sending RSVP Aggregate Messages RSVP Aggregate messages are sent hop by hop between RSVP-capable neighbors as "raw" IP datagrams with protocol number 46. Raw IP datagrams are also intended to be used between an end system and the first/last hop router, although it is also possible to encapsulate RSVP messages as UDP datagrams for end-system communication that cannot perform raw network I/O. RSVP Aggregate messages should not be used if the next-hop RSVP neighbor does not support RSVP Aggregate messages. Methods for discovering such information include 1) Manual configuration. 2) Observing the Aggregate-capable bit (see below) in the received RSVP messages. Support for RSVP Aggregate message is optional. While it might help in scaling RSVP and in reducing processing overhead and bandwidth consumption, a node is not required to transmit every standard RSVP message in an Aggregate message. A node must always be ready to receive standard RSVP messages. The IP source address is local system that originated the Aggregate Swallow, editor [Page 35]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 message. The IP destination address is the next-hop node for which the submessages are intended. These addresses need not be identical if submessages are sent as standard RSVP messages. For example, the IP source address of Path and PathTear messages is the address of the sender it describes, while the IP destination address is the DestAddress for the session. These end-to-end addresses are overridden by hop-by-hop addresses while encapsulated in an Aggregate message. These addresses can easily be restored from the SENDER_TEMPLATE and SESSION objects within Path and PathTear messages. For Path and PathTear messages, the next-hop node can be learned by looking up DestAddress in forwarding table. RSVP Aggregate messages do not require the Router Alert IP option [RFC 2113] in their IP headers. This is because Aggregate messages are addressed directly to RSVP neighbors. Each RSVP Aggregate message must occupy exactly one IP datagram. If it exceeds the MTU, the datagram is fragmented by IP and reassembled at the recipient node. A single RSVP Aggregate message cannot exceed the maximum IP datagram size, approximately 64K bytes. 5.4. Receiving RSVP Aggregate Messages If the local system does not recognize or does not wish to accept an Aggregate message, the received messages shall be discarded without further analysis. The receiver next compares the IP TTL with which an Aggregate message is sent to the TTL with which it is received. If a non-RSVP hop is detected, the number of non-RSVP hops is recorded. It is used later in processing of sub-messages. Next, the receiver verifies the version number and checksum of the RSVP aggregate message. Discard message if any mismatch is found. Start decapsulating individual sub-messages. Each sub-message has its own complete message length and authentication information. Process each sub-message according to procedures in RFC 2209. Swallow, editor [Page 36]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 5.5. Forwarding RSVP Aggregate Messages RSVP Aggregate messages could be forwarded by routers in non-RSVP cloud. Aggregate messages shall not be forwarded RSVP routers. When individual submessages are being forwarded, they can be encapsulated in another aggregate message before sending to the next-hop neighbor. The Send_TTL field in the submessages should be decremented properly before transmission. 5.6. Aggregate-capable bit An additional bit is added to RSVP common header, which is defined in RFC2205 [1]. 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Vers | Flags | Msg Type | RSVP Checksum | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Send_TTL | (Reserved) | RSVP Length | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Flags: 4 bits 0x01: Aggregate capable If set, indicates to RSVP neighbors that this node is willing and capable of receiving aggregate messages. This bit is meaningful only between adjacent RSVP neighbors. 6. Tear Confirm The failure of an LSP Tunnel may result in loss of data. Locally decapsulating the packet and routing is not recommended as this has the potential of inducing routing loops and may also violate policy. It is thus desirable to make the teardown function reliable. Due to the overhead involved in refreshes, administrators may desire to set the refreseh timers longer. The Tear Confirm mechanism provides a means of ensuring timely teardown without the necessity of setting short refresh timers. This has the effect of both rapidly notifying the source that the tunnel is inoperative and of freeing the LSP's resources so they may be reallocated. Swallow, editor [Page 37]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 The reliable teardown is accomplished via a combination of (a) including the CONFIRM object in the Resv Tear message, and (b) new ResvTear Confirm message. The confirmations occur hop by hop. When the CONFIRM object is included in the ResvTear message, the code should: (a) set a boolean in the upstream RDB indicating that it is present only for retransmission purposes, and set a relatively short retransmission interval. The RDB remains subject to normal timeout mechanisms. If the node is a host, then the RDB should be timed out in a period based on its previous retransmission timer, i.e. (K + 0.5)*1.5*R, where R is the retransmission timer and K is a small integer with the default value of 3. (b) whenever the retransmission timer fires, if the boolean is FALSE, do as now: send a refreshing Resv message upstream and set a longish timer. If it is TRUE, however, set a relatively short timer and send an Resv Tear message with the CONFIRM object (c) when such an ResvTear message is received, similarly set the CONFIRM object and the boolean in the RDB if it exists, and send an Resv Tear message. If the RDB does not exist, reply ResvTear Confirm. (d) in the sender system, when the ResvTear with CONFIRM object is received, tear down the RDB and reply ResvTear Confirm to the NHOP. Do so whether the RDB exists on receipt or not. (e) in the non-sender systems, when the Resv Tear Confirm is received, tear down the RDB and forward Resv Tear Confirm to the next NHOP. Resv Tear Confirm is identical in content to the Resv Confirm, except that it results in the RDB being torn down, not established. The value of the MessageType for the Resv Tear Confirm message is 10 (need to get an official one from the IANA). (We may also want to define a new CONFIRM object with a C-Type >= 192 so that nodes which do not recognize the confirm object in the ResvTear message will not drop the message and will pass the object on.) Swallow, editor [Page 38]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 7. Security Considerations We assume that the security procedures defined for RSVP will handle any security issues that arise from coupling label switching with RSVP. For example, mechanisms that are used to authenticate RSVP resource reservation requests may also be used to authenticate requests to establish explicitly routed label switched paths. It may be desirable to enable the setup of ER-LSPs without enabling general purpose resource reservations. This would be done using the policy mechanisms defined for RSVP. It is likely that explicitly routed paths would often be setup only within a single administrative domain, and thus RSVP requests from outside the domain would be ignored. 8. Acknowledgments This document contains ideas as well as text which which have appeared in previous Internet Drafts. The editors/authors of the current draft wish to thank the authors of those drafts. They are Steven Blake, Bruce Davie, Roch Guerin, Sanjay Kamat, Yakov Rekhter, Eric Rosen, and Arun Viswanathan. 9. References [1] Braden, R. et al. Resource ReSerVation Protocol (RSVP) -- Version 1, Functional Specification, RFC 2205, September 1997. [2] Rosen, E. et al. A Proposed Architecture for MPLS, Internet Draft, draft-ietf-mpls-arch-00.txt, August 1997. [3] Davie, B. et al. Explicit Route Support in MPLS, Internet Draft, draft-davie-mpls-explicit-routes-00.txt, November 1997 [4] Davie, B. et al. Use of Label Switching With RSVP, Internet Draft, draft-ietf-mpls-rsvp-00.txt, March 1998. [5] Guerin, R. et al. Setting up Reservations on Explicit Paths using RSVP, Internet Draft, draft-guerin-expl-path-rsvp-01.txt, November 1997. [6] Awduche, D. et al. Requirements for Traffic Engineering over MPLS, Internet Draft, draft-awduche-mpls-traffic-eng-00.txt, April 1998. [7] Wroclawski, J. Specification of the Controlled-Load Network Element Service, RFC 2211, September 1997. Swallow, editor [Page 39]
Internet Draft draft-swallow-mpls-rsvp-trafeng-00.txt August 1998 [8] Rosen, E. MPLS Label Stack Encoding. Internet Draft, draft-ietf-mpls-label-encaps-01.txt, February 1998. Swallow, editor [Page 40]