Network Working Group A. Farrel (Editor) Internet Draft Old Dog Consulting Category: Standards Track Expires: July 2007 A. Satyanarayana Cisco Systems, Inc. A. Iwata N. Fujita NEC Corporation G. Ash AT&T January 2007 Crankback Signaling Extensions for MPLS and GMPLS RSVP-TE Status of this Memo By submitting this Internet-Draft, each author represents that any applicable patent or other IPR claims of which he or she is aware have been or will be disclosed, and any of which he or she becomes aware will be disclosed, in accordance with Section 6 of BCP 79. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet- Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt. The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html. Abstract In a distributed, constraint-based routing environment, the information used to compute a path may be out of date. This means that Multiprotocol Label Switching (MPLS) and Generalized MPLS (GMPLS) Traffic Engineered (TE) Label Switched Path (LSP) setup requests may be blocked by links or nodes without sufficient resources. Crankback is a scheme whereby setup failure information is returned from the point of failure to allow new setup attempts to be made avoiding the blocked resources. Crankback can also be applied to LSP recovery to indicate the location of the failed link or node. A. Farrel et al. Page 1 draft-ietf-ccamp-crankback-06.txt January 2007 This document specifies crankback signaling extensions for use in MPLS signaling using RSVP-TE as defined in "RSVP-TE: Extensions to RSVP for LSP Tunnels", RFC3209, and GMPLS signaling as defined in "Generalized Multi-Protocol Label Switching (GMPLS) Signaling Functional Description", RFC3473. These extensions mean that the LSP setup request can be retried on an alternate path that detours around blocked links or nodes. This offers significant improvements in the successful setup and recovery ratios for LSPs, especially in situations where a large number of setup requests are triggered at the same time. Table of Contents Section A : Problem Statement 1. Terminology......................................................3 1.1. Control Plane and Data Plane Separation........................3 2. Introduction and Framework.......................................4 2.1. Background.....................................................4 2.2. Repair and Recovery............................................5 2.3. Interaction with TE Flooding Mechanisms .......................6 3. Discussion: Explicit Versus Implicit Re-routing Indications......6 4. Required Operation...............................................7 4.1. Resource Failure or Unavailability.............................7 4.2. Computation of an Alternate Path...............................8 4.2.1 Information Required for Re-routing...........................8 4.2.2 Signaling a New Route.........................................9 4.3. Persistence of Error Information...............................9 4.4. Handling Re-route Failure.....................................10 4.5. Limiting Re-routing Attempts..................................10 5. Existing Protocol Support for Crankback Re-routing..............11 5.1. RSVP-TE ......................................................11 5.2. GMPLS-RSVP-TE ................................................12 Section B : Solution 6. Control of Crankback Operation..................................12 6.1. Requesting Crankback and Controlling In-Network Re-routing....12 6.2. Action on Detecting a Failure.................................13 6.3. Limiting Re-routing Attempts..................................13 6.3.1 New Status Codes for Re-routing..............................13 6.4. Protocol Control of Re-routing Behavior.......................14 7. Reporting Crankback Information.................................14 7.1. Required Information..........................................14 7.2. Protocol Extensions...........................................14 7.3 Guidance for Use of IF_ID Error Spec TLVs......................18 7.3.1 General Principles...........................................18 7.3.2 Error Report TLVs............................................19 7.3.3 Fundamental Crankback TLVs...................................19 7.3.4 Additional Crankback TLVs....................................20 A. Farrel et al. Page 2 draft-ietf-ccamp-crankback-06.txt January 2007 7.3.5 Grouping TLVs by Failure Location............................21 7.3.6 Alternate Path identification................................22 7.4. Action on Receiving Crankback Information.....................22 7.4.1 Re-route Attempts............................................22 7.4.2 Location Identifiers of Blocked Links or Nodes...............23 7.4.3 Locating Errors within Loose or Abstract Nodes...............23 7.4.4 When Re-routing Fails........................................23 7.4.5 Aggregation of Crankback Information.........................24 7.5. Notification of Errors........................................24 7.5.1 ResvErr Processing...........................................24 7.5.2 Notify Message Processing....................................25 7.6. Error Values..................................................25 7.7. Backward Compatibility........................................25 8. LSP Recovery Considerations.....................................26 8.1. Upstream of the Fault.........................................26 8.2. Downstream of the Fault.......................................27 9. IANA Considerations.............................................27 9.1. Error Codes...................................................27 9.2. IF_ID_ERROR_SPEC TLVs.........................................27 9.3. LSP_ATTRIBUTES Object.........................................28 10. Security Considerations........................................28 11. Acknowledgments................................................29 12. Normative References...........................................30 13. Informational References.......................................30 14. Authors' Addresses.............................................31 A. Experience of Crankback in TDM-based Networks..................34 Section A : Problem Statement 1. 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 [RFC2119]. 1.1. Control Plane and Data Plane Separation Throughout this document the processes and techniques are described as though the control plane and data plane elements that comprise a Label Switching Router (LSR) are coresident and related in a one-to-one manner. This is a convenience of documentation only. It should be noted that GMPLS LSRs may be decomposed such that the control plane components are not physically collocated. Further, one presence in the control plane may control more than one LSR in the data plane. These points have several consequences with respect to this document: A. Farrel et al. Page 3 draft-ietf-ccamp-crankback-06.txt January 2007 o The nodes, links and resources that are reported as in error, are data plane entities. o The nodes, areas and ASs that report that they have attempted re-routing, are control plane entities. o Where a single control plane entity is responsible for more than one data plane LSR, crankback signaling may be implicit in just the same way as LSP establishment signaling may be. The above points may be considered self-evident, but are stated here for absolute clarity. The stylistic convenience of referring to both the control plane element responsible for a single LSR and the data plane component of that LSR simply as "the LSR", should not be taken to mean that this document is applicable only to a collocated one-to-one relationship. Further, in the majority case the control plane and data plane components are related in a 1:1 ratio and are usually collocated. 2. Introduction and Framework 2.1. Background RSVP-TE (RSVP Extensions for LSP Tunnels) [RFC3209] can be used for establishing explicitly routed LSPs in an MPLS network. Using RSVP-TE, resources can also be reserved along a path to guarantee and/or control QoS for traffic carried on the LSP. To designate an explicit path that satisfies QoS guarantees, it is necessary to discern the resources available to each link or node in the network. For the collection of such resource information, routing protocols, such as OSPF and IS-IS , can be extended to distribute additional state information [RFC2702]. Explicit paths can be computed based on the distributed information at the LSR (ingress) initiating an LSP and signaled as Explicit Routes during LSP establishment. Explicit Routes may contain 'loose hops' and 'abstract nodes' that convey routing through a collection of nodes. This mechanism may be used to devolve parts of the path computation to intermediate nodes such as area border LSRs. In a distributed routing environment, however, the resource information used to compute a constraint-based path may be out of date. This means that a setup request may be blocked, for example, because a link or node along the selected path has insufficient resources. In RSVP-TE, a blocked LSP setup may result in a PathErr message sent to the ingress, or a ResvErr sent to the egress (terminator). These messages may result in the LSP setup being abandoned. In A. Farrel et al. Page 4 draft-ietf-ccamp-crankback-06.txt January 2007 Generalized MPLS [RC3473] the Notify message may additionally be used to expedite notification of failures of existing LSPs to ingress and egress LSRs, or to a specific "repair point" - an LSR responsible for performing protection or restoration. These existing mechanisms provide a certain amount of information about the path of the failed LSP. Generalized MPLS [RFC3471] and [RFC3473] extends MPLS into networks that manage Layer 2, TDM and lambda resources as well as packet resources. Thus, crankback routing is also useful in GMPLS networks. In a network without wavelength converters, setup requests are likely to be blocked more often than in a conventional MPLS environment because the same wavelength must be allocated at each Optical Cross-Connect on an end-to-end explicit path. This makes crankback routing all the more important in certain GMPLS networks. 2.2. Repair and Recovery If the ingress LSR or intermediate area border LSR knows the location of the blocked link or node, it can designate an alternate path and then reissue the setup request. Determination of the identity of the blocked link or node can be achieved by the mechanism known as crankback routing [PNNI, ASH1]. In RSVP-TE, crankback signaling requires notifying the upstream LSR of the location of the blocked link or node. In some cases this requires more information than is currently available in the signaling protocols. On the other hand, various recovery schemes for link or node failures have been proposed in [RFC3469] and include fast re-routing. These schemes rely on the existence of a protecting LSP to protect the working LSP, but if both the working and protecting paths fail it is necessary to re-establish the LSP on an end-to-end basis avoiding the known failures. Similarly, fast re-routing by establishing a recovery path on demand after failure requires computation of a new LSP that avoids the known failures. End-to-end recovery for alternate routing requires the location of the failed link or node. Crankback routing schemes could be used to notify the upstream LSRs of the location of the failure. Furthermore, in situations where many link or node failures occur at the same time, the difference between the distributed routing information and the real-time network state becomes much greater than in normal LSP setups. LSP recovery might, therefore, be performed with inaccurate information, which is likely to cause setup blocking. Crankback routing could improve failure recovery in these situations. The requirement for end-to-end allocation of lambda resources in GMPLS networks without wavelength converters means that end-to-end A. Farrel et al. Page 5 draft-ietf-ccamp-crankback-06.txt January 2007 recovery may be the only way to recover from LSP failures. This is because segment protection may be much harder to achieve in networks of photonic cross-connects where a particular lambda may already be in use on other links: End-to-end protection offers the choice of use of another lambda, but this choice is not available in segment protection. This requirement makes crankback re-routing particularly useful in a GMPLS network, in particular in dynamic LSP re-routing cases (no protecting LSP pre-establishment). 2.3. Interaction with TE Flooding Mechanisms GMPLS uses IGPs (OSPF and IS-IS) to flood traffic engineering (TE) information that is used to construct a traffic engineering database (TED) which acts as a data source for path computation. Crankback signaling is not intended to supplement or replace the normal operation of the TE flooding mechanism, since these mechanisms are independent of each other. That is, information gathered from crankback signaling may be applied to compute an alternate path for the LSP for which the information was signaled, but the information is not intended to be used to influence the computation of the paths of other LSPs. Any requirement to rapidly flood updates about resource availability so that they may be applied as deltas to the TED and utilized in future path computations are out of scope of this document. 3. Discussion: Explicit Versus Implicit Re-routing Indications There have been problems in service provider networks when "inferring" from indirect information that re-routing is allowed. This document proposes the use of an explicit re-routing indication that explicitly authorizes re-routing, and contrasts it with the inferred or implicit re-routing indication that has previously been used. Various existing protocol options and exchanges including the error values of PathErr message [RFC2205, RFC3209] and the Notify message [RFC3473] allow an implementation to infer a situation where re-routing can be performed. This allows for recovery from network errors or resource contention. However, such inference of recovery signaling is not always desirable since it may be doomed to failure. For example, experience of using release messages in TDM-based networks for analogous implicit and explicit re-routing indications purposes provides some guidance. This background information is given in Appendix A. A. Farrel et al. Page 6 draft-ietf-ccamp-crankback-06.txt January 2007 It is certainly the case that with topology information distribution as performed with routing protocols such as OSPF, the ingress LSR could infer the re-routing condition. However, convergence of topology information using routing protocols is typically slower than the expected LSP setup times. One of the reasons for crankback is to avoid the overhead of available-link-bandwidth flooding, and to more efficiently use local state information to direct alternate routing at the path computation point. [ASH1] shows how event-dependent-routing can just use crankback, and not available-link-bandwidth flooding, to decide on the re-route path in the network through "learning models". Reducing this flooding reduces overhead and can lead to the ability to support much larger AS sizes. Therefore, the use of alternate routing should be based on an explicit indication, and it is best to know the following information separately: - where blockage/congestion occurred - whether alternate routing "should" be attempted. 4. Required Operation Section 2 identifies some of the circumstances under which crankback may be useful. Crankback routing is performed as described in the following procedures, when an LSP setup request is blocked along the path, or when an existing LSP fails. 4.1. Resource Failure or Unavailability When an LSP setup request is blocked due to unavailable resources, an error message response with the location identifier of the blockage should be returned to the LSR initiating the LSP setup (ingress LSR), the area border LSR, the AS border LSR, or to some other repair point. This error message carries an error specification according to [RFC3209] - this indicates the cause of the error and the node/link on which the error occurred. Crankback operation may require further information as detailed in sections 4.2.1 and 7. A repair point (for example, an ingress LSR) that receives crankback information resulting from the failure of an established LSP may apply local policy to govern how it attempts repair of the LSP. For example, it may prioritise repair attempts between multiple LSPs that have failed, and it may consider LSPs that have been locally repaired ([RFC4090]) to be less urgent candidates for end-to-end repair. Further, there is a likelihood that other LSRs are also attempting LSP repair for LSPs affected by the same fault which may give rise to A. Farrel et al. Page 7 draft-ietf-ccamp-crankback-06.txt January 2007 resource contention within the network, so an LSR may stagger its repair attempts in order to reduce the chance of resource contention. 4.2. Computation of an Alternate Path In a flat network without partitioning of the routing topology, when the ingress LSR receives the error message it computes an alternate path around the blocked link or node to satisfy QoS guarantees using link state information about the network. If an alternate path is found, a new LSP setup request is sent over this path. On the other hand, in a network partitioned into areas such as with OSPF, the area border LSR may intercept and terminate the error response, and perform alternate (re-)routing within the downstream area. In a third scenario, any node within an area may act as a repair point. In this case, each LSR behaves much as an area border LSR as described above. It can intercept and terminate the error response, and perform alternate routing. This may be particularly useful where domains of computation are applied within the (partitioned) network, where such domains are not coincident on the routing partition boundaries. However if, all nodes in the network perform re-routing it is possible to spend excessive network and CPU resources on re-routing attempts that would be better made only at designated re-routing nodes. This scenario is somewhat like 'MPLS fast re-route' [RFC4090], in which any node in the MPLS domain can establish 'local repair' LSPs upon failure notification. 4.2.1 Information Required for Re-routing In order to correctly compute a route that avoids the blocking problem, a repair point LSR must gather as much crankback information as possible. Ideally, the repair node will be given the node, link and reason for the failure. The reason for the failure may provide an important discriminator to help decide what action should be taken. For example, a failure that indicates "No Route to Destination" is likely to give rise to a new path computation excluding the reporting LSR, but the reason "Temporary Control Plane Congestion" might lead to a simple retry after a suitable pause. However, even this information may not be enough to help with re-computation. Consider for instance an explicit route that contains a non-explicit abstract node or a loose hop. In this case, the failed node and link is not necessarily enough to tell the repair point which hop in the explicit route has failed. The crankback information needs to indicate where, within the explicit route, the problem has occurred. A. Farrel et al. Page 8 draft-ietf-ccamp-crankback-06.txt January 2007 4.2.2 Signaling a New Route If the crankback information can be used to compute a new route avoiding the failed/blocking network resource, the route can be signaled as an Explicit Route. However, it may be that the repair point does not have sufficient topology information to compute an Explicit Route that is guaranteed to avoid the failed link or node. In this case, Route Exclusions [EXCLUDE] may be particularly helpful. To achieve this, [EXCLUDE] allows the crankback information to be presented as route exclusions to force avoidance of the failed node, link or resource. 4.3. Persistence of Error Information The repair point LSR that computes the alternate path should store the location identifiers of the blockages indicated in the error message until the LSP is successfully established by downstream LSRs or until the repair point LSR abandons re-routing attempts. Since crankback signaling information may be returned to the same repair point LSR more than once while establishing a specific LSP, the repair point LSR SHOULD maintain a history table of all experienced blockages for this LSP (at least until the routing protocol updates the state of this information) so that the resulting path computation(s) can detour all blockages. If a second error response is received by a repair point (while it is performing crankback re-routing) it should update the history table that lists all experienced blockages, and use the entire gathered information when making a further re-routing attempt. Note that the purpose of this history table is to correlate information when repeated retry attempts are made by the same LSR. For example, suppose that an attempt is made to route from A through B, and B returns a failure with crankback information, an attempt may be made to route from A through C, and this may also fail with the return of crankback information - the next attempt SHOULD NOT be to route from A through B, and this may be achieved by use of the history table. The history table can be discarded by the signaling controller for A if the LSP is successfully established through A. The history table MAY be retained after the signaling controller for A sends an error upstream, however it is questionable what value this provides since a future retry as a result of crankback re-routing should not attempt to route through A. If the history information is retained for a longer period it SHOULD be discarded after a local timeout has expired. This timer is required so that the repair point does not apply the history table to an attempt by the ingress to re-establish A. Farrel et al. Page 9 draft-ietf-ccamp-crankback-06.txt January 2007 a failed LSP, but to allow the history table to be available for use in re-routing attempts before the ingress declares the LSP as failed. It is RECOMMENDED that the repair point LSR discard the history table using a timer no larger than the LSP retry timer configured on the ingress LSR. The correlation of the timers between the ingress and repair point LSRs is typically by manual configuration of timers local to each LSR, and is outside the scope of this document. It is not intended that the information in the history table be used to supplement the TED for the computation of paths of other LSPs. 4.4. Handling Re-route Failure Multiple blockages (for the same LSP) may occur, and successive setup retry attempts may fail. Retaining error information from previous attempts ensures that there is no thrashing of setup attempts, and knowledge of the blockages increases with each attempt. It may be that after several retries, a given repair point is unable to compute a path to the destination (that is, the egress of the LSP) that avoids all of the blockages. In this case, it must pass an error indication message upstream. It is most useful to the upstream nodes (and in particular to the ingress LSR) that may as repair points for the LSP setup, if the error indication message identifies all of the downstream blockages and also the repair point that was unable to compute an alternate path. 4.5. Limiting Re-routing Attempts It is important to prevent endless repetition of LSP setup attempts using crankback routing information after error conditions are signaled, or during periods of high congestion. It may also be useful to reduce the number of retries, since failed retries will increase setup latency and degrade performance by increasing the amount of signaling processing and message exchanges within the network. The maximum number of crankback re-routing attempts allowed may be limited in a variety of ways. This document allows an LSR to limit the retries per LSP, and assumes that such a limit will be applied either as a per node configuration for those LSRs that are capable of re-routing, or as a network-wide configuration value. When the number of retries at a particular LSR is exceeded, the LSR reports the failure in an upstream direction until it reaches the next repair point where further re-routing attempts may be attempted, or reaches the ingress which may act as a repair point or declare the LSP as faioled. It is important that the crankback information provided indicates that routing back through this node will not succeed - this situation is similar to that in section 4.4. A. Farrel et al. Page 10 draft-ietf-ccamp-crankback-06.txt January 2007 5. Existing Protocol Support for Crankback Re-routing Crankback re-routing is appropriate for use with RSVP-TE. 1) LSP establishment may fail because of an inability to route, perhaps because links are down. In this case a PathErr message is returned to the ingress. 2) LSP establishment may fail because resources are unavailable. This is particularly relevant in GMPLS where explicit label control may be in use. Again, a PathErr message is returned to the ingress. 3) Resource reservation may fail during LSP establishment, as the Resv is processed. If resources are not available on the required link or at a specific node, a ResvErr message is returned to the egress node indicating "Admission Control failure" [RFC2205]. The egress is allowed to change the FLOWSPEC and try again, but in the event that this is not practical or not supported (particularly in the non-PSC context), the egress LSR may choose to take any one of the following actions. - Ignore the situation and allow recovery to happen through Path refresh message and refresh timeout [RFC2205]. - Send a PathErr message towards the ingress indicating "Admission Control failure". Note that in multi-area/AS networks, the ResvErr might be intercepted and acted on at an area/AS border router. 4) It is also possible to make resource reservations on the forward path as the Path message is processed. This choice is compatible with LSP setup in GMPLS networks [RFC3471], [RFC4373]. In this case if resources are not available, a PathErr message is returned to ingress indicating "Admission Control failure". Crankback information would be useful to an upstream node (such as the ingress) if it is supplied on a PathErr or a Notify message that is sent upstream. 5.1. RSVP-TE In RSVP-TE a failed LSP setup attempt results in a PathErr message returned upstream. The PathErr message carries an ERROR_SPEC object, which indicates the node or interface reporting the error and the reason for the failure. Crankback re-routing can be performed explicitly avoiding the node or interface reported. A. Farrel et al. Page 11 draft-ietf-ccamp-crankback-06.txt January 2007 5.2. GMPLS-RSVP-TE GMPLS extends the error reporting described above by allowing LSRs to report the interface that is in error in addition to the identity of the node reporting the error. This further enhances the ability of a re-computing node to route around the error. GMPLS introduces a targeted Notify message that may be used to report LSP failures direct to a selected node. This message carries the same error reporting facilities as described above. The Notify message may be used to expedite the propagation of error notifications, but in a network that offers crankback routing at multiple nodes there would need to be some agreement between LSRs as to whether PathErr or Notify provides the stimulus for crankback operation. This agreement is constrained by the re-routing behavior selection (as listed in section 6.4). Otherwise, multiple nodes might attempt to repair the LSP at the same time, because: 1) these messages can flow through different paths before reaching the ingress LSR, and 2) the destination of the Notify message might not be the ingress LSR. Section B : Solution 6. Control of Crankback Operation 6.1. Requesting Crankback and Controlling In-Network Re-routing When a request is made to set up an LSP tunnel, the ingress LSR should specify whether it wants crankback information to be collected in the event of a failure, and whether it requests re-routing attempts by any or specific intermediate nodes. For this purpose, a Re-routing Flag field is added to the protocol setup request messages. The corresponding values are mutually exclusive. No Re-routing The ingress node MAY attempt re-routing after failure. Intermediate nodes SHOULD NOT attempt re-routing after failure. Nodes detecting failures MUST report an error and MAY supply crankback information. This is the default and backwards compatible option. End-to-end Re-routing The ingress node MAY attempt re-routing after failure. Intermediate nodes SHOULD NOT attempt re-routing after failure. Nodes detecting failures MUST report an error and SHOULD supply crankback information. A. Farrel et al. Page 12 draft-ietf-ccamp-crankback-06.txt January 2007 Boundary Re-routing Intermediate nodes MAY attempt re-routing after failure only if they are Area Border Routers or AS Border Routers. The boundary (ABR/ASBR) can either decide to forward the error message upstream to the ingress LSR or try to select another egress boundary LSR. Other intermediate nodes SHOULD NOT attempt re-routing. Nodes detecting failures MUST report an error and SHOULD supply crankback information. Segment-based Re-routing Any node MAY attempt re-routing after it receives an error report and before it passes the error report further upstream. Nodes detecting failures MUST report an error and SHOULD supply full crankback information. 6.2. Action on Detecting a Failure A node that detects the failure to setup an LSP or the failure of an established LSP SHOULD act according to the Re-routing Flag passed on the LSP setup request. If Segment-based Re-routing is allowed, or if Boundary Re-routing is allowed and the detecting node is an ABR or ASBR, the detecting node MAY immediately attempt to re-route. If End-to-end Re-routing is indicated, or if Segment-based or Boundary Re-routing is allowed and the detecting node chooses not to make re-routing attempts (or has exhausted all possible re-routing attempts), the detecting node MUST return a protocol error indication and SHOULD include full crankback information. 6.3. Limiting Re-routing Attempts Each repair point SHOULD apply a locally configurable limit to the number of attempts it makes to re-route an LSP. This helps to prevent excessive network usage in the event of significant faults, and allows back-off to other repair points which may have a better chance of routing around the problem. 6.3.1 New Status Codes for Re-routing An error code/value of "Routing Problem"/"Re-routing limit exceeded" (24/TBD) is used to identify that a node has abandoned crankback re-routing because it has reached a threshold for retry attempts. A node receiving an error response with this status code MAY also attempt crankback re-routing, but it is RECOMMENDED that such attempts be limited to the ingress LSR. A. Farrel et al. Page 13 draft-ietf-ccamp-crankback-06.txt January 2007 6.4. Protocol Control of Re-routing Behavior The LSP_ATTRIBUTES object defined in [RFC4420] is used on Path messages to convey the Re-Routing Flag described in section 5.1. Three bits are defined for inclusion in the LSP Attributes TLV as follows. The bit numbers below are suggested and actual values are to be assigned by IANA. Bit Name and Usage Number 1 End-to-end re-routing desired. This flag indicates the end-to-end re-routing behavior for an LSP under establishment. This MAY also be used for specifying the behavior of end-to-end LSP recovery for established LSPs. 2 Boundary re-routing desired. This flag indicates the boundary re-routing behavior for an LSP under establishment. This MAY also be used for specifying the segment-based LSP recovery through nested crankback for established LSPs. The boundary ABR/ASBR can either decide to forward the PathErr message upstream to an upstream boundary ABR/ASBR or to the ingress LSR. Alternatively, it can try to select another egress boundary LSR. 3 Segment-based re-routing desired. This flag indicates the segment-based re-routing behavior for an LSP under establishment. This MAY also be used to specify the segment-based LSP recovery for established LSPs. 7. Reporting Crankback Information 7.1. Required Information As described above, full crankback information SHOULD indicate the node, link and other resources, which have been attempted but have failed because of allocation issues or network failure. The default crankback information SHOULD include the interface and the node address. Any address reported in such crankback information SHOULD be an address that was distributed by the routing protocols (OSPF and ISIS) in their TE link state advertisements. However, some additional information such as component link identifiers is additional to this. 7.2. Protocol Extensions [RFC3473] defines an IF_ID ERROR_SPEC object that can be used on PathErr, ResvErr and Notify messages to convey the information A. Farrel et al. Page 14 draft-ietf-ccamp-crankback-06.txt January 2007 carried in the Error Spec Object defined in [RFC3209]. Additionally, the IF_ID ERROR_SPEC Object has scope for carrying TLVs that identify the link associated with the error. The TLVs for use with this object are defined in [RFC3471], and are listed below. They are used to identify links in the IF_ID RSVP_HOP Object, and in the IF_ID ERROR_SPEC object to identify the failed resource which is usually the downstream resource from the reporting node. Type Length Format Description -------------------------------------------------------------------- 1 8 IPv4 Addr. IPv4 (Interface address) 2 20 IPv6 Addr. IPv6 (Interface address) 3 12 Compound IF_INDEX (Interface index) 4 12 Compound COMPONENT_IF_DOWNSTREAM (Component interface) 5 12 Compound COMPONENT_IF_UPSTREAM (Component interface) Note that TLVs 4 and 5 are obsoleted by [RFC4201] and SHOULD NOT be used to identify component interfaces in IF_ID ERROR_SPEC objects. In order to facilitate reporting of crankback information, the following additional TLVs are defined. Note that the Type values shown here are only suggested values - final values are TBD and to be determined by IETF consensus. Type Length Format Description -------------------------------------------------------------------- 6 var See below DOWNSTREAM_LABEL (GMPLS label) 7 var See below UPSTREAM_LABEL (GMPLS label) 8 8 See below NODE_ID (TE Router ID) 9 x See below OSPF_AREA (Area ID) 10 x See below ISIS_AREA (Area ID) 11 8 See below AUTONOMOUS_SYSTEM (Autonomous system) 12 var See below ERO_CONTEXT (ERO subobject) 13 var See below ERO_NEXT_CONTEXT (ERO subobjects) 14 8 IPv4 Addr. PREVIOUS_HOP_IPv4 (Node address) 15 20 IPv6 Addr. PREVIOUS_HOP_IPv6 (Node address) 16 8 IPv4 Addr. INCOMING_IPv4 (Interface address) 17 20 IPv6 Addr. INCOMING_IPv6 (Interface address) 18 12 Compound INCOMING_IF_INDEX (Interface index) 19 var See below INCOMING_DOWN_LABEL (GMPLS label) 20 var See below INCOMING_UP_LABEL (GMPLS label) 21 8 See below REPORTING_NODE_ID (Router ID) 22 x See below REPORTING_OSPF_AREA (Area ID) 23 x See below REPORTING_ISIS_AREA (Area ID) 24 8 See below REPORTING_AS (Autonomous system) 25 var See below PROPOSED_ERO (ERO subobjects) 26 var See below NODE_EXCLUSIONS (List of nodes) A. Farrel et al. Page 15 draft-ietf-ccamp-crankback-06.txt January 2007 27 var See below LINK_EXCLUSIONS (List of interfaces) For types 1, 2 and 3 the format of the Value field is already defined in [RFC3471]. For types 14 and 16, they format of the Value field is the same as for type 1. For types 15 and 17, the format of the Value field is the same as for type 2. For type 18 the format of the Value field is the same as for type 3. For types 6, 7, 19 and 20 the length field is variable and the Value field is a label as defined in [RFC3471]. As with all uses of labels, it is assumed that any node that can process the label information knows the syntax and semantics of the label from the context. Note that all TLVs are zero-padded to a multiple four octets so that if a label is not itself a multiple of four octets it must be disambiguated from the trailing zero pads by knowledge derived from the context. For types 8 and 21 the Value field has the 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 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Router ID | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Router ID: 32 bits The TE Router ID (TLV type 8) or the Router ID (TLV type 21) used to identify the node within the IGP. For types 9 and 22 the Value field has the 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 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | OSPF Area Identifier | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ OSPF Area Identifier The 4-octet area identifier for the node. This identifies the area where the failure has occurred. A. Farrel et al. Page 16 draft-ietf-ccamp-crankback-06.txt January 2007 For types 10 and 23 the Value field has the 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 | ISIS Area Identifier | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ~ ISIS Area Identifier (continued) ~ +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Length Length of the actual (non-padded) IS-IS Area Identifier in octets. Valid values are from 2 to 11 inclusive. ISIS Area Identifier The variable-length IS-IS area identifier. Padded with trailing zeroes to a four-octet boundary. For types 11 and 24 the Value field has the 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 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Autonomous System Number | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Autonomous System Number: 32 bits The AS Number of the associated Autonomous System. Note that if 16-bit AS numbers are in use, the low order bits (16 through 31) should be used and the high order bits (0 through 15) should be set to zero. For types 12, 13 and 25 the Value field has the 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 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | ~ ERO Subobjects ~ | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ERO Subobjects: A sequence of ERO subobjects. Any ERO subobjects are allowed whether defined in [RFC3209], [RFC3473] or other documents. Note that ERO subobjects contain their own types and lengths. A. Farrel et al. Page 17 draft-ietf-ccamp-crankback-06.txt January 2007 For type 26 the Value field has the 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 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | ~ Node Identifiers ~ | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Node Identifiers: A sequence of TLVs as defined here of types 1, 2 or 8 that indicates downstream nodes that have already participated in crankback attempts and have been declared unusable for the current LSP setup attempt. Note that an interface identifier may be used to identify a node. For type 27 the Value field has the 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 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | ~ Link Identifiers ~ | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Link Identifiers: A sequence of TLVs as defined here of the same format as type 1, 2 or 3 TLVs that indicate incoming interfaces at downstream nodes that have already participated in crankback attempts and have been declared unusable for the current LSP setup attempt. 7.3 Guidance for Use of IF_ID ERROR_SPEC TLVs 7.3.1 General Principles If crankback is not being used, inclusion of an IF-ID ERROR_SPEC object in PathErr, ResvErr and Notify messages follows the processing rules defined in [RFC3473] and [RFC4201]. A sender MAY include additional TLVs of types 6 through 27 to report crankback information for informational/monitoring purposes. If crankback is being used, the sender of a PathErr, ResvErr or Notify message MUST use the IF_ID ERROR_SPEC object and MUST include at least one of the TLVs in the range 1 through 3 as described in [RFC3473], [RFC4201], and the previous paragraph. Additional TLVs SHOULD also be included to report further information. The following A. Farrel et al. Page 18 draft-ietf-ccamp-crankback-06.txt January 2007 section gives advice on which TLVs should be used under different circumstances, and which TLVs must be supported by LSRs. Note that all such additional TLVs are optional and MAY be omitted. Inclusion of the optional TLVs SHOULD be performed where doing so helps to facilitate error reporting and crankback. The TLVs fall into three categories: those that are essential to report the error, those that provide additional information that is or may be fundamental to the utility of crankback, and those that provide additional information that may be useful for crankback in some circumstances. Note that all LSRs MUST be prepared to receive and forward any TLV as per [RFC3473]. This includes TLVs of type 4 or 5 as defined in [RFC3473] and obsoleted by [RFC4201]. There is, however, no requirement for an LSR to actively process any but the TLVs defined in [RFC3473]. An LSR that proposes to perform crankback re-routing SHOULD support receipt and processing of all of the fundamental crankback TLVs, and is RECOMMENDED to support the receipt and processing of the additional crankback TLVs. It should be noted, however, that some assumptions about the TLVs that will be used MAY be made based on the deployment scenarios. For example, a router that is deployed in a single-area network does not need to support the receipt and processing of TLV types 22 and 23. Those TLVs might be inserted in an IF_ID ERROR_SPEC object, but would not need to be processed by the receiver of a PathErr message. 7.3.2 Error Report TLVs Error Report TLVs are those in the range 1 through 3. (Note that the obsoleted TLVs 4 and 5 may be considered in this category, but SHOULD NOT be used.) As stated above, when crankback information is reported, the IF_ID ERROR_SPEC object MUST be used. When the IF_ID ERROR_SPEC object is used, at least one of the TLVs in the range 1 through 3 MUST be present. The choice of which TLV to use will be dependent on the circumstance of the error and device capabilities. For example, a device that does not support IPv6 will not need the ability to create a TLV of type 2. Note, however, that such a device MUST still be prepared to receive and process all error report TLVs. 7.3.3 Fundamental Crankback TLVs Many of the TLVs report the specific resource that has failed. For example, TLV type 1 can be used to report that the setup attempt was blocked by some form of resource failure on a specific interface identified by the IP address supplied. TLVs in this category are 1 through 11, although TLVs 4 and 5 may be considered to be excluded from this category by dint of having been obsoleted. A. Farrel et al. Page 19 draft-ietf-ccamp-crankback-06.txt January 2007 These TLVs SHOULD be supplied whenever the node detecting and reporting the failure with crankback information has the information available. (Note that some of these TLVs MUST be included as described in the previous two sections.) The TLVs of type 8, 9, 10 and 11 MAY, however, be omitted according to local policy and relevance of the information. 7.3.4 Additional Crankback TLVs Some TLVs help to locate the fault within the context of the path of the LSP that was being set up. TLVs of types 12, 13, 14 and 15 help to set the context of the error within the scope of an explicit path that has loose hops or non-precise abstract nodes. The ERO context information is not always a requirement, but a node may notice that it is a member of the next hop in the ERO (such as a loose or non-specific abstract node) and deduce that its upstream neighbor may have selected the path using next hop routing. In this case, providing the ERO context will be useful to the upstream node that performs re-routing. Note the distinction between TLVs 12 and 13 is the distinction between "this is the hop I was trying to satisfy when I failed" and "this is the next hop I was trying to reach when I failed. Reporting nodes SHOULD also supply TLVs from the range 12 through 20 as appropriate for reporting the error. The reporting nodes MAY also supply TLVs from the range 21 through 27. Note that in deciding whether a TLV in the range 12 through 20 "is appropriate", the reporting node should consider amongst other things, whether the information is pertinent to the cause of the failure. For example, when a cross-connection fails it may be that the outgoing interface is faulted, in which case only the interface (for example, TLV type 1) needs to be reported, but if the problem is that the incoming interface cannot be connected to the outgoing interface because of temporary or permanent cross-connect limitations, the node should also include reference to the incoming interface (for example, TLV type 16). Four TLVs (21, 22, 23 and 24) allow the location of the reporting node to be expanded upon. These TLVs would not be included if the information is not of use within the local system, but might be added by ABRs relaying the error. Note that the Reporting Node ID (TLV 21) need not be included if the IP address of the reporting node as indicated in the ERROR_SPEC itself, is sufficient to fully identify the node. The last three TLVs (25, 26, and 27) provide additional information for recomputation points. The reporting node (or some node forwarding A. Farrel et al. Page 20 draft-ietf-ccamp-crankback-06.txt January 2007 the error) MAY make suggestions about how the error could have been avoided, for example by supplying a partial ERO that would cause the LSP to be successfully set up if it were used. As the error propagates back upstream and as crankback routing is attempted and fails, it is beneficial to collect lists of failed nodes and links so that they will not be included in further computations performed at upstream nodes. These lists may also be factored into route exclusions [EXCLUDE]. Note that there is no ordering requirement on any of the TLVs within the IF_ID Error Spec, and no implication should be drawn from the ordering of the TLVs in a received IF_ID Error Spec. The decision of precisely which TLV types a reporting node includes is dependent on the specific capabilities of the node, and is outside the scope of this document. 7.3.5 Grouping TLVs by Failure Location Further guidance as to the inclusion of crankback TLVs can be given by grouping the TLVs according to the location of the failure and the context within which it is reported. For example, a TLV that reports an area identifier would only need to be included as the crankback error report transits an area boundary. Resource Failure 6 DOWNSTREAM_LABEL 7 UPSTREAM_LABEL Interface failures 1 IPv4 2 IPv6 3 IF_INDEX 4 COMPONENT_IF_DOWNSTREAM (obsoleted) 5 COMPONENT_IF_UPSTREAM (obsoleted) 12 ERO_CONTEXT 13 ERO_NEXT_CONTEXT 14 PREVIOUS_HOP_IPv4 15 PREVIOUS_HOP_IPv6 16 INCOMING_IPv4 17 INCOMING_IPv6 18 INCOMING_IF_INDEX 19 INCOMING_DOWN_LABEL 20 INCOMING_UP_LABEL Node failures 8 NODE_ID 21 REPORTING_NODE_ID Area failures 9 OSPF_AREA 10 ISIS_AREA 22 REPORTING_OSPF_AREA A. Farrel et al. Page 21 draft-ietf-ccamp-crankback-06.txt January 2007 23 REPORTING_ISIS_AREA 25 PROPOSED_ERO 26 NODE_EXCLUSIONS 27 LINK_EXCLUSIONS AS failures 11 AUTONOMOUS_SYSTEM 24 REPORTING_AS Although discussion of aggregation of crankback information is out of the scope of this document, it should be noted that this topic is closely aligned to the information presented here. Aggregation is discussed further in section 7.4.5. 7.3.6 Alternate Path Identification No new object is used to distinguish between Path/Resv messages for an alternate LSP. Thus, the alternate LSP uses the same SESSION and SENDER_TEMPLATE/FILTER_SPEC objects as the ones used for the initial LSP under re-routing. 7.4. Action on Receiving Crankback Information 7.4.1 Re-route Attempts As described in section 3, a node receiving crankback information in a PathErr must first check to see whether it is allowed to perform re-routing. This is indicated by the Re-routing Flags in the LSP_ATTRIBUTES object during LSP setup request. If a node is not allowed to perform re-routing it should forward the PathErr message, or if it is the ingress report the LSP as having failed. If re-routing is allowed, the node should attempt to compute a path to the destination using the original (received) explicit path and excluding the failed/blocked node/link. The new path should be added to an LSP setup request as an explicit route and signaled. LSRs performing crankback re-routing should store all received crankback information for an LSP until the LSP is successfully established or until the node abandons its attempts to re-route the LSP. On the next crankback re-routing path computation attempt, the LSR should exclude all the failed nodes and links and resources reported from previous attempts. It is an implementation decision whether the crankback information is discarded immediately upon successful LSP establishment or retained for a period in case the LSP fails. A. Farrel et al. Page 22 draft-ietf-ccamp-crankback-06.txt January 2007 7.4.2 Location Identifiers of Blocked Links or Nodes In order to compute an alternate path by crankback re-routing, it is necessary to identify the blocked links or nodes and their locations. The common identifier of each link or node in an MPLS network should be specified. Both protocol-independent and protocol-dependent identifiers may be specified. Although a general identifier that is independent of other protocols is preferable, there are a couple of restrictions on its use as described in the following subsection. In link state protocols such as OSPF and IS-IS, each link and node in a network can be uniquely identified. For example, by the context of a TE Router ID and the Link ID. If the topology and resource information obtained by OSPF advertisements is used to compute a constraint-based path, the location of a blockage can be represented by such identifiers. Note that, when the routing-protocol-specific link identifiers are used, the Re-routing Flag on the LSP setup request must have been set to show support for boundary or segment-based re-routing. In this document, we specify routing protocol specific link and node identifiers for OSPFv2, OSPFv3, and IS-IS for IPv4 and IPv6. These identifiers may only be used if segment-based re-routing is supported, as indicated by the Routing Behavior flag on the LSP setup request. 7.4.3 Locating Errors within Loose or Abstract Nodes The explicit route on the original LSP setup request may contain a loose or an Abstract Node. In these cases, the crankback information may refer to links or nodes that were not in the original explicit route. In order to compute a new path, the repair point may need to identify the pair of hops (or nodes) in the explicit route between which the error/blockage occurred. To assist this, the crankback information reports the top two hops of the explicit route as received at the reporting node. The first hop will likely identify the node or the link, the second hop will identify a 'next' hop from the original explicit route. 7.4.4 When Re-routing Fails When a node cannot or chooses not to perform crankback re-routing it must forward the PathErr message further upstream. However, when a node was responsible for expanding or replacing the explicit route as the LSP setup was processed it MUST update the A. Farrel et al. Page 23 draft-ietf-ccamp-crankback-06.txt January 2007 crankback information with regard to the explicit route that it received. Only if this is done will the upstream nodes stand a chance of successfully routing around the problem. 7.4.5 Aggregation of Crankback Information When a setup blocking error or an error in an established LSP occurs and crankback information is sent in an error notification message, some node upstream may choose to attempt crankback re-routing. If that node's attempts at re-routing fail the node will accumulate a set of failure information. When the node gives up it MUST propagate the failure message further upstream and include crankback information when it does so. Including a full list of all failures that have occurred due to multiple crankback failures by multiple repair point LSRs downstream could lead to too much information signaled using the protocol extensions described in this document. A compression mechanism for such information is available using TLVs 26 and 27. These TLVs allow for a more concise accumulation of failure information as crankback failures are propagated upstream. Aggregation may involve reporting all links from a node as unusable by flagging the node as unusable, flagging an ABR as unusable when there is no downstream path available, or including a TLV of type 9 which results in the exclusion of the entire area, and so on. The precise details of how aggregation of crankback information is performed are beyond the scope of this document. 7.5. Notification of Errors 7.5.1 ResvErr Processing As described above, the resource allocation failure for RSVP-TE may occur on the reverse path when the Resv message is being processed. In this case, it is still useful to return the received crankback information to the ingress LSR. However, when the egress LSR receives the ResvErr message, per [RFC2205] it still has the option of re-issuing the Resv with different resource requirements (although not on an alternate path). When a ResvErr carrying crankback information is received at an egress LSR, the egress LSR MAY ignore this object and perform the same actions as for any other ResvErr. However, if the egress LSR supports the crankback extensions defined in this document, and after all local recovery procedures have failed, it SHOULD generate a PathErr message carrying the crankback information and send it to the ingress LSR. If a ResvErr reports on more than one FILTER_SPEC (because the Resv A. Farrel et al. Page 24 draft-ietf-ccamp-crankback-06.txt January 2007 carried more than one FILTER_SPEC) then only one set of crankback information should be present in the ResvErr and it should apply to all FILTER_SPEC carried. In this case, it may be necessary per [RFC2205] to generate more than one PathErr. 7.5.2 Notify Message Processing [RFC3473] defines the Notify message to enhance error reporting in RSVP-TE networks. This message is not intended to replace the PathErr and ResvErr messages. The Notify message is sent to addresses requested on the Path and Resv messages. These addresses could (but need not) identify the ingress and egress LSRs respectively. When a network error occurs, such as the failure of link hardware, the LSRs that detect the error MAY send Notify messages to the requested addresses. The type of error that causes a Notify message to be sent is an implementation detail. In the event of a failure, an LSR that supports [RFC3473] and the crankback extensions defined in this document MAY choose to send a Notify message carrying crankback information. This would ensure a speedier report of the error to the ingress/egress LSRs. 7.6. Error Values Error values for the Error Code "Admission Control Failure" are defined in [RFC2205]. Error values for the error code "Routing Problem" are defined in [RFC3209] and [RFC3473]. A new error value is defined for the error code "Routing Problem". "Re-routing limit exceeded" indicates that re-routing has failed because the number of crankback re-routing attempts has gone beyond the predetermined threshold at an individual LSR. 7.7. Backward Compatibility It is recognized that not all nodes in an RSVP-TE network will support the extensions defined in this document. It is important that an LSR that does not support these extensions can continue to process a PathErr, ResvErr or Notify message even if it carries the newly defined IF_ID ERROR_SPEC information (TLVs). This document introduces no backward compatibility issues provided that existing implementations conform to the TLV processing rules defined in [RFC3471] and [RFC3473]. A. Farrel et al. Page 25 draft-ietf-ccamp-crankback-06.txt January 2007 8. LSP Recovery Considerations LSP recovery is performed to recover an established LSP when a failure occurs along the path. In the case of LSP recovery, the extensions for crankback re-routing explained above can be applied for improving performance. This section gives an example of applying the above extensions to LSP recovery. The goal of this example is to give a general overview of how this might work, and not to give a detailed procedure for LSP recovery. Although there are several techniques for LSP recovery, this section explains the case of on-demand LSP recovery, which attempts to set up a new LSP on demand after detecting an LSP failure. 8.1. Upstream of the Fault When an LSR detects a fault on an adjacent downstream link or node, a PathErr message is sent upstream. In GMPLS, the ERROR_SPEC object may carry a Path_State_Remove_Flag indication. Each LSR receiving the message then releases the corresponding LSP. (Note that if the state removal indication is not present on the PathErr message, the ingress node MUST issue a PathTear message to cause the resources to be released.) If the failed LSP has to be recovered at an upstream LSR, the IF_ID ERROR SPEC that includes the location information of the failed link or node is included in the PathErr message. The ingress, intermediate area border LSR, or indeed any repair point permitted by the Re-routing Flags, that receives the PathErr message can terminate the message and then perform alternate routing. In a flat network, when the ingress LSR receives the PathErr message with the IF_ID ERROR_SPEC TLVs, it computes an alternate path around the blocked link or node satisfying the QoS guarantees. If an alternate path is found, a new Path message is sent over this path toward the egress LSR. In a network segmented into areas, the following procedures can be used. As explained in Section 6.4, the LSP recovery behavior is indicated in the Flags field of the LSP_ATTRIBUTES object of the Path message. If the Flags indicate "End-to-end re-routing", the PathErr message is returned all the way back to the ingress LSR, which may then issue a new Path message along another path, which is the same procedure as in the flat network case above. If the Flags field indicates Boundary re-routing, the ingress area border LSR MAY terminate the PathErr message and then perform alternate routing within the area for which the area border LSR is the ingress LSR. If the Flags field indicates segment-based re-routing, any node MAY A. Farrel et al. Page 26 draft-ietf-ccamp-crankback-06.txt January 2007 apply the procedures described above for Boundary re-routing. 8.2. Downstream of the Fault This section only applies to errors that occur after an LSP has been established. Note that an LSR that generates a PathErr with Path_State_Remove Flag SHOULD also send a PathTear downstream to clean up the LSP. A node that detects a fault and is downstream of the fault MAY send a PathErr and/or Notify message containing an IF_ID ERROR SPEC that includes the location information of the failed link or node, and MAY send a PathTear to clean up the LSP at all other downstream nodes. However, if the reservation style for the LSP is Shared Explicit (SE) the detecting LSR MAY choose not to send a PathTear - this leaves the downstream LSP state in place and facilitates make-before-break repair of the LSP re-utilizing downstream resources. Note that if the detecting node does not send a PathTear immediately then unused sate will timeout according to the normal rules of [RFC2205]. At a well-known merge point, an ABR or an ASBR, a similar decision might also be made so as to better facilitate make-before-break repair. In this case a received PathTear might be 'absorbed' and not propagated further downstream for an LSP that has SE reservation style. Note, however, that this is a divergence from the protocol and might severely impact normal tear-down of LSPs. 9. IANA Considerations 9.1 Error Codes IANA maintains a registry called "RSVP Parameters" with a subregistry called "Error Codes and Globally-Defined Error Value Sub-Codes". This subregistry includes the RSVP-TE "Routing Problem" error code that is defined in [RFC3209]. IANA is requested to assign a new error value for the "Routing Problem" error code as follows: 17 Re-routing limit exceeded. The value of 17 is suggested and is for confirmation by IANA. 9.2 IF_ID_ERROR_SPEC TLVs The IF_ID_ERROR_SPEC TLV type values defined in [RFC3471] are maintained by IANA in the "Interface_ID Types" sub-registry of the "GMPLS Signaling Parameters" registry. A. Farrel et al. Page 27 draft-ietf-ccamp-crankback-06.txt January 2007 IANA is requested to make new assignments from this subregistry for the new TLV types defined in Section 7.2 of this document. The text 'see below' may be replaced by 'see RFC' within the subregistry to give a clear reference to the location of the definition of the TLV format. 9.3 LSP_ATTRIBUTES Object IANA maintains an "RSVP TE Parameters" registry with an "Attributes Flags" subregistry. IANA is requested to make three new allocations from this registry as listed in Section 6.4. These bits are defined for inclusion in the LSP Attributes TLV of the LSP_ATTRIBUTES. The values shown are suggested and are for confirmation by IANA. 10. Security Considerations The RSVP-TE trust model assumes that RSVP-TE neighbors and peers trust each other to exchange ligitimate and non-malicious messages. This assumption is necessary in order that the signaling protocol can function. Note that this trust model is assumed to cascade. That is, if an LSR trusts its neighbors, it extends this trust to all LSRs that its neighbor trusts. This means that the trust model is usually applied across the whole network to create a trust domain. Authentication of neighbor identity is already a standard provision of RSVP-TE, as is the protection of messages against tampering and spoofing. Refer to [RFC2205], [RFC3209], and [RFC3473] for a description of applicable security considerations. These considerations and mechanisms are applicable to hop-by-hop message exchanges (such as used for crankback propagation on PathErr messages) and directed message exchanges (such as used for crankback propagation on Notify messages). Key management may also be used with RSVP-TE to help to protect against impersonation and also against message content falsification. This requires the maintenance, exchange, and configuration of keys on each LSR. Note that such maintenance may be especially onerous to operators, hence it is important to limit the number of keys while ensuring the required level of security. This document does not introduce any protocol elements or message exchanges that change the operation of RSVP-TE security. A. Farrel et al. Page 28 draft-ietf-ccamp-crankback-06.txt January 2007 However, it should be noted that crankback is envisaged as an inter-domain mechanism, and as such it is likely that crankback information is exchanged over trust domain borders. In these cases it is expected that the information from within a neighboring domain would be of little or no value to the node performing crankback re-routing and would be ignored. In any case, it is highly likely that the reporting domain will have applied some form of information aggregation in order to preserve the confirentiality of its network topology. The issue of a direct attack by one domain upon another domain is possible and domain administrators should apply policies to protect their domains against the results of another domain attempting to thrash LSPs by allowing them to set up before reporting them as failed. On the whole, it is expected that commercial contracts between trust domains will provide a degree of protection. A more serious threat might arise if a domain reports that neither it nor its downstream neighbor can provide a path to the destination. Such a report could be bogus in that the reporting domain might not have allowed the downstream domain the chance to attempt to provide a path. (Note that the same problem does not arise for nodes within a domain because of the trust model.) This type of malicious behavior is hard to overcome, but may be detected by use of indirect path computation requests send direct to the falsely reported domain using mechanisms such as the Path Computation Element [RFC4655]. Note that a separate document describing inter-domain MPLS and GMPLS security considerations will be produced. Finally, it should be noted that while the extensions in this document introduce no new security holes in the protocols, should a malicious user gain protocol access to the network, the crankback information might be used to prevent establishment of valid LSPs. Thus, the existing security features available in RSVP-TE should be carefully considered by all deployers and SHOULD be made available by all implementations that offer cranback. Note that the implementation of re-routing attempt thresholds are also particularly useful in this context. 11. Acknowledgments We would like to thank Juha Heinanen and Srinivas Makam for their review and comments, and Zhi-Wei Lin for his considered opinions. Thanks, too, to John Drake for encouraging us to resurrect this document and consider the use of the IF_ID ERROR SPEC object. Thanks for a welcome and very thorough review by Dimitri Papadimitriou. Stephen Shew made useful comments for clarification through the ITU-T liaison process. A. Farrel et al. Page 29 draft-ietf-ccamp-crankback-06.txt January 2007 Simon Marshall-Unitt made contributions to this draft. SecDir review was provided by Tero Kivinen. Thanks to Ross Callon for useful discussions of prioritization of crankback re-routing attempts. 12. Normative References [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. [RFC2205] R. Braden, et al., "Resource ReSerVation Protocol (RSVP) Version 1 Functional Specification", RFC2205, September 1997. [RFC3209] D. Awduche, et al., "RSVP-TE: Extensions to RSVP for LSP Tunnels", RFC3209, December 2001. [RFC3471] P. Ashwood-Smith and L. Berger, et al., "Generalized MPLS - Signaling Functional Description", RFC 3471, January 2003. [RFC3473] L. Berger, et al., "Generalized MPLS Signaling - RSVP-TE Extensions", RFC 3473, January 2003. [RFC4420] Farrel, A., et al, "Encoding of Attributes for Multiprotocol Label Switching (MPLS) Label Switched Path (LSP) Establishment Using RSVP-TE", RFC 4420, February 2006. 13. Informational References [ASH1] G. Ash, ITU-T Recommendations E.360.1 --> E.360.7, "QoS Routing & Related Traffic Engineering Methods for IP-, ATM-, & TDM-Based Multiservice Networks", May, 2002. [RFC2702] D. Awduche, et al., "Requirements for Traffic Engineering Over MPLS", RFC2702, September 1999. [RFC3469] V. Sharma, et al., "Framework for MPLS-based Recovery", RFC 3469, February 2003. [RFC4090] Ping Pan, et al, "Fast Reroute Extensions to RSVP-TE for LSP Tunnels", RFC 4090, May 2005. [RFC4655] Farrel, A., Vasseur, J.-P., and J. Ash, "A Path Computation Element (PCE)-Based Architecture", RFC 4655, August 2006. A. Farrel et al. Page 30 draft-ietf-ccamp-crankback-06.txt January 2007 [G8080] ITU-T Recommendation G.808/Y.1304, Architecture for the Automatically Switched Optical Network (ASON), November 2001. For information on the availability of this document, please see http://www.itu.int. [EXCLUDE] C-Y. Lee, A. Farrel and S. De Cnodder, "Exclude Routes - Extension to RSVP-TE", draft-ietf-ccamp-rsvp-te-exclude-route, work in progress. [PNNI] ATM Forum, "Private Network-Network Interface Specification Version 1.0 (PNNI 1.0)", , May 1996. 14. Authors' Addresses Adrian Farrel (Editor) Old Dog Consulting Phone: +44 (0) 1978 860944 EMail: adrian@olddog.co.uk Arun Satyanarayana Cisco Systems, Inc. 170 West Tasman Dr. San Jose, CA 95134 Phone: +1 408 853-3206 Email: asatyana@cisco.com Atsushi Iwata NEC Corporation System Platforms Research Laboratories 1753 Shimonumabe Nakahara-ku, Kawasaki, Kanagawa, 211-8666, JAPAN Phone: +81-(44)-396-2744 Fax: +81-(44)-431-7612 E-Mail: a-iwata@ah.jp.nec.com Norihito Fujita NEC Corporation System Platforms Research Laboratories 1753 Shimonumabe Nakahara-ku, Kawasaki, Kanagawa, 211-8666, JAPAN Phone: +81-(44)-396-2091 Fax: +81-(44)-431-7644 E-Mail: n-fujita@bk.jp.nec.com A. Farrel et al. Page 31 draft-ietf-ccamp-crankback-06.txt January 2007 Gerald R. Ash AT&T Room MT D5-2A01 200 Laurel Avenue Middletown, NJ 07748, USA Phone: (+1) 732-420-4578 Fax: (+1) 732-368-8659 EMail: gash@att.com Intellectual Property Statement The IETF takes no position regarding the validity or scope of any Intellectual Property Rights 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; nor does it represent that it has made any independent effort to identify any such rights. Information on the procedures with respect to rights in RFC documents can be found in BCP 78 and BCP 79. Copies of IPR disclosures made to the IETF Secretariat 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 implementers or users of this specification can be obtained from the IETF on-line IPR repository at http://www.ietf.org/ipr. The IETF invites any interested party to bring to its attention any copyrights, patents or patent applications, or other proprietary rights that may cover technology that may be required to implement this standard. Please address the information to the IETF at ietf- ipr@ietf.org. Disclaimer of Validity This document and the information contained herein are provided on an "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY, THE IETF TRUST AND THE INTERNET ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. A. Farrel et al. Page 32 draft-ietf-ccamp-crankback-06.txt January 2007 Full Copyright Statement Copyright (C) The IETF Trust (2007). This document is subject to the rights, licenses and restrictions contained in BCP 78, and except as set forth therein, the authors retain all their rights. A. Farrel et al. Page 33 draft-ietf-ccamp-crankback-06.txt January 2007 Appendix A. Experience of Crankback in TDM-based Networks Experience of using release messages in TDM-based networks for analogous repair and re-routing purposes provides some guidance. One can use the receipt of a release message with a cause value (CV) indicating "link congestion" to trigger a re-routing attempt at the originating node. However, this sometimes leads to problems. *--------------------* *-----------------* | | | | | N2 ----------- N3-|--|----- AT--- EO2 | | | | \| | / | | | | | |--|- / | | | | | | | \/ | | | | | | | /\ | | | | | |--|- \ | | | | | /| | \ | | | N1 ----------- N4-|--|----- EO1 | | | | | *--------------------* *-----------------* A-1 A-2 Figure 1. Example of network topology Figure 1 illustrates four examples based on service-provider experiences with respect to crankback (i.e., explicit indication) versus implicit indication through a release with CV. In this example, N1, N2,N3, and N4 are located in one area (A-1), and AT, EO1, and EO2 are in another area (A-2). Note that two distinct areas are used in this example to expose the issues clearly. In fact, the issues are not limited to multi-area networks, but arise whenever path computation is distributed throughout the network. For example where loose routes, AS routes or path computation domains are used. 1. A connection request from node N1 to EO1 may route to N4 and then find "all circuits busy". N4 returns a release message to N1 with CV34 indicating all circuits busy. Normally, a node such as N1 is programmed to block a connection request when receiving CV34, although there is good reason to try to alternate route the connection request via N2 and N3. Some service providers have implemented a technique called route advance (RA), where if a node that is RA capable receives a release message with CV34, it will use this as an implicit re-route indication and try to find an alternate route for the connection request if possible. In this example, alternate route N1-N2-N3-EO1 can be tried and may well succeed. A. Farrel et al. Page 34 draft-ietf-ccamp-crankback-06.txt January 2007 2. Suppose a connection request goes from N2 to N3 to AT trying to reach EO2 and is blocked at link AT-EO2. Node AT returns a CV34 and with RA, N2 may try to re-route N2-N1-N4-AT-EO2, but of course this fails again. The problem is that N2 does not realize where this blocking occurred based on the CV34, and in this case there is no point in further alternate routing. 3. However, in another case of a connection request from N2 to E02, suppose that link N3-AT is blocked. In this case N3 should return crankback information (and not CV34) so that N2 can alternate route to N1-N4-AT-EO2, which may well be successful. 4. In a final example, for a connection request from EO1 to N2, EO1 first tries to route the connection request directly to N3. However, node N3 may reject the connection request even if there is bandwidth available on link N3-EO1 (perhaps for priority routing considerations, e.g., reserving bandwidth for high priority connection requests). However, when N3 returns CV34 in the release message, EO1 blocks the connection request (a normal response to CV34 especially if E01-N4 is already known blocked) rather than trying to alternate route through AT-N3-N2, which might be successful. If N3 returns crankback information, EO1 could respond by trying the alternate route. It is certainly the case that with topology exchange, such as OSPF, the ingress LSR could infer the re-routing condition. However, convergence of routing information is typically slower than the expected LSP setup times. One of the reasons for crankback is to avoid the overhead of available-link-bandwidth flooding, and to more efficiently use local state information to direct alternate routing at the ingress-LSR. [ASH1] shows how event-dependent-routing can just use crankback, and not available-link-bandwidth flooding, to decide on the re-route path in the network through "learning models". Reducing this flooding reduces overhead and can lead to the ability to support much larger AS sizes. Therefore, the alternate routing should be indicated based on an explicit indication (as in examples 3 and 4), and it is best to know the following information separately: a) where blockage/congestion occurred (as in examples 1-2), and b) whether alternate routing "should" be attempted even if there is no "blockage" (as in example 4). A. Farrel et al. Page 35