SRv6 Midpoint Protection

The current mechanism, e.g., TI-LFA (), allows local repair actions on the direct neighbors of the failed node or link to temporarily route traffic to the destination. This mechanism could not work properly when the failure happens in the destination point. In SRv6 TE, the IPv6 destination address in the outer IPv6 header could be the segment endpoint of the TE path rather than the destination of the TE path (). When the endpoint fails, local repair couldn't work on the direct neighbor of the failed endpoint either. This document defines midpoint protection for SRv6 TE path, which enables other nodes on the network to perform endpoint behaviors instead of the faulty node, update the IPv6 destination address to the other endpoint, and choose the next hop based on the new destination address.

When an endpoint node fails, the packet needs to bypass the failed endpoint node and be forwarded to the next endpoint node of the failed endpoint. Only endpoint node can porcess SRH, Therefore, only endpoint nodes can perform midpoint protection. There are two stages or time periods after an endpoint node fails. The first is the time period from the failure until the IGP converges on the failure. The second is the time period after the IGP converges on the failure. During the first time period, the packet will be sent to the direct neighbor of the failed endpoint node. After detecting the failure of its interface to the failed endpoint node, the neighbor forwards the packets around the failed endpoint node. It changes the IPv6 destination address with the IPv6 address of the next endpoint node (or the last or other reasonable endpoint node) which could avoid going through the failed endpoint. During the second time period. There is no route to the failed endpoint node after the IGP converges. When a previous hop node of the failed endpoint node finds out that there is no route to the IPv6 destination address (of the failed endpoint node), it changes the IPv6 destination address with the IPv6 address of the next endpoint node. Note that the previous hop node may not be the direct neighbor of the failed endpoint node.

The topology in illustrates an example of network topology with SRv6 enabled on each node.

In this document, an end SID at node n with locator block B is represented as B:n. An end.x SID at node n towards node k with locator block B is represented as B:n:k. A SID list is represented as <S1, S2, S3> where S1 is the first SID to visit, S2 is the second SID to visit and S3 is the last SID to visit along the SRv6 TE path. In the reference topology, suppose that Node N1 is an ingress node of SRv6 TE path going through N3 and N4. Node N1 steers a packet into a segment list < B:2, B:3, B:4>. When node N3 fails, the packet needs to bypass the failed endpoint node and be forwarded to the next endpoint node after the failed endpoint in the TE path. When outbound interface failure happens in the Repair Node (which is not limited to the previous hop node of the failed endpoint node), it performs the proxy forwarding as follows: During the first time period (i.e., before the IGP converges), node N2 (direct neighbor of N3) as a Repair Node forwards the packets around the failed endpoint N3 after detecting the failure of the outbound interface to the endpoint B:3. It changes the IPv6 destination address with the next sid B:4. N2 detects the failure of outbound interface to B:4 in the current route, it could use the normal Ti-LFA repair path to forward the packet, because it is not directly connected to the node N4. N2 encapsulates the packet with the segment list < B:5, B:6> as a repair path. During the second time period (i.e., after the IGP converges), node N2 does not have any route to the failed endpoint N3 in its FIB. Node N2, as a Repair Node, forwards the packets around the failed endpoint N3 to the next endpoint node (e.g., N4) directly. There is no need to check whether the failed endpoint node is directly connected to N2. N2 changes the IPv6 destination address with the next sid B:4. Since IGP has completed convergence, it forwards packets directly based on the IGP SPF path.

A node N protecting the failure of an endpoint node on a SRv6 path may be one of the following types: a transit node: The transit node cannot process SRH. Therefore, only Ti-LFA can be executed on the transit node, but not midpoint protection. an endpoint node: the destination address (DA) of the packet received by N is a N's local END SID. an endpoint x node (i.e., an endpoint with cross-connect node): the destination address (DA) of the packet received by N is a N's local End.X SID with an array of layer 3 adjacencies. This section describes the behavior of each of these nodes as a repair node for the two time periods after the endpoint node fails.

When the Repair Node is an endpoint node, it provides fast protections for the failure through executing the following procedure after looking up the FIB for the updated DA.

When the Repair Node is an endpoint x node, it provides fast protections for the failure through executing the following procedure after updating DA.

SRv6 Midpoint Protection provides a mechanism to bypass a failed endpoint. But in some scenarios, some important functions may be implemented in the bypassed failed endpoints that should not be bypassed, such as firewall functionality or In-situ Flow Information Telemetry of a specified path. Therefore, a mechanism is needed to indicate whether an endpoint can be bypassed or not. provides method to determine whether enbale SRv6 midpoint protection or not by defining a "no bypass" flag for the SIDs in IGP.

This section reviews security considerations related to SRv6 Midpoint protection processing discussed in this document.To ensure that the Repair node does not modify the SRH header Encapsulated by nodes outside the SRv6 Domain.Only the segment within the SRH is same domain as the repair node. So it is necessary to check the skipped segment have same block as repair node.

This document makes no request of IANA. Note to RFC Editor: this section may be removed on publication as an RFC.

The authors would like to thank Bruno Decraene, Jeff Tantsura, Ketan Talaulikar and Parag Kaneriya for their comments to this work.