SRv6 Network Programming

Segment Routing leverages the source routing paradigm. An ingress node steers a packet through a ordered list of instructions, called segments. Each one of these instructions represents a function to be called at a specific location in the network. A function is locally defined on the node where it is executed and may range from simply moving forward in the segment list to any complex user-defined behavior. The network programming consists in combining segment routing functions, both simple and complex, to achieve a networking objective that goes beyond mere packet routing. This document defines the SRv6 Network Programming concept and aims at standardizing the main segment routing functions to enable the creation of interoperable overlays with underlay optimization and service programming. The companion document illustrates the concepts defined in this document. Familiarity with the Segment Routing Header is assumed.

SRH is the abbreviation for the Segment Routing Header. We assume that the SRH may be present multiple times inside each packet. NH is the abbreviation of the IPv6 next-header field. NH=SRH means that the next-header field is 43 with routing type 4. When there are multiple SRHs, they must follow each other: the next-header field of all SRH, except the last one, must be SRH. The effective next-header (ENH) is the next-header field of the IP header when no SRH is present, or is the next-header field of the last SRH. In this version of the document, we assume that there are no other extension headers than the SRH. These will be lifted in future versions of the document. SID: A Segment Identifier which represents a specific segment in segment routing domain. The SID type used in this document is IPv6 address (also referenced as SRv6 Segment or SRv6 SID). 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 SR path. (SA,DA) (S3, S2, S1; SL) represents an IPv6 packet with: IPv6 header with source address SA, destination addresses DA and SRH as next-header SRH with SID list <S1, S2, S3> with SegmentsLeft = SL Note the difference between the <> and () symbols: <S1, S2, S3> represents a SID list where S1 is the first SID and S3 is the last SID to traverse. (S3, S2, S1; SL) represents the same SID list but encoded in the SRH format where the rightmost SID in the SRH is the first SID and the leftmost SID in the SRH is the last SID. When referring to an SR policy in a high-level use-case, it is simpler to use the <S1, S2, S3> notation. When referring to an illustration of the detailed packet behavior, the (S3, S2, S1; SL) notation is more convenient. The payload of the packet is omitted. SRH[SL] represents the SID pointed by the SL field in the first SRH. In our example, SRH[2] represents S1, SRH[1] represents S2 and SRH[0] represents S3. FIB is the abbreviation for the forwarding table. A FIB lookup is a lookup in the forwarding table. When a packet is intercepted on a wire, it is possible that SRH[SL] is different from the DA.

An SRv6 Segment is a 128-bit value. "SID" (abbreviation for Segment Identifier) is often used as a shorter reference for "SRv6 Segment". An SRv6-capable node N maintains a "My SID Table". This table contains all the SRv6 segments explicitly instantiated at node N. N is the parent node for these SIDs. A local SID of N can be an IPv6 address associated to a local interface of N but it is not mandatory. Nor is the "My SID table" populated by default with all IPv6 addresses defined on node N. In most use-cases, a local SID will NOT be an address associated to a local interface of N. A local SID of N could be routed to N but it does not have to be. Most often, it is routed to N via a shorter-mask prefix. Let's provide a classic illustration. Node N is configured with a loopback0 interface address of A:1::/32 originated in its IGP. Node N is configured with two SIDs: B:1:100:: and B:2:101::. The entry A:1:: is not defined explicitly as an SRv6 SID and hence does not appear in the "My SID Table". The entries B:1:100:: and B:2:101:: are defined explicitly as SRv6 SIDs and hence appear in the "My SID Table". The network learns about a path to B:1::/32 via the IGP and hence a packet destined to B:1:100:: would be routed up to N. The network does not learn about a path to B:2::/32 via the IGP and hence a packet destined to B:2:101:: would not be routed up to N. A packet could be steered to a non-routed SID B:2:101:: by using a SID list <...,B:1:100::,B:2:101::,...> where the non-routed SID is preceded by a routed SID to the same node. This is similar to the local vs global segments in SR-MPLS. Every SRv6 SID instantiated has a specific instruction bound to it. This information is stored in the "My SID Table". The "My SID Table" has three main purposes: Define which SIDs are explicitly instantiated on that node Specify which instruction is bound to each of the instantiated SIDs Store the parameters associated with such instruction (i.e. OIF, NextHop, VRF,...) We represent an SRv6 SID as LOC:FUNCT where LOC is the L most significant bits and FUNCT is the 128-L least significant bits. L is called the locator length and is flexible. Each operator is free to use the locator length it chooses. Most often the LOC part of the SID is routable and leads to the node which instantiates that SID. The FUNCT part of the SID is an opaque identification of a local function bound to the SID. The FUNCT value zero is invalid. Often, for simplicity of illustration, we will use a locator length of 32 bits. This is just an example. Implementations must not assume any a priori prefix length. A function may require additional arguments that would be placed immediately after the FUNCT. In such case, the SRv6 SID will have the form LOC:FUNCT:ARGS::. For this reason, the "My SID Table" matches on a per longest-prefix-match basis. These arguments may vary on a per-packet basis and may contain information related to the flow, service, or any other information required by the function associated to the SRv6 SID. A node may receive a packet with an SRv6 SID in the DA without an SRH. In such case the packet should still be processed by the Segment Routing engine.

Each entry of the "My SID Table" indicates the function associated with the local SID and its parameters. We define hereafter a set of well-known functions that can be associated with a SID.

The list is not exhaustive. In practice, any function can be attached to a local SID: e.g. a node N can bind a SID to a local VM or container which can apply any complex function on the packet. We call N the node who has an explicitly instantiated SID S and we detail the function that N binds to S. At the end of this section we also present some flavours of these well-known functions.

The Endpoint function ("End" for short) is the most basic function. When N receives a packet whose IPv6 DA is S and S is a local End SID, N does:

0 2. decrement SL 3. update the IPv6 DA with SRH[SL] 4. FIB lookup on the updated DA ;; Ref1 5. forward accordingly to the matched entry ;; Ref2 6. ELSE IF NH!=SRH 7. Send an ICMP parameter problem message; drop the packet ;; Ref3 8. ELSE 9. drop the packet ]]> Ref1: The End function performs the FIB lookup in the forwarding table associated to the ingress interface Ref2: If the FIB lookup matches a multicast state, then the related RPF check must be considered successful Ref3: ICMP error is sent to the source address with error code (TBD by IANA) "SR Upper-layer Header Error" and pointer set to the NH. A local SID could be bound to a function which authorizes the decapsulation of an outer header (e.g. IPinIP) or the punting of the packet to TCP, UDP or any other protocol. This however needs to be explicitly defined in the function bound to the local SID. By default, a local SID bound to the well-known function "End"neither allows the decapsulation of an outer header nor the cleanup of an SRH. As a consequence, an End SID cannot be the last SID of an SRH and cannot be the DA of a packet without SRH. This is the SRv6 instantiation of a Prefix SID.

The "Endpoint with cross-connect to an array of layer-3 adjacencies" function (End.X for short) is a variant of the End function. When N receives a packet destined to S and S is a local End.X SID, N does:

0 2. decrement SL 3. update the IPv6 DA with SRH[SL] 4. forward to layer-3 adjacency bound to the SID S ;; Ref1 6. ELSE IF NH!=SRH 7. Send an ICMP parameter problem message; drop the packet ;; Ref2 8. ELSE 9. drop the packet ]]> Ref1: If an array of adjacencies is bound to the End.X SID, then one entry of the array is selected based on a hash of the packet's header. Ref2: ICMP error is sent to the source address with error code (TBD by IANA) "SR Upper-layer Header Error" and pointer set to the NH. The End.X function is required to express any traffic-engineering policy. This is the SRv6 instantiation of an Adjacency SID. If a node N has 30 outgoing interfaces to 30 neighbors, usually the operator would explicitly instantiate 30 End.X SIDs at N: one per layer-3 adjacency to a neighbor. Potentially, more End.X could be explicitly defined (groups of layer-3 adjacencies to the same neighbor or to different neighbors). Note that with SR-MPLS, an AdjSID is typically preceded by a PrefixSID. This is unlikely in SRv6 as most likely an End.X SID is globally routed to N. Note that if N has an outgoing interface bundle I to a neighbor Q made of 10 member links, N may allocate up to 11 End.X local SIDs for that bundle: one for the bundle itself and then up to one for each member link. This is the equivalent of the L2-Link Adj SID in SR-MPLS .

The "Endpoint with specific IPv6 table lookup" function (End.T for short) is a variant of the End function. When N receives a packet destined to S and S is a local End.T SID, N does:

0 ;; Ref1 2. decrement SL 3. update the IPv6 DA with SRH[SL] 4. lookup the next segment in IPv6 table T associated with the SID 5. forward via the matched table entry 6. ELSE IF NH!=SRH 7. Send an ICMP parameter problem message; drop the packet ;; Ref2 8. ELSE 9. drop the packet ]]> Ref1: The End.T SID must not be the last SID Ref2: ICMP error is sent to the source address with error code (TBD by IANA) "SR Upper-layer Header Error" and pointer set to the NH. The End.T is used for multi-table operation in the core.

The "Endpoint with decapsulation and Layer-2 cross-connect to OIF" function (End.DX2 for short) is a variant of the endpoint function. When N receives a packet destined to S and S is a local End.DX2 SID, N does:

0 2. drop the packet ;; Ref1 3. ELSE IF ENH=59 ;; Ref2 4. pop the (outer) IPv6 header and its extension headers 5. forward the resulting frame to OIF bound to the SID S 6. ELSE 7. Send an ICMP parameter problem message ;; Ref3 8. drop the packet ]]> Ref1: An End.DX2 SID must always be the last SID, or it can be the Destination Address of an IPv6 packet with no SRH header. Ref2: We conveniently reuse the next-header value 59 allocated to IPv6 No Next Header . When the SID corresponds to function End.DX2 and the Next-Header value is 59, we know that an Ethernet frame is in the payload without any further header. Ref3: ICMP error is sent to the source address with error code (TBD by IANA) "SR Upper-layer Header Error" and pointer set to the NH. An End.DX2 function could be customized to expect a specific VLAN format and rewrite the egress VLAN header before forwarding on the outgoing interface. One of the applications of the End.DX2 function is the L2VPN/EVPN VPWS use-case.

The "Endpoint with decapsulation and specific VLAN table lookup" function (End.DX2V for short) is a variant of the endpoint function. When N receives a packet destined to S and S is a local End.DX2V SID, N does:

0 2. drop the packet ;; Ref1 3. ELSE IF ENH = 59 ;; Ref2 4. pop the (outer) IPv6 header and its extension headers 5. lookup the exposed inner VLANs in L2 table T 6. forward via the matched table entry 7. ELSE 8. Send an ICMP parameter problem message ;; Ref3 9. drop the packet ]]> Ref1: An End.DX2V SID must always be the last SID, or it can be the Destination Address of an IPv6 packet with no SRH header. Ref2: We conveniently reuse the next-header value 59 allocated to IPv6 No Next Header . When the SID corresponds to function End.DX2V and the Next-Header value is 59, we know that an Ethernet frame is in the payload without any further header. Ref3: ICMP error is sent to the source address with error code (TBD by IANA) "SR Upper-layer Header Error" and pointer set to the NH. An End.DX2V function could be customized to expect a specific VLAN format and rewrite the egress VLAN header before forwarding on the outgoing interface. The End.DX2V is used for EVPN Flexible cross-connect use-cases.

The "Endpoint with decapsulation and specific unicast MAC L2 table lookup" function (End.DT2U for short) is a variant of the endpoint function. When N receives a packet destined to S and S is a local End.DT2U SID, N does:

0 2. drop the packet ;; Ref1 3. ELSE IF ENH = 59 ;; Ref2 4. pop the (outer) IPv6 header and its extension headers 5. learn the exposed inner MAC SA in L2 table T ;; Ref3 6. lookup the exposed inner MAC DA in L2 table T 7. IF matched entry in table T 8. forward via the matched table T entry 9. ELSE 10. forward via all L2OIF entries in table T 11. ELSE 12. Send an ICMP parameter problem message ;; Ref4 13. drop the packet ]]> Ref1: An End.DT2U SID must always be the last SID, or it can be the Destination Address of an IPv6 packet with no SRH header. Ref2: We conveniently reuse the next-header value 59 allocated to IPv6 No Next Header . When the SID corresponds to function End.DT2U and the Next-Header value is 59, we know that an Ethernet frame is in the payload without any further header. Ref3: In EVPN, the learning of the exposed inner MAC SA is done via control plane. Ref4: ICMP error is sent to the source address with error code (TBD by IANA) "SR Upper-layer Header Error" and pointer set to the NH. The End.DT2U is used for EVPN Bridging unicast use cases.

The "Endpoint with decapsulation and specific L2 table flooding" function (End.DT2M for short) is a variant of the endpoint function. This function may take an argument: "Arg.FE2". It is an argument specific to EVPN ESI filtering. It is used to exclude a specific OIF (or set of OIFs) from L2 table T flooding. When N receives a packet destined to S and S is a local End.DT2M SID, N does:

0 2. drop the packet ;; Ref1 3. ELSE IF ENH = 59 ;; Ref2 4. pop the (outer) IPv6 header and its extension headers 3. learn the exposed inner MAC SA in L2 table T ;; Ref3 4. forward on all L2OIF excluding the one specified in Arg.FE2 5. ELSE 6. Send an ICMP parameter problem message ;; Ref4 7. drop the packet ]]> Ref1: An End.DT2M SID must always be the last SID, or it can be the Destination Address of an IPv6 packet with no SRH header. Ref2: We conveniently reuse the next-header value 59 allocated to IPv6 No Next Header . When the SID corresponds to function End.DT2M and the Next-Header value is 59, we know that an Ethernet frame is in the payload without any further header. Ref3: In EVPN, the learning of the exposed inner MAC SA is done via control plane Ref4: ICMP error is sent to the source address with error code (TBD by IANA) "SR Upper-layer Header Error" and pointer set to the NH. The End.DT2M is used for EVPN Bridging BUM use-case with ESI filtering capability.

The "Endpoint with decapsulation and cross-connect to an array of IPv6 adjacencies" function (End.DX6 for short) is a variant of the End.X function. When N receives a packet destined to S and S is a local End.DX6 SID, N does:

0 2. drop the packet ;; Ref1 3. ELSE IF ENH = 41 ;; Ref2 4. pop the (outer) IPv6 header and its extension headers 5. forward to layer-3 adjacency bound to the SID S ;; Ref3 6. ELSE 7. Send an ICMP parameter problem message ;; Ref4 8. drop the packet ]]> Ref1: The End.DX6 SID must always be the last SID, or it can be the Destination Address of an IPv6 packet with no SRH header. Ref2: 41 refers to IPv6 encapsulation as defined by IANA allocation for Internet Protocol Numbers Ref3: Selected based on a hash of the packet's header (at least SA, DA, Flow Label) Ref4: ICMP error is sent to the source address with error code (TBD by IANA) "SR Upper-layer Header Error" and pointer set to the NH. One of the applications of the End.DX6 function is the L3VPNv6 use-case where a FIB lookup in a specific tenant table at the egress PE is not required. This would be equivalent to the per-CE VPN label in MPLS .

The "Endpoint with decapsulation and cross-connect to an array of IPv4 adjacencies" function (End.DX4 for short) is a variant of the End.X functions. When N receives a packet destined to S and S is a local End.DX4 SID, N does:

0 2. drop the packet ;; Ref1 3. ELSE IF ENH = 4 ;; Ref2 4. pop the (outer) IPv6 header and its extension headers 5. forward to layer-3 adjacency bound to the SID S ;; Ref3 6. ELSE 7. Send an ICMP parameter problem message ;; Ref4 8. drop the packet ]]> Ref1: The End.DX4 SID must always be the last SID, or it can be the Destination Address of an IPv6 packet with no SRH header. Ref2: 4 refers to IPv4 encapsulation as defined by IANA allocation for Internet Protocol Numbers Ref3: Selected based on a hash of the packet's header (at least SA, DA, Flow Label) Ref4: ICMP error is sent to the source address with error code (TBD by IANA) "SR Upper-layer Header Error" and pointer set to the NH. One of the applications of the End.DX4 function is the L3VPNv4 use-case where a FIB lookup in a specific tenant table at the egress PE is not required. This would be equivalent to the per-CE VPN label in MPLS .

The "Endpoint with decapsulation and specific IPv6 table lookup" function (End.DT6 for short) is a variant of the End function. When N receives a packet destined to S and S is a local End.DT6 SID, N does:

0 2. drop the packet ;; Ref1 3. ELSE IF ENH = 41 ;; Ref2 4. pop the (outer) IPv6 header and its extension headers 5. lookup the exposed inner IPv6 DA in IPv6 table T 6. forward via the matched table entry 7. ELSE 8. Send an ICMP parameter problem message ;; Ref3 9. drop the packet ]]> Ref1: the End.DT6 SID must always be the last SID, or it can be the Destination Address of an IPv6 packet with no SRH header. Ref2: 41 refers to IPv6 encapsulation as defined by IANA allocation for Internet Protocol Numbers Ref3: ICMP error is sent to the source address with error code (TBD by IANA) "SR Upper-layer Header Error" and pointer set to the NH. One of the applications of the End.DT6 function is the L3VPNv6 use-case where a FIB lookup in a specific tenant table at the egress PE is required. This would be equivalent to the per-VRF VPN label in MPLS. Note that an End.DT6 may be defined for the main IPv6 table in which case and End.DT6 supports the equivalent of an IPv6inIPv6 decaps (without VPN/tenant implication).

The "Endpoint with decapsulation and specific IPv4 table lookup" function (End.DT4 for short) is a variant of the End function. When N receives a packet destined to S and S is a local End.DT4 SID, N does:

0 2. drop the packet ;; Ref1 3. ELSE IF ENH = 4 ;; Ref2 4. pop the (outer) IPv6 header and its extension headers 5. lookup the exposed inner IPv4 DA in IPv4 table T 6. forward via the matched table entry 7. ELSE 8. Send an ICMP parameter problem message ;; Ref3 9. drop the packet ]]> Ref1: the End.DT4 SID must always be the last SID, or it can be the Destination Address of an IPv6 packet with no SRH header. Ref2: 4 refers to IPv4 encapsulation as defined by IANA allocation for Internet Protocol Numbers Ref3: ICMP error is sent to the source address with error code (TBD by IANA) "SR Upper-layer Header Error" and pointer set to the NH. One of the applications of the End.DT4 is the L3VPNv4 use-case where a FIB lookup in a specific tenant table at the egress PE is required. This would be equivalent to the per-VRF VPN label in MPLS. Note that an End.DT4 may be defined for the main IPv4 table in which case and End.DT4 supports the equivalent of an IPv4inIPv6 decaps (without VPN/tenant implication).

The "Endpoint with decapsulation and specific IP table lookup" function (End.DT46 for short) is a variant of the End.DT4 and End.DT6 functions. When N receives a packet destined to S and S is a local End.DT46 SID, N does:

0 2. drop the packet ;; Ref1 3. ELSE IF ENH = 4 ;; Ref2 4. pop the (outer) IPv6 header and its extension headers 5. lookup the exposed inner IPv4 DA in IPv4 table T 6. forward via the matched table entry 7. ELSE IF ENH = 41 ;; Ref2 8. pop the (outer) IPv6 header and its extension headers 9. lookup the exposed inner IPv6 DA in IPv6 table T 10. forward via the matched table entry 11. ELSE 12. Send an ICMP parameter problem message ;; Ref3 13. drop the packet ]]> Ref1: the End.DT46 SID must always be the last SID, or it can be the Destination Address of an IPv6 packet with no SRH header. Ref2: 4 and 41 refer to IPv4 and IPv6 encapsulation respectively as defined by IANA allocation for Internet Protocol Numbers Ref3: ICMP error is sent to the source address with error code (TBD by IANA) "SR Upper-layer Header Error" and pointer set to the NH. One of the applications of the End.DT46 is the L3VPN use-case where a FIB lookup in a specific IP tenant table at the egress PE is required. This would be equivalent to the per-VRF VPN label in MPLS . Note that an End.DT46 may be defined for the main IP table in which case and End.DT46 supports the equivalent of an IPinIPv6 decaps (without VPN/tenant implication).

The "Endpoint bound to an SRv6 Policy" is a variant of the End function. When N receives a packet destined to S and S is a local End.B6.Insert SID, N does:

0 ;; Ref1 2. do not decrement SL nor update the IPv6 DA with SRH[SL] 3. insert a new SRH, in between the IPv6 header and the received SRH 4. set the IPv6 DA to the first segment of the SRv6 Policy 5. forward according to the first segment of the SRv6 Policy 6. ELSE 7. Send an ICMP parameter problem message ;; Ref2 8. drop the packet ]]> Ref1: An End.B6.Insert SID, by definition, is never the last SID. Ref2: ICMP error is sent to the source address with error code (TBD by IANA) "SR Upper-layer Header Error" and pointer set to the NH. Note: Instead of the term "insert", "push" may also be used. The End.B6.Insert function is required to express scalable traffic-engineering policies across multiple domains. This is the SRv6 instantiation of a Binding SID.

This is an optimization of the End.B6.Insert function. End.B6.Insert.Red will reduce the size of the SRH by one segment by avoiding the insertion of the first SID in the pushed SRH. In this way, the first segment is only introduced in the DA and the packet is forwarded according to it. Note that SL value is initially pointing to a non-existing segment in the SRH.

This is a variation of the End.B6.Insert behavior where the SRv6 Policy also includes an IPv6 Source Address A. When N receives a packet destined to S and S is a local End.B6.Encaps SID, N does:

0 2. decrement SL and update the IPv6 DA with SRH[SL] 3. push an outer IPv6 header with its own SRH 4. set the outer IPv6 SA to A 5. set the outer IPv6 DA to the first segment of the SRv6 Policy 6. set outer payload length, trafic class and flow label ;; Ref1,2 7. update the Next-Header value ;; Ref1 8. decrement inner Hop Limit or TTL ;; Ref1 9. forward according to the first segment of the SRv6 Policy 10. ELSE 11. Send an ICMP parameter problem message ;; Ref3 12. drop the packet ]]> Ref 1: As described in (Generic Packet Tunneling in IPv6 Specification) Ref 2: As described in (IPv6 Flow Label Specification) Ref3: ICMP error is sent to the source address with error code (TBD by IANA) "SR Upper-layer Header Error" and pointer set to the NH. Instead of simply inserting an SRH with the policy (End.B6), this behavior also adds an outer IPv6 header. The source address defined for the outer header does not have to be a local SID of the node. The SRH MAY be omitted when the SRv6 Policy only contains one segment and there is no need to use any flag, tag or TLV.

This is an optimization of the End.B6.Encaps function. End.B6.Encaps.Red will reduce the size of the SRH by one segment by avoiding the insertion of the first SID in the outer SRH. In this way, the first segment is only introduced in the DA and the packet is forwarded according to it. Note that SL value is initially pointing to a non-existing segment in the SRH. The SRH MAY be omitted when the SRv6 Policy only contains one segment and there is no need to use any flag, tag or TLV.

The "Endpoint bound to an SR-MPLS Policy" is a variant of the End.B6 function. When N receives a packet destined to S and S is a local End.BM SID, N does:

0 ;; Ref1 2. decrement SL and update the IPv6 DA with SRH[SL] 3. push an MPLS label stack on the received packet 4. forward according to L1 5. ELSE 6. Send an ICMP parameter problem message ;; Ref2 7. drop the packet ]]> Ref1: an End.BM SID, by definition, is never the last SID. Ref2: ICMP error is sent to the source address with error code (TBD by IANA) "SR Upper-layer Header Error" and pointer set to the NH. The End.BM function is required to express scalable traffic-engineering policies across multiple domains where some domains support the MPLS instantiation of Segment Routing. This is an SRv6 instantiation of an SR-MPLS Binding SID.

The "Endpoint in search of a target in Table T" function (End.S for short) is a variant of the End function. When N receives a packet destined to S and S is a local End.S SID, N does:

Ref1: By definition, an End.S SID cannot be the last SID, as the last SID is the targeted object. Ref2: ICMP error is sent to the source address with error code (TBD by IANA) "SR Upper-layer Header Error" and pointer set to the NH. The End.S function is required in information-centric networking (ICN) use-cases where the last SID in the SRv6 SID list represents a targeted object. If the identification of the object would require more than 128 bits, then obvious customization of the End.S function may either use multiple SIDs or a TLV of the SR header to encode the searched object ID.

Generally, any SR-aware application can be bound to an SRv6 SID. This application could represent anything from a small piece of code focused on topological/tenant function to a larger process focusing on higher-level applications (e.g. video compression, transcoding etc.). The ways in which an SR-aware application binds itself on a local SID depends on the operating system. Let us consider an SR-aware application running on a Linux operating system. A possible approach is to associate an SRv6 SID to a target (virtual) interface, so that packets with IP DA corresponding to the SID will be sent to the target interface. In this approach, the SR-aware application can simply listen to all packets received on the interface. A different approach for the SR-aware app is to listen to packets received with a specific SRv6 SID as IPv6 DA on a given transport port (i.e. corresponding to a TCP or UDP socket). In this case, the app can read the SRH information with a getsockopt Linux system call and can set the SRH information to be added to the outgoing packets with a setsocksopt system call.

defines a set of additional functions in order to enable non SR-aware applications to be associated with an SRv6 SID.

We present the PSP and USP variants of the functions End, End.X and End.T. For each of these functions these variants can be enabled or disabled either individually or together.

After the instruction 'update the IPv6 DA with SRH[SL]' is executed, the following instructions must be added:

Ref1: The received SRH had SL=1. When the last SID is written in the DA, the End, End.X and End.T functions with the PSP flavour pop the first (top-most) SRH. Subsequent stacked SRH's may be present but are not processed as part of the function.

We insert at the beginning of the pseudo-code the following instructions:

Ref1: The next header is an SRH header Ref2: Typically SL of the exposed SRH is > 0 and hence the restarting of the complete function would lead to decrement SL, update the IPv6 DA with SRH[SL], FIB lookup on updated DA and forward accordingly to the matched entry.

We insert at the beginning of the pseudo-code the following instructions:

Ref1: In case that the USD flavor is applied on an End.X function, the packet is forwarded to the layer-3 adjancency bound to SID S without any lookup.

We define hereafter the set of basic transit behaviors. These behaviors are not bound to a SID and they correspond to source SR nodes or transit nodes .

This list can be expanded in case any new functionality requires it.

As per , if a node N receives a packet (A, S2)(S3, S2, S1; SL=2) and S2 is neither a local address nor a local SID of N then N forwards the packet without inspecting the SRH. This means that N treats the following two packets with the same performance: (A, S2) (A, S2)(S3, S2, S1; SL=2) A transit node does not need to count by default the amount of transit traffic with an SRH extension header. This accounting might be enabled as an optional behavior. A transit node MUST include the outer flow label in its ECMP load-balancing hash .

Node N receives two packets P1=(A, B2) and P2=(A,B2)(B3, B2, B1; SL=1). B2 is neither a local address nor SID of N. N steers the transit packets P1 and P2 into an SRv6 Policy with one SID list <S1, S2, S3>. The "T.Insert" transit insertion behavior is defined as follows:

Ref1: The received IPv6 DA is placed as last SID of the inserted SRH. Ref1bis: The SRH is inserted before any other IPv6 Routing Extension Header. After the T.Insert behavior, P1 and P2 respectively look like: (A, S1) (B2, S3, S2, S1; SL=3) (A, S1) (B2, S3, S2, S1; SL=3) (B3, B2, B1; SL=1)

The T.Insert.Red behavior is an optimization of the T.Insert behavior. It is defined as follows:

T.Insert.Red will reduce the size of the SRH by one segment by avoiding the insertion of the first SID in the pushed SRH. In this way, the first segment is only introduced in the DA and the packet is forwarded according to it. Note that SL value is initially pointing to a non-existing segment in the SRH. After the T.Insert.Red behavior, P1 and P2 respectively look like: (A, S1) (B2, S3, S2; SL=3) (A, S1) (B2, S3, S2; SL=3) (B3, B2, B1; SL=1)

Node N receives two packets P1=(A, B2) and P2=(A,B2)(B3, B2, B1; SL=1). B2 is neither a local address nor SID of N. N steers the transit packets P1 and P2 into an SR Encapsulation Policy with a Source Address T and a Segment list <S1, S2, S3>. The T.Encaps transit encapsulation behavior is defined as follows:

After the T.Encaps behavior, P1 and P2 respectively look like: (T, S1) (S3, S2, S1; SL=2) (A, B2) (T, S1) (S3, S2, S1; SL=2) (A, B2) (B3, B2, B1; SL=1) The T.Encaps behavior is valid for any kind of Layer-3 traffic. This behavior is commonly used for L3VPN with IPv4 and IPv6 deployments. The SRH MAY be omitted when the SRv6 Policy only contains one segment and there is no need to use any flag, tag or TLV. Ref 1: As described in (Generic Packet Tunneling in IPv6 Specification) Ref 2: As described in (IPv6 Flow Label Specification)

The T.Encaps.Red behavior is an optimization of the T.Encaps behavior. It is defined as follows:

Ref 1: As described in (Generic Packet Tunneling in IPv6 Specification) Ref 2: As described in (IPv6 Flow Label Specification) T.Encaps.Red will reduce the size of the SRH by one segment by avoiding the insertion of the first SID in the SRH of the pushed IPv6 packet. In this way, the first segment is only introduced in the DA and the packet is forwarded according to it. Note that SL value is initially pointing to a non-existing segment in the SRH. After the T.Encaps.Red behavior, P1 and P2 respectively look like: (T, S1) (S3, S2; SL=2) (A, B2) (T, S1) (S3, S2; SL=2) (A, B2) (B3, B2, B1; SL=1) The SRH MAY be omitted when the SRv6 Policy only contains one segment and there is no need to use any flag, tag or TLV.

While T.Encaps encapsulates the received IP packet, T.Encaps.L2 encapsulates the received L2 frame (i.e. the received ethernet header and its optional VLAN header is in the payload of the outer packet). If the outer header is pushed without SRH, then the DA must be a SID of type End.DX2, End.DX2V, End.DT2U or End.DT2M and the next-header must be 59 (IPv6 NoNextHeader). The received Ethernet frame follows the IPv6 header and its extension headers. Else, if the outer header is pushed with an SRH, then the last SID of the SRH must be of type End.DX2, End.DX2V, End.DT2U or End.DT2M and the next-header of the SRH must be 59 (IPv6 NoNextHeader). The received Ethernet frame follows the IPv6 header and its extension headers. The SRH MAY be omitted when the SRv6 Policy only contains one segment and there is no need to use any flag, tag or TLV.

The T.Encaps.L2.Red behavior is an optimization of the T.Encaps.L2 behavior. T.Encaps.L2.Red will reduce the size of the SRH by one segment by avoiding the insertion of the first SID in the SRH of the pushed IPv6 packet. In this way, the first segment is only introduced in the DA and the packet is forwarded according to it. Note that SL value is initially pointing to a non-existing segment in the SRH. The SRH MAY be omitted when the SRv6 Policy only contains one segment and there is no need to use any flag, tag or TLV.

Any SRv6 capable node SHOULD implement the following set of combined counters (packets and bytes): CNT-1: Per entry of the "My SID Table", traffic that matched that SID and was processed correctly. CNT-2: Per SRv6 Policy, traffic steered into it and processed correctly. Furthermore, an SRv6 capable node maintains an aggregate counter CNT-3 tracking the IPv6 traffic that was received with a destination address matching a local interface address that is not a locally instantiated SID and the next-header is SRH with SL>0. We remind that this traffic is dropped as an interface address is not a local SID by default. A SID must be explicitly instantiated.

When a flow-based selection within a set needs to be performed, the source address, the destination address and the flow-label MUST be included in the flow-based hash. This occurs when the destination address is updated, a FIB lookup is performed and multiple ECMP paths exist to the updated destination address. This occurs when End.X, End.DX4, or End.DX6 are bound to an array of adjacencies. This occurs when the packet is steered in an SR policy whose selected path has multiple SID lists .

defines the OAM behavior for SRv6. This includes the definition of the SRH Flag 'O-bit', as well as additional OAM Endpoint functions.

We use the following terminology: An internal node is a node part of the domain of trust. A border router is an internal node at the edge of the domain of trust. An external interface is an interface of a border router towards another domain. An internal interface is an interface entirely within the domain of trust. The internal address space is the IP address block dedicated to internal interfaces. An internal SID is a SID instantiated on an internal node. The internal SID space is the IP address block dedicated to internal SIDs. External traffic is traffic received from an external interface to the domain of trust. Internal traffic is traffic that originates and ends within the domain of trust. The purpose of this section is to document how a domain of trust can operate SRv6-based services for internal traffic while preventing any external traffic from accessing the internal SRv6-based services. It is expected that future documents will detail enhanced security mechanisms for SRv6 (e.g. how to allow external traffic to leverage internal SRv6 services).

An SRv6 router MUST support an ACL on the external interface that drops any traffic with SA or DA in the internal SID space. A provider would generally do this for its internal address space to prevent access to internal addresses and in order to prevent spoofing. The technique is extended to the local SID space. The typical counters of an ACL are expected.

An SRv6 router MUST support an ACL with the following behavior:

This prevents access to locally instantiated SIDs from outside the operator's infrastructure. Note that this ACL may not be enabled in all cases. For example, specific SIDs can be used to provide resources to devices that are outside of the operator's infrastructure. The typical counters of an ACL are expected.

As per the End definition, an SRv6 router MUST only implement the End behavior on a local IPv6 address if that address has been explicitly enabled as an SRv6 SID. This address may or may not be associated with an interface. This address may or may not be routed. The only thing that matters is that the local SID must be explicitly instantiated and explicitly bound to a function. Packets received with destination address representing a local interface that has not been enabled as an SRv6 SID MUST be dropped.

In an SDN environment, one expects the controller to explicitly provision the SIDs and/or discover them as part of a service discovery function. Applications residing on top of the controller could then discover the required SIDs and combine them to form a distributed network program. The concept of "SRv6 network programming" refers to the capability for an application to encode any complex program as a set of individual functions distributed through the network. Some functions relate to underlay SLA, others to overlay/tenant, others to complex applications residing in VM and containers. The specification of the SRv6 control-plane is outside the scope of this document. We limit ourselves to a few important observations.

The End, End.T and End.X SIDs express topological functions and hence are expected to be signaled in the IGP together with the flavours PSP, USP and USD. The presence of SIDs in the IGP do not imply any routing semantics to the addresses represented by these SIDs. The routing reachability to an IPv6 address is solely governed by the classic, non-SID-related, IGP information. Routing is not governed neither influenced in any way by a SID advertisement in the IGP. These three SIDs provide important topological functions for the IGP to build FRR/TI-LFA solution and for TE processes relying on IGP LSDB to build SR policies.

BGP-LS is expected to be the key service discovery protocol. Every node is expected to advertise via BGP-LS its SRv6 capabilities (e.g. how many SIDs in can insert as part of an T.Insert behavior) and any locally instantiated SID .

The End.DX4, End.DX6, End.DT4, End.DT6, End.DT46, End.DX2, End.DX2V, End.DT2U and End.DT2M SIDs are expected to be signaled in BGP .

The following table summarizes which SIDs are signaled in which signaling protocol. IGP BGP-LS BGP IP/VPN/EVPN End (PSP, USP, USD) X X End.X (PSP, USP, USD) X X End.T (PSP, USP, USD) X X End.DX2 X X End.DX2V X X End.DT2U X X End.DT2M X X End.DX6 X X X End.DX4 X X X End.DT6 X X X End.DT4 X X X End.DT46 X X X End.B6.Insert X End.B6.Insert.Red X End.B6.Encaps X End.B6.Encaps.Red X End.B6.BM X End.S X The following table summarizes which transit capabilities are signaled in which signaling protocol. IGP BGP-LS BGP IP/VPN/EVPN T X T.Insert X X T.Insert.Red X X T.Encaps X X T.Encaps.Red X X T.Encaps.L2 X T.Encaps.L2.Red X The previous table describes generic capabilities. It does not describe specific instantiated SR policies. For example, a BGP-LS advertisement of the T capability of node N would indicate that node N supports the basic transit behavior. The T.Insert behavior would describe the capability of node N to perform a T.Insert behavior, specifically it would describe how many SIDs could be inserted by N without significant performance degradation. Same for T.Encaps (the number is potentially lower as the overhead of the additional outer IP header is accounted). The reader should also remember that any specific instantiated SR policy is always assigned a Binding SID. They should remember that BSIDs are advertised in BGP-LS as shown in . Hence, it is normal that only focuses on the generic capabilities related to T.Insert and T.Encaps as advertises the specific instantiated BSID properties.

This document requests the following new IANA registries: A new top-level registry "Segment-routing with IPv6 dataplane (SRv6) Parameters" to be created under IANA Protocol registries. This registry is being defined to serve as a top-level registry for keeping all other SRv6 sub-registries. A sub-registry "SRv6 Endpoint Behaviors" to be defined under top-level "Segment-routing with IPv6 dataplane (SRv6) Parameters" registry. This sub-registry maintains 16-bit identifiers for the SRv6 Endpoint behaviors. The range of the registry is 0-65535 (0x0000 - 0xFFFF) and has the following registration rules and allocation policies: Range Hex Registration proceadure Notes 0 0x0000 Reserved Invalid 1-32767 0x0001-0x7FFF IETF review Draft Specifications 32768-49151 0x8000-0xBFFF Reserved for experimental use 49152-65534 0xC000-0xFFFE Reserved for private use 65535 0xFFFF Reserved Opaque The initial registrations for the "Draft Specifications" portion of the sub-registry are as follows: Value Hex Endpoint function Reference 1 0x0001 End (no PSP, no USP) [This.ID] 2 0x0002 End with PSP [This.ID] 3 0x0003 End with USP [This.ID] 4 0x0004 End with PSP&USP [This.ID] 5 0x0005 End.X (no PSP, no USP) [This.ID] 6 0x0006 End.X with PSP [This.ID] 7 0x0007 End.X with USP [This.ID] 8 0x0008 End.X with PSP&USP [This.ID] 9 0x0009 End.T (no PSP, no USP) [This.ID] 10 0x000A End.T with PSP [This.ID] 11 0x000B End.T with USP [This.ID] 12 0x000C End.T with PSP&USP [This.ID] 13 0x000D End.B6 [This.ID] 14 0x000E End.B6.Encaps [This.ID] 15 0x000F End.BM [This.ID] 16 0x0010 End.DX6 [This.ID] 17 0x0011 End.DX4 [This.ID] 18 0x0012 End.DT6 [This.ID] 19 0x0013 End.DT4 [This.ID] 20 0x0014 End.DT46 [This.ID] 21 0x0015 End.DX2 [This.ID] 22 0x0016 End.DX2V [This.ID] 23 0x0017 End.DT2U [This.ID] 24 0x0018 End.DT2M [This.ID] 25 0x0019 End.S [This.ID] 26 0x001A End.B6.Red [This.ID] 27 0x001B End.B6.Encaps.Red [This.ID] 28 0x001C End with USD [This.ID] 29 0x001D End with PSP&USD [This.ID] 30 0x001E End with USP&USD [This.ID] 31 0x001F End with PSP, USP & USD [This.ID] 32 0x0020 End.X with USD [This.ID] 33 0x0021 End.X with PSP&USD [This.ID] 34 0x0022 End.X with USP&USD [This.ID] 35 0x0023 End.X with PSP, USP & USD [This.ID] 36 0x0024 End.T with USD [This.ID] 37 0x0025 End.T with PSP&USD [This.ID] 38 0x0026 End.T with USP&USD [This.ID] 39 0x0027 End.T with PSP, USP & USD [This.ID]

We are working on a extension of this document to provide Yang modelling for all the functionality described in this document. This work is ongoing in .

The authors would like to acknowledge Stefano Previdi, Dave Barach, Mark Townsley, Peter Psenak, Thierry Couture, Kris Michielsen, Paul Wells, Robert Hanzl, Dan Ye, Gaurav Dawra, Faisal Iqbal, Jaganbabu Rajamanickam, David Toscano, Asif Islam, Jianda Liu, Yunpeng Zhang, Jiaoming Li, Narendra A.K, Mike Mc Gourty, Bhupendra Yadav, Sherif Toulan, Satish Damodaran, John Bettink, Kishore Nandyala Veera Venk, Jisu Bhattacharya and Saleem Hafeez.

Daniel Bernier Bell Canada Canada Email: daniel.bernier@bell.ca Dirk Steinberg Steinberg Consulting Germany Email: dws@dirksteinberg.de Robert Raszuk Bloomberg LP United States of America Email: robert@raszuk.net Bruno Decraene Orange Frence Email: bruno.decraene@orange.com Bart Peirens Proximus Belgium Email: bart.peirens@proximus.com Hani Elmalky Ericsson United States of America Email: hani.elmalky@gmail.com Prem Jonnalagadda Barefoot Networks United States of America Email: prem@barefootnetworks.com Milad Sharif Barefoot Networks United States of America Email: msharif@barefootnetworks.com David Lebrun Google Belgium Email: dlebrun@google.com Stefano Salsano Universita di Roma "Tor Vergata" Italy Email: stefano.salsano@uniroma2.it Ahmed AbdelSalam Gran Sasso Science Institute Italy Email: ahmed.abdelsalam@gssi.it Gaurav Naik Drexel University United States of America Email: gn@drexel.edu Arthi Ayyangar Arista United States of America Email: arthi@arista.com Satish Mynam Innovium Inc. United States of America Email: smynam@innovium.com Wim Henderickx Nokia Belgium Email: wim.henderickx@nokia.com Shaowen Ma Juniper Singapore Email: mashao@juniper.net Ahmed Bashandy Individual United States of America Email: abashandy.ietf@gmail.com Francois Clad Cisco Systems, Inc. France Email: fclad@cisco.com Kamran Raza Cisco Systems, Inc. Canada Email: skraza@cisco.com Darren Dukes Cisco Systems, Inc. Canada Email: ddukes@cisco.com Patrice Brissete Cisco Systems, Inc. Canada Email: pbrisset@cisco.com Zafar Ali Cisco Systems, Inc. United States of America Email: zali@cisco.com