| MPLS Working Group | S. Bryant, Ed. |
| Internet-Draft | Huawei |
| Intended status: Standards Track | A. Farrel, Ed. |
| Expires: February 12, 2018 | J. Drake |
| Juniper Networks | |
| August 11, 2017 |
A Unified Approach to IP Segment Routing
draft-bryant-mpls-unified-ip-sr-01
Segment routing is a source routed forwarding method that allows packets to be steered through a network on paths other than the shortest path derived from the routing protocol. The approach uses information encoded in the packet header to partially or completely specify the route the packet takes through the network, and does not make use of a signaling protocol to pre-install paths in the network.
Two different encapsulations have been defined to enable segment routing in an MPLS network and in an IPv6 network. While acknowledging that there is a strong need to support segment routing in both environments, this document defines a converged, unified approach to segment routing that enables a single mechanism to be applied in both types of network. The resulting approach is also applicable to IPv4 networks without the need for any changes to the IPv4 specification.
This document makes no changes to the segment routing architecture and builds on existing protocol mechanisms such as the encapsulation of MPLS within UDP defined in RFC 7510.
No new procedures are introduced, but existing mechanisms are combined to achieve the desired result.
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].
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at http://datatracker.ietf.org/drafts/current/.
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."
This Internet-Draft will expire on February 12, 2018.
Copyright (c) 2017 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
Segment routing (SR) [I-D.ietf-spring-segment-routing] is a source routed forwarding method that allows packets to be steered through a network on paths other than the shortest path derived from the routing protocol. SR also allows the packets to be steered through a set of packet processing functions along that path. SR uses information encoded in the packet header to partially or completely specify the route the packet takes through the network and does not make use of a signaling protocol to pre-install paths in the network.
MPLS-SPRING [I-D.ietf-spring-segment-routing-mpls] (also known as MPLS Segment Routing or MPLS-SR) encodes the route the packet takes through the network and the instructions to be applied to the packet as it transits the network by imposing a stack of MPLS label entries on the packet.
The approach to IPv6 segment routing (SR) described in [I-D.ietf-6man-segment-routing-header] proposes that the segment routing instruction list is encoded as an ordered list of 128-bit IPv6 addresses that is carried in a new IPv6 extension header: the Source Routing Header (SRH). This approach can be challenging to implement in some types of forwarder, particularly where a large number of instructions/segments are needed to specify the required behaviour. Furthermore, the approach does not allow the use of SR techniques in legacy IPv4 networks. In this document we describe a method for running SR in IP networks that has a lower overhead and that uses an MPLS label stack carried in UDP as a method of encoding the segment routing instructions to be executed as the packet traverses the network. We call this Unified Segment Routing (USR) because the same instruction encoding method can be applied to MPLS, IPv4, and IPv6.
The format defined in this document uses 32 bits per additional instruction compared to the 128 bits for the method described in [I-D.ietf-6man-segment-routing-header]. The methods are further compared in Section 9.
The sequence of 32 bit units, one for each instruction, is called the Segment Routing Instruction Stack (SRIS). Each basic unit is encoded as an MPLS label stack entry and the segment routing instructions (i.e., the Segment Identifiers, SIDs) are encoded in the 20 bit MPLS Label fields. This is a hardware convenience rather than an indication of the use of MPLS as a forwarding protocol and the MPLS protocol stack, and in particular the MPLS control protocols, do not need to be deployed. It is a hardware convenience because many hardware components are already able to perform lookups based on MPLS labels.
In summary, the processing described in this document is a combination of normal MPLS-over-UDP behavior as described in [RFC7510], MPLS-SR lookup and label-pop behavior as described in [I-D.ietf-spring-segment-routing-mpls], and normal IP forwarding. No new procedures are introduced, but existing mechanisms are combined to achieve the desired result.
The method defined is a complementary way of running SR in an IP network that can be used alongside or interchangeably with that defined in [I-D.ietf-6man-segment-routing-header]. Implementers and deployers should consider the benefits and drawbacks of each method and select the approach most suited to their needs.
The USR protocol stack is shown in Figure 1.
+---------------------+
| |
| IP Header |
| |
+---------------------+
| |
| UDP Header |
| |
+---------------------+
| |
| Segment Routing |
| Instruction Stack |
~ ~
~ ~
| |
+---------------------+
| |
| Payload |
~ ~
~ ~
| |
+---------------------+
Figure 1: Packet Encapsulation
The payload may be of any type that, with an appropriate convergence layer, can be carried over a packet network. It is anticipated that the most common packet types will be IPv4, IPv6, native MPLS, and pseudowires [RFC3985].
Preceding the Payload is the Segment Routing Instruction Stack (SRIS) that carries the sequence of instructions to be executed on the packet as it traverses the network. This is the Segment Identifier (SID) stack that is the ordered list of segments described in [I-D.ietf-spring-segment-routing].
Preceding the SRIS is a UDP header. The UDP header is included to:
Preceding the UDP header is the IP header which may be IPv4 or IPv6.
The SRIS consists of a sequence of Segment Identifiers as described in [I-D.ietf-spring-segment-routing] encoded as an MPLS label stack as described in [I-D.ietf-spring-segment-routing-mpls].
The top SRIS entry is the next instruction to be executed. When the node to which this instruction is directed has processed the instruction it is removed (popped) from the SRIS, and the next instruction processed.
Each instruction is encoded in a single Label Stack Entry (LSE) as shown in Figure 2. The structure of the LSE is unchanged from [RFC3032].
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
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Instruction | TC |S| TTL |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Instruction: Label Value, 20 bits
TC: Traffic Class, 3 bits
S: Bottom of Stack, 1 bit
TTL: Time to Live, 8 bits
Figure 2: SRIS Label Stack Entry
As with [I-D.ietf-spring-segment-routing-mpls] a 32 bit LSE is used to carry each SR instruction. The instruction itself is carried in the 20 bit Label Value field. The TC field has the normal meaning as defined in [RFC3032] and modified in [RFC5462]. The S bit has bottom of stack semantics defined in [RFC3032]. TTL is discussed in Section 3.1.
The setting of the TTL is application specific, but the following operational consideration should be born in mind. In SR the size of the label stack may be increased within a single routing domain by various operations such as the pushing of a binding SID. Furthermore in SR packets are not necessarily constrained to travel on the shortest path with that routing domain. Consideration therefore has to be given to possibility of a forwarding loop. To mitigate against this it is RECOMMENDED that the TTL is continuously decremented as the packet passes through the SR network regardless of any other changes to the network layer encapsulation.
The procedures defined in [RFC7510] are followed. RFC7510 specifies the values to be used in the UDP Source Port, Destination Port, and Checksum fields.
An administrative domain, or set of administrative domains that are sufficiently well managed and monitored to be able to safely use IP segment routing is likely to comply with the requirements called out in [RFC7510] to permit operation with a zero checksum over IPv6. However each operator needs to validate the decision on whether or not to use a UDP checksum for themselves.
The [RFC7510] UDP header may be carried over IPv4 or over IPv6.
The IP source address is the address of the encapsulating device. The IP destination address is implied by the instruction at the top of the instruction stack.
If IPv4 is in use, fragmentation is not permitted.
There are six type of node in an SR domain:
The following sub-sections describe the processing behavior in each case.
In summary, the processing is a combination of normal MPLS-over-UDP behavior as described in [RFC7510], MPLS-SR lookup and label-pop behavior as described in [I-D.ietf-spring-segment-routing-mpls], and normal IP forwarding. No new procedures are introduced, but existing mechanisms ae combined to achieve the desired result.
The descriptions in the following sections represent the functional behavior. Optimizations on this behavior may be possible in implementations.
Domain ingress nodes receive packets from outside the domain and encapsulate them to be forwarded across the domain. Received packets may already be MPLS-SR packets (in the case of connecting two MPLS-SR networks across a native IP network), or may be IP or MPLS packets.
In the latter case, the packet is classified by the domain ingress node and an MPLS-SR stack is imposed. In the former case the MPLS-SR stack is already in the packet. The top entry in the stack is popped from the stack and retained for use below.
The packet is then encapsulated in UDP with the destination port set to 6635 to indicate "MPLS-UDP" as described in [RFC7510]. The source UDP port is set randomly or to provide entropy as described in [RFC7510].
The packet is then encapsulated in IP for transmission across the network. The IP source address is set to the domain ingress node, and the destination address is set to the address corresponding to the label that was previously popped from the stack.
This corresponds to sending the packet out of a virtual interface that corresponds to a virtual link between the ingress node and the next hop SR node realized by a UDP tunnel.
The packet is then sent into the IP network and is routed according to the local FIB and applying hashing to resolve any ECMP choices.
A legacy transit node is an IP router that has no SR capabilities. When such a router receives an MPLS-SR-in-UDP packet it will carry out normal TTL processing and if the packet is still live it will forward it as it would any other UDP-in-IP packet. The packet will be routed toward the destination indicated in the packet header using the local FIB and applying hashing to resolve any ECMP choices.
If the packet is mistakenly addressed to the legacy router, the UDP tunnel will be terminated and the packet will be discarded either because the MPLS-in-UDP port is not supported or because the uncovered top label has not been allocated. This is, however, a misconnection and should not occur unless there is a routing error.
Just because a node is SR capable and receives an MPLS-SR-in-UDP packet does not mean that it performs SR processing on the packet. Only routers identified by SIDs in the SR stack need to do such processing.
Routers that are not addressed by the destination address in the IP header simply treat the packet as a normal UDP-in-IP packet carrying out normal TTL processing and if the packet is still live routing the packet according to the local FIB and applying hashing to resolve any ECMP choices.
This is important because it means that the SR stack can be kept relatively small and the packet can be steered through the network using shortest path first routing between selected SR nodes.
When a router receives an MPLS-SR-in-UDP packet that is addressed to it, it acts as follows:
NOTE: This section needs a correction to the PHP behaviour since in SR this depends on a flag in the SID advertisement. This will be corrected in a future revision of this text.
The penultimate SR transit node is only different from the SR transit node described in Section 5.4 because it pops the final MPLS-SR SID from the stack. In order to avoid confusion at the egress, the router replaces the popped SR label with an explicit null label (label value 0 [RFC3032]). The packet is then encapsulated and sent as described in Section 5.4.
NOTE: This section may also need changing depending on any correction to the text in the previous section. This will also be addressed in a future revision of this document.
The domain egress strips the IP and UDP headers, pops the explicit null label, and forwards the payload packet according to its type and the local routing/forwarding mechanisms.
As previously noted, the procedures described in this document may be used to connect islands of SR functionality across an IP backbone, or can provide SR function within a native IP network. This section briefly expounds upon those two deployment modes.
Figure 3 shows two SR domains interconnected by an IP network. The procedures described in this document are deployed at border routers R1 and R2 and packets are carried across the backbone network in a UDP tunnel.
R1 acts as the domain ingress as described in Section 5.1. It takes the MPLS-SR packet from the SR domain, pops the top label and uses it to identify its peer border router R2. R1 then encapsulates the packet in UDP in IP and sends it toward R2.
Routers within the IP network simply forward the packet using normal IP routing.
R2 acts as a domain egress router as described in Section 5.6. It receives a packet that is addressed to it, strips the IP and UDP headers, and acts on the payload SR label stack to continue to route the packet.
________________________
______ ( ) ______
( ) ( IP Network ) ( )
( ) ( ) ( )
( -------- -------- )
( | Border | SR-in-UDP Tunnel | Border | )
( SR | Router |========================| Router | SR )
( | R1 | | R2 | )
( -------- -------- )
( ) ( ) ( )
(______) ( ) (______)
(________________________)
Figure 3: SR in UDP to Tunnel Between SR Sites
Figure 4 shows the procedures defined in this document to provide SR function across an IP network.
R1 receives a native packet and classifies it, determining that it should be sent on the SR path R2-R3-R4-R5. It imposes a label stack accordingly and then acts as a domain ingress as described in Section 5.1. It pops the label for R2, and encapsulates the packet in UDP in IP, sets the IP source to R1 and the IP destination to R2, and sends the packet into the IP network.
Routers Ra and Rb are transit routers that simply forward the packets using normal IP forwarding. They may be legacy transit routers (see Section 5.2) or on-path pass-through SR nodes (see Section 5.3).
R2 is an SR transit nodes as described in Section 5.4. It receives a packet addressed to it, strips the IP and UDP headers, and processes the SR label stack. It pops the top label and uses it to identify the next SR hop which is R3. R2 then encapsulates the packet in UDP in IP setting the IP source to R2 and the IP destination to R3.
Rc, Rd, and Re are transit routers and perform as Ra and Rb.
R3 is an SR transit node and performs as R2.
R4 is a penultimate SR transit node as described in Section 5.5. It receives a packet addressed to it, strips the IP and UDP headers, and processes the SR label stack. It pops the top label and uses it to identify the next SR hop which is R5. This was the last label in the stack so R4 includes an explicit null label before encapsulating the packet in UDP in IP setting the IP source to R4 and the IP destination to R5.
NOTE may also need adjustment to line up with the PHP text.
R5 is the domain egress as described in Section 5.6. It receives a packet addressed to it, strips the IP and UDP headers, and pops the explicit null label before forwarding the payload packet.
__________________________________
__( IP Network )__
__( )__
( -- -- -- )
-------- -- -- |R2| -- |R3| -- |R4| -- --------
| Ingress| |Ra| |Rb| | | |Rc| | | |Rd| | | |Re| | Egress |
--->| Router |===========| |======| |======| |======| Router |--->
| R1 | | | | | | | | | | | | | | | | | | R5 |
-------- -- -- | | -- | | -- | | -- --------
(__ -- -- -- __)
(__ __)
(__________________________________)
Figure 4: SR Within an IP Network
The method of advertising the tunnel encapsulation capability of a router using IS-IS or OSPF are specified in [I-D.ietf-isis-encapsulation-cap] and [I-D.ietf-ospf-encapsulation-cap] respectively. No changes to those procedures are needed in support of this work.
OAM at the payload layer follows the normal OAM procedures for the payload. To the payload the whole SR network looks like a tunnel.
OAM in the IP domain follows the normal IP procedures. This can only be carried out between on the IP hops between pairs of SR nodes.
OAM between instruction processing entities i.e. at the SR layer uses the procedures documented for MPLS.
The format described in [I-D.ietf-6man-segment-routing-header] (referred to here as SRv6) requires an initial 36 octet IPv6 header: no encoding is provided for operation in an IPv4 network. USR requires either an initial 36 octet IPv6 header or an initial 20 octet IPv4 header.
As previously noted, the method defined is a complementary way of running SR in an IPv6 network that can be used alongside or interchangeably with SRv6. Implementers and deployers should consider the benefits and drawbacks of each method and select the approach most suited to their needs.
The security consideration of [I-D.ietf-spring-ipv6-use-cases] and [RFC7510] apply.
It is difficult for an attacker to pass a raw MPLS encoded packet into a network and operators have considerable experience at excluding such packets at the network boundaries.
It is easy for an ingress node to detect any attempt to smuggle IP packet into the network since it would see that the UDP destination port was set to MPLS. SR packets not having a destination address terminating in the network would be transparently carried and would pose no security risk to the network under consideration.
This document makes no IANA requests.
This draft was partly inspired by [I-D.xu-mpls-unified-source-routing-instruction], and we acknowledge the following authors of version -02 of that draft: Robert Raszuk, Uma Chunduri, Luis M. Contreras, Luay Jalil, Hamid Assarpour, Gunter Van De Velde, Jeff Tantsura, and Shaowen Ma.
Thanks to Joel Halpern, Bruno Decraene, Loa Andersson, Ron Bonica, and Eric Rosen for their insightful comments on this draft.