<- RFC Index (9001..9100)
RFC 9015
Internet Engineering Task Force (IETF) A. Farrel
Request for Comments: 9015 Old Dog Consulting
Category: Standards Track J. Drake
ISSN: 2070-1721 E. Rosen
Juniper Networks
J. Uttaro
AT&T
L. Jalil
Verizon
June 2021
BGP Control Plane for the Network Service Header in Service Function
Chaining
Abstract
This document describes the use of BGP as a control plane for
networks that support service function chaining. The document
introduces a new BGP address family called the "Service Function
Chain (SFC) Address Family Identifier / Subsequent Address Family
Identifier" (SFC AFI/SAFI) with two Route Types. One Route Type is
originated by a node to advertise that it hosts a particular instance
of a specified service function. This Route Type also provides
"instructions" on how to send a packet to the hosting node in a way
that indicates that the service function has to be applied to the
packet. The other Route Type is used by a controller to advertise
the paths of "chains" of service functions and give a unique
designator to each such path so that they can be used in conjunction
with the Network Service Header (NSH) defined in RFC 8300.
This document adopts the service function chaining architecture
described in RFC 7665.
Status of This Memo
This is an Internet Standards Track document.
This document is a product of the Internet Engineering Task Force
(IETF). It represents the consensus of the IETF community. It has
received public review and has been approved for publication by the
Internet Engineering Steering Group (IESG). Further information on
Internet Standards is available in Section 2 of RFC 7841.
Information about the current status of this document, any errata,
and how to provide feedback on it may be obtained at
https://www.rfc-editor.org/info/rfc9015.
Copyright Notice
Copyright (c) 2021 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
(https://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.
Table of Contents
1. Introduction
1.1. Requirements Language
1.2. Terminology
2. Overview
2.1. Overview of Service Function Chaining
2.2. Control Plane Overview
3. BGP SFC Routes
3.1. Service Function Instance Route (SFIR)
3.1.1. SFIR Pool Identifier Extended Community
3.1.2. MPLS Mixed Swapping/Stacking Extended Community
3.2. Service Function Path Route (SFPR)
3.2.1. The SFP Attribute
3.2.2. General Rules for the SFP Attribute
4. Mode of Operation
4.1. Route Targets
4.2. Service Function Instance Routes
4.3. Service Function Path Routes
4.4. Classifier Operation
4.5. Service Function Forwarder Operation
4.5.1. Processing with "Gaps" in the SI Sequence
5. Selection within Service Function Paths
6. Looping, Jumping, and Branching
6.1. Protocol Control of Looping, Jumping, and Branching
6.2. Implications for Forwarding State
7. Advanced Topics
7.1. Correlating Service Function Path Instances
7.2. Considerations for Stateful Service Functions
7.3. VPN Considerations and Private Service Functions
7.4. Flow Specification for SFC Classifiers
7.5. Choice of Data Plane SPI/SI Representation
7.5.1. MPLS Representation of the SPI/SI
7.6. MPLS Label Swapping/Stacking Operation
7.7. Support for MPLS-Encapsulated NSH Packets
8. Examples
8.1. Example Explicit SFP with No Choices
8.2. Example SFP with Choice of SFIs
8.3. Example SFP with Open Choice of SFIs
8.4. Example SFP with Choice of SFTs
8.5. Example Correlated Bidirectional SFPs
8.6. Example Correlated Asymmetrical Bidirectional SFPs
8.7. Example Looping in an SFP
8.8. Example Branching in an SFP
8.9. Examples of SFPs with Stateful Service Functions
8.9.1. Forward and Reverse Choice Made at the SFF
8.9.2. Parallel End-to-End SFPs with Shared SFF
8.9.3. Parallel End-to-End SFPs with Separate SFFs
8.9.4. Parallel SFPs Downstream of the Choice
8.10. Examples Using IPv6 Addressing
8.10.1. Example Explicit SFP with No Choices
8.10.2. Example SFP with Choice of SFIs
8.10.3. Example SFP with Open Choice of SFIs
8.10.4. Example SFP with Choice of SFTs
9. Security Considerations
10. IANA Considerations
10.1. New BGP AF/SAFI
10.2. "SFP attribute" BGP Path Attribute
10.3. "SFP Attribute TLVs" Registry
10.4. "SFP Association Type" Registry
10.5. "Service Function Chaining Service Function Types"
Registry
10.6. Flow Specification for SFC Classifiers
10.7. New BGP Transitive Extended Community Type
10.8. "SFC Extended Community Sub-Types" Registry
10.9. New SPI/SI Representation Sub-TLV
10.10. "SFC SPI/SI Representation Flags" Registry
11. References
11.1. Normative References
11.2. Informative References
Acknowledgements
Contributors
Authors' Addresses
1. Introduction
As described in [RFC7498], the delivery of end-to-end services can
require a packet to pass through a series of Service Functions (SFs)
-- e.g., WAN and application accelerators, Deep Packet Inspection
(DPI) engines, firewalls, TCP optimizers, and server load balancers
-- in a specified order; this is termed "service function chaining".
There are a number of issues associated with deploying and
maintaining service function chaining in production networks, which
are described below.
Historically, if a packet needed to travel through a particular
service chain, the nodes hosting the service functions of that chain
were placed in the network topology in such a way that the packet
could not reach its ultimate destination without first passing
through all the service functions in the proper order. This need to
place the service functions at particular topological locations
limited the ability to adapt a service function chain to changes in
network topology (e.g., link or node failures), network utilization,
or offered service load. These topological restrictions on where the
service functions could be placed raised the following issues:
1. The process of configuring or modifying a service function chain
is operationally complex and may require changes to the network
topology.
2. Alternate or redundant service functions may need to be co-
located with the primary service functions.
3. When there is more than one path between source and destination,
forwarding may be asymmetric, and it may be difficult to support
bidirectional service function chains using simple routing
methodologies and protocols without adding mechanisms for traffic
steering or traffic engineering.
In order to address these issues, the service function chaining
architecture describes service function chains that are built in
their own overlay network (the service function overlay network),
coexisting with other overlay networks, over a common underlay
network [RFC7665]. A service function chain is a sequence of service
functions through which packet flows that satisfy specified criteria
will pass.
This document describes the use of BGP as a control plane for
networks that support service function chaining. The document
introduces a new BGP address family called the "Service Function
Chain (SFC) Address Family Identifier / Subsequent Address Family
Identifier" (SFC AFI/SAFI) with two Route Types. One Route Type is
originated by a node to advertise that it hosts a particular instance
of a specified service function. This Route Type also provides
"instructions" on how to send a packet to the hosting node in a way
that indicates that the service function has to be applied to the
packet. The other Route Type is used by a controller (a centralized
network component responsible for planning and coordinating service
function chaining within the network) to advertise the paths of
"chains" of service functions and give a unique designator to each
such path so that they can be used in conjunction with the Network
Service Header (NSH) [RFC8300].
This document adopts the service function chaining architecture
described in [RFC7665].
1.1. Requirements Language
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and
"OPTIONAL" in this document are to be interpreted as described in BCP
14 [RFC2119] [RFC8174] when, and only when, they appear in all
capitals, as shown here.
1.2. Terminology
This document uses the following terms from [RFC7665]:
* Bidirectional Service Function Chain
* Classifier
* Service Function (SF)
* Service Function Chain (SFC)
* Service Function Forwarder (SFF)
* Service Function Instance (SFI)
* Service Function Path (SFP)
* SFC branching
Additionally, this document uses the following terms from [RFC8300]:
* Network Service Header (NSH)
* Service Index (SI)
* Service Path Identifier (SPI)
This document introduces the following terms:
Service Function Instance Route (SFIR): A new BGP Route Type
advertised by the node that hosts an SFI to describe the SFI and
to announce the way to forward a packet to the node through the
underlay network.
Service Function Overlay Network: The logical network comprised of
classifiers, SFFs, and SFIs that are connected by paths or tunnels
through underlay transport networks.
Service Function Path Route (SFPR): A new BGP Route Type originated
by controllers to advertise the details of each SFP.
Service Function Type (SFT): An indication of the function and
features of an SFI.
2. Overview
This section provides an overview of service function chaining in
general and the control plane defined in this document. After
reading this section, readers may find it helpful to look through
Section 8 for some simple worked examples.
2.1. Overview of Service Function Chaining
In [RFC8300], a Service Function Chain (SFC) is an ordered list of
Service Functions (SFs). A Service Function Path (SFP) is an
indication of which instances of SFs are acceptable to be traversed
in an instantiation of an SFC in a service function overlay network.
The Service Path Identifier (SPI) is a 24-bit number that identifies
a specific SFP, and a Service Index (SI) is an 8-bit number that
identifies a specific point in that path. In the context of a
particular SFP (identified by an SPI), an SI represents a particular
service function and indicates the order of that SF in the SFP.
Within the context of a specific SFP, an SI references a set of one
or more SFs. Each of those SFs may be supported by one or more
Service Function Instances (SFIs). Thus, an SI may represent a
choice of SFIs of one or more service function types. By deploying
multiple SFIs for a single SF, one can provide load balancing and
redundancy.
A special functional element, called a "classifier", is located at
each ingress point to a service function overlay network. It assigns
the packets of a given packet flow to a specific SFP. This may be
done by comparing specific fields in a packet's header with local
policy, which may be customer/network/service specific. The
classifier picks an SFP and sets the SPI accordingly; it then sets
the SI to the value of the SI for the first hop in the SFP, and then
prepends a Network Service Header (NSH) [RFC8300] containing the
assigned SPI/SI to that packet. Note that the classifier and the
node that hosts the first SF in an SFP need not be located at the
same point in the service function overlay network.
Note that the presence of the NSH can make it difficult for nodes in
the underlay network to locate the fields in the original packet that
would normally be used to constrain equal-cost multipath (ECMP)
forwarding. Therefore, it is recommended that the node prepending
the NSH also provide some form of entropy indicator that can be used
in the underlay network. How this indicator is generated and
supplied, and how an SFF generates a new entropy indicator when it
forwards a packet to the next SFF, are out of the scope of this
document.
The Service Function Forwarder (SFF) receives a packet from the
previous node in an SFP, removes the packet's link layer or tunnel
encapsulation, and hands the packet and the NSH to the SFI for
processing. The SFI has no knowledge of the SFP.
When the SFF receives the packet and the NSH back from the SFI, it
must select the next SFI along the path using the SPI and SI in the
NSH and potentially choosing between multiple SFIs (possibly of
different SFTs), as described in Section 5. In the normal case, the
SPI remains unchanged, and the SI will have been decremented to
indicate the next SF along the path. But other possibilities exist
if the SF makes other changes to the NSH through a process of
reclassification:
* The SI in the NSH may indicate:
- A previous SF in the path; this is known as "looping" (see
Section 6).
- An SF further down the path; this is known as "jumping" (again
see Section 6).
* The SPI and the SI may point to an SF on a different SFP; this is
known as "branching" (see Section 6).
Such modifications are limited to within the same service function
overlay network. That is, an SPI is known within the scope of
service function overlay network. Furthermore, the new SI value is
interpreted in the context of the SFP identified by the SPI.
As described in [RFC8300], an SPI that is unknown or not valid is
treated as an error, and the SFF drops the packet; such errors should
be logged, and such logs are subject to rate limits.
Also, as described in [RFC8300], an SFF receiving an SI that is
unknown in the context of the SPI can reduce the value to the next
meaningful SI value in the SFP indicated by the SPI. If no such
value exists, or if the SFF does not support reducing the SI, the SFF
drops the packet and should log the event; such logs are also subject
to rate limits.
The SFF then selects an SFI that provides the SF denoted by the SPI/
SI and forwards the packet to the SFF that supports that SFI.
[RFC8300] makes it clear that the intended scope is for use within a
single provider's operational domain.
This document adopts the service function chaining architecture
described in [RFC7665] and adds a control plane to support the
functions, as described in Section 2.2. An essential component of
this solution is the controller. This is a network component
responsible for planning SFPs within the network. It gathers
information about the availability of SFIs and SFFs, instructs the
control plane about the SFPs to be programmed, and instructs the
classifiers how to assign traffic flows to individual SFPs.
2.2. Control Plane Overview
To accomplish the function described in Section 2.1, this document
introduces the Service Function Type (SFT), which is the category of
SF that is supported by an SFF (such as "firewall"). An IANA
registry of service function types is introduced in Section 10.5 and
is consistent with types used in other work, such as [BGP-LS-SR]. An
SFF may support SFs of multiple different SFTs, and it may support
multiple SFIs of each SF.
The registry of SFT values (see Section 10.5) is split into three
ranges with assignment policies per [RFC8126]:
* The special-purpose SFT values range is assigned through Standards
Action. Values in that range are used for special SFC operations
and do not apply to the types of SF that may form part of the SFP.
* The First Come First Served range tracks assignments of SFT values
made by any party that defines an SF type. Reference through an
Internet-Draft is desirable, but not required.
* The Private Use range is not tracked by IANA and is primarily
intended for use in private networks where the meaning of the SFT
values is locally tracked and under the control of a local
administrator.
It is envisaged that the majority of SFT values used will be assigned
from the First Come First Served space in the registry. This will
ensure interoperability, especially in situations where software and
hardware from different vendors are deployed in the same networks, or
when networks are merged. However, operators of private networks may
choose to develop their own SFs and manage the configuration and
operation of their network through their own list of SFT values.
This document also introduces a new BGP AFI/SAFI (values 31 and 9,
respectively) for "SFC Routes". Two SFC Route Types are defined by
this document: the Service Function Instance Route (SFIR) and the
Service Function Path Route (SFPR). As detailed in Section 3, the
Route Type is indicated by a subfield in the Network Layer
Reachability Information (NLRI).
* The SFIR is advertised by the node that provides access to the
service function instance (i.e., the SFF). The SFIR describes a
particular instance of a particular SF (i.e., an SFI) and the way
to forward a packet to it through the underlay network, i.e., IP
address and encapsulation information.
* The SFPRs are originated by controllers. One SFPR is originated
for each SFP. The SFPR specifies:
A. the SPI of the path,
B. the sequence of SFTs and/or SFIs of which the path consists,
and
C. for each such SFT or SFI, the SI that represents it in the
identified path.
This approach assumes that there is an underlay network that provides
connectivity between SFFs and controllers and that the SFFs are
grouped to form one or more service function overlay networks through
which SFPs are built. We assume that the controllers have BGP
connectivity to all SFFs and all classifiers within each service
function overlay network.
When choosing the next SFI in a path, the SFF uses the SPI and SI as
well as the SFT to choose among the SFIs, applying, for example, a
load-balancing algorithm or direct knowledge of the underlay network
topology, as described in Section 4.
The SFF then encapsulates the packet using the encapsulation
specified by the SFIR of the selected SFI and forwards the packet.
See Figure 1.
Thus, the SFF can be seen as a portal in the underlay network through
which a particular SFI is reached.
Figure 1 shows a reference model for the service function chaining
architecture. There are four SFFs (SFF-1 through SFF-4) connected by
tunnels across the underlay network. Packets arrive at a classifier
and are channeled along SFPs to destinations reachable through SFF-4.
SFF-1 and SFF-4 each have one instance of one SF attached (SFa and
SFe). SFF-2 has two types of SF attached: one instance of one (SFc)
and three instances of the other (SFb). SFF-3 has just one instance
of an SF (SFd), but in this case, the type of SFd is the same type as
SFb (SFTx).
This figure demonstrates how load balancing can be achieved by
creating several SFPs that satisfy the same SFC. Suppose an SFC
needs to include SFa, an SF of type SFTx, and SFc. A number of SFPs
can be constructed using any instance of SFb or using SFd. Load
balancing may be applied at two places:
* The classifier may distribute different flows onto different SFPs
to share the load in the network and across SFIs.
* SFF-2 may distribute different flows (on the same SFP) to
different instances of SFb to share the processing load.
Note that, for convenience and clarity, Figure 1 shows only a few
tunnels between SFFs. There could be a full mesh of such tunnels, or
more likely, a selection of tunnels connecting key SFFs to enable the
construction of SFPs and balance load and traffic in the network.
Further, the figure does not show any controllers; these would each
have BGP connectivity to the classifier and all of the SFFs.
Packets
| | |
------------
| |
| Classifier |
| |
------+-----
|
---+--- --------- -------
| | Tunnel | | | |
| SFF-1 |===============| SFF-2 |=========| SFF-4 |
| | | | | |
| | -+-----+- | |
| | ,,,,,,,,,,,,,,/,, \ | |
| | ' .........../. ' ..\...... | |
| | ' : SFb / : ' : \ SFc : | |
| | ' : ---+- : ' : --+-- : | |
| | ' : -| SFI | : ' : | SFI | : | |
| | ' : -| ----- : ' : ----- : | |
| | ' : | ----- : ' ......... | |
| | ' : ----- : ' | |
| | ' ............. ' | |--- Dests
| | ' ' | |--- Dests
| | ' ......... ' | |
| | ' : ----- : ' | |
| | ' : | SFI | : ' | |
| | ' : --+-- : ' | |
| | ' :SFd | : ' | |
| | ' ....|.... ' | |
| | ' | ' | |
| | ' SFTx | ' | |
| | ',,,,,,,,|,,,,,,,,' | |
| | | | |
| | ---+--- | |
| | | | | |
| |======| SFF-3 |====================| |
---+--- | | ---+---
| ------- |
....|.... ....|....
: | SFa: : | SFe:
: --+-- : : --+-- :
: | SFI | : : | SFI | :
: ----- : : ----- :
......... .........
Figure 1: The Service Function Chaining Architecture Reference Model
As previously noted, [RFC8300] makes it clear that the mechanisms it
defines are intended for use within a single provider's operational
domain. This reduces the requirements on the control plane function.
Section 5.2 of [RFC7665] sets out the functions provided by a control
plane for a service function chaining network. The functions are
broken down into six items, the first four of which are completely
covered by the mechanisms described in this document:
1. Visibility of all SFs and the SFFs through which they are
reached.
2. Computation of SFPs and programming into the network.
3. Selection of SFIs explicitly in the SFP or dynamically within the
network.
4. Programming of SFFs with forwarding path information.
The fifth and sixth items in the list in RFC 7665 concern the use of
metadata. These are more peripheral to the control plane mechanisms
defined in this document but are discussed in Section 4.4.
3. BGP SFC Routes
This document defines a new AFI/SAFI for BGP, known as "SFC", with an
NLRI that is described in this section.
The format of the SFC NLRI is shown in Figure 2.
+---------------------------------------+
| Route Type (2 octets) |
+---------------------------------------+
| Length (2 octets) |
+---------------------------------------+
| Route Type specific (variable) |
+---------------------------------------+
Figure 2: The Format of the SFC NLRI
The "Route Type" field determines the encoding of the rest of the
Route Type specific SFC NLRI.
The "Length" field indicates the length, in octets, of the "Route
Type specific" field of the SFC NLRI.
This document defines the following Route Types:
1. Service Function Instance Route (SFIR)
2. Service Function Path Route (SFPR)
An SFIR is used to identify an SFI. An SFPR defines a sequence of
SFs (each of which has at least one instance advertised in an SFIR)
that form an SFP.
The detailed encoding and procedures for these Route Types are
described in subsequent sections.
The SFC NLRI is carried in BGP [RFC4271] using BGP Multiprotocol
Extensions [RFC4760] with an Address Family Identifier (AFI) of 31
and a Subsequent Address Family Identifier (SAFI) of 9. The "NLRI"
field in the MP_REACH_NLRI/MP_UNREACH_NLRI attribute contains the SFC
NLRI, encoded as specified above.
In order for two BGP speakers to exchange SFC NLRIs, they MUST use
BGP capabilities advertisements to ensure that they both are capable
of properly processing such NLRIs. This is done as specified in
[RFC4760], by using capability code 1 (Multiprotocol BGP) with an AFI
of 31 and a SAFI of 9.
The "nexthop" field of the MP_REACH_NLRI attribute of the SFC NLRI
MUST be set to a loopback address of the advertising SFF.
3.1. Service Function Instance Route (SFIR)
Figure 3 shows the Route Type specific NLRI of the SFIR.
+--------------------------------------------+
| Route Distinguisher (RD) (8 octets) |
+--------------------------------------------+
| Service Function Type (2 octets) |
+--------------------------------------------+
Figure 3: SFIR Route Type Specific NLRI
[RFC4364] defines a Route Distinguisher (RD) as consisting of a two-
byte "Type" field and a six-byte "Value" field, and it defines RD
types 0, 1, and 2. In this specification, the RD (used for the SFIR)
MUST be of type 0, 1, or 2.
If two SFIRs are originated from different administrative domains
(within the same provider's operational domain), they MUST have
different RDs. In particular, SFIRs from different VPNs (for
different service function overlay networks) MUST have different RDs,
and those RDs MUST be different from any non-VPN SFIRs.
The SFT identifies the functions/features an SF can offer, e.g.,
classifier, firewall, load balancer. There may be several SFIs that
can perform a given service function. Each node hosting an SFI MUST
originate an SFIR for each type of SF that it hosts (as indicated by
the SFT value), and it MAY advertise an SFIR for each instance of
each type of SF. A minimal advertisement allows construction of
valid SFPs and leaves the selection of SFIs to the local SFF; a
detailed advertisement may have scaling concerns but allows a
controller that constructs an SFP to make an explicit choice of SFI.
Note that a node may advertise all its SFIs of one SFT in one shot
using normal BGP UPDATE packing. That is, all of the SFIRs in an
Update share a common Tunnel Encapsulation and Route Target (RT)
attribute. See also Section 3.2.1.
The SFIR representing a given SFI will contain an NLRI with "RD"
field set to an RD as specified above, and with the "SFT" field set
to identify that SFI's SFT. The values for the "SFT" field are taken
from a registry administered by IANA (see Section 10). A BGP UPDATE
containing one or more SFIRs MUST also include a tunnel encapsulation
attribute [RFC9012]. If a data packet needs to be sent to an SFI
identified in one of the SFIRs, it will be encapsulated as specified
by the tunnel encapsulation attribute and then transmitted through
the underlay network.
Note that the tunnel encapsulation attribute MUST contain sufficient
information to allow the advertising SFF to identify the overlay or
VPN network that a received packet is transiting. This is because
the [SPI, SI] in a received packet is specific to a particular
overlay or VPN network.
3.1.1. SFIR Pool Identifier Extended Community
This document defines a new transitive Extended Community [RFC4360]
of type 0x0b called the "SFC Extended Community". When used with
Sub-Type 1, this is called the "SFIR Pool Identifier extended
community". It MAY be included in SFIR advertisements, and it is
used to indicate the identity of a pool of SFIRs to which an SFIR
belongs. Since an SFIR may be a member of more than one pool,
multiple of these extended communities may be present on a single
SFIR advertisement.
SFIR pools allow SFIRs to be grouped for any purpose. Possible uses
include control plane scalability and stability. A pool identifier
may be included in an SFPR to indicate a set of SFIs that are
acceptable at a specific point on an SFP (see Sections 3.2.1.3 and
4.3).
The SFIR Pool Identifier Extended Community is encoded in 8 octets as
shown in Figure 4.
+--------------------------------------------+
| Type = 0x0b (1 octet) |
+--------------------------------------------+
| Sub-Type = 1 (1 octet) |
+--------------------------------------------+
| SFIR Pool Identifier value (6 octets) |
+--------------------------------------------+
Figure 4: The SFIR Pool Identifier Extended Community
The SFIR Pool Identifier value is encoded in a 6-octet field in
network byte order, and the value is unique within the scope of an
overlay network. This means that pool identifiers need to be
centrally managed, which is consistent with the assignment of SFIs to
pools.
3.1.2. MPLS Mixed Swapping/Stacking Extended Community
As noted in Section 3.1.1, this document defines a new transitive
Extended Community of type 0x0b called the "SFC Extended Community".
When used with Sub-Type 2, this is called the "MPLS Mixed Swapping/
Stacking Labels Extended Community". The community is encoded as
shown in Figure 5. It contains a pair of MPLS labels: an SFC Context
Label and an SF Label, as described in [RFC8595]. Each label is 20
bits encoded in a 3-octet (24-bit) field with 4 trailing bits that
MUST be set to zero.
+--------------------------------------------+
| Type = 0x0b (1 octet) |
+--------------------------------------------|
| Sub-Type = 2 (1 octet) |
+--------------------------------------------|
| SFC Context Label (3 octets) |
+--------------------------------------------|
| SF Label (3 octets) |
+--------------------------------------------+
Figure 5: The MPLS Mixed Swapping/Stacking Labels Extended Community
Note that it is assumed that each SFF has one or more globally unique
SFC Context Labels and that the context-label space and the SPI-
address space are disjoint. In other words, a label value cannot be
used to indicate both an SFC context and an SPI, and it can be
determined from knowledge of the label spaces whether a label
indicates an SFC context or an SPI.
If an SFF supports SFP Traversal with an MPLS Label Stack, it MUST
include this Extended Community with the SFIRs that it advertises.
See Section 7.6 for a description of how this Extended Community is
used.
3.2. Service Function Path Route (SFPR)
Figure 6 shows the Route Type specific NLRI of the SFPR.
+-----------------------------------------------+
| Route Distinguisher (RD) (8 octets) |
+-----------------------------------------------+
| Service Path Identifier (SPI) (3 octets) |
+-----------------------------------------------+
Figure 6: SFPR Route Type Specific NLRI
[RFC4364] defines a Route Distinguisher (RD) as consisting of a two-
byte "Type" field and a six-byte "Value" field, and it defines RD
types 0, 1, and 2. In this specification, the RD (used for the SFPR)
MUST be of type 0, 1, or 2.
All SFPs MUST be associated with an RD. The association of an SFP
with an RD is determined by provisioning. If two SFPRs are
originated from different controllers, they MUST have different RDs.
Additionally, SFPRs from different VPNs (i.e., in different service
function overlay networks) MUST have different RDs, and those RDs
MUST be different from any non-VPN SFPRs.
The Service path identifier is defined in [RFC8300] and is the value
to be placed in the "Service Path Identifier" field of the NSH of any
packet sent on this SFP. It is expected that one or more controllers
will originate these routes in order to configure a service function
overlay network.
The SFP is described in a new BGP Path attribute, the SFP attribute.
Section 3.2.1 shows the format of that attribute.
3.2.1. The SFP Attribute
[RFC4271] defines BGP Path attributes. This document introduces a
new Optional Transitive Path attribute called the "SFP attribute",
with value 37. The first SFP attribute MUST be processed, and
subsequent instances MUST be ignored.
The common fields of the SFP attribute are set as follows:
* The Optional bit is set to 1 to indicate that this is an optional
attribute.
* The Transitive bit is set to 1 to indicate that this is a
transitive attribute.
* The Extended Length bit is set if the length of the SFP attribute
is encoded in one octet (set to 0) or two octets (set to 1), as
described in [RFC4271].
* The Attribute Type Code is set to 37.
The content of the SFP attribute is a series of Type-Length-Value
(TLV) constructs. Some TLVs may include Sub-TLVs. All TLVs and Sub-
TLVs have a common format:
Type: A single octet indicating the type of the SFP attribute TLV.
Values are taken from the registry described in Section 10.3.
Length: A two-octet field indicating the length of the data
following the "Length" field, counted in octets.
Value: The contents of the TLV.
The formats of the TLVs defined in this document are shown in the
following sections. The presence rules and meanings are as follows.
* The SFP attribute contains a sequence of zero or more Association
TLVs. That is, the Association TLV is OPTIONAL. Each Association
TLV provides an association between this SFPR and another SFPR.
Each associated SFPR is indicated using the RD with which it is
advertised (we say the SFPR-RD to avoid ambiguity).
* The SFP attribute contains a sequence of one or more Hop TLVs.
Each Hop TLV contains all of the information about a single hop in
the SFP.
* Each Hop TLV contains an SI value and a sequence of one or more
SFT TLVs. Each SFT TLV contains an SFI reference for each
instance of an SF that is allowed at this hop of the SFP for the
specific SFT. Each SFI is indicated using the RD with which it is
advertised (we say the SFIR-RD to avoid ambiguity).
Section 6 of [RFC4271] describes the handling of malformed BGP
attributes, or those that are in error in some way. [RFC7606]
revises BGP error handling specifically for the UPDATE message,
provides guidelines for the authors of documents defining new
attributes, and revises the error-handling procedures for a number of
existing attributes. This document introduces the SFP attribute and
so defines error handling as follows:
* When parsing a message, an unknown Attribute Type Code or a length
that suggests that the attribute is longer than the remaining
message is treated as a malformed message, and the "treat-as-
withdraw" approach is used as per [RFC7606].
* When parsing a message that contains an SFP attribute, the
following cases constitute errors:
1. Optional bit is set to 0 in the SFP attribute.
2. Transitive bit is set to 0 in the SFP attribute.
3. Unknown "TLV Type" field found in the SFP attribute.
4. TLV length that suggests the TLV extends beyond the end of the
SFP attribute.
5. Association TLV contains an unknown SFPR-RD.
6. No Hop TLV found in the SFP attribute.
7. No Sub-TLV found in a Hop TLV.
8. Unknown SFIR-RD found in an SFT TLV.
* The errors listed above are treated as follows:
1, 2, 4, 6, 7: The attribute MUST be treated as malformed and the
"treat-as-withdraw" approach used as per [RFC7606].
3: Unknown TLVs MUST be ignored, and message processing MUST
continue.
5, 8: The absence of an RD with which to correlate is nothing
more than a soft error. The receiver SHOULD store the
information from the SFP attribute until a corresponding
advertisement is received.
3.2.1.1. The Association TLV
The Association TLV is an optional TLV in the SFP attribute. It MAY
be present multiple times. Each occurrence provides an association
with another SFP as advertised in another SFPR. The format of the
Association TLV is shown in Figure 7.
+--------------------------------------------+
| Type = 1 (1 octet) |
+--------------------------------------------|
| Length (2 octets) |
+--------------------------------------------|
| Association Type (1 octet) |
+--------------------------------------------|
| Associated SFPR-RD (8 octets) |
+--------------------------------------------|
| Associated SPI (3 octets) |
+--------------------------------------------+
Figure 7: The Format of the Association TLV
The fields are as follows:
* "Type" is set to 1 to indicate an Association TLV.
* "Length" indicates the length in octets of the "Association Type"
and "Associated SFPR-RD" fields. The value of the "Length" field
is 12.
* The "Association Type" field indicates the type of association.
The values are tracked in an IANA registry (see Section 10.4).
Only one value is defined in this document: Type 1 indicates
association of two unidirectional SFPs to form a bidirectional
SFP. An SFP attribute SHOULD NOT contain more than one
Association TLV with Association Type 1; if more than one is
present, the first one MUST be processed, and subsequent instances
MUST be ignored. Note that documents that define new association
types must also define the presence rules for Association TLVs of
the new type.
* The Associated SFPR-RD contains the RD of the associated SFP as
advertised in an SFPR.
* The Associated SPI contains the SPI of the associated SFP as
advertised in an SFPR.
Association TLVs with unknown Association Type values SHOULD be
ignored. Association TLVs that contain an Associated SFPR-RD value
equal to the RD of the SFPR in which they are contained SHOULD be
ignored. If the Associated SPI is not equal to the SPI advertised in
the SFPR indicated by the Associated SFPR-RD, then the Association
TLV SHOULD be ignored. In all three of these cases, an
implementation MAY reject the SFP attribute as malformed and use the
"treat-as-withdraw" approach per [RFC7606]; however, implementors are
cautioned that such an approach may make an implementation less
flexible in the event of future extensions to this protocol.
Note that when two SFPRs reference each other using the Association
TLV, one SFPR advertisement will be received before the other.
Therefore, processing of an association MUST NOT be rejected simply
because the Associated SFPR-RD is unknown.
Further discussion of correlation of SFPRs is provided in
Section 7.1.
3.2.1.2. The Hop TLV
There is one Hop TLV in the SFP attribute for each hop in the SFP.
The format of the Hop TLV is shown in Figure 8. At least one Hop TLV
MUST be present in an SFP attribute.
+--------------------------------------------+
| Type = 2 (1 octet) |
+--------------------------------------------|
| Length (2 octets) |
+--------------------------------------------|
| Service Index (1 octet) |
+--------------------------------------------|
| Hop Details (variable) |
+--------------------------------------------+
Figure 8: The Format of the Hop TLV
The fields are as follows:
* "Type" is set to 2 to indicate a Hop TLV.
* "Length" indicates the length, in octets, of the "Service Index"
and "Hop Details" fields.
* The Service Index is defined in [RFC8300] and is the value found
in the "Service Index" field of the NSH that an SFF will use to
look up to which next SFI a packet is to be sent.
* The "Hop Details" field consists of a sequence of one or more Sub-
TLVs.
Each hop of the SFP may demand that a specific type of SF is
executed, and that type is indicated in Sub-TLVs of the Hop TLV. At
least one Sub-TLV MUST be present. This document defines the SFT
Sub-TLV (see Section 3.2.1.3) and the MPLS Swapping/Stacking Sub-TLV
(see Section 3.2.1.4); other Sub-TLVs may be defined in future. The
SFT Sub-TLV provides a list of which types of SF are acceptable at a
specific hop, and for each type it allows a degree of control to be
imposed on the choice of SFIs of that particular type. The MPLS
Swapping/Stacking Sub-TLV indicates the type of SFC encoding to use
in an MPLS label stack.
If no Hop TLV is present in an SFP attribute, it is a malformed
attribute.
3.2.1.3. The SFT Sub-TLV
The SFT Sub-TLV MAY be included in the list of Sub-TLVs of the Hop
TLV. The format of the SFT Sub-TLV is shown in Figure 9. The Hop
Sub-TLV contains a list of SFIR-RD values each taken from the
advertisement of an SFI. Together they form a list of acceptable
SFIs of the indicated type.
+--------------------------------------------+
| Type = 3 (1 octet) |
+--------------------------------------------|
| Length (2 octets) |
+--------------------------------------------|
| Service Function Type (2 octets) |
+--------------------------------------------|
| SFIR-RD List (variable) |
+--------------------------------------------+
Figure 9: The Format of the SFT Sub-TLV
The fields are as follows:
* "Type" is set to 3 to indicate an SFT Sub-TLV.
* "Length" indicates the length, in octets, of the "Service Function
Type" and "SFIR-RD List" fields.
* The SFT value indicates the category (type) of SF that is to be
executed at this hop. The types are as advertised for the SFs
supported by the SFFs. SFT values in the range 1-31 are special-
purpose SFT values and have meanings defined by the documents that
describe them -- the value "Change Sequence" is defined in
Section 6.1 of this document.
* The hop description is further qualified beyond the specification
of the SFTs by listing, for each SFT in each hop, the SFIs that
may be used at the hop. The SFIs are identified using the SFIR-
RDs from the advertisements of the SFIs in the SFIRs. Note that
if the list contains one or more SFIR Pool Identifiers, then for
each, the SFIR-RD list is effectively expanded to include the
SFIR-RD of each SFIR advertised with that SFIR Pool Identifier.
An SFIR-RD of value zero has special meaning, as described in
Section 5. Each entry in the list is eight octets long, and the
number of entries in the list can be deduced from the value of the
"Length" field.
* Note that an SFIR-RD is of type 0, 1, or 2 (as described in
Section 3.1). Thus, the high-order octet of an RD found in an
SFIR-RD List always has a value of 0x00. However, the high-order
octet of an SFIR Pool Identifier (an Extended Community with
"Type" field 0x0b) will always have a nonzero value. This allows
the node processing the SFIR-RD list to distinguish between the
two types of list entry.
3.2.1.4. MPLS Swapping/Stacking Sub-TLV
The MPLS Swapping/Stacking Sub-TLV (Type value 4) is a zero-length
Sub-TLV that is OPTIONAL in the Hop TLV and is used when the data
representation is MPLS (see Section 7.5). When present, it indicates
to the classifier imposing an MPLS label stack that the current hop
is to use an {SFC Context Label, SF label} rather than an {SPI, SF}
label pair. See Section 7.6 for more details.
3.2.1.5. SFP Traversal With MPLS Label Stack TLV
The SFP Traversal With MPLS Label Stack TLV (Type value 5) is a zero-
length TLV that can be carried in the SFP attribute and indicates to
the classifier and the SFFs on the SFP that an MPLS label stack with
label swapping/stacking is to be used for packets traversing the SFP.
All of the SFFs specified at each of the SFP's hops MUST have
advertised an MPLS Mixed Swapping/Stacking Extended Community (see
Section 3.1.2) for the SFP to be considered usable.
3.2.2. General Rules for the SFP Attribute
It is possible for the same SFI, as described by an SFIR, to be used
in multiple SFPRs.
When two SFPRs have the same SPI but different SFPR-RDs, there can be
three cases:
1. Two or more controllers are originating SFPRs for the same SFP.
In this case, the content of the SFPRs is identical, and the
duplication is to ensure receipt and provide controller
redundancy.
2. There is a transition in content of the advertised SFP, and the
advertisements may originate from one or more controllers. In
this case, the content of the SFPRs will be different.
3. The reuse of an SPI may result from a configuration error.
There is no way in any of these cases for the receiving SFF to know
which SFPR to process, and the SFPRs could be received in any order.
At any point in time, when multiple SFPRs have the same SPI but
different SFPR-RDs, the SFF MUST use the SFPR with the numerically
lowest SFPR-RD when interpreting the RDs as 8-octet integers in
network byte order. The SFF SHOULD log this occurrence to assist
with debugging.
Furthermore, a controller that wants to change the content of an SFP
is RECOMMENDED to use a new SPI and so create a new SFP onto which
the classifiers can transition packet flows before the SFPR for the
old SFP is withdrawn. This avoids any race conditions with SFPR
advertisements.
Additionally, a controller SHOULD NOT reuse an SPI after it has
withdrawn the SFPR that used it until at least a configurable amount
of time has passed. This timer SHOULD have a default of one hour.
4. Mode of Operation
This document describes the use of BGP as a control plane to create
and manage a service function overlay network.
4.1. Route Targets
The main feature introduced by this document is the ability to create
multiple service function overlay networks through the use of Route
Targets (RTs) [RFC4364].
Every BGP UPDATE containing an SFIR or SFPR carries one or more RTs.
The RT carried by a particular SFIR or SFPR is determined by the
provisioning of the route's originator.
Every node in a service function overlay network is configured with
one or more import RTs. Thus, each SFF will import only the SFPRs
with matching RTs, allowing the construction of multiple service
function overlay networks or the instantiation of SFCs within a Layer
3 Virtual Private Network (L3VPN) or Ethernet VPN (EVPN) instance
(see Section 7.3). An SFF that has a presence in multiple service
function overlay networks (i.e., one that imports more than one RT)
will usually maintain separate forwarding state for each overlay
network.
4.2. Service Function Instance Routes
The SFIR (see Section 3.1) is used to advertise the existence and
location of a specific SFI; it consists of:
* The RT as just described.
* A Service Function Type (SFT) that is the type of service function
that is provided (such as "firewall").
* A Route Distinguisher (RD) that is unique to a specific overlay.
4.3. Service Function Path Routes
The SFPR (see Section 3.2) describes a specific path of an SFC. The
SFPR contains the Service Path Identifier (SPI) used to identify the
SFP in the NSH in the data plane. It also contains a sequence of
Service Indexes (SIs). Each SI identifies a hop in the SFP, and each
hop is a choice between one or more SFIs.
As described in this document, each SFP route is identified in the
service function overlay network by an RD and an SPI. The SPI is
unique within a single VPN instance supported by the underlay
network.
The SFPR advertisement comprises:
* An RT as described in Section 4.1.
* A tuple that identifies the SFPR.
- An RD that identifies an advertisement of an SFPR.
- The SPI that uniquely identifies this path within the VPN
instance distinguished by the RD. This SPI also appears in the
NSH.
* A series of SIs. Each SI is used in the context of a particular
SPI and identifies one or more SFs (distinguished by their SFTs).
For each SF, it identifies a set of SFIs that instantiate the SF.
The values of the SI indicate the order in which the SFs are to be
executed in the SFP that is represented by the SPI.
* The SI is used in the NSH to identify the entries in the SFP.
Note that the SI values have meaning only relative to a specific
path. They have no semantic other than to indicate the order of
SFs within the path and are assumed to be monotonically decreasing
from the start to the end of the path [RFC8300].
* Each SI is associated with a set of one or more SFIs that can be
used to provide the indexed SF within the path. Each member of
the set comprises:
- The RD used in an SFIR advertisement of the SFI.
- The SFT that indicates the type of function as used in the same
SFIR advertisement of the SFI.
This may be summarized as follows, where the notations "SFPR-RD" and
"SFIR-RD" are used to distinguish the two different RDs, and where
"*" indicates a multiplier:
RT, {SFPR-RD, SPI}, m * {SI, {n * {SFT, p * SFIR-RD} } }
Where:
RT: Route Target
SFPR-RD: The Route Descriptor of the SFPR advertisement
SPI: Service Path Identifier used in the NSH
m: The number of hops in the SFP
n: The number of choices of SFT for a specific hop
p: The number of choices of SFI for a given SFT in a specific hop
SI: Service Index used in the NSH to indicate a specific hop
SFT: The Service Function Type used in the same advertisement of the
SFIR
SFIR-RD: The Route Descriptor used in an advertisement of the SFIR
That is, there can be multiple SFTs at a given hop, as described in
Section 5.
Note that the values of SI are from the set {255, ..., 1} and are
monotonically decreasing within the SFP. SIs MUST appear in order
within the SFPR (i.e., monotonically decreasing) and MUST NOT appear
more than once. Gaps MAY appear in the sequence, as described in
Section 4.5.1. Malformed SFPRs MUST be discarded and MUST cause any
previous instance of the SFPR (same SFPR-RD and SPI) to be discarded.
Note that if the SFIR-RD list in an SFT TLV contains one or more SFIR
Pool Identifiers, then in the above expression, "p" is the sum of the
number of individual SFIR-RD values and the sum for each SFIR Pool
Identifier of the number of SFIRs advertised with that SFIR Pool
Identifier. In other words, the list of SFIR-RD values is
effectively expanded to include the SFIR-RD of each SFIR advertised
with each SFIR Pool Identifier in the SFIR-RD list.
The choice of SFI is explained further in Section 5. Note that an
SFIR-RD value of zero has special meaning, as described in that
section.
4.4. Classifier Operation
As shown in Figure 1, the classifier is a component that is used to
assign packets to an SFP.
The classifier is responsible for determining to which packet flow a
packet belongs. The mechanism it uses to achieve that classification
is out of the scope of this document but might include inspection of
the packet header. The classifier has been instructed (by the
controller or through some other configuration mechanism -- see
Section 7.4) which flows are to be assigned to which SFPs, and so it
can impose an NSH on each packet and initialize the NSH with the SPI
of the selected SFP and the SI of its first hop.
Note that instructions delivered to the classifier may include
information about the metadata to encode (and the format for that
encoding) on packets that are classified by the classifier to a
particular SFP. As mentioned in Section 2.2, this corresponds to the
fifth element of control plane functionality described in [RFC7665].
Such instructions fall outside the scope of this specification (but
see Section 7.4), as do instructions to other service function
chaining elements on how to interpret metadata (as described in the
sixth element of control plane functionality described in [RFC7665]).
4.5. Service Function Forwarder Operation
Each packet sent to an SFF is transmitted encapsulated in an NSH.
The NSH includes an SPI and SI: the SPI indicates the SFPR
advertisement that announced the SFP; the tuple SPI/SI indicates a
specific hop in a specific path and maps to the RD/SFT of a
particular SFIR advertisement.
When an SFF gets an SFPR advertisement, it will first determine
whether to import the route by examining the RT. If the SFPR is
imported, the SFF then determines whether it is on the SFP by looking
for its own SFIR-RDs or any SFIR-RD with value zero in the SFPR. For
each occurrence in the SFP, the SFF creates forwarding state for
incoming packets and forwarding state for outgoing packets that have
been processed by the specified SFI.
The SFF creates local forwarding state for packets that it receives
from other SFFs. This state makes the association between the SPI/SI
in the NSH of the received packet and one or more specific local
SFIs, as identified by the SFIR-RD/SFT. If there are multiple local
SFIs that match, this is because a single advertisement was made for
a set of equivalent SFIs, and the SFF may use local policy (such as
load balancing) to determine to which SFI to forward a received
packet.
The SFF also creates next-hop forwarding state for packets received
back from the local SFI that need to be forwarded to the next hop in
the SFP. There may be a choice of next hops, as described in
Section 4.3. The SFF could install forwarding state for all
potential next hops or it could choose to only install forwarding
state for a subset of the potential next hops. If a choice is made,
then it will be as described in Section 5.
The installed forwarding state may change over time, reacting to
changes in the underlay network and the availability of particular
SFIs. Note that the forwarding state describes how one SFF sends
packets to another SFF, but not how those packets are routed through
the underlay network. SFFs may be connected by tunnels across the
underlay, or packets may be sent addressed to the next SFF and routed
through the underlay. In any case, transmission across the underlay
requires encapsulation of packets with a header for transport in the
underlay network.
Note that SFFs only create and store forwarding state for the SFPs on
which they are included. They do not retain state for all SFPs
advertised.
An SFF may also install forwarding state to support looping, jumping,
and branching. The protocol mechanism for explicit control of
looping, jumping, and branching uses a specific reserved SFT value at
a given hop of an SFPR and is described in Section 6.1.
4.5.1. Processing with "Gaps" in the SI Sequence
The behavior of an SF, as described in [RFC8300], is to decrement the
value of the "SI" field in the NSH by one before returning a packet
to the local SFF for further processing. This means that there is a
good reason to assume that the SFP is composed of a series of SFs,
each indicated by an SI value one less than the previous.
However, there is an advantage to having nonsuccessive SIs in an SPI.
Consider the case where an SPI needs to be modified by the insertion
or removal of an SF. In the latter case, this would lead to a "gap"
in the sequence of SIs, and in the former case, this could only be
achieved if a gap already existed into which the new SF with its new
SI value could be inserted. Otherwise, all "downstream" SFs would
need to be renumbered.
Now, of course, such renumbering could be performed, but it would
lead to a significant disruption to the SFC as all the SFFs along the
SFP were "reprogrammed". Thus, to achieve dynamic modification of an
SFP (and even in-service modification), it is desirable to be able to
make these modifications without changing the SIs of the elements
that were present before the modification. This will produce much
more consistent/predictable behavior during the convergence period,
where otherwise the change would need to be fully propagated.
Another approach says that any change to an SFP simply creates a new
SFP that can be assigned a new SPI. All that would be needed would
be to give a new instruction to the classifier, and traffic would be
switched to the new SFP that contains the new set of SFs. This
approach is practical but neglects to consider that the SFP may be
referenced by other SFPs (through "branch" instructions) and used by
many classifiers. In those cases, the corresponding configuration
resulting from a change in SPI may have wide ripples and create scope
for errors that are hard to trace.
Therefore, while this document requires that the SI values in an SFP
are monotonically decreasing, it makes no assumption that the SI
values are sequential. Configuration tools may apply that rule, but
they are not required to. To support this, an SFF SHOULD process as
follows when it receives a packet:
* If the SI indicates a known entry in the SFP, the SFF MUST process
the packet as normal, looking up the SI and determining to which
local SFI to deliver the packet.
* If the SI does not match an entry in the SFP, the SFF MUST reduce
the SI value to the next (smaller) value present in the SFP and
process the packet using that SI.
* If there is no smaller SI (i.e., if the end of the SFP has been
reached), the SFF MUST treat the SI value as not valid, as
described in [RFC8300].
This makes the behavior described in this document a superset of the
function in [RFC8300]. That is, an implementation that strictly
follows RFC 8300 in performing SI decrements in units of one is
perfectly in line with the mechanisms defined in this document.
SFF implementations MAY choose to only support contiguous SI values
in an SFP. Such an implementation will not support receiving an SI
value that is not present in the SFP and will discard the packets as
described in [RFC8300].
5. Selection within Service Function Paths
As described in Section 2, the SPI/SI in the NSH passed back from an
SFI to the SFF may leave the SFF with a choice of next-hop SFTs and a
choice of SFIs for each SFT. That is, the SPI indicates an SFPR, and
the SI indicates an entry in that SFPR. Each entry in an SFPR is a
set of one or more SFT/SFIR-RD pairs. The SFF MUST choose one of
these, identify the SFF that supports the chosen SFI, and send the
packet to that next-hop SFF.
The choice be may offered for load balancing across multiple SFIs, or
for discrimination between different actions necessary at a specific
hop in the SFP. Different SFT values may exist at a given hop in an
SFP to support several cases:
* There may be multiple instances of similar service functions that
are distinguished by different SFT values. For example, firewalls
made by vendor A and vendor B may need to be identified by
different SFT values because, while they have similar
functionality, their behavior is not identical. Then, some SFPs
may limit the choice of SF at a given hop by specifying the SFT
for vendor A, but other SFPs might not need to control which
vendor's SF is used and so can indicate that either SFT can be
used.
* There may be an obvious branch needed in an SFP, such as the
processing after a firewall where admitted packets continue along
the SFP, but suspect packets are diverted to a "penalty box". In
this case, the next hop in the SFP will be indicated with two
different SFT values.
In the typical case, the SFF chooses a next-hop SFF by looking at the
set of all SFFs that support the SFs identified by the SI (that set
having been advertised in individual SFIR advertisements), finding
the one or more that are "nearest" in the underlay network, and
choosing between next-hop SFFs using its own load-balancing
algorithm.
An SFI may influence this choice process by passing additional
information back, along with the packet and NSH. This information
may influence local policy at the SFF to either cause it to favor a
next-hop SFF (perhaps selecting one that is not nearest in the
underlay) or influence the load-balancing algorithm.
This selection applies to the normal case but also applies in the
case of looping, jumping, and branching (see Section 6).
Suppose an SFF in a particular service function overlay network
(identified by a particular import RT, RT-z) needs to forward an NSH-
encapsulated packet whose SPI is SPI-x and whose SI is SI-y. It does
the following:
1. It looks for an installed SFPR that carries RT-z and has SPI-x in
its NLRI. If there is none, then such packets cannot be
forwarded.
2. From the SFP attribute of that SFPR, it finds the Hop TLV with SI
value set to SI-y. If there is no such Hop TLV, then such
packets cannot be forwarded.
3. It then finds the "relevant" set of SFIRs by going through the
list of SFT TLVs contained in the Hop TLV as follows:
A. An SFIR is relevant if it carries RT-z, the SFT in its NLRI
matches the SFT value in one of the SFT TLVs, and the RD
value in its NLRI matches an entry in the list of SFIR-RDs in
that SFT TLV.
B. If an entry in the SFIR-RD list of an SFT TLV contains the
value zero, then an SFIR is relevant if it carries RT-z and
the SFT in its NLRI matches the SFT value in that SFT TLV.
That is, any SFIR in the service function overlay network
defined by RT-z and with the correct SFT is relevant.
C. If a pool identifier is in use, then an SFIR is relevant if
it is a member of the pool.
Each of the relevant SFIRs identifies a single SFI and contains a
tunnel encapsulation attribute that specifies how to send a packet to
that SFI. For a particular packet, the SFF chooses a particular SFI
from the set of relevant SFIRs. This choice is made according to
local policy.
A typical policy might be to figure out the set of SFIs that are
closest and load balance among them. But this is not the only
possible policy.
Thus, at any point in time when an SFF selects its next hop, it
chooses from the intersection of the set of next-hop RDs contained in
the SFPR and the RDs contained in the SFF's local set of SFIRs (i.e.,
according to the determination of "relevance", above). If the
intersection is null, the SFPR is unusable. Similarly, when this
condition applies on the controller that originated the SFPR, it
SHOULD either withdraw the SFPR or re-advertise it with a new set of
RDs for the affected hop.
6. Looping, Jumping, and Branching
As described in Section 2, an SFI or an SFF may cause a packet to
"loop back" to a previous SF on a path in order that a sequence of
functions may be re-executed. This is simply achieved by replacing
the SI in the NSH with a higher value, instead of decreasing it as
would normally be the case, to determine the next hop in the path.
Section 2 also describes how an SFI or SFF may cause a packet to
"jump forward" to an SF on a path that is not the immediate next SF
in the SFP. This is simply achieved by replacing the SI in the NSH
with a lower value than would be achieved by decreasing it by the
normal amount.
A more complex option to move packets from one SFP to another is
described in [RFC8300] and Section 2, where it is termed "branching".
This mechanism allows an SFI or SFF to make a choice of downstream
treatments for packets based on local policy and the output of the
local SF. Branching is achieved by changing the SPI in the NSH to
indicate the new path and setting the SI to indicate the point in the
path at which the packets enter.
Note that the NSH does not include a marker to indicate whether a
specific packet has been around a loop before. Therefore, the use of
NSH metadata [RFC8300] may be required in order to prevent infinite
loops.
6.1. Protocol Control of Looping, Jumping, and Branching
If the SFT value in an SFT TLV in an SFPR has the special-purpose SFT
value "Change Sequence" (see Section 10), then this is an indication
that the SFF may make a loop, jump, or branch according to local
policy and information returned by the local SFI.
In this case, the SPI and SI of the next hop are encoded in the eight
bytes of an entry in the SFIR-RD list as follows:
3 bytes SPI
1 byte SI
4 bytes Reserved (SHOULD be set to zero and ignored)
If the SI in this encoding is not part of the SFPR indicated by the
SPI in this encoding, then this is an explicit error that SHOULD be
detected by the SFF when it parses the SFPR. The SFPR SHOULD NOT
cause any forwarding state to be installed in the SFF, and packets
received with the SPI that indicates this SFPR SHOULD be silently
discarded.
If the SPI in this encoding is unknown, the SFF SHOULD NOT install
any forwarding state for this SFPR but MAY hold the SFPR pending
receipt of another SFPR that does use the encoded SPI.
If the SPI matches the current SPI for the path, this is a loop or
jump. In this case, if the SI is greater than or equal to the
current SI, it is a loop. If the SPI matches and the SI is less than
the next SI, it is a jump.
If the SPI indicates another path, this is a branch, and the SI
indicates the point at which to enter that path.
The Change Sequence SFT is just another SFT that may appear in a set
of SFI/SFT tuples within an SI and is selected as described in
Section 5.
Note that special-purpose SFTs MUST NOT be advertised in SFIRs. If
such an SFIR is received, it SHOULD be ignored.
6.2. Implications for Forwarding State
Support for looping and jumping requires that the SFF has forwarding
state established to an SFF that provides access to an instance of
the appropriate SF. This means that the SFF must have seen the
relevant SFIR advertisements and mush have known that it needed to
create the forwarding state. This is a matter of local configuration
and implementation; for example, an implementation could be
configured to install forwarding state for specific looping/jumping.
Support for branching requires that the SFF has forwarding state
established to an SFF that provides access to an instance of the
appropriate entry SF on the other SFP. This means that the SFF must
have seen the relevant SFIR and SFPR advertisements and known that it
needed to create the forwarding state. This is a matter of local
configuration and implementation; for example, an implementation
could be configured to install forwarding state for specific
branching (identified by SPI and SI).
7. Advanced Topics
This section highlights several advanced topics introduced elsewhere
in this document.
7.1. Correlating Service Function Path Instances
It is often useful to create bidirectional SFPs to enable packet
flows to traverse the same set of SFs, but in the reverse order.
However, packets on SFPs in the data plane (per [RFC8300]) do not
contain a direction indicator, so each direction must use a different
SPI.
As described in Section 3.2.1.1, an SFPR can contain one or more
correlators encoded in Association TLVs. If the Association Type
indicates "Bidirectional SFP", then the SFP advertised in the SFPR is
one direction of a bidirectional pair of SFPs, where the other in the
pair is advertised in the SFPR with RD as carried in the "Associated
SFPR-RD" field of the Association TLV. The SPI carried in the
"Associated SPI" field of the Association TLV provides a cross-check
against the SPI advertised in the SFPR with RD as carried in the
"Associated SFPR-RD" field of the Association TLV.
As noted in Section 3.2.1.1, when SFPRs reference each other, one
SFPR advertisement will be received before the other. Therefore,
processing of an association will require that the first SFPR not be
rejected simply because the Associated SFPR-RD it carries is unknown.
However, the SFP defined by the first SFPR is valid and SHOULD be
available for use as a unidirectional SFP, even in the absence of an
advertisement of its partner.
Furthermore, in error cases where SFPR-a associates with SFPR-b, but
SFPR-b associates with SFPR-c such that a bidirectional pair of SFPs
cannot be formed, the individual SFPs are still valid and SHOULD be
available for use as unidirectional SFPs. An implementation SHOULD
log this situation, because it represents a controller error.
Usage of a bidirectional SFP may be programmed into the classifiers
by the controller. Alternatively, a classifier may look at incoming
packets on a bidirectional packet flow, extract the SPI from the
received NSH, and look up the SFPR to find the reverse-direction SFP
to use when it sends packets.
See Section 8 for an example of how this works.
7.2. Considerations for Stateful Service Functions
Some service functions are stateful. That means that they build and
maintain state derived from configuration or the packet flows that
they handle. In such cases, it can be important or necessary that
all packets from a flow continue to traverse the same instance of a
service function so that the state can be leveraged and does not need
to be regenerated.
In the case of bidirectional SFPs, it may be necessary to traverse
the same instances of a stateful service function in both directions.
A firewall is a good example of such a service function.
This issue becomes a concern where there are multiple parallel
instances of a service function and a determination of which one to
use could normally be left to the SFF as a load-balancing or local-
policy choice.
For the forward-direction SFP, the concern is that the same choice of
SF is made for all packets of a flow under normal network conditions.
It may be possible to guarantee that the load-balancing functions
applied in the SFFs are stable and repeatable, but a controller that
constructs SFPs might not want to trust to this. The controller can,
in these cases, build a number of more specific SFPs, each traversing
a specific instance of the stateful SFs. In this case, the load-
balancing choice can be left up to the classifier. Thus, the
classifier selects which instance of a stateful SF is used by a
particular flow by selecting the SFP that the flow uses.
For bidirectional SFPs where the same instance of a stateful SF must
be traversed in both directions, it is not enough to leave the choice
of SFI as a local choice, even if the load balancing is stable,
because coordination would be required between the decision points in
the forward and reverse directions, and this may be hard to achieve
in all cases except where it is the same SFF that makes the choice in
both directions.
Note that this approach necessarily increases the amount of SFP state
in the network (i.e., there are more SFPs). It is possible to
mitigate this effect by careful construction of SFPs built from a
concatenation of other SFPs.
Section 8.9 includes some simple examples of SFPs for stateful SFs.
7.3. VPN Considerations and Private Service Functions
Likely deployments include reserving specific instances of SFs for
specific customers or allowing customers to deploy their own SFs
within the network. Building SFs in such environments requires that
suitable identifiers be used to ensure that SFFs distinguish which
SFIs can be used and which cannot.
This problem is similar to a problem in the way that VPNs are
supported and is solved in a similar way. The "RT" field is used to
indicate a set of SFs from which all choices must be made.
7.4. Flow Specification for SFC Classifiers
[RFC8955] defines a set of BGP routes that can be used to identify
the packets in a given flow using fields in the header of each
packet, and a set of actions -- encoded as Extended Communities --
that can be used to disposition those packets. This document enables
the use of these mechanisms by SFC classifiers by defining a new
action Extended Community called "Flow Specification for SFC
Classifiers", identified by the value 0x0d. Note that implementation
of this section of this specification will be controllers or
classifiers communicating with each other directly for the purpose of
instructing the classifier how to place packets onto an SFP. So that
the implementation of classifiers can be kept simple, and to avoid
the confusion between the purposes of different Extended Communities,
a controller MUST NOT include other action Extended Communities at
the same time as a "Flow Specification for SFC Classifiers" Extended
Community. A "Flow Specification for SFC Classifiers" Traffic
Filtering Action Extended Community advertised with any other Traffic
Filtering Action Extended Community MUST be treated as malformed in
line with [RFC8955] and result in the flow-specification UPDATE
message being handled as "treat-as-withdraw", according to [RFC7606],
Section 2.
To put the flow specification into context, when multiple service
function chaining overlays are present in one network, each FlowSpec
update MUST be tagged with the route target of the overlay or VPN
network for which it is intended.
This Extended Community is encoded as an 8-octet value, as shown in
Figure 10.
1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Type=0x80 | Sub-Type=0x0d | SPI |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| SPI (cont.) | SI | SFT |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 10: The Format of the Flow Specification for SFC
Classifiers Extended Community
The Extended Community contains the Service Path Identifier (SPI),
Service Index (SI), and Service Function Type (SFT), as defined
elsewhere in this document. Thus, each action extended community
defines the entry point (not necessarily the first hop) into a
specific SFP. This allows, for example, different flows to enter the
same SFP at different points.
Note that, according to [RFC8955], a given flow-specification update
may include multiple of these action Extended Communities. If a
given action extended community does not contain an installed SFPR
with the specified {SPI, SI, SFT}, it MUST NOT be used for
dispositioning the packets of the specified flow.
The normal case of packet classification for service function
chaining will see a packet enter the SFP at its first hop. In this
case, the SI in the Extended Community is superfluous, and the SFT
may also be unnecessary. To allow these cases to be handled, a
special meaning is assigned to an SI of zero (not a valid value) and
an SFT of zero (a reserved value in the registry -- see
Section 10.5).
* If an SFC Classifiers Extended Community is received with SI = 0,
then it means that the first hop of the SFP indicated by the SPI
MUST be used.
* If an SFC Classifiers Extended Community is received with SFT = 0,
then there are two subcases:
- If there is a choice of SFT in the hop indicated by the value
of the SI (including SI = 0), then SFT = 0 means there is a
free choice of which SFT to use, according to local policy).
- If there is no choice of SFT in the hop indicated by the value
of SI, then SFT = 0 means that the value of the SFT at that
hop, as indicated in the SFPR for the indicated SPI, MUST be
used.
One of the filters that the flow specification may describe is the
VPN to which the traffic belongs. Additionally, as noted above, to
put the indicated SPI into context when multiple SFC overlays are
present in one network, each FlowSpec update MUST be tagged with the
route target of the overlay or VPN network for which it is intended.
Note that future extensions might be made to the Flow Specification
for SFC Classifiers Extended Community to provide instruction to the
classifier about what metadata to add to packets that it classifies
for forwarding on a specific SFP; however, that is outside the scope
of this document.
7.5. Choice of Data Plane SPI/SI Representation
This document ties together the control and data planes of a service
function chaining overlay network through the use of the SPI/SI that
is nominally carried in the NSH of a given packet. However, in order
to handle situations in which the NSH is not ubiquitously deployed,
it is also possible to use alternative data plane representations of
the SPI/SI by carrying the identical semantics in other protocol
fields, such as MPLS labels [RFC8595].
This document defines a new Sub-TLV for the tunnel encapsulation
attribute [RFC9012], the SPI/SI Representation Sub-TLV of type 16.
This Sub-TLV MAY be present in each Tunnel TLV contained in a tunnel
encapsulation attribute when the attribute is carried by an SFIR.
The "Value" field of this Sub-TLV is a two-octet field of flags
numbered counting from the most significant bit, each of which
describes how the originating SFF expects to see the SPI/SI
represented in the data plane for packets carried in the tunnels
described by the Tunnel TLV.
The following bits are defined by this document and are tracked in an
IANA registry described in Section 10.10:
Bit 0: If this bit is set, the NSH is to be used to carry the SPI/SI
in the data plane.
Bit 1: If this bit is set, two labels in an MPLS label stack are to
be used as described in Section 7.5.1.
If a given Tunnel TLV does not contain an SPI/SI Representation Sub-
TLV, then it MUST be processed as if such a Sub-TLV is present with
Bit 0 set and no other bits set. That is, the absence of the Sub-TLV
SHALL be interpreted to mean that the NSH is to be used.
If a given Tunnel TLV contains an SPI/SI Representation Sub-TLV with
a "Value" field that has no flag set, then the tunnel indicated by
the Tunnel TLV MUST NOT be used for forwarding SFC packets. If a
given Tunnel TLV contains an SPI/SI Representation Sub-TLV with both
bit 0 and bit 1 set, then the tunnel indicated by the Tunnel TLV MUST
NOT be used for forwarding SFC packets. The meaning and rules for
the presence of other bits is to be defined in future documents, but
implementations of this specification MUST set other bits to zero and
ignore them on receipt.
If a given Tunnel TLV contains more than one SPI/SI Representation
Sub-TLV, then the first one MUST be considered and subsequent
instances MUST be ignored.
Note that the MPLS representation of the logical NSH may be used even
if the tunnel is not an MPLS tunnel. Conversely, MPLS tunnels may be
used to carry other encodings of the logical NSH (specifically, the
NSH itself). It is a requirement that both ends of a tunnel over the
underlay network know that the tunnel is used for service function
chaining and know what form of NSH representation is used. The
signaling mechanism described here allows coordination of this
information.
7.5.1. MPLS Representation of the SPI/SI
If bit 1 is set in the SPI/SI Representation Sub-TLV, then labels in
the MPLS label stack are used to indicate SFC forwarding and
processing instructions to achieve the semantics of a logical NSH.
The label stack is encoded as shown in [RFC8595].
7.6. MPLS Label Swapping/Stacking Operation
When a classifier constructs an MPLS label stack for an SFP, it
starts with that SFP's last hop. If the last hop requires an {SPI,
SI} label pair for label swapping, it pushes the SI (set to the SI
value of the last hop) and the SFP's SPI onto the MPLS label stack.
If the last hop requires a {context label, SFI label} label pair for
label stacking, it selects a specific SFIR and pushes that SFIR's SFI
label and context label onto the MPLS label stack.
The classifier then moves sequentially back through the SFP one hop
at a time. For each hop, if the hop requires an {SPI, SI} and there
is an {SPI, SI} at the top of the MPLS label stack, the SI is set to
the SI value of the current hop. If there is not an {SPI, SI} at the
top of the MPLS label stack, it pushes the SI (set to the SI value of
the current hop) and the SFP's SPI onto the MPLS label stack.
If the hop requires a {context label, SFI label}, it selects a
specific SFIR and pushes that SFIR's SFI label and context label onto
the MPLS label stack.
7.7. Support for MPLS-Encapsulated NSH Packets
[RFC8596] describes how to transport SFC packets using the NSH over
an MPLS transport network. Signaling that this approach is in use is
supported by this document as follows:
* A "BGP Tunnel Encapsulation Attribute" Sub-TLV is included with
the codepoint 10 (representing "MPLS Label Stack") from the "BGP
Tunnel Encapsulation Attribute Sub-TLVs" registry defined in
[RFC9012].
* An "SFP Traversal With MPLS Label Stack" TLV is included
containing an "SPI/SI Representation" Sub-TLV with bit 0 set and
bit 1 cleared.
In this case, the MPLS label stack constructed by the SFF to forward
a packet to the next SFF on the SFP will consist of the labels needed
to reach that SFF, and if label stacking is used, it will also
include the labels advertised in the MPLS Label Stack Sub-TLV and the
labels remaining in the stack needed to traverse the remainder of the
SFP.
8. Examples
Most of the examples in this section use IPv4 addressing. But there
is nothing special about IPv4 in the mechanisms described in this
document, and they are equally applicable to IPv6. A few examples
using IPv6 addressing are provided in Section 8.10.
Assume we have a service function overlay network with four SFFs
(SFF1, SFF2, SFF3, and SFF4). The SFFs have addresses in the
underlay network as follows:
SFF1 192.0.2.1
SFF2 192.0.2.2
SFF3 192.0.2.3
SFF4 192.0.2.4
Each SFF provides access to some SFIs from the four SFTs, SFT=41,
SFT=42, SFT=43, and SFT=44, as follows:
SFF1 SFT=41 and SFT=42
SFF2 SFT=41 and SFT=43
SFF3 SFT=42 and SFT=44
SFF4 SFT=43 and SFT=44
The service function network also contains a controller with address
198.51.100.1.
This example service function overlay network is shown in Figure 11.
--------------
| Controller |
| 198.51.100.1 | ------ ------ ------ ------
-------------- | SFI | | SFI | | SFI | | SFI |
|SFT=41| |SFT=42| |SFT=41| |SFT=43|
------ ------ ------ ------
\ / \ /
--------- ---------
---------- | SFF1 | | SFF2 |
Packet --> | | |192.0.2.1| |192.0.2.2|
Flows --> |Classifier| --------- --------- -->Dest
| | -->
---------- --------- ---------
| SFF3 | | SFF4 |
|192.0.2.3| |192.0.2.4|
--------- ---------
/ \ / \
------ ------ ------ ------
| SFI | | SFI | | SFI | | SFI |
|SFT=42| |SFT=44| |SFT=43| |SFT=44|
------ ------ ------ ------
Figure 11: Example Service Function Overlay Network
The SFFs advertise routes to the SFIs they support. These
advertisements contain RDs that are set according to the network
operator's configuration model. In all of these IPv4 examples, we
use RDs of Type 1 such that the available six octets are partitioned
as four octets for the IPv4 address of the advertising SFF, and two
octets that are a local index of the SFI. This scheme is chosen
purely for convenience of documentation, and an operator is totally
free to use any other scheme so long as it conforms to the
definitions of SFIR and SFPR in Sections 3.1 and 3.2.
Thus, we see the following SFIRs advertised:
RD = 192.0.2.1/1, SFT = 41
RD = 192.0.2.1/2, SFT = 42
RD = 192.0.2.2/1, SFT = 41
RD = 192.0.2.2/2, SFT = 43
RD = 192.0.2.3/7, SFT = 42
RD = 192.0.2.3/8, SFT = 44
RD = 192.0.2.4/5, SFT = 43
RD = 192.0.2.4/6, SFT = 44
Note that the addressing used for communicating between SFFs is taken
from the tunnel encapsulation attribute of the SFIR and not from the
SFIR-RD.
8.1. Example Explicit SFP with No Choices
Consider the following SFPR.
SFP1: RD = 198.51.100.1/101, SPI = 15,
[SI = 255, SFT = 41, RD = 192.0.2.1/1],
[SI = 250, SFT = 43, RD = 192.0.2.2/2]
The SFP consists of an SF of Type 41 located at SFF1, followed by an
SF of Type 43 located at SFF2. This path is fully explicit, and each
SFF is offered no choice in forwarding packets along the path.
SFF1 will receive packets on the path from the classifier and will
identify the path from the SPI (15). The initial SI will be 255, and
so SFF1 will deliver the packets to the SFI for SFT 41.
When the packets are returned to SFF1 by the SFI, the SI will be
decreased to 250 for the next hop. SFF1 has no flexibility in the
choice of SFF to support the next-hop SFI and will forward the packet
to SFF2, which will send the packets to the SFI that supports SFT 43
before forwarding the packets to their destinations.
8.2. Example SFP with Choice of SFIs
SFP2: RD = 198.51.100.1/102, SPI = 16,
[SI = 255, SFT = 41, RD = 192.0.2.1/1],
[SI = 250, SFT = 43, {RD = 192.0.2.2/2,
RD = 192.0.2.4/5 } ]
In this example, the path also consists of an SF of Type 41 located
at SFF1, and this is followed by an SF of Type 43. However, in this
case, the SI = 250 contains a choice between the SFI located at SFF2
and the SFI located at SFF4.
SFF1 will receive packets on the path from the classifier and will
identify the path from the SPI (16). The initial SI will be 255, and
so SFF1 will deliver the packets to the SFI for SFT 41.
When the packets are returned to SFF1 by the SFI, the SI will be
decreased to 250 for the next hop. SFF1 now has a choice of next-hop
SFFs to execute the next hop in the path. It can either forward
packets to SFF2 or SFF4 to execute a function of Type 43. It uses
its local load-balancing algorithm to make this choice. The chosen
SFF will send the packets to the SFI that supports SFT 43 before
forwarding the packets to their destinations.
8.3. Example SFP with Open Choice of SFIs
SFP3: RD = 198.51.100.1/103, SPI = 17,
[SI = 255, SFT = 41, RD = 192.0.2.1/1],
[SI = 250, SFT = 44, RD = 0]
In this example, the path also consists of an SF of Type 41 located
at SFF1, and this is followed by an SI with an RD of zero and SF of
Type 44. This means that a choice can be made between any SFF that
supports an SFI of Type 44.
SFF1 will receive packets on the path from the classifier and will
identify the path from the SPI (17). The initial SI will be 255, and
so SFF1 will deliver the packets to the SFI for SFT 41.
When the packets are returned to SFF1 by the SFI, the SI will be
decreased to 250 for the next hop. SFF1 now has a free choice of
next-hop SFFs to execute the next hop in the path, selecting between
all SFFs that support SFs of Type 44. Looking at the SFIRs it has
received, SFF1 knows that SF Type 44 is supported by SFF3 and SFF4.
SFF1 uses its local load-balancing algorithm to make this choice.
The chosen SFF will send the packets to the SFI that supports SFT 44
before forwarding the packets to their destinations.
8.4. Example SFP with Choice of SFTs
SFP4: RD = 198.51.100.1/104, SPI = 18,
[SI = 255, SFT = 41, RD = 192.0.2.1/1],
[SI = 250, {SFT = 43, RD = 192.0.2.2/2,
SFT = 44, RD = 192.0.2.3/8 } ]
This example provides a choice of SF type in the second hop in the
path. The SI of 250 indicates a choice between SF Type 43 located at
SF2 and SF Type 44 located at SF3.
SFF1 will receive packets on the path from the classifier and will
identify the path from the SPI (18). The initial SI will be 255, and
so SFF1 will deliver the packets to the SFI for SFT 41.
When the packets are returned to SFF1 by the SFI, the SI will be
decreased to 250 for the next hop. SFF1 now has a free choice of
next-hop SFFs to execute the next hop in the path, selecting between
all SFFs that support an SF of Type 43 and SFF3, which supports an SF
of Type 44. These may be completely different functions that are to
be executed dependent on specific conditions, or they may be similar
functions identified with different type identifiers (such as
firewalls from different vendors). SFF1 uses its local policy and
load-balancing algorithm to make this choice and may use additional
information passed back from the local SFI to help inform its
selection. The chosen SFF will send the packets to the SFI that
supports the chosen SFT before forwarding the packets to their
destinations.
8.5. Example Correlated Bidirectional SFPs
SFP5: RD = 198.51.100.1/105, SPI = 19,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/106, Assoc-SPI = 20,
[SI = 255, SFT = 41, RD = 192.0.2.1/1],
[SI = 250, SFT = 43, RD = 192.0.2.2/2]
SFP6: RD = 198.51.100.1/106, SPI = 20,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/105, Assoc-SPI = 19,
[SI = 254, SFT = 43, RD = 192.0.2.2/2],
[SI = 249, SFT = 41, RD = 192.0.2.1/1]
This example demonstrates correlation of two SFPs to form a
bidirectional SFP, as described in Section 7.1.
Two SFPRs are advertised by the controller. They have different SPIs
(19 and 20), so they are known to be separate SFPs, but they both
have Association TLVs with Association Type set to 1, indicating
bidirectional SFPs. Each has an "Associated SFPR-RD" field
containing the value of the other SFPR-RD to correlate the two SFPs
as a bidirectional pair.
As can be seen from the SFPRs in this example, the paths are
symmetric: the hops in SFP5 appear in the reverse order in SFP6.
8.6. Example Correlated Asymmetrical Bidirectional SFPs
SFP7: RD = 198.51.100.1/107, SPI = 21,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/108, Assoc-SPI = 22,
[SI = 255, SFT = 41, RD = 192.0.2.1/1],
[SI = 250, SFT = 43, RD = 192.0.2.2/2]
SFP8: RD = 198.51.100.1/108, SPI = 22,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/107, Assoc-SPI = 21,
[SI = 254, SFT = 44, RD = 192.0.2.4/6],
[SI = 249, SFT = 41, RD = 192.0.2.1/1]
Asymmetric bidirectional SFPs can also be created. This example
shows a pair of SFPs with distinct SPIs (21 and 22) that are
correlated in the same way as in the example in Section 8.5.
However, unlike in that example, the SFPs are different in each
direction. Both paths include a hop of SF Type 41, but SFP7 includes
a hop of SF Type 43 supported at SFF2, while SFP8 includes a hop of
SF Type 44 supported at SFF4.
8.7. Example Looping in an SFP
SFP9: RD = 198.51.100.1/109, SPI = 23,
[SI = 255, SFT = 41, RD = 192.0.2.1/1],
[SI = 250, SFT = 44, RD = 192.0.2.4/5],
[SI = 245, {SFT = 1, RD = {SPI=23, SI=255, Rsv=0},
SFT = 42, RD = 192.0.2.3/7 } ]
Looping and jumping are described in Section 6. This example shows
an SFP that contains an explicit loop-back instruction that is
presented as a choice within an SFP hop.
The first two hops in the path (SI = 255 and SI = 250) are normal.
That is, the packets will be delivered to SFF1 and SFF4 in turn for
execution of SFs of Type 41 and 44, respectively.
The third hop (SI = 245) presents SFF4 with a choice of next hop. It
can either forward the packets to SFF3 for an SF of Type 42 (the
second choice) or it can loop back.
The loop-back entry in the SFPR for SI = 245 is indicated by the
special-purpose SFT value 1 ("Change Sequence"). Within this hop,
the RD is interpreted as encoding the SPI and SI of the next hop (see
Section 6.1). In this case, the SPI is 23, which indicates that this
is a loop or branch, i.e., the next hop is on the same SFP. The SI
is set to 255; this is a higher number than the current SI (245),
indicating a loop.
SFF4 must make a choice between these two next hops. The packet will
be either forwarded to SFF3 with the NSH SI decreased to 245 or
looped back to SFF1 with the NSH SI reset to 255. This choice will
be made according to local policy, information passed back by the
local SFI, and details in the packets' metadata that are used to
prevent infinite looping.
8.8. Example Branching in an SFP
SFP10: RD = 198.51.100.1/110, SPI = 24,
[SI = 254, SFT = 42, RD = 192.0.2.3/7],
[SI = 249, SFT = 43, RD = 192.0.2.2/2]
SFP11: RD = 198.51.100.1/111, SPI = 25,
[SI = 255, SFT = 41, RD = 192.0.2.1/1],
[SI = 250, SFT = 1, RD = {SPI=24, SI=254, Rsv=0}]
Branching follows a similar procedure to that for looping (and
jumping), as shown in Section 8.7. However, there are two SFPs
involved.
SFP10 shows a normal path with packets forwarded to SFF3 and SFF2 for
execution of service functions of Type 42 and 43, respectively.
SFP11 starts as normal (SFF1 for an SF of Type 41), but then SFF1
processes the next hop in the path and finds a "Change Sequence"
special-purpose SFT. The "SFIR-RD" field includes an SPI of 24,
which indicates SFP10, not the current SFP. The SI in the SFIR-RD is
254, so SFF1 knows that it must set the SPI/SI in the NSH to 24/254
and send the packets to the appropriate SFF, as advertised in the
SFPR for SFP10 (that is, SFF3).
8.9. Examples of SFPs with Stateful Service Functions
This section provides some examples to demonstrate establishing SFPs
when there is a choice of service functions at a particular hop, and
where consistency of choice is required in both directions. The
scenarios that give rise to this requirement are discussed in
Section 7.2.
8.9.1. Forward and Reverse Choice Made at the SFF
Consider the topology shown in Figure 12. There are three SFFs
arranged neatly in a line, and the middle one (SFF2) supports three
SFIs all of SFT 42. These three instances can be used by SFF2 to
load balance so that no one instance is swamped.
------ ------ ------ ------ ------
| SFI | | SFIa | | SFIb | | SFIc | | SFI |
|SFT=41| |SFT=42| |SFT=42| |SFT=42| |SFT=43|
------ ------\ ------ /------ ------
\ \ | / /
--------- --------- ---------
---------- | SFF1 | | SFF2 | | SFF3 |
--> | |..|192.0.2.1|...|192.0.2.2|...|192.0.2.3|-->
--> |Classifier| --------- --------- ---------
| |
----------
Figure 12: Example Where Choice Is Made at the SFF
This leads to the following SFIRs being advertised.
RD = 192.0.2.1/11, SFT = 41
RD = 192.0.2.2/11, SFT = 42 (for SFIa)
RD = 192.0.2.2/12, SFT = 42 (for SFIb)
RD = 192.0.2.2/13, SFT = 42 (for SFIc)
RD = 192.0.2.3/11, SFT = 43
The controller can create a single forward SFP (SFP12), giving SFF2
the choice of which SFI to use to provide a function of SFT 42, as
follows. The load-balancing choice between the three available SFIs
is assumed to be within the capabilities of the SFF, and if the SFs
are stateful, it is assumed that the SFF knows this and arranges load
balancing in a stable, flow-dependent way.
SFP12: RD = 198.51.100.1/112, SPI = 26,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/113, Assoc-SPI = 27,
[SI = 255, SFT = 41, RD = 192.0.2.1/11],
[SI = 254, SFT = 42, {RD = 192.0.2.2/11,
192.0.2.2/12,
192.0.2.2/13 }],
[SI = 253, SFT = 43, RD = 192.0.2.3/11]
The reverse SFP (SFP13) in this case may also be created as shown
below, using association with the forward SFP and giving the load-
balancing choice to SFF2. This is safe, even in the case that the
SFs of Type 42 are stateful, because SFF2 is doing the load balancing
in both directions and can apply the same algorithm to ensure that
packets associated with the same flow use the same SFI regardless of
the direction of travel.
SFP13: RD = 198.51.100.1/113, SPI = 27,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/112, Assoc-SPI = 26,
[SI = 255, SFT = 43, RD = 192.0.2.3/11],
[SI = 254, SFT = 42, {RD = 192.0.2.2/11,
192.0.2.2/12,
192.0.2.2/13 }],
[SI = 253, SFT = 41, RD = 192.0.2.1/11]
How an SFF knows that an attached SFI is stateful is out of the scope
of this document. It is assumed that this will form part of the
process by which SFIs are registered as local to SFFs. Section 7.2
provides additional observations about the coordination of the use of
stateful SFIs in the case of bidirectional SFPs.
In general, the problems of load balancing and the selection of the
same SFIs in both directions of a bidirectional SFP can be addressed
by using sufficiently precisely specified SFPs (specifying the exact
SFIs to use) and suitable programming of the classifiers at each end
of the SFPs to make sure that the matching pair of SFPs are used.
8.9.2. Parallel End-to-End SFPs with Shared SFF
The mechanism described in Section 8.9.1 might not be desirable
because of the functional assumptions it places on SFF2 to be able to
load balance with suitable flow identification, stability, and
equality in both directions. Instead, it may be desirable to place
the responsibility for flow classification in the classifier and let
it determine load balancing with the implied choice of SFIs.
Consider the network graph as shown in Figure 12 and with the same
set of SFIRs as listed in Section 8.9.1. In this case, the
controller could specify three forward SFPs with their corresponding
associated reverse SFPs. Each bidirectional pair of SFPs uses a
different SFI for the SF of Type 42. The controller can instruct the
classifier how to place traffic on the three bidirectional SFPs, or
it can treat them as a group, leaving the classifier responsible for
balancing the load.
SFP14: RD = 198.51.100.1/114, SPI = 28,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/117, Assoc-SPI = 31,
[SI = 255, SFT = 41, RD = 192.0.2.1/11],
[SI = 254, SFT = 42, RD = 192.0.2.2/11],
[SI = 253, SFT = 43, RD = 192.0.2.3/11]
SFP15: RD = 198.51.100.1/115, SPI = 29,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/118, Assoc-SPI = 32,
[SI = 255, SFT = 41, RD = 192.0.2.1/11],
[SI = 254, SFT = 42, RD = 192.0.2.2/12],
[SI = 253, SFT = 43, RD = 192.0.2.3/11]
SFP16: RD = 198.51.100.1/116, SPI = 30,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/119, Assoc-SPI = 33,
[SI = 255, SFT = 41, RD = 192.0.2.1/11],
[SI = 254, SFT = 42, RD = 192.0.2.2/13],
[SI = 253, SFT = 43, RD = 192.0.2.3/11]
SFP17: RD = 198.51.100.1/117, SPI = 31,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/114, Assoc-SPI = 28,
[SI = 255, SFT = 43, RD = 192.0.2.3/11],
[SI = 254, SFT = 42, RD = 192.0.2.2/11],
[SI = 253, SFT = 41, RD = 192.0.2.1/11]
SFP18: RD = 198.51.100.1/118, SPI = 32,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/115, Assoc-SPI = 29,
[SI = 255, SFT = 43, RD = 192.0.2.3/11],
[SI = 254, SFT = 42, RD = 192.0.2.2/12],
[SI = 253, SFT = 41, RD = 192.0.2.1/11]
SFP19: RD = 198.51.100.1/119, SPI = 33,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/116, Assoc-SPI = 30,
[SI = 255, SFT = 43, RD = 192.0.2.3/11],
[SI = 254, SFT = 42, RD = 192.0.2.2/13],
[SI = 253, SFT = 41, RD = 192.0.2.1/11]
8.9.3. Parallel End-to-End SFPs with Separate SFFs
While the examples in Sections 8.9.1 and 8.9.2 place the choice of
SFI as subtended from the same SFF, it is also possible that the SFIs
are each subtended from a different SFF, as shown in Figure 13. In
this case, it is harder to coordinate the choices for forward and
reverse paths without some form of coordination between SFF1 and
SFF3. Therefore, it would be normal to consider end-to-end parallel
SFPs, as described in Section 8.9.2.
------
| SFIa |
|SFT=42|
------
------ |
| SFI | ---------
|SFT=41| | SFF5 |
------ ..|192.0.2.5|..
| ..: --------- :..
---------.: :.---------
---------- | SFF1 | --------- | SFF3 |
--> | |..|192.0.2.1|....| SFF6 |....|192.0.2.3| -->
--> |Classifier| ---------: |192.0.2.6| :---------
| | : --------- : |
---------- : | : ------
: ------ : | SFI |
:.. | SFIb | ..: |SFT=43|
:.. |SFT=42| ..: ------
: ------ :
:.---------.:
| SFF7 |
|192.0.2.7|
---------
|
------
| SFIc |
|SFT=42|
------
Figure 13: Second Example with Parallel End-to-End SFPs
In this case, five SFIRs are advertised as follows:
RD = 192.0.2.1/11, SFT = 41
RD = 192.0.2.5/11, SFT = 42 (for SFIa)
RD = 192.0.2.6/11, SFT = 42 (for SFIb)
RD = 192.0.2.7/11, SFT = 42 (for SFIc)
RD = 192.0.2.3/11, SFT = 43
In this case, the controller could specify three forward SFPs with
their corresponding associated reverse SFPs. Each bidirectional pair
of SFPs uses a different SFF and SFI for the middle hop (for an SF of
Type 42). The controller can instruct the classifier how to place
traffic on the three bidirectional SFPs, or it can treat them as a
group, leaving the classifier responsible for balancing the load.
SFP20: RD = 198.51.100.1/120, SPI = 34,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/123, Assoc-SPI = 37,
[SI = 255, SFT = 41, RD = 192.0.2.1/11],
[SI = 254, SFT = 42, RD = 192.0.2.5/11],
[SI = 253, SFT = 43, RD = 192.0.2.3/11]
SFP21: RD = 198.51.100.1/121, SPI = 35,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/124, Assoc-SPI = 38,
[SI = 255, SFT = 41, RD = 192.0.2.1/11],
[SI = 254, SFT = 42, RD = 192.0.2.6/11],
[SI = 253, SFT = 43, RD = 192.0.2.3/11]
SFP22: RD = 198.51.100.1/122, SPI = 36,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/125, Assoc-SPI = 39,
[SI = 255, SFT = 41, RD = 192.0.2.1/11],
[SI = 254, SFT = 42, RD = 192.0.2.7/11],
[SI = 253, SFT = 43, RD = 192.0.2.3/11]
SFP23: RD = 198.51.100.1/123, SPI = 37,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/120, Assoc-SPI = 34,
[SI = 255, SFT = 43, RD = 192.0.2.3/11],
[SI = 254, SFT = 42, RD = 192.0.2.5/11],
[SI = 253, SFT = 41, RD = 192.0.2.1/11]
SFP24: RD = 198.51.100.1/124, SPI = 38,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/121, Assoc-SPI = 35,
[SI = 255, SFT = 43, RD = 192.0.2.3/11],
[SI = 254, SFT = 42, RD = 192.0.2.6/11],
[SI = 253, SFT = 41, RD = 192.0.2.1/11]
SFP25: RD = 198.51.100.1/125, SPI = 39,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/122, Assoc-SPI = 36,
[SI = 255, SFT = 43, RD = 192.0.2.3/11],
[SI = 254, SFT = 42, RD = 192.0.2.7/11],
[SI = 253, SFT = 41, RD = 192.0.2.1/11]
8.9.4. Parallel SFPs Downstream of the Choice
The mechanism of parallel SFPs demonstrated in Section 8.9.3 is
perfectly functional and may be practical in many environments.
However, there may be scaling concerns because of the large amount of
state (knowledge of SFPs -- i.e., SFPR advertisements retained) if
there is a very large number of possible SFIs (for example, tens of
instances of the same stateful SF) or if there are multiple choices
of stateful SF along a path. This situation may be mitigated using
SFP fragments that are combined to form the end-to-end SFPs.
The example presented here is necessarily simplistic but should
convey the basic principle. The example presented in Figure 14 is
similar to that in Section 8.9.3 but with an additional first hop.
------
| SFIa |
|SFT=43|
------
------ ------ |
| SFI | | SFI | ---------
|SFT=41| |SFT=42| | SFF5 |
------ ------ ..|192.0.2.5|..
| | ..: --------- :..
--------- ---------.: :.---------
------ | SFF1 | | SFF2 | --------- | SFF3 |
-->|Class-|.|192.0.2.1|.|192.0.2.2|....| SFF6 |....|192.0.2.3|-->
-->| ifier| --------- ---------: |192.0.2.6| :---------
------ : --------- : |
: | : ------
: ------ : | SFI |
:.. | SFIb | ..: |SFT=44|
:.. |SFT=43| ..: ------
: ------ :
:.---------.:
| SFF7 |
|192.0.2.7|
---------
|
------
| SFIc |
|SFT=43|
------
Figure 14: Example with Parallel SFPs Downstream of Choice
The six SFIs are advertised as follows:
RD = 192.0.2.1/11, SFT = 41
RD = 192.0.2.2/11, SFT = 42
RD = 192.0.2.5/11, SFT = 43 (for SFIa)
RD = 192.0.2.6/11, SFT = 43 (for SFIb)
RD = 192.0.2.7/11, SFT = 43 (for SFIc)
RD = 192.0.2.3/11, SFT = 44
SFF2 is the point at which a load-balancing choice must be made. So
"tail-end" SFPs are constructed as follows. Each takes in a
different SFF that provides access to an SF of Type 43.
SFP26: RD = 198.51.100.1/126, SPI = 40,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/130, Assoc-SPI = 44,
[SI = 255, SFT = 43, RD = 192.0.2.5/11],
[SI = 254, SFT = 44, RD = 192.0.2.3/11]
SFP27: RD = 198.51.100.1/127, SPI = 41,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/131, Assoc-SPI = 45,
[SI = 255, SFT = 43, RD = 192.0.2.6/11],
[SI = 254, SFT = 44, RD = 192.0.2.3/11]
SFP28: RD = 198.51.100.1/128, SPI = 42,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/132, Assoc-SPI = 46,
[SI = 255, SFT = 43, RD = 192.0.2.7/11],
[SI = 254, SFT = 44, RD = 192.0.2.3/11]
Now an end-to-end SFP with load-balancing choice can be constructed
as follows. The choice made by SFF2 is expressed in terms of
entering one of the three "tail-end" SFPs.
SFP29: RD = 198.51.100.1/129, SPI = 43,
[SI = 255, SFT = 41, RD = 192.0.2.1/11],
[SI = 254, SFT = 42, RD = 192.0.2.2/11],
[SI = 253, {SFT = 1, RD = {SPI=40, SI=255, Rsv=0},
RD = {SPI=41, SI=255, Rsv=0},
RD = {SPI=42, SI=255, Rsv=0} } ]
Now, despite the load-balancing choice being made elsewhere than at
the initial classifier, it is possible for the reverse SFPs to be
well constructed without any ambiguity. The three reverse paths
appear as follows.
SFP30: RD = 198.51.100.1/130, SPI = 44,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/126, Assoc-SPI = 40,
[SI = 255, SFT = 44, RD = 192.0.2.4/11],
[SI = 254, SFT = 43, RD = 192.0.2.5/11],
[SI = 253, SFT = 42, RD = 192.0.2.2/11],
[SI = 252, SFT = 41, RD = 192.0.2.1/11]
SFP31: RD = 198.51.100.1/131, SPI = 45,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/127, Assoc-SPI = 41,
[SI = 255, SFT = 44, RD = 192.0.2.4/11],
[SI = 254, SFT = 43, RD = 192.0.2.6/11],
[SI = 253, SFT = 42, RD = 192.0.2.2/11],
[SI = 252, SFT = 41, RD = 192.0.2.1/11]
SFP32: RD = 198.51.100.1/132, SPI = 46,
Assoc-Type = 1, Assoc-RD = 198.51.100.1/128, Assoc-SPI = 42,
[SI = 255, SFT = 44, RD = 192.0.2.4/11],
[SI = 254, SFT = 43, RD = 192.0.2.7/11],
[SI = 253, SFT = 42, RD = 192.0.2.2/11],
[SI = 252, SFT = 41, RD = 192.0.2.1/11]
8.10. Examples Using IPv6 Addressing
This section provides several examples using IPv6 addressing. As
will be seen from the examples, there is nothing special or clever
about using IPv6 addressing rather than IPv4 addressing.
The reference network for these IPv6 examples is based on that
described at the top of Section 8 and shown in Figure 11.
Assume we have a service function overlay network with four SFFs
(SFF1, SFF3, SFF3, and SFF4). The SFFs have addresses in the
underlay network as follows:
SFF1 2001:db8::192:0:2:1
SFF2 2001:db8::192:0:2:2
SFF3 2001:db8::192:0:2:3
SFF4 2001:db8::192:0:2:4
Each SFF provides access to some SFIs from the four service function
types SFT=41, SFT=42, SFT=43, and SFT=44, just as before:
SFF1 SFT=41 and SFT=42
SFF2 SFT=41 and SFT=43
SFF3 SFT=42 and SFT=44
SFF4 SFT=43 and SFT=44
The service function network also contains a controller with address
2001:db8::198:51:100:1.
This example service function overlay network is shown in Figure 15.
------------------------
| Controller |
| 2001:db8::198:51:100:1 |
------------------------
------ ------ ------ ------
| SFI | | SFI | | SFI | | SFI |
|SFT=41| |SFT=42| |SFT=41| |SFT=43|
------ ------ ------ ------
\ / \ /
------------------- -------------------
| SFF1 | | SFF2 |
|2001:db8::192:0:2:1| |2001:db8::192:0:2:2|
------------------- -------------------
----------
Packet --> | | -->
Flows --> |Classifier| -->Dest
| | -->
----------
------------------- -------------------
| SFF3 | | SFF4 |
|2001:db8::192:0:2:3| |2001:db8::192:0:2:4|
------------------- -------------------
/ \ / \
------ ------ ------ ------
| SFI | | SFI | | SFI | | SFI |
|SFT=42| |SFT=44| |SFT=43| |SFT=44|
------ ------ ------ ------
Figure 15: Example Service Function Overlay Network
The SFFs advertise routes to the SFIs they support. These
advertisements contain RDs that are set according to the network
operator's configuration model. Note that in an IPv6 network, the RD
is not large enough to contain the full IPv6 address, as only six
octets are available. So, in all of these IPv6 examples, we use RDs
of Type 1 such that the available six octets are partitioned as four
octets for an IPv4 address of the advertising SFF, and two octets
that are a local index of the SFI. Furthermore, we have chosen an
IPv6 addressing scheme so that the low-order four octets of the IPv6
address match an IPv4 address of the advertising node. This scheme
is chosen purely for convenience of documentation, and an operator is
totally free to use any other scheme so long as it conforms to the
definitions of SFIR and SFPR in Sections 3.1 and 3.2.
Observant readers will notice that this makes the BGP advertisements
shown in these examples exactly the same as in the previous examples.
All that is different is that the advertising SFFs and controller
have IPv6 addresses.
Thus, we see the following SFIRs advertised.
The SFFs advertise routes to the SFIs they support. So we see the
following SFIRs:
RD = 192.0.2.1/1, SFT = 41
RD = 192.0.2.1/2, SFT = 42
RD = 192.0.2.2/1, SFT = 41
RD = 192.0.2.2/2, SFT = 43
RD = 192.0.2.3/7, SFT = 42
RD = 192.0.2.3/8, SFT = 44
RD = 192.0.2.4/5, SFT = 43
RD = 192.0.2.4/6, SFT = 44
Note that the addressing used for communicating between SFFs is taken
from the tunnel encapsulation attribute of the SFIR and not from the
SFIR-RD.
8.10.1. Example Explicit SFP with No Choices
Consider the following SFPR similar to that in Section 8.1.
SFP1: RD = 198.51.100.1/101, SPI = 15,
[SI = 255, SFT = 41, RD = 192.0.2.1/1],
[SI = 250, SFT = 43, RD = 192.0.2.2/2]
The SFP consists of an SF of Type 41 located at SFF1, followed by an
SF of Type 43 located at SFF2. This path is fully explicit, and each
SFF is offered no choice in forwarding a packet along the path.
SFF1 will receive packets on the path from the classifier and will
identify the path from the SPI (15). The initial SI will be 255, and
so SFF1 will deliver the packets to the SFI for SFT 41.
When the packets are returned to SFF1 by the SFI, the SI will be
decreased to 250 for the next hop. SFF1 has no flexibility in the
choice of SFF to support the next-hop SFI and will forward the packet
to SFF2, which will send the packets to the SFI that supports SFT 43
before forwarding the packets to their destinations.
8.10.2. Example SFP with Choice of SFIs
SFP2: RD = 198.51.100.1/102, SPI = 16,
[SI = 255, SFT = 41, RD = 192.0.2.1/1],
[SI = 250, SFT = 43, {RD = 192.0.2.2/2,
RD = 192.0.2.4/5 } ]
In this example, like that in Section 8.2, the path also consists of
an SF of Type 41 located at SFF1, and this is followed by an SF of
Type 43; but in this case, the SI = 250 contains a choice between the
SFI located at SFF2 and the SFI located at SFF4.
SFF1 will receive packets on the path from the classifier and will
identify the path from the SPI (16). The initial SI will be 255, and
so SFF1 will deliver the packets to the SFI for SFT 41.
When the packets are returned to SFF1 by the SFI, the SI will be
decreased to 250 for the next hop. SFF1 now has a choice of next-hop
SFFs to execute the next hop in the path. It can either forward
packets to SFF2 or SFF4 to execute a function of Type 43. It uses
its local load-balancing algorithm to make this choice. The chosen
SFF will send the packets to the SFI that supports SFT 43 before
forwarding the packets to their destinations.
8.10.3. Example SFP with Open Choice of SFIs
SFP3: RD = 198.51.100.1/103, SPI = 17,
[SI = 255, SFT = 41, RD = 192.0.2.1/1],
[SI = 250, SFT = 44, RD = 0]
In this example, like that in Section 8.3, the path also consists of
an SF of Type 41 located at SFF1, and this is followed by an SI with
an RD of zero and SF of Type 44. This means that a choice can be
made between any SFF that supports an SFI of Type 44.
SFF1 will receive packets on the path from the classifier and will
identify the path from the SPI (17). The initial SI will be 255, and
so SFF1 will deliver the packets to the SFI for SFT 41.
When the packets are returned to SFF1 by the SFI, the SI will be
decreased to 250 for the next hop. SFF1 now has a free choice of
next-hop SFFs to execute the next hop in the path, selecting between
all SFFs that support SFs of Type 44. Looking at the SFIRs it has
received, SFF1 knows that SF Type 44 is supported by SFF3 and SFF4.
SFF1 uses its local load-balancing algorithm to make this choice.
The chosen SFF will send the packets to the SFI that supports SFT 44
before forwarding the packets to their destinations.
8.10.4. Example SFP with Choice of SFTs
SFP4: RD = 198.51.100.1/104, SPI = 18,
[SI = 255, SFT = 41, RD = 192.0.2.1/1],
[SI = 250, {SFT = 43, RD = 192.0.2.2/2,
SFT = 44, RD = 192.0.2.3/8 } ]
This example, similar to that in Section 8.4, provides a choice of SF
type in the second hop in the path. The SI of 250 indicates a choice
between SF Type 43 located through SF2 and SF Type 44 located at SF3.
SFF1 will receive packets on the path from the classifier and will
identify the path from the SPI (18). The initial SI will be 255, and
so SFF1 will deliver the packets to the SFI for SFT 41.
When the packets are returned to SFF1 by the SFI, the SI will be
decreased to 250 for the next hop. SFF1 now has a free choice of
next-hop SFFs to execute the next hop in the path, selecting between
all SFFs that support an SF of Type 43 and SFF3, which supports an SF
of Type 44. These may be completely different functions that are to
be executed dependent on specific conditions, or they may be similar
functions identified with different type identifiers (such as
firewalls from different vendors). SFF1 uses its local policy and
load-balancing algorithm to make this choice, and it may use
additional information passed back from the local SFI to help inform
its selection. The chosen SFF will send the packets to the SFI that
supports the chosen SFT before forwarding the packets to their
destinations.
9. Security Considerations
The mechanisms in this document use BGP for the control plane.
Hence, techniques such as those discussed in [RFC5925] can be used to
help authenticate BGP sessions and, thus, the messages between BGP
peers, making it harder to spoof updates (which could be used to
install bogus SFPs or advertise false SIs) or withdrawals.
Further discussion of security considerations for BGP may be found in
the BGP specification itself [RFC4271] and the security analysis for
BGP [RFC4272]. [RFC5925] contains a discussion of the
inappropriateness of the TCP MD5 signature option for protecting BGP
sessions. [RFC6952] includes an analysis of BGP keying and
authentication issues.
Additionally, this document depends on other documents that specify
BGP Multiprotocol Extensions and the documents that define the
attributes that are carried by BGP UPDATEs of the SFC AFI/SAFI.
[RFC4760] observes that the use of AFI/SAFI does not change the
underlying security issues inherent in the existing BGP. Relevant
additional security measures are considered in [RFC9012].
This document does not fundamentally change the security behavior of
BGP deployments, which depend considerably on the network operator's
perception of risk in their network. It may be observed that the
application of the mechanisms described in this document is scoped to
a single domain, as implied by [RFC8300] and noted in Section 2.1 of
this document. Applicability of BGP within a single domain may
enable a network operator to make easier and more consistent
decisions about what security measures to apply, and the domain
boundary, which BGP enforces by definition, provides a safeguard that
prevents leakage of SFC programming in either direction at the
boundary.
Service function chaining provides a significant attack opportunity;
packets can be diverted from their normal paths through the network,
packets can be made to execute unexpected functions, and the
functions that are instantiated in software can be subverted.
However, this specification does not change the existence of service
function chaining, and security issues specific to service function
chaining are covered in [RFC7665] and [RFC8300].
This document defines a control plane for service function chaining.
Clearly, this provides an attack vector for a service function
chaining system, as an attack on this control plane could be used to
make the system misbehave. Thus, the security of the BGP system is
critically important to the security of the whole service function
chaining system. The control plane mechanisms are very similar to
those used for BGP/MPLS IP VPNs as described in [RFC4364], and so the
security considerations in that document (Section 13) provide good
guidance for securing service function chaining systems reliant on
this specification. Of particular relevance is the need to securely
distinguish between messages intended for the control of different
SFC overlays, which is similar to the need to distinguish between
different VPNs. Section 19 of [RFC7432] also provides useful
guidance on the use of BGP in a similar environment.
Note that a component of a service function chaining system that uses
the procedures described in this document also requires
communications between a controller and the service function chaining
network elements (specifically the SFFs and classifiers). This
communication covers instructing the classifiers using BGP mechanisms
(see Section 7.4); therefore, the use of BGP security is strongly
recommended. But it also covers other mechanisms for programming the
classifier and instructing the SFFs and SFs (for example, to bind SFs
to an SFF, and to cause the establishment of tunnels between SFFs).
This document does not cover these latter mechanisms, and so their
security is out of scope, but it should be noted that these
communications provide an attack vector on the service function
chaining system, and so attention must be paid to ensuring that they
are secure.
There is an intrinsic assumption in service function chaining systems
that nodes that announce support for specific SFs actually offer
those functions and that SFs are not, themselves, attacked or
subverted. This is particularly important when the SFs are
implemented as software that can be updated. Protection against this
sort of concern forms part of the security of any service function
chaining system and so is outside the scope of the control plane
mechanisms described in this document.
Similarly, there is a vulnerability if a rogue or subverted
controller announces SFPs, especially if that controller "takes over"
an existing SFP and changes its contents. This corresponds to a
rogue BGP speaker entering a routing system, or even a Route
Reflector becoming subverted. Protection mechanisms, as above,
include securing BGP sessions and protecting software loads on the
controllers.
In an environment where there is concern that rogue controllers might
be introduced to the network and inject false SFPRs or take over and
change existing SFPRs, it is RECOMMENDED that each SFF and classifier
be configured with the identities of authorized controllers. Thus,
the announcement of an SFPR by any other BGP peer would be rejected.
Lastly, note that Section 3.2.2 makes two operational suggestions
that have implications for the stability and security of the
mechanisms described in this document:
* That modifications to active SFPs not be made.
* That SPIs not be immediately reused.
10. IANA Considerations
10.1. New BGP AF/SAFI
IANA maintains the "Address Family Numbers" registry. IANA has
assigned a new Address Family Number from the "Standards Action"
range called "BGP SFC" (31), with this document as a reference.
IANA maintains the "Subsequent Address Family Identifiers (SAFI)
Parameters" registry. IANA has assigned a new SAFI value from the
"Standards Action" range called "BGP SFC" (9), with this document as
a reference.
10.2. "SFP attribute" BGP Path Attribute
IANA maintains a registry of "Border Gateway Protocol (BGP)
Parameters" with a subregistry of "BGP Path Attributes". IANA has
assigned a new Path attribute called "SFP attribute" with a value of
37 and with this document as a reference.
10.3. "SFP Attribute TLVs" Registry
IANA maintains a registry of "Border Gateway Protocol (BGP)
Parameters". IANA has created a new subregistry called the "SFP
Attribute TLVs" registry.
Valid values are in the range 0 to 65535.
* Values 0 and 65535 are marked "Reserved".
* Values 1 through 65534 are to be assigned according to the "First
Come First Served" policy [RFC8126].
This document is a reference for this registry.
The registry tracks:
* Type
* Name
* Reference
* Registration Date
The registry is initially populated as follows:
+======+=========================+===========+===================+
| Type | Name | Reference | Registration Date |
+======+=========================+===========+===================+
| 1 | Association TLV | RFC 9015 | 2020-09-02 |
+------+-------------------------+-----------+-------------------+
| 2 | Hop TLV | RFC 9015 | 2020-09-02 |
+------+-------------------------+-----------+-------------------+
| 3 | SFT TLV | RFC 9015 | 2020-09-02 |
+------+-------------------------+-----------+-------------------+
| 4 | MPLS Swapping/Stacking | RFC 9015 | 2020-09-02 |
+------+-------------------------+-----------+-------------------+
| 5 | SFP Traversal With MPLS | RFC 9015 | 2020-09-02 |
+------+-------------------------+-----------+-------------------+
Table 1: SFP Attribute TLVs Subregistry Initial Contents
10.4. "SFP Association Type" Registry
IANA maintains a registry of "Border Gateway Protocol (BGP)
Parameters". IANA has created a new subregistry called the "SFP
Association Type" registry.
Valid values are in the range 0 to 65535.
* Values 0 and 65535 are marked "Reserved".
* Values 1 through 65534 are assigned according to the "First Come
First Served" policy [RFC8126].
This document is given as a reference for this registry.
The new registry tracks:
* Association Type
* Name
* Reference
* Registration Date
The registry should initially be populated as follows:
+==================+===================+===========+============+
| Association Type | Name | Reference | Date |
+==================+===================+===========+============+
| 1 | Bidirectional SFP | RFC 9015 | 2020-09-02 |
+------------------+-------------------+-----------+------------+
Table 2: SFP Association Type Subregistry Initial Contents
10.5. "Service Function Chaining Service Function Types" Registry
IANA has created a new top-level registry called "Service Function
Chaining Service Function Types".
Valid values are in the range 0 to 65535.
* Values 0 and 65535 are marked "Reserved".
* Values 1 through 31 are to be assigned by "Standards Action"
[RFC8126] and are referred to as the "special-purpose SFT values".
* Values 32 through 64495 are to be assigned according to the "First
Come First Served" policy [RFC8126].
* Values 64496 through 65534 are for Private Use and are not to be
recorded by IANA.
This document is given as a reference for this registry.
The registry tracks:
* Value
* Name
* Reference
* Registration Date
The registry is initially populated as follows.
+=============+===================+=============+============+
| Value | Name | Reference | Date |
+=============+===================+=============+============+
| 0 | Reserved | RFC 9015 | 2020-09-02 |
+-------------+-------------------+-------------+------------+
| 1 | Change Sequence | RFC 9015 | 2020-09-02 |
+-------------+-------------------+-------------+------------+
| 2-31 | Unassigned | | |
+-------------+-------------------+-------------+------------+
| 32 | Classifier | RFC 9015, | 2020-09-02 |
| | | [BGP-LS-SR] | |
+-------------+-------------------+-------------+------------+
| 33 | Firewall | RFC 9015, | 2020-09-02 |
| | | [BGP-LS-SR] | |
+-------------+-------------------+-------------+------------+
| 34 | Load balancer | RFC 9015, | 2020-09-02 |
| | | [BGP-LS-SR] | |
+-------------+-------------------+-------------+------------+
| 35 | Deep packet | RFC 9015, | 2020-09-02 |
| | inspection engine | [BGP-LS-SR] | |
+-------------+-------------------+-------------+------------+
| 36 | Penalty box | RFC 9015, | 2020-09-02 |
| | | [RFC8300] | |
+-------------+-------------------+-------------+------------+
| 37 | WAN accelerator | RFC 9015, | 2020-09-02 |
| | | [RFC7665], | |
| | | [RFC8300] | |
+-------------+-------------------+-------------+------------+
| 38 | Application | RFC 9015, | 2020-09-02 |
| | accelerator | [RFC7665] | |
+-------------+-------------------+-------------+------------+
| 39 | TCP optimizer | RFC 9015, | 2020-09-02 |
| | | [RFC7665] | |
+-------------+-------------------+-------------+------------+
| 40 | Network Address | RFC 9015, | 2020-09-02 |
| | Translator | [RFC7665] | |
+-------------+-------------------+-------------+------------+
| 41 | NAT44 | RFC 9015, | 2020-09-02 |
| | | [RFC7665], | |
| | | [RFC3022] | |
+-------------+-------------------+-------------+------------+
| 42 | NAT64 | RFC 9015, | 2020-09-02 |
| | | [RFC7665], | |
| | | [RFC6146] | |
+-------------+-------------------+-------------+------------+
| 43 | NPTv6 | RFC 9015, | 2020-09-02 |
| | | [RFC7665], | |
| | | [RFC6296] | |
+-------------+-------------------+-------------+------------+
| 44 | Lawful intercept | RFC 9015, | 2020-09-02 |
| | | [RFC7665] | |
+-------------+-------------------+-------------+------------+
| 45 | HOST_ID injection | RFC 9015, | 2020-09-02 |
| | | [RFC7665] | |
+-------------+-------------------+-------------+------------+
| 46 | HTTP header | RFC 9015, | 2020-09-02 |
| | enrichment | [RFC7665] | |
+-------------+-------------------+-------------+------------+
| 47 | Caching engine | RFC 9015, | 2020-09-02 |
| | | [RFC7665] | |
+-------------+-------------------+-------------+------------+
| 48-64495 | Unassigned | | |
+-------------+-------------------+-------------+------------+
| 64496-65534 | Reserved for | | |
| | Private Use | | |
+-------------+-------------------+-------------+------------+
| 65535 | Reserved, not to | RFC 9015 | 2020-09-02 |
| | be allocated | | |
+-------------+-------------------+-------------+------------+
Table 3: Service Function Chaining Service Function Types
Registry Initial Contents
10.6. Flow Specification for SFC Classifiers
IANA maintains a registry of "Border Gateway Protocol (BGP) Extended
Communities" with a subregistry of "Generic Transitive Experimental
Use Extended Community Sub-Types". IANA has assigned a new subtype
as follows:
"Flow Specification for SFC Classifiers" with a value of 0x0d and
with this document as the reference.
10.7. New BGP Transitive Extended Community Type
IANA maintains a registry of "Border Gateway Protocol (BGP) Extended
Communities" with a subregistry of "BGP Transitive Extended Community
Types". IANA has assigned a new type as follows:
SFC (Sub-Types are defined in the "SFC Extended Community Sub-
Types" registry) with a value of 0x0b and with this document as
the reference.
10.8. "SFC Extended Community Sub-Types" Registry
IANA maintains a registry of "Border Gateway Protocol (BGP)
Parameters". IANA has created a new subregistry called the "SFC
Extended Community Sub-Types" registry.
IANA has included the following note:
| This registry contains values of the second octet (the "Sub-
| Type" field) of an extended community when the value of the
| first octet (the "Type" field) is set to 0x0b.
The allocation policy for this registry is First Come First Served.
Valid values are 0 to 255. The value 0 is reserved and should not be
allocated.
IANA has populated this registry with the following entries:
+==========+==========================+===========+============+
| Sub-Type | Name | Reference | Date |
| Value | | | |
+==========+==========================+===========+============+
| 0 | Reserved | RFC 9015 | |
+----------+--------------------------+-----------+------------+
| 1 | SFIR pool identifier | RFC 9015 | 2020-09-02 |
+----------+--------------------------+-----------+------------+
| 2 | MPLS Label Stack Mixed | RFC 9015 | 2020-09-02 |
| | Swapping/Stacking Labels | | |
+----------+--------------------------+-----------+------------+
| 3-255 | Unassigned | | |
+----------+--------------------------+-----------+------------+
Table 4: SFC Extended Community Sub-Types Subregistry
Initial Contents
10.9. New SPI/SI Representation Sub-TLV
IANA has assigned a codepoint from the "BGP Tunnel Encapsulation
Attribute Sub-TLVs" registry for the "SPI/SI Representation Sub-TLV"
with a value of 16 and with this document as the reference.
10.10. "SFC SPI/SI Representation Flags" Registry
IANA maintains the "BGP Tunnel Encapsulation Attribute Sub-TLVs"
registry and has created an associated registry called the "SFC SPI/
SI Representation Flags" registry.
Bits are to be assigned by Standards Action. The field is 16 bits
long, and bits are counted from the most significant bit as bit zero.
IANA has populated the registry as follows:
+=======+=================+===========+
| Value | Name | Reference |
+=======+=================+===========+
| 0 | NSH data plane | RFC 9015 |
+-------+-----------------+-----------+
| 1 | MPLS data plane | RFC 9015 |
+-------+-----------------+-----------+
Table 5: SFC SPI/SI Representation
Flags Registry Initial Contents
11. References
11.1. Normative References
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119,
DOI 10.17487/RFC2119, March 1997,
<https://www.rfc-editor.org/info/rfc2119>.
[RFC4271] Rekhter, Y., Ed., Li, T., Ed., and S. Hares, Ed., "A
Border Gateway Protocol 4 (BGP-4)", RFC 4271,
DOI 10.17487/RFC4271, January 2006,
<https://www.rfc-editor.org/info/rfc4271>.
[RFC4360] Sangli, S., Tappan, D., and Y. Rekhter, "BGP Extended
Communities Attribute", RFC 4360, DOI 10.17487/RFC4360,
February 2006, <https://www.rfc-editor.org/info/rfc4360>.
[RFC4364] Rosen, E. and Y. Rekhter, "BGP/MPLS IP Virtual Private
Networks (VPNs)", RFC 4364, DOI 10.17487/RFC4364, February
2006, <https://www.rfc-editor.org/info/rfc4364>.
[RFC4760] Bates, T., Chandra, R., Katz, D., and Y. Rekhter,
"Multiprotocol Extensions for BGP-4", RFC 4760,
DOI 10.17487/RFC4760, January 2007,
<https://www.rfc-editor.org/info/rfc4760>.
[RFC7432] Sajassi, A., Ed., Aggarwal, R., Bitar, N., Isaac, A.,
Uttaro, J., Drake, J., and W. Henderickx, "BGP MPLS-Based
Ethernet VPN", RFC 7432, DOI 10.17487/RFC7432, February
2015, <https://www.rfc-editor.org/info/rfc7432>.
[RFC7606] Chen, E., Ed., Scudder, J., Ed., Mohapatra, P., and K.
Patel, "Revised Error Handling for BGP UPDATE Messages",
RFC 7606, DOI 10.17487/RFC7606, August 2015,
<https://www.rfc-editor.org/info/rfc7606>.
[RFC7665] Halpern, J., Ed. and C. Pignataro, Ed., "Service Function
Chaining (SFC) Architecture", RFC 7665,
DOI 10.17487/RFC7665, October 2015,
<https://www.rfc-editor.org/info/rfc7665>.
[RFC8126] Cotton, M., Leiba, B., and T. Narten, "Guidelines for
Writing an IANA Considerations Section in RFCs", BCP 26,
RFC 8126, DOI 10.17487/RFC8126, June 2017,
<https://www.rfc-editor.org/info/rfc8126>.
[RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC
2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174,
May 2017, <https://www.rfc-editor.org/info/rfc8174>.
[RFC8300] Quinn, P., Ed., Elzur, U., Ed., and C. Pignataro, Ed.,
"Network Service Header (NSH)", RFC 8300,
DOI 10.17487/RFC8300, January 2018,
<https://www.rfc-editor.org/info/rfc8300>.
[RFC8595] Farrel, A., Bryant, S., and J. Drake, "An MPLS-Based
Forwarding Plane for Service Function Chaining", RFC 8595,
DOI 10.17487/RFC8595, June 2019,
<https://www.rfc-editor.org/info/rfc8595>.
[RFC8596] Malis, A., Bryant, S., Halpern, J., and W. Henderickx,
"MPLS Transport Encapsulation for the Service Function
Chaining (SFC) Network Service Header (NSH)", RFC 8596,
DOI 10.17487/RFC8596, June 2019,
<https://www.rfc-editor.org/info/rfc8596>.
[RFC8955] Loibl, C., Hares, S., Raszuk, R., McPherson, D., and M.
Bacher, "Dissemination of Flow Specification Rules",
RFC 8955, DOI 10.17487/RFC8955, December 2020,
<https://www.rfc-editor.org/info/rfc8955>.
[RFC9012] Patel, K., Van de Velde, G., Sangli, S., and J. Scudder,
"The BGP Tunnel Encapsulation Attribute", RFC 9012,
DOI 10.17487/RFC9012, April 2021,
<https://www.rfc-editor.org/info/rfc9012>.
11.2. Informative References
[BGP-LS-SR]
Dawra, G., Filsfils, C., Talaulikar, K., Clad, F.,
Bernier, D., Uttaro, J., Decraene, B., Elmalky, H., Xu,
X., Guichard, J., and C. Li, "BGP-LS Advertisement of
Segment Routing Service Segments", Work in Progress,
Internet-Draft, draft-dawra-idr-bgp-ls-sr-service-
segments-05, 15 February 2021,
<https://tools.ietf.org/html/draft-dawra-idr-bgp-ls-sr-
service-segments-05>.
[RFC3022] Srisuresh, P. and K. Egevang, "Traditional IP Network
Address Translator (Traditional NAT)", RFC 3022,
DOI 10.17487/RFC3022, January 2001,
<https://www.rfc-editor.org/info/rfc3022>.
[RFC4272] Murphy, S., "BGP Security Vulnerabilities Analysis",
RFC 4272, DOI 10.17487/RFC4272, January 2006,
<https://www.rfc-editor.org/info/rfc4272>.
[RFC5925] Touch, J., Mankin, A., and R. Bonica, "The TCP
Authentication Option", RFC 5925, DOI 10.17487/RFC5925,
June 2010, <https://www.rfc-editor.org/info/rfc5925>.
[RFC6146] Bagnulo, M., Matthews, P., and I. van Beijnum, "Stateful
NAT64: Network Address and Protocol Translation from IPv6
Clients to IPv4 Servers", RFC 6146, DOI 10.17487/RFC6146,
April 2011, <https://www.rfc-editor.org/info/rfc6146>.
[RFC6296] Wasserman, M. and F. Baker, "IPv6-to-IPv6 Network Prefix
Translation", RFC 6296, DOI 10.17487/RFC6296, June 2011,
<https://www.rfc-editor.org/info/rfc6296>.
[RFC6952] Jethanandani, M., Patel, K., and L. Zheng, "Analysis of
BGP, LDP, PCEP, and MSDP Issues According to the Keying
and Authentication for Routing Protocols (KARP) Design
Guide", RFC 6952, DOI 10.17487/RFC6952, May 2013,
<https://www.rfc-editor.org/info/rfc6952>.
[RFC7498] Quinn, P., Ed. and T. Nadeau, Ed., "Problem Statement for
Service Function Chaining", RFC 7498,
DOI 10.17487/RFC7498, April 2015,
<https://www.rfc-editor.org/info/rfc7498>.
Acknowledgements
Thanks to Tony Przygienda, Jeff Haas, and Andy Malis for helpful
comments, and to Joel Halpern for discussions that improved this
document. Yuanlong Jiang provided a useful review and caught some
important issues. Stephane Litkowski did an exceptionally good and
detailed Document Shepherd review.
Andy Malis contributed text that formed the basis of Section 7.7.
Brian Carpenter and Martin Vigoureux provided useful reviews during
IETF Last Call. Thanks also to Sheng Jiang, Med Boucadair, Ravi
Singh, Benjamin Kaduk, Roman Danyliw, Adam Roach, Alvaro Retana,
Barry Leiba, and Murray Kucherawy for review comments. Ketan
Talaulikar provided helpful discussion of the SFT codepoint registry.
Ron Bonica kept us honest on the difference between an RD and an RT;
Benjamin Kaduk kept us on message about the difference between an RD
and an Extended Community.
Contributors
Stuart Mackie
Juniper Networks
Email: wsmackie@juinper.net
Keyur Patel
Arrcus, Inc.
Email: keyur@arrcus.com
Avinash Lingala
AT&T
Email: ar977m@att.com
Authors' Addresses
Adrian Farrel
Old Dog Consulting
Email: adrian@olddog.co.uk
John Drake
Juniper Networks
Email: jdrake@juniper.net
Eric Rosen
Juniper Networks
Email: erosen52@gmail.com
Jim Uttaro
AT&T
Email: ju1738@att.com
Luay Jalil
Verizon
Email: luay.jalil@verizon.com