Processing of the Hop-by-Hop Options Header

Abstract

This document describes the processing of the Hop-by-Hop Options Header (HBH) in today's routers in the aspects of standards specification, common implementations, and default operations. This document outlines the reasons why the Hop-by-Hop Options Header is rarely utilized in current networks. In addition, this document describes how the HBH could be used as a powerful mechanism allowing deployment and operations of new services requiring a more optimized way to leverage network resources of an infrastructure. The Hop-by-Hop Options Header is taken into consideration by several network operators as a valuable container for carrying the information facilitating the introduction of new services. The processing requirments of the HBH and the migration strategies are also suggested.¶

1. Introduction

Due to historical reasons, such as incapable ASICs, limited IPv6 deployments, and few service requirements, the most common Hop-by-Hop Options header (HBH) processing implementation is that the node sends the IPv6 packets with the Hop-by-Hop Options header to the control plane of the node. The option type of each option carried within the Hop-by-Hop Options header will not even be examined before the packet is sent to the control plane. Very often, such processing behavior is the default configuration or, even worse, is the only behavior of the ipv6 implementation of the node.¶

Such default processing behavior of the Hop-by-Hop Options header could result in various unpleasant effects such as a risk of Denial of Service (DoS) attack on the router control plane and inconsistent packet drops due to rate limiting on the interface between the router control plane and forwarding plane, which will impact the normal end-to-end IP forwarding of the network services.¶

This actually introduced a circular problem:¶

-> An implementation problem caused HBH to become a DoS vector.¶

-> Because HBH is a DoS vector, network operators deployed ACLs that discard packets containing HBH.¶

-> Because network operators deployed ACLs that discard packets containing HBH, network designers stopped defining new HBH Options.¶

-> Because network designers stopped defining new HBH Options, the community was not motivated to fix the implementation problem that cause HBH to become a DoS vector.¶

The purpose of this draft is to break the cycle described above, fixing the problem that caused HBH not actually being utilized in operators' networks so to allow a better leverage of the HBH capability.¶

Driven by the wide deployments of IPv6 and ever-emerging new services, the Hop-by-Hop Options Header is taken as a valuable container for carrying the information to facilitate these new services.¶

This document suggests the desired processing behavior and the migration strategies towards it.¶

2. Modern Router Architecture

Modern router architecture design maintains a strict separation of the router control plane and its forwarding plane [RFC6192], as shown in Figure 1. Either the control plane or the forwarding plane is composed of both software and hardware, but each plane is responsible for different functions.¶

                             +----------------+
                             | Router Control |
                             |     Plane      |
                             +------+-+-------+
                                    | |
                                 Interface Z
                                    | |
                             +------+-+-------+
                             |   Forwarding   |
               Interface X ==[     Plane      ]== Interface Y
                             +----------------+

                     Figure 1. Modern Router Architecture

The router control plane supports routing and management functions, handling packets destined to the device as well as building and sending packets originated locally on the device, and also drives the programming of the forwarding plane. The router control plane is generally realized in software on general-purpose processors, and its hardware is usually not optimized for high-speed packet handling. Because of the wide range of functionality, it is more susceptible to security vulnerabilities and a more likely a target for a DoS attack.¶

The forwarding plane is typically responsible for receiving a packet on an incoming interface, performing a lookup to identify the packet's next hop and determine the outgoing interface towards the destination, and forwarding the packet out through the appropriate outgoing interface. Typically, forwarding plane functionality is realized in high-performance Application Specific Integrated Circuits (ASICs) or Network Processors (NPs) that are capable of handling very high packet rates.¶

The router control plane interfaces with its forwarding plane through the Interface Z, as shown in the Figure 1, and the forwarding plane connects to other network devices via Interfaces such as X and Y. Since the router control plane is vulnerable to the DoS attack, usually a traffic filtering mechanism is implemented on Interface Z in order to block unwanted traffic. In order to protect the router control plane, a rate-limiting mechanism is always implemented on this interface. However, such rate limiting mechanism will always cause inconsistent packet drops, which will impact the normal IP forwarding.¶

Semiconductor chip technology has advanced significantly in the last decade, and as such the widely used network processing and forwarding process can now not only forward packets at line speed, but also easily support other feature processing such as QoS for DiffServ/ MPLS, Access Control List (ACL), Firewall, and Deep Packet Inspection (DPI).¶

A Network Processing Unit (NPU) is a non-ASIC based Integrated Circuit (IC) that is programmable through software. It performs all packet header operations between the physical layer interface and the switching fabric such as packet parsing and forwarding, modification, and forwarding. Many equipment vendors implement these functions in fixed function ASICs rather than using "off-the-shelf" NPUs, because of proprietary algorithms.¶

Classification Co-processor is a specialized processor that can be used to lighten the processing load on an NPU by handling the parsing and classification of incoming packets such as IPv6 extended header HBH options processing. This advancement enables network processors to do the general process to handle simple control messages for traffic management, such as signaling for hardware programming, congestion state report, OAM, etc. Industry trend is for intelligent multi-core CPU hardware using modern NPUs for forwarding packets at line rate while still being able to perform other complex tasks such as HBH forwarding options processing without having to punt to the control plane.¶

Many of the packet-processing devices employed in modern switch and router designs are fixed-function ASICs to handle proprietary functions. While these devices can be very efficient for the set of functions they are designed for, they can be very inflexible. There is a tradeoff of price, performance and flexibility when vendors make a choice to use a fixed function ASIC as opposed to NPU. Due to the inflexibility of the fixed function ASIC, tasks that require additional processing such as IPv6 HBH header processing must be punted to the control plane. This problem is still a challenge today and is the reason why operators to protect against control plane DOS attack vector must drop or ignore HBH options. As industry shifts to Merchant Silicon based NPU evolution from fixed function ASIC, the gap will continue to close increasing the viability ubiquitous HBH use cases due to now processing in the forwarding plane.¶

Most modern routers maintain a strict separation between forwarding plane and control plane hardware. Forwarding plane bandwidth and resources are plentiful, while control plane bandwidth and resources are constrained. In order to protect scarce control plane resources, routers enforce policies that restrict access from the forwarding plane to the control plane. Effective policies address packets containing the HBH Options Extension header, because HBH control options require access from the forwarding plane to the control plane. Many network operators perceive HBH Options to be a breach of the separation between the forwarding and control planes. In this case HBH control options would be required to be punted to control plane by fixed function ASICs as well as NPUs.¶

The maximum length of an HBH Options header is 2,048 bytes. A source node can encode hundreds of options in 2,048 bytes. With today's technology it would be cost prohibitive to be able to process hundreds of options with either NPU or proprietary fixed function ASIC.¶

While [RFC8200] required that all nodes must examine and process the Hop-by-Hop Options header, it is now expected that nodes along a packet's delivery path only examine and process the Hop-by-Hop Options header if explicitly configured to do so. This can be beneficial in cases where transit nodes are legacy hardware and the destination endpoint PE is newer NPU based hardware that can process HBH in the forwarding plane.¶

IPv6 Extended Header limitations that need to be addressed to make HBH processing more efficient and viable in the forwarding plane:¶

[RFC8504] defines the IPv6 node requirements and how to protect a node from excessive header chain and excessive header options with various limitations that can be defined on a node. [RFC8883] defines ICMPv6 Errors for discarding packets due to processing limits. Per [RFC8200] HBH options must be processed serially. However, an implementation of options processing can be made to be done with more parallelism in serial processing grouping of similar options to be processed in parallel.¶

The IPv6 standard does not currently limit the header chain length or number of options that can be encoded.¶

Each Option is encoded in a TLV and so processing of a long list of TLVs is expensive. Zero data length encoded options TLVs are a valid option. A DOS vector could be easily generated by encoding 1000 HBH options (Zero data length) in a standard 1500 MTU packet. So now imagine if you have a Christmas tree long header chain to parse each with many options.¶

3. Specification of RFC 8200

[RFC8200] defines several IPv6 extension header types, including the Hop-by-Hop (HBH) Options header. As specified in [RFC8200], the Hop-by-Hop (HBH) Options header is used to carry optional information that will be examined and processed by every node along a packet's delivery path, and it is identified by a Next Header value of zero in the IPv6 header.¶

The Hop-by-Hop (HBH) Options header contains the following fields:¶

-- Next Header: 8-bit selector, identifies the type of header immediately following the Hop-by-Hop Options header.¶

-- Hdr Ext Len: 8-bit unsigned integer, the length of the Hop-by-Hop Options header in 8-octet units, not including the first 8 octets.¶

-- Options: Variable-length field, of length such that the complete Hop-by-Hop Options header is an integer multiple of 8 octets long.¶

The Hop-by-Hop (HBH) Options header carries a variable number of "options" that are encoded in the format of type-length-value (TLV).¶

The highest-order two bits (i.e., the ACT bits) of the Option Type specify the action that must be taken if the processing IPv6 node does not recognize the Option Type. The third-highest-order bit (i.e., the CHG bit) of the Option Type specifies whether or not the Option Data of that option can change en route to the packet's final destination.¶

While [RFC2460] required that all nodes must examine and process the Hop-by-Hop Options header, with [RFC8200] it is expected that nodes along a packet's delivery path only examine and process the Hop-by-Hop Options header if explicitly configured to do so. It means that the HBH processing behavior in a node depends on its configuration.¶

However, in the current [RFC8200], there is no explicit specification of the possible configurations. Therefore, the nodes may be configured to ignore the Hop-by-Hop Options header, drop packets containing a Hop-by-Hop Options header, or assign packets containing a Hop-by-Hop Options header to the control plane [RFC8200]. Because of these likely uncertain processing behaviors, new hop-by-hop options are not recommended.¶

4. Common Implementations

In the current common implementations, once an IPv6 packet, with its Next Header field set to 0, arrives at a node, it will be directly sent to the control plane of the node. With such implementations, the value of the Next Header field in the IPv6 header is the only trigger for the default processing behavior. The option type of each option carried within the Hop-by-Hop Options header will not even be examined before the packet is sent to the control plane.¶

Very often, such processing behavior is the default configuration on the node, which is embedded in the implementation and cannot be changed or reconfigured.¶

Another critical component of IPv6 HBH processing, in some cases overlooked, is the operator core network which can be designed to use the global Internet routing table for internet traffic and in other cases use an overlay MPLS VPN to carry Internet traffic.¶

In the global Internet routing table scenario where only an underlay global routing table exists, and no VPN overlay carrying customer Internet traffic, the IPv6 HBH options can be used as a DOS attack vector for both the operator nodes, adjacent inter-as peer nodes as well as customer nodes along a path.¶

In a case where the Internet routing table is carried in a MPLS VPN overlay payload, the HBH options header does not impact the operator underlay framework and only impacts the VPN overlay payload and thus the operator underlay topmost label global table routing FEC LSP instantiation is not impacted as the operator underlay is within the operators closed domain.¶

However, HBH options DOS attack vector in the VPN overlay can still impact the customer CE destination end nodes as well as other adjacent inter-as operators that only use underlay global Internet routing table. In an operator closed domain where MPLS VPN overlay is utilized to carry internet traffic, the operator has full control of the underlay and IPv6 Extended header chain length as well as the number of HBH options encoded.¶

In the global routing table scenario for Internet traffic there is no way to control the IPv6 Extended header chain lenghth as well as the number of HBH options encoded.¶

5. Operators' Typical Processing

To mitigate this DoS vulnerability, many operators deployed Access Control Lists (ACLs) that discard all packets containing HBH Options.¶

[RFC6564] shows the Reports from the field indicating that some IP routers deployed within the global Internet are configured either to ignore or to drop packets having a hop-by-hop header. As stated in [RFC7872], many network operators perceive HBH Options to be a breach of the separation between the forwarding and control planes. Therefore, several network operators configured their nodes so as to discard all packets containing the HBH Options Extension Header, while others configured nodes to forward the packet but to ignore the HBH Options. [RFC7045] also states that hop-by-hop options are not handled by many high-speed routers or are processed only on a control plane.¶

Due to such behaviors observed and described in these specifications, new hop-by-hop options are not recommended in [RFC8200] hence the usability of HBH options is severely limited.¶

6. New Services

As IPv6 is being rapidly and widely deployed worldwide, more and more applications and network services are migrating to or directly adopting IPv6. More and more new services that require HBH are emerging and the HBH Options header is going to be utilized by the new services in various scenarios.¶

In-situ OAM (IOAM) with IPv6 encapsulation [I-D.ietf-ippm-ioam-ipv6-options] is one of the examples. IOAM in IPv6 is used to enhance diagnostics of IPv6 networks and complements other mechanisms, such as the IPv6 Performance and Diagnostic Metrics Destination Option described in [RFC8250]. The IOAM data fields are encapsulated in "option data" fields of the Hop-by-Hop Options header if Pre-allocated Tracing Option, Incremental Tracing Option, or Proof of Transit Option are carried [I-D.ietf-ippm-ioam-data], that is, the IOAM performs per hop.¶

Alternate Marking Method can be used as the passive performance measurement tool in an IPv6 domain. The AltMark Option is defined as a new IPv6 extension header option to encode alternate marking technique and Hop-by-Hop Options Header is considered [I-D.ietf-6man-ipv6-alt-mark].¶

The Minimum Path MTU Hop-by-Hop Option is defined in [I-D.ietf-6man-mtu-option] to record the minimum Path MTU along the forward path between a source host to a destination host. This Hop-by-Hop option is intended to be used in environments like Data Centers and on paths between Data Centers as well as other environments including the general Internet. It provides a useful tool for allowing to better take advantage of paths able to support a large Path MTU.¶

As more services start utilizing the HBH Options header, more packets containing HBH Options are going to be injected into the networks. According to the current common configuration in most network deployments, all the packets of the new services are going to be sent to the control plane of the nodes, with the possible consequence of causing a DoS on the control plane. The packets will be dropped and the normal IP forwarding may be severely impacted. The deployment of new network services involving multi-vendor interoperability will become impossible.¶

7. Requirements

• The HBH options header MUST NOT become a possible DDoS Vector.¶
• HBH options SHOULD be designed so that they don't reduce the probability of packet delivery. For example, an intermediate node may discard a packet because it contains more HBH options than the node can process.¶
• HBH processing MUST be efficient. That is, it MUST be possible to produce implementations that perform well at a reasonable cost.¶
• The Router Alert Option MUST NOT impact the processing of other HBH options that should be processed more quickly.¶
• HBH Options MAY influence how a packet is forwarded. However, with the exception of the Router Alert Option, an HBH Option MUST NOT cause control plane state to be created, modified or destroyed on the processing node. As per [RFC6398], protocol developers SHOULD avoid future use of the Router Alert Option.¶
• More requirements are to be added.¶

8. Migration Strategies

In order to achieve the desired processing behavior of the HBH options header and facilitate the ever-emerging new services to be deployed in operators' networks across multiple vendors' devices, the migration can happen in three parts as described below:¶

1. The source of the HBH options header encapsulation.¶

The information to be carried in the HBH options header needs to be first categorized and encapsulated into either control options or forwarding options, and then encapsulated in different packets.¶

2. The nodes within the network.¶

The nodes within the network are updated to the proposed behavior introduced in the previous section.¶

3. The edge nodes of the network.¶

The edge nodes should check whether the packet contains an HBH header with control or forwarding option. Packets with a control option may still be filtered and dropped while packets with forwarding option SHOULD be allowed by the ACL.¶

If it is certain that there is no harm that can be introduced by the HBH control options to the nodes and the services, they can also be allowed.¶

Note: During the migration stage, the nodes that are not yet updated will stay with their existing configurations.¶

11. Acknowledgements

The authors would like to acknowledge Ron Bonica, Fred Baker, Bob Hinden, Stefano Previdi, and Donald Eastlake for their valuable review and comments.¶

12. References