| Internet-Draft | Extension Header Limits | October 2022 |
| Herbert | Expires 26 April 2023 | [Page] |
This specification defines various limits that may be applied to receiving, sending, and otherwise processing packets that contain IPv6 extension headers. The need for such limits is pragmatic to facilitate interoperability amongst hosts and routers in the presence of extension headers and thereby increasing the feasibility of deployment of extension headers. The limits described herein establish the minimum baseline of support for use of extension headers in the Internet. If it is known that all communicating parties for a particular commumication, including end hosts and any intermediate nodes in the path, are capable of supporting more than the baseline then these default limits may be freely exceeded.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 26 April 2023.¶
Copyright (c) 2022 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 Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
Extension headers are a core component of the IPv6 protocol as specified in [RFC8200]. IPv6 extension headers were originally defined with few restrictions. For instance, there is no specified limit on the number of extension headers a packet may have, nor is there a limit on the length in bytes of extension headers in a packet (other than being limited by the MTU). Similarly, variable length extension headers typically do not have prescribed limits such as limits on the number of Hop-by-Hop or Destination options in a packet. The lack of limits essentially requires implementations to handle every conceivable usage of the protocol, including a myriad of use cases those are obviously outside the realm of ever being realistic or useful in real world deployment.¶
The lack of limits and the requirements for supporting a virtually open-ended protocol have led to a significant lack of support and deployment of extension headers [RFC7872]. Instead of attempting to satisfy the protocol requirements concerning extension headers, some router and middlebox vendors have opted to either invent and apply their own ad hoc limits, relegate packets with extension headers to slow path processing, or have gone so far as to summarily discard all packets with extension headers [RFC9098]. The net result of this situation is that deployment and use of extension headers is underwhelming to the extent that they are sometimes considered unusable on the Internet, and hence IPv6 extension headers have not lived up to their potential as the extensibility mechanism of IPv6.¶
As an example, consider that Hop-by-Hop Options and Destination Options have no limit on how many options may be placed in a packet, nor any limits as to how many options a receiver must process. A single 1500 byte MTU sized packet could legally contain a Hop-by-Hop Options extension header with over seven hundred two byte options. There is no use case for this other than a Denial of Service attack where an attacker simply creates packets with hundreds of small unknown Hop-by-Hop Options with the two high order bits in the option type set to 00 meaning to skip the unknown option. Any node in the path that attempts to dutifully process all these options per the requirements of [RFC2460] would be easily overwhelmed by the processing needed to parse these options (this is true for both hardware or software implementations).¶
This specification describes various limits that hosts and intermediate nodes may apply to the processing of extension headers. The goal of establishing limits is to narrow the requirements to better match reasonable use cases thereby facilitating practical implementation. Subsequently, this increases the viability of extension headers as the extensibility mechanism of IPv6.¶
Some of the problems of unlimited extension headers have been addressed in certain aspects.¶
[RFC8200] relaxed the requirement that all nodes in the path must process Hop-by-Hop Options to be:¶
Section 5.3 of [RFC8504] defines a number of limits that hosts may apply to processing extensions. For instance:¶
[RFC8883] defines a set of ICMP errors that my be sent if a limit concerning extension headers is exceeded and a node discards a packet as a result. This RFC allows both hosts and routers to send such messages (effectively acknowledging that some routers drop packets with extension headers even though such behavior is non-conformant with [RFC8200]).¶
[RFC7872] presents real-world data regarding the extent to which packets with IPv6 Extension Headers (EHs) are dropped in the Internet, and [RFC8200] summarizes the operational implications of IPv6 extension headers, and attempts to analyze reasons why packets with IPv6 extension headers are often dropped in the public Internet.¶
The robustness principle, or Postel's Law, can be stated as "Be conservative in what you send, liberal in what you receive". This section considers the limits defined in this specification with respect to the robustness principle.¶
The limits on sending extension headers are well aligned with the send clause of the robustness principle. A sender of extension headers is generally constrained in its use of extension headers. Most of these limits are assumed to be the default to apply in an arbitrary environment such as the public Internet, that is they can be considered "baseline limits". These limits may be relaxed if a sender has a priori information that all possible nodes in path will properly handle packets that exceed the baseline limits. In particular, if a sender is sending in a limited domain, it might be known that all nodes in the limited domain have sufficient capabilities to handle packets exceeding the baseline limits.¶
Considering the receive clause of the robustness principle, this specification recommends that receivers accept all packets with extension headers, however they may ignore extension headers or options within extension headers (in accordance with [RFC8200]). In particular, the philosophy of this specification is that intermediate nodes should not drop packets on the basis that they don't have sufficient capabilities to process all the headers in a packet. As such, intermediate nodes may define arbitrarily restrictive limits on what they process with regards to extension headers as long as the action taken when those limits are exceeded is to ignore items beyond the limit. Hosts are more constrained in this regard since they generally can't correctly process a packet without processing all the headers, so when limits are exceeded on a host, packets should be dropped. It should be noted that hosts stacks inherently have more processing capabilities than intermediate nodes, so it is expected that they should be able to support higher limits for processing extension headers.¶
This specification does specify one hard requirement for receiving nodes, namely nodes must be able to properly handle packets having an IPv6 header chain length up to 104 bytes. This requirement acknowledges that some intermediate nodes perform deep packet inspection to extract information from transport layer headers [RFC9098]. Often a node that requires parsing transport layer information will have a fixed sized "parsing buffer" to contain packet headers. If the transport layer headers within a packet are beyond the extent of the parsing buffer then an implementation might take some detrimental action such as arbitrarily dropping packets. To this end, this specification requires that any intermediate node that requires access to to transport layer header must minimally be able to parse at least 128 bytes of headers, from which the 104 byte limit for the IP header chain is derived.¶
This specification considers extension header limits in three dimensions: 1) The types of nodes that may process extension headers and the requirements specific to each type, 2) The types of limits that may be applied, 3) The action taken when a limit is exceeded.¶
For the purposes of describing handling of extension headers this specification considers three types of node in an IPv6 network:¶
The limits and requirements for handling extension headers defined in this specification fall in the following categories:¶
[RFC8504] defines limits that may be defined for the length of an extension header. Those limits are extended to be applicable to intermediate nodes. [RFC8883] defines ICMP Parameter Problem codes that may be sent when an extension header is exceeded.¶
A node may establish a limit on the size Hop-by-Hop or Destination options. Conceivably, such a limit could apply to all option types, or length limits may be specific to individual options. [RFC8883] defines ICMP Parameter Problem codes that may be sent when an option length limit is exceeded.¶
A node may define a limit on the number of extension headers it will process. Although [RFC8200] only defines four types of extension headers, it does not preclude the same type of extension header being present multiple times. A limit on the number of extension headers could be useful to disallow packets that contain multiple instances of the same extension header.¶
Limits may be established for the number of options sent or received (specifically applicable to Hop-by-Hop options and Destination options). The need for this limit arises from the fact that [RFC8200] does not specify a limit. Requiring nodes to process packets with tens or hundreds of options has no foreseeable use cases in deployment except as a denial of service attack. [RFC8504] has proposed such a limit for host processing of Hop-by-Hop and Destination options with a default of eight options. This specification extends that limit to be applicable to intermediate nodes. Specific limits may be established for number of non-padding options or the number of all options including padding.¶
To derive a limit for the total number options in an extension header, one can assume that at most one padding option is used between two non-padding options (an explicit limit on consecutive padding options is described below). With this assumption, we can extrapolate a reasonable limit on the number of all options that should be twice the limit of the number of non-padding options. Per [RFC8504], the recommended default limit for number of non-padding options is eight, so this specification establishes a maximum default limit of sixteen options including padding options. The choice of sixteen options as a default limit attempts to strikes a balance between allowing extensibility and maintaining reasonable expectations for node processing requirements.¶
With regards to extensibility, it is observed that in the almost thirty year history of IPv6 there are only thirteen defined non-deprecated Destination options and Hop-by-Hop options and three temporary assigned options. Current evidence suggests that having more than one Destination option or Hop-by-Hop option in a packet is rare, and extrapolating that point with the rate of new options being defined suggests a limit of eight non-padding options allows for sufficient extensibility in the foreseeable future.¶
With regards to processing requirements, TLVs, such as Hop-by-Hop options and Destination options, have historically been considered difficult to process efficiently due to their serial processing requirements and combinatorial nature. TLV processing has been a particularly acute problem for ASIC based hardware devices. Recently, there is a strong trend in programmable implementation even in high performance routers using emerging programming frameworks such as PANDA and P4. Programmable implementations are better equipped to handle TLVs, at least for a reasonably small number. It might also be pointed out that the need to efficiently process TLVs exists in other protocols, for instance processing TCP requires processing of TLVs in the form of TCP options which are an intrinsic part of the protocol.¶
[RFC8200] defines PAD1 and PADN options that respectively provide one byte or N bytes of padding in an extension header. The purpose of padding is to properly align the following non-padding option to its expected alignment, or to add padding after the last Destination or Hop-by-Hop option so that the length of the extension header is a multiple of eight bytes as required by [RFC8200]. [RFC8504] defines limits on number of bytes used for consecutive padding where the amount of padding between options or at the end of the extension header is no more than seven bytes; this limit is sufficient to align any following data after the padding to eight bytes. These limits are extended to be applicable to intermediate nodes.¶
This specification allows a receiving node to set a requirement that consecutive padding options are not present in a packet; which in turn requires a sender to not place consecutive padding options in a packet. The rationale for this limit is that a PAD1 or PADN option is able to provide one to 257 bytes of padding, so a single padding option is sufficient for any expected use case of padding. When the sender creates options, it can compute the amount of padding necessary to satisfy the alignment requirements of the following data. If one byte of padding is needed a PAD1 option is used, if more than one byte of padding is needed then an appropriate PADN option is used.¶
Intermediate nodes often perform deep packet inspection (DPI) in order to implement various functions in the network. Routers perform DPI when they inspect packets beyond the IPv6 header or beyond Hop-by-Hop options if they are present. Some router implementations must inspect the transport layer headers in order to process and forward the packet, and if the transport layer headers are not readable a packet might be dropped. Even if a transport layer header is in plain text within a packet, some devices may not be capable of reading it if the header is too deep in the packet.¶
Hardware devices often have constraints on how much of the headers in a packet can be parsed for DPI. A typical design is that some portion of the beginning of a received packet is loaded into a memory buffer for header parsing (i.e. the parsing buffer). The size of this parsing buffer is often fixed per device.¶
To derive a size limit on the IPv6 header chain, we need to take into account headers in a packet that might be subject to DPI which include the link layer header through at least the pertinent fields of the transport layer header. The most common required information is the transport layer port numbers which typically occupy the first four bytes of the transport headers (in TCP, UDP, SCTP, DCCP, etc.). Inspection of port numbers may be needed for stateless load balancing as well as port filtering. There are middleboxes that may need to inspect more of transport layer headers or the transport payload, however those can be considered specialized devices that perform work beyond simple packet forwarding and filtering and hence should have more capabilities for DPI.¶
In addition to limits on the length of the IP header chain, it is conceivable that there could be a limit on the length of the whole header chain. The whole header chain would comprise the IPv6 header chain as well as any headers that are part of network encapsulation that precede the innermost transport layer. The definition of such a limit is out of scope for this document, however [RFC8883] defines an ICMP error to send when a limit on size of an aggregate header chain is exceeded.¶
This document specifies that the minimum supported limit for IPv6 header chains is 104 bytes. The value is derived by assuming that nodes have the ability to process at least the first 128 bytes of a packet (that is they have a parsing buffer that can contain at least 128 bytes). The 128 byte parsing buffer would be expected to at least contain:¶
This scheme thus establishes a requirement that all Internet devices are capable of correctly processing packets with up to sixty-four bytes of extension headers, and subsequently it establishes a requirement that a host shouldn't send packets with more than sixty-four bytes of extension headers. Note that this establishes a global baseline requirement across the Internet, within a limited domain higher limits could be applied.¶
128 bytes is likely the minimal useful parsing buffer size in deployment today. Devices performing a very narrow DPI could conceptually use a smaller parsing buffer, for instance that could be as small as sixty-four bytes which accommodates an L2 header, IPv6 header, and eight bytes of transport header; however, such a device would be extremely limited in capabilities and if they do exist they are likely legacy devices that will eventually be decommissioned. Many routers now have the capability to perform DPI into encapsulation headers which implies they already have a larger parsing buffer than this baseline minimum.¶
Similar to limiting the number of options allowing in a packet, setting a limit for IP header length chain is a tradeoff between extensibility and feasible implementation.¶
For extensibility, the pertinent extension headers contributing to the sixty-four byte limit are mostly Hop-by-Hop and Destination options. The Routing Header extension header is really intended for limited domains and not the Internet (for instance, the SRv6 Routing Header is confined to a Segment Routing Domain) and therefore would be subject to a domain specific limit for IP header chain length. Encryption Header may be used on the Internet, however encryption obfuscates the encapsulated transport headers such that such that intermediate nodes can't inspect them regardless of their position in a packet. Fragmentation may be used in the Internet, however only the first fragment of a fragmented packet might contain transport layer headers that could be read by an Intermediate node. In any case, the Fragment Header is only four bytes so that would not be a particularly large portion of a sixty-four byte limit.¶
The Authentication Header is usable on the Internet and does allow the transport layer headers to be in readable in plain text. The Authentication Header is relatively large, typically thirty-two bytes or more, so it would contribute significantly to a limit on IP header chain length; however, the use of Authentication Header without encryption is currently rare on the Internet.¶
Individual Hop-by-Hop Destination Options may also be categorized as being intended for use over the Internet or just in limited domains. For instance, the IOAM Hop-by-Hop option is intended for use in limited domains.¶
Paring this down, the types extension headers and Destination and Hop-by-Hop options that might be used outside of limited domains are fairly limited. Options that are intended for use over the public Internet could be defined to be small and compact to promote not exceeding a sixty-four byte limit on extension headers, whereas options constrained to a limited domain could be larger since larger limits can be assumed.¶
For each limit that is defined, an action is specified for when the limit is exceeded. The appropriate action depends on whether the processing node is the destination host, an intermediate destination, or an intermediate node. For a destination host, the typical action to take when a limit is exceeded is to discard the packet. This is appropriate since the destination host is required to process all of the headers in a packet, and if a limit is exceeded then it cannot process the packet so there is no other alternative but to discard.¶
For intermediate nodes, the typical action to take when a limit is exceeded is to stop processing headers at the point the limit is reached and to forward the packet on. [RFC8200] allows that an intermediate node may not process the Hop-by-Hop Options extension headers, therefore an intermediate node may ignore all of the Hop-by-Hop options in a packet. This specification expands on that requirement to allow an intermediate node to process some arbitrary subset of consecutive Hop-by-Hop options in the TLV list and to ignore the following ones. In the case of an egregious violation of a limit, for instance an attacker sends three hundred options in a packet, the destination host can decide if the appropriate response is to drop (the destination host must process all options). Note that this provision motivates the sender to place Hop-by-Hop Options in the packet so that those considered more important are placed first. It should also be noted that [RFC8504] sets a default limit of eight; this specification adds a counterpart for sending hosts that they shouldn't send more than eight Hop-by-Hop options.¶
Intermediate destinations have characteristics of both hosts and intermediate modes. If a limit is exceeded related to Hop-by-Hop options then the suggested action in this specification is to assume the same processing of limits as intermediate nodes. If limits are exceeded that affect the processing specific to an intermediate destination, such as limits on Destination options before the Routing header, then the action should be to discard packet.¶
This section lists the normative requirements related to sending and processing extension headers.¶
The set of limits that a node may apply when processing extension headers include:¶
The requirements are:¶
Per [RFC8200], a host node that receives a packet with an extension headers must process all the extension headers in the packet before accepting the payload and processing the payload.¶
As described in [RFC8504] a host may establish limits on the processing of extension headers. This specification reiterates and updates those requirements to allow for a host to send an RFC8883 error if a limit has been exceeded.¶
Additional host requirements for receive.¶
The following requirements are established for intermediate nodes and intermediate destination nodes that receive and process packets with extension header.¶
The following are requirements specific to intermediate destinations pertaining to the processing of Destination Options before the Routing header. For processing Hop-by-Hop options at an intermediate destination the requirements for processing them at an intermediate node are assumed.¶