Internet-Draft | Path MTU Option | September 2021 |
Hinden & Fairhurst | Expires 3 April 2022 | [Page] |
This document specifies a new IPv6 Hop-by-Hop option that is used to record the minimum Path MTU along the forward path between a source host to a destination host. The recorded value can then be communicated back to the source using the return Path MTU field in the option.¶
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 3 April 2022.¶
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.¶
This draft proposes a new IPv6 Hop-by-Hop Option to record the minimum Maximum Transmission Unit (MTU) along the forward path between a source and a destination host. The source host creates a packet with this option and initializes the Min-PMTU field with the value of the MTU for the outbound link that will be used to forward the packet towards the destination host.¶
At each subsequent hop where the option is processed, the router compares the value of the Min-PMTU Field in the option and the MTU of its outgoing link. If the MTU of the link is less than the Min-PMTU, it rewrites the value in the option data with the smaller value. When the packet arrives at the destination host, the host can send the value of the minimum reported MTU for the path back to the source host using the Rtn-PMTU field in the option. The source host can then use this value as input to the method that sets the Path MTU (PMTU) used by upper layer protocols.¶
This method has the potential to complete Path MTU discovery in a single round trip time, even over paths that have successive links each with a lower MTU.¶
The figure below illustrates the operation of the method. In this case, the path between the source host and the destination host comprises three links, the sender has a link MTU of size MTU-S, the link between routers R1 and R2 has an MTU of size 9000 bytes, and the final link to the destination has an MTU of size MTU-D.¶
Three scenarios are described:¶
Scenario 1, considers all links to have an 9000 byte MTU and the method is supported by both routers. The initial Min-PMTU is not modified along the path, and therefore the PMTU is 9000 bytes.¶
Scenario 2, considers the link to the destination host (MTU-D) to have an MTU of 1500 bytes. This is the smallest MTU, router R2 updates the Min-PMTU to 1500 bytes and the method correctly updates the PMTU to 1500 bytes. Had there been another smaller MTU at a link further along the path that also supports the method, the lower MTU would also have been detected.¶
Scenario 3, considers the case where the router preceding the smallest link (R2) does not support the method, and the link to the destination host (MTU-D) has an MTU of 1500 bytes. Therefore, router R2 does not update the Min-PMTU to 1500 bytes. The method then fails to detect the actual PMTU.¶
In Scenarios 2 and 3, a lower PMTU would also fail to be detected in the case where PMTUD had been used and an ICMPv6 Packet Too Big (PTB) message had not been delivered to the sender [RFC8201].¶
These scenarios are summarized in the table below. "H" in R1 and/or R2 columns means the router understands the Minimum Path MTU Hop-by-Hop option.¶
IPv6 as specified in [RFC8200] allows nodes to optionally process the Hop-by-Hop header. Specifically, from Section 4:¶
The Hop-by-Hop Options header is not inserted or deleted, but may be examined or processed by any node along a packet's delivery path, until the packet reaches the node (or each of the set of nodes, in the case of multicast) identified in the Destination Address field of the IPv6 header. The Hop-by-Hop Options header, when present, must immediately follow the IPv6 header. Its presence is indicated by the value zero in the Next Header field of the IPv6 header.¶
NOTE: While [RFC2460] 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.¶
The Hop-by-Hop Option defined in this document is designed to take advantage of this property of how Hop-by-Hop options are processed. Nodes that do not support this Option SHOULD ignore them. This can mean that the Min-PMTU value does not account for all links along a path.¶
The current state of Path MTU Discovery on the Internet is problematic. The mechanisms defined in [RFC8201] are known to not work well in all environments. It fails to work in various cases, including when nodes in the middle of the network do not send ICMP PTB messages, or rate-limited ICMP messages, or do not have a return path to the source host.¶
This results in many transport connections being configured to use smaller packets (e.g., 1280 bytes) by default and makes it difficult to take advantage of paths with a larger PMTU where they do exist. Applications that send large packets are forced to use IPv6 Fragmentation [RFC8200], which can reduce the reliability of Internet communication [RFC8900].¶
Encapsulations and network-layer tunnels further reduce the payload size available for a transport to use. Also, some use-cases increase packet overhead, for example, Network Virtualization Using Generic Routing Encapsulation (NVGRE) [RFC7637] encapsulates L2 packets in an outer IP header and does not allow IP Fragmentation.¶
Sending larger packets can improve host performance, e.g., avoiding limits to packet processing by the packet rate. For example, the packet per second rate required to reach wire speed on a 10G link with 1280 byte packets is about 977K packets per second (pps), vs. 139K pps for 9000 byte packets.¶
The purpose of this draft is to improve the situation by defining a mechanism that does not rely on reception of ICMPv6 Packet Too Big messages from nodes in the middle of the network. Instead, this provides information to the destination host about the minimum Path MTU, and sends this information back to the source host. This is expected to work better than the current RFC8201-based mechanisms.¶
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.¶
The Path MTU option is designed for environments where there is control over the hosts and nodes that connect them, and where there is more than one MTU size in use. For example, in Data Centers and on paths between Data Centers, to allow hosts to better take advantage of a path that is able to support a large PMTU.¶
The design of the option is sufficiently simple that it can be executed on a router's fast path. A successful experiment depends on both implementation by host and router vendors and deployment by operators. The contained use-case of connections within and between Data Centers could be a driver for deployment.¶
The method could also be useful in other environments, including the general Internet, and offers advantage when this Hop-by-Hop Option is supported on all paths. The method is more robust when used to probe the path using packets that do not carry application data and when also paired with a method such as Packetization Layer PMTUD [RFC4821] or Datagram PLPMTUD [RFC8899].¶
The Minimum Path MTU Hop-by-Hop Option has the following format:¶
NOTE: The encoding of the final two octets (Rtn-PMTU and R-Flag) could be implemented by a mask of the latest received Min-PMTU value with 0xFFFE, discarding the right-most bit and then performing a logical 'OR' with the R-Flag value of the sender. This encoding fits in the minimum-sized HBH Option header.¶
Routers that are not configured to support Hop-by-Hop Options SHOULD ignore this option and SHOULD forward the packet [RFC8200].¶
Routers that support Hop-by-Hop Options, but that are not configured to support this option SHOULD ignore the option and SHOULD forward the packet.¶
Routers that support this option SHOULD compare the value of the Min-PMTU field with the MTU configured for the outgoing link. If the MTU of the outgoing link is less than the Min-PMTU, the router rewrites the Min-PMTU in the Option to use the smaller value. (The router processing is performed without checking the valid range of the Min-PMTU or the Rtn-PMTU fields.)¶
A router MUST ignore and MUST NOT change the Rtn-PMTU field or the R-Flag in the option.¶
Discussion:¶
The design of this option makes it feasible to be implemented within the fast path of a router, because the processing requirements are minimal.¶
The PMTU entry associated with the destination in the host's Destination cache [RFC4861] SHOULD be updated after detecting a change using the IPv6 Minimum Path MTU Hop-by-Hop Option. This cached value can be used by other flows that share the host's Destination cache.¶
The value in the host Destination cache SHOULD be used by PLPMTUD to select an initial PMTU for a flow. The cached PMTU is only increased by PLPMTUD when the PL determines the path actually supports a larger PMTU [RFC4821] [RFC8899].¶
When requested to send an IPv6 packet with the Minimum Path MTU option, the source host includes the option in an outgoing packet. The source host MUST fill the Min-PMTU field with the MTU configured for the link over which it will send the packet on the next hop towards the destination host.¶
It sets the R Flag if it wishes the remote host to return the discovered PMTU value.¶
When a host includes the option in a packet it sends, the host SHOULD set the Rtn-PMTU field to the previously cached value of the received Minimum Path MTU for the flow in the Rtn-PMTU field (see Section 6.2.2.3). If this value is not set (for example, because there is no cached reported Min-PMTU value), the Rtn-PMTU field value MUST be set to zero.¶
The source host MAY request the destination host to return the reported Min-PMTU value by setting the R-Flag in the option of an outgoing packet. The R-Flag SHOULD NOT be set when the Minimum Path MTU Option was sent solely to provide requested feedback on the return Path MTU to avoid each response generating another response.¶
The destination host controls when to send a packet with this option in response to a R-flag, as well as which packets to include it in. The destination host MAY limit the rate at which it sends these packets.¶
The normal sequence of operation of the R-Flag using the terminology from the diagram in Figure 1 is:¶
This Hop-by-Hop option is intended to be used with a path MTU discovery method.¶
Section 4.1 of [RFC9000] describes different types of PMTU Probe, depending on whether the probe packets carry application data. When the path is expected to support use of the option, the PMTU Probe can be sent on packets that include application data, but needs to be robust to potential loss of the packet with the possibility that retransmission might be needed. Using a PMTU Probe on packets that do not carry application data will avoid the need for loss recovery if a router on the path later drops packets that set this option. This avoids the transport needing to retransmit a lost packet that includes this option.¶
The upper layer protocol can request the Minimum Path MTU option to be included in an outgoing IPv6 packet. A transport protocol (or upper layer protocol) can include this option only on specific packets used to test the path. This option does not need to be included in all packets belonging to a flow.¶
NOTE: Including this option in a large packet (e.g., one larger than the present PMTU) is not likely to be useful, since the large packet would itself be dropped by any link along the path with a smaller MTU, preventing the Min-PMTU information from reaching the destination host.¶
Discussion:¶
In the case of TCP, the option could be included in a packet that carries a TCP segment sent after the connection is established. A segment without data could be used, to avoid the need to retransmit this data if the probe packet is lost. The discovered value can be used to inform PLPMTUD [RFC4821].¶
NOTE: A TCP SYN can also negotiate the Maximum Segment Size (MSS), which acts as an upper limit to the packet size that can be sent by a TCP sender. If this option were to be included in a TCP SYN, it could increase the probability that the SYN segment is lost when routers on the path drop packets with this option (see Section 6.2.2.6), which could have an unwanted impact on the result of racing options [I-D.ietf-taps-arch] or feature negotiation.¶
The use with datagram transport protocols (e.g., UDP) is harder to characterize because applications using datagram transports range from very short-lived (low data-volume applications) exchanges, to longer (bulk) exchanges of packets between the source and destination hosts [RFC8085].¶
Simple-exchange protocols (i.e., low data-volume applications [RFC8085] that only send one or a few packets per transaction), might assume that the PMTU is symmetrical. That is, the PMTU is the same in both directions, or at least not smaller for the return path. This optimization does not hold when the paths are not symmetric.¶
A datagram transport can utilise DPLPMTUD [RFC8899]. For example, QUIC (see section 14.3 of [RFC9000]), can use DPLPMTUD to determine whether the path to a destination will support a desired maximum datagram size. When using the IPv6 MinPMTU HBH option, the option could be added to an additional QUIC PMTU Probe that is of minimal size (or one no larger than the currently supported PMTU size). Once the return Path MTU value in the Min PMTU HBH option has been learned, DPLPMTUD can be triggered to test for a larger PLPMTU using an appropriately sized PLPMTU Probe Packet (see section 5.3.1 of [RFC8899]).¶
The use of this option with DNS and DNSSEC over UDP is expected to work for paths where the PMTU is symmetric. The DNS server will learn the PMTU from the DNS query messages. If the Rtn-PMTU value is smaller, then a large DNSSEC response might be dropped and the known problems with PMTUD will then occur. DNS and DNSSEC over transport protocols that can carry the PMTU ought to work.¶
This method also can be used with Anycast to discover the PMTU of the path, but the use needs to be aware that the Anycast binding might change.¶
An upper layer protocol (e.g., transport endpoint) using this option needs to provide protection from data injection attacks by off-path devices [RFC8085]. This requires a method to assure that the information in the Option Data is provided by a node on the path. This validates that the packet forms a part of an existing flow, using context available at the upper layer. For example, a TCP connection or UDP application that maintains the related state and uses a randomized ephemeral port would provide this basic validation to protect from off-path data injection, see Section 5.1 of [RFC8085]. IPsec [RFC4301] and TLS [RFC8446] provide greater assurance.¶
The upper layer discards any received packet when the packet validation fails. When packet validation fails, the upper layer MUST also discard the associated Option Data from the minimum Path MTU option without further processing.¶
For a connection-oriented upper layer protocol, caching of the received Min-PMTU could be implemented by saving the value in the connection context at the transport layer. A connection-less upper layer (e.g., one using UDP), requires the upper layer protocol to cache the value for each flow it uses.¶
A destination host that receives a Minimum Path MTU Option with the R-Flag SHOULD include the Minimum Path MTU option in the next outgoing IPv6 packet for the corresponding flow.¶
A simple mechanism could only include this option (with the Rtn-PMTU field set) the first time this option is received or when it notifies a change in the Minimum Path MTU. This limits the number of packets including the option packets that are sent. However, this does not provide robustness to packet loss or recovery after a sender loses state.¶
Discussion:¶
Some upper layer protocols send packets less frequently than the rate at which the host receives packets. This provides less frequent feedback of the received Rtn-PMTU value. However, a host always sends the most recent Rtn-PMTU value.¶
The Rtn-PMTU field provides an indication of the PMTU from on-path routers. It does not necessarily reflect the actual PMTU between the sender and destination. Care therefore needs to be exercised in using the Rtn-PMTU value. Specifically:¶
IPv6 requires that every link in the Internet have an MTU of 1280 octets or greater. A node MUST ignore a Rtn-PMTU value less than 1280 octets [RFC8200].¶
To avoid unintentional dropping of packets that exceed the actual PMTU (e.g., Scenario 3 in Section 1.1), the source host can delay increasing the PMTU until a probe packet with the size of the Rtn-PMTU value has been successfully acknowledged by the upper layer, confirming that the path supports the larger PMTU. This probing increases robustness, but adds one additional path round trip time before the PMTU is updated. This use resembles that of PTB messages in section 4.6 of DPLPMTUD [RFC8899] (with the important difference that a PTB message can only seek to lower the PMTU, whereas this option could trigger a probe packet to seek to increase the PMTU.)¶
Section 5.2 of [RFC8201] provides guidance on the caching of PMTU information and also the relation to IPv6 flow labels. Implementations should consider the impact of Equal Cost Multipath (ECMP) [RFC6438]. Specifically, whether a PMTU ought to be maintained for each transport endpoint, or for each network address.¶
Path characteristics can change and the actual PMTU could increase or decrease over time. For instance, following a path change when packets are forwarded over a link with a different MTU than that previously used. To bound the delay in discovering an increase in the actual PMTU, a host with a link MTU larger than the current PMTU SHOULD periodically send the Minimum Path MTU Option with the R-bit set. DPLPMTUD provides recommendations concerning how this could be implemented (see Section 5.3 of [RFC8899]). Since the option consumes less capacity than a full-sized probe packet, there can be advantage in using this to detect a change in the path characteristics.¶
There is evidence that some middleboxes drop packets that include Hop-by-Hop options. For example, a firewall might drop a packet that carries an unknown extension header or option. This practice is expected to decrease as an option becomes more widely used. It could result in generation of an ICMPv6 message indicating the problem. This could be used to (temporarily) suspend use of this option.¶
A middlebox that silently discards a packet with this option results in dropping of any packet using the option. This dropping can be avoided by appropriate configuration in a controlled environment, such as within a data centre, but needs to be considered for Internet usage. Section 6.2 recommends that this option is not used on packets where loss might adversely impact performance.¶
IANA has assigned and registered an IPv6 Hop-by-Hop Option type with Temporary status from the "Destination Options and Hop-by-Hop Options" registry [IANA-HBH]. This assignment is shown in Section 5.¶
IANA is requested to update this registry to point to this document and remove the Temporary status.¶
This section discusses the security considerations. It first reviews router option processing. It then reviews host processing when receiving this option at the network layer. It then considers two ways in which the Option Data can be processed, followed by two approaches for using the Option Data. Finally, it discusses middlebox implications related to use in the general Internet.¶
This option shares the characteristics of all other IPv6 Hop by Hop Options, in that if not supported at line rate it could be used to degrade the performance of a router. This option, while simple, is no different to other uses of IPv6 Hop-by-Hop options.¶
It is common for routers to ignore the Hop-by-Hop Option header or drop packets containing a Hop-by-Hop Option header. Routers implementing IPv6 according to [RFC8200] only examine and process the Hop-by-Hop Options header if explicitly configured to do so.¶
A malicious attacker can forge a packet directed at a host that carries the minimum Path MTU option. By design, the fields of this IP option can be modified by the network.¶
For comparison, the ICMPv6 Packet Too Big message used in [RFC8201] Path MTU Discovery, the source host has an inherent trust relationship with the destination host including this option. This trust relationship can be used to help verify the option. ICMPv6 Packet Too Big messages are sent from any router on the path to the destination host, the source host has no prior knowledge of these routers (except for the first hop router).¶
Reception of this packet will require processing as the network stack parses the packet before the packet is delivered to the upper layer protocol. This network layer option processing is normally completed before any upper layer protocol delivery checks are performed.¶
The network layer does not normally have sufficient information to validate that the packet carrying an option originated from the destination (or an on-path node). It also does not typically have sufficient context to demultiplex the packet to identify the related transport flow. This can mean that any changes resulting from reception of the option applies to all flows between a pair of endpoints.¶
These considerations are no different to other uses of Hop-by-Hop options, and this is the use case for PMTUD. The following section describes a mitigation for this attack.¶
Transport protocols should be designed to provide protection from data injection attacks by off-path devices and mechanisms should be described in the Security Considerations for each transport specification (see Section 5.1 of the UDP Guidelines [RFC8085]). For example, a TCP or UDP application that maintains the related state and uses a randomized ephemeral port would provide basic protection. TLS [RFC8446] or IPsec [RFC4301] provide cryptographic authentication. An upper layer protocol that validates each received packet discards any packet when this validation fails. In this case, the host MUST also discard the associated Option Data from the minimum Path MTU option without further processing (Section 6.2.2).¶
A network node on the path has visibility of all packets it forwards. By observing the network packet payload, the node might be able to construct a packet that might be validated by the destination host. Such a node would also be able to drop or limit the flow in other ways that could be potentially more disruptive. Authenticating the packet, for example, using IPsec [RFC4301] or TLS [RFC8446] mitigates this attack.¶
The simplest way to utilize the Rtn-PMTU value is to directly use this to update the PMTU. This approach results in a set of security issues when the option carries malicious data:¶
A direct update of the PMTU using the Rtn-PMTU value could result in an attacker inflating or reducing the size of the host PMTU for the destination. Forcing a reduction in the PMTU can decrease the efficiency of network use, might increase the number of packets/fragments required to send the same volume of payload data, and prevents sending an unfragmented datagram larger than the PMTU. Increasing the PMTU can result in black-holing (see Section 1.1 of [RFC8899]) when the source sends packets larger than the actual PMTU. This persists until the PMTU is next updated.¶
The method can be used to solicit a response from the destination host. A malicious attacker could forge a packet that causes the sender to add the option to a packet sent to the source. A forged value of Rtn-PMTU in the Option Data might also impact the remote endpoint, as described in the previous bullet. This persists until a valid minimum Path MTU option is received. This attack could be mitigated by limiting the sending of the minimum Path MTU option in reply to incoming packets that carry the option.¶
Another way to utilize the Rtn-PMTU value is to indirectly trigger a probe to determine if the path supports a PMTU of size Rtn-PMTU. This approach needs context for the flow, and hence assumes an upper layer protocol that validates the packet that carries the option (see Section 8.3). This is the case when used in combination with DPLPMTUD [RFC8899]. A set of security considerations result when an option carries malicious data:¶
There is evidence that some middleboxes drop packets that include Hop-by-Hop options. For example, a firewall might drop a packet that carries an unknown extension header or option. This practice is expected to decrease as the option becomes more widely used. Methods to address this are discussed in Section 6.2.2.6.¶
When a forged packet cause a packet to be sent including the minimum Path MTU option, and the return path does not forward packets with this option, the packet will be dropped Section 6.2.2.6. This attack is mitigated by validating the option data before use and by limiting the rate of responses generated. An upper layer could further mitigate the impact by responding to a R-Flag by including the option in a packet that does not carry application data.¶
This section describes the experimental goals of this specification.¶
A successful deployment of the method depends upon several components being implemented and deployed:¶
Experience from deployment is an expected input to any decision to progress this specification from Experimental to IETF Standards Track. Appropriate inputs might include:¶
At the time this document was published there are two known implementations of the Path MTU Hop-by-Hop option. These are:¶
A similar mechanism was proposed for IPv4 in 1988 in [RFC1063] by Jeff Mogul, C. Kent, Craig Partridge, and Keith McCloghire. It was later obsoleted in 1990 by [RFC1191] the current deployed approach to Path MTU Discovery.¶
Helpful comments were received from Tom Herbert, Tom Jones, Fred Templin, Ole Troan, Tianran Zhou, Jen Linkova, Brian Carpenter, Peng Shuping, Mark Smith, Fernando Gont, Michael Dougherty, Erik Kline, and other members of the 6MAN working group.¶
draft-ietf-6man-mtu-option-11, 2021-September-30¶
draft-ietf-6man-mtu-option-10, 2021-September-27¶
draft-ietf-6man-mtu-option-09, 2021-September-23¶
draft-ietf-6man-mtu-option-08, 2021-September-7¶
draft-ietf-6man-mtu-option-07, 2021-August-31¶
draft-ietf-6man-mtu-option-06, 2021-August-7¶
draft-ietf-6man-mtu-option-05, 2021-April-28¶
draft-ietf-6man-mtu-option-04, 2020-Oct-23¶
draft-ietf-6man-mtu-option-03, 2020-Sept-14¶
draft-ietf-6man-mtu-option-02, 2020-March-9¶
draft-ietf-6man-mtu-option-01, 2019-September-13¶
draft-ietf-6man-mtu-option-00, 2019-August-9¶
draft-hinden-6man-mtu-option-02, 2019-July-5¶
draft-hinden-6man-mtu-option-01, 2019-March-05¶
draft-hinden-6man-mtu-option-00, 2018-Oct-16¶