| Internet-Draft | avoid-fragmentation | December 2023 |
| Fujiwara & Vixie | Expires 14 June 2024 | [Page] |
The widely deployed EDNS0 feature in the DNS enables a DNS receiver to indicate its received UDP message size capacity which supports the sending of large UDP responses by a DNS server. Large DNS/UDP responses are fragmented, and IP fragmentation has exposed weaknesses in application protocols. It is possible to avoid IP fragmentation in DNS by limiting response size where possible, and signaling the need to upgrade from UDP to TCP transport where necessary. This document specifies techniques to avoid IP fragmentation in DNS.¶
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 14 June 2024.¶
Copyright (c) 2023 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.¶
DNS has an EDNS0 [RFC6891] mechanism. The widely deployed EDNS0 feature in the DNS enables a DNS receiver to indicate its received UDP message size capacity which supports the sending of large UDP responses by a DNS server. DNS over UDP relies on IP fragmentation when the EDNS buffer size is set to a value larger than the path MTU.¶
Fragmented DNS UDP responses have systemic weaknesses, which expose the requestor to DNS cache poisoning from off-path attackers. (See Section 7.3 for references and details.)¶
[RFC8900] summarized that IP fragmentation introduces fragility to Internet communication. The transport of DNS messages over UDP should take account of the observations stated in that document.¶
TCP avoids fragmentation by segmenting data into packets that are smaller than or equal to the Maximum Segment Size (MSS). As for each transmitted segment, the size of the IP and TCP headers is known, and the IP packet size can be chosen to keep it below the other end's MSS and path MTU. This takes advantage of the elasticity of TCP's packetizing process as to how much queued data will fit into the next segment. In contrast, DNS over UDP has little datagram size elasticity and lacks insight into IP header and option size, and so must make more conservative estimates about available UDP payload space.¶
This document specifies various techniques to avoid IP fragmentation of UDP packets in DNS. In contrast, a path MTU that deviates from the recommended value can be obtained through static configuration, server routing hints, or a future discovery protocol. However, addressing this falls outside the scope of this document and may be the subject of future specifications.¶
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 BCP14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶
"Requestor" refers to the side that sends a request. "Responder" refers to an authoritative server, recursive resolver or other DNS component that responds to questions. (Quoted from EDNS0 [RFC6891])¶
"Path MTU" is the minimum link MTU of all the links in a path between a source node and a destination node. (Quoted from [RFC8201])¶
In this document, the term "Path MTU discovery" includes both Classical Path MTU discovery [RFC1191], [RFC8201], and Packetization Layer Path MTU discovery [RFC8899].¶
Many of the specialized terms used in this document are defined in DNS Terminology [RFC8499].¶
These recommendations are intended for nodes with global IP addresses on the Internet. Private networks or local networks are out of the scope of this document.¶
The methods to avoid IP fragmentation in DNS are described below:¶
R1. UDP responders SHOULD send DNS responses without "Fragment header" [RFC8200] on IPv6.¶
R2. UDP responders MAY set IP "Don't Fragment flag (DF) bit" [RFC0791] on IPv4.¶
At the time of writing, most DNS server software did not set the DF bit for IPv4, and many OS kernel constraints make it difficult to set the DF bit in all cases. Best Current Practice documents should not specify what is currently impossible, so R2, which is setting the DF bit, is "MAY" rather than "SHOULD".¶
R3. UDP responders SHOULD compose response packets that fit in the minimum of the offered requestor's maximum UDP payload size [RFC6891], the interface MTU, and the RECOMMENDED maximum DNS/UDP payload size 1400.¶
R4. If the UDP responder detects an immediate error indicating that the UDP packet cannot be sent beyond the path MTU size (EMSGSIZE), the UDP responder MAY recreate response packets fit in path MTU size, or with the TC bit set.¶
R5. UDP responders SHOULD limit the response size when UDP responders are located on small MTU (<1500) networks.¶
The cause and effect of the TC bit are unchanged from EDNS0 [RFC6891].¶
R6. UDP requestors SHOULD limit the requestor's maximum UDP payload size to the RECOMMENDED size of 1400 or a smaller size.¶
R7. UDP requestors MAY drop fragmented DNS/UDP responses without IP reassembly to avoid cache poisoning attacks.¶
R8. DNS responses may be dropped by IP fragmentation. Upon a timeout, to avoid resolution failures, UDP requestors MAY retry using TCP or UDP with a smaller EDNS requestor's maximum UDP payload size per local policy.¶
Large DNS responses are typically the result of zone configuration. Zone operators SHOULD seek configurations resulting in small responses. For example,¶
R9. Use a smaller number of name servers (13 may be too large)¶
R10. Use a smaller number of A/AAAA RRs for a domain name¶
R11. Use minimal-responses configuration: Some implementations have a 'minimal responses' configuration option that causes DNS servers to make response packets smaller, containing only mandatory and required data (Appendix B).¶
R12. Use a smaller signature / public key size algorithm for DNSSEC. Notably, the signature sizes of ECDSA and EdDSA are smaller than those usually used for RSA.¶
Prior research [Fujiwara2018] has shown that some authoritative servers ignore the EDNS0 requestor's maximum UDP payload size, and return large UDP responses.¶
It is also well known that some authoritative servers do not support TCP transport.¶
Such non-compliant behavior cannot become implementation or configuration constraints for the rest of the DNS. If failure is the result, then that failure must be localized to the non-compliant servers.¶
This document requests no IANA actions.¶
If the Don't Fragment (DF) bit is not set, on-path fragmentation may happen on IPv4, and be vulnerable as shown in Section 7.3. To avoid this, recommendation R7 should be used to discard the fragmented responses and retry by TCP.¶
In the future, recommendation R2 could be changed from "MAY" to "SHOULD".¶
When avoiding fragmentation, a DNS/UDP requestor behind a small-MTU network may experience UDP timeouts which would reduce performance and which may lead to TCP fallback. This would indicate prior reliance upon IP fragmentation, which is universally considered to be harmful to both the performance and stability of applications, endpoints, and gateways. Avoiding IP fragmentation will improve operating conditions overall, and the performance of DNS/TCP has increased and will continue to increase.¶
If a UDP response packet is dropped (for any reason), it increases the attack window for poisoning the requestor's cache.¶
"Fragmentation Considered Poisonous" [Herzberg2013] proposed effective off-path DNS cache poisoning attack vectors using IP fragmentation. "IP fragmentation attack on DNS" [Hlavacek2013] and "Domain Validation++ For MitM-Resilient PKI" [Brandt2018] proposed that off-path attackers can intervene in path MTU discovery [RFC1191] to perform intentionally fragmented responses from authoritative servers. [RFC7739] stated the security implications of predictable fragment identification values.¶
DNSSEC is a countermeasure against cache poisoning attacks that use IP fragmentation. However, DNS delegation responses are not signed with DNSSEC, and DNSSEC does not have a mechanism to get the correct response if an incorrect delegation is injected. This is a denial-of-service vulnerability that can yield failed name resolutions. If cache poisoning attacks can be avoided, DNSSEC validation failures will be avoided.¶
In Section 3.2 (Message Side Guidelines) of UDP Usage Guidelines [RFC8085] we are told that an application SHOULD NOT send UDP datagrams that result in IP packets that exceed the Maximum Transmission Unit (MTU) along the path to the destination.¶
A DNS message receiver cannot trust fragmented UDP datagrams primarily due to the small amount of entropy provided by UDP port numbers and DNS message identifiers, each of which being only 16 bits in size, and both likely being in the first fragment of a packet, if fragmentation occurs. By comparison, TCP protocol stack controls packet size and avoids IP fragmentation under ICMP NEEDFRAG attacks. In TCP, fragmentation should be avoided for performance reasons, whereas for UDP, fragmentation should be avoided for resiliency and authenticity reasons.¶
The author would like to specifically thank Paul Wouters, Mukund Sivaraman, Tony Finch, Hugo Salgado, Peter van Dijk, Brian Dickson, Puneet Sood, Jim Reid, Petr Spacek, Andrew McConachie, Joe Abley, Daisuke Higashi, Joe Touch and Wouter Wijngaards for extensive review and comments.¶
There are many discussions for default path MTU size and requestor's maximum UDP payload size.¶
The minimum MTU for an IPv6 interface is 1280 octets (see Section 5 of [RFC8200]). So, we can use it as the default path MTU value for IPv6. The corresponding minimum MTU for an IPv4 interface is 68 (60 + 8) [RFC0791].¶
Most of the Internet and especially the inner core has an MTU of at least 1500 octets. Maximum DNS/UDP payload size for IPv6 on MTU 1500 ethernet is 1452 (1500 minus 40 (IPv6 header size) minus 8 (UDP header size)). To allow for possible IP options and distant tunnel overhead, the authors' recommendation of default maximum DNS/UDP payload size is 1400.¶
[RFC4035] defines that "A security-aware name server MUST support the EDNS0 message size extension, MUST support a message size of at least 1220 octets". Then, the smallest number of the maximum DNS/UDP payload size is 1220.¶
In order to avoid IP fragmentation, [DNSFlagDay2020] proposed that the UDP requestors set the requestor's payload size to 1232, and the UDP responders compose UDP responses so they fit in 1232 octets. The size 1232 is based on an MTU of 1280, which is required by the IPv6 specification [RFC8200], minus 48 octets for the IPv6 and UDP headers.¶
[Huston2021] analyzed the result of [DNSFlagDay2020] and reported that their measurements suggest that in the interior of the Internet between recursive resolvers and authoritative servers the prevailing MTU is at 1,500 and there is no measurable signal of use of smaller MTUs in this part of the Internet, and proposed that their measurements suggest setting the EDNS0 requestor's UDP payload size to 1472 octets for IPv4, and 1452 octets for IPv6.¶
Some implementations have a "minimal responses" configuration setting/option that causes a DNS server to make response packets smaller, containing only mandatory and required data.¶
Under the minimal-responses configuration, a DNS server composes responses containing only nessesary RRs. For delegations, see [RFC9471]. In case of a non-existent domain name or non-existent type, the authority section will contain an SOA record and the answer section is empty. (defined in Section 2 of [RFC2308]).¶
Some resource records (MX, SRV, SVCB, HTTTPS) require additional A, AAAA, and SVCB records in the Additional Section defined in [RFC1035], [RFC2782] and [I-D.ietf-dnsop-svcb-https].¶
In addition, if the zone is DNSSEC signed and a query has the DNSSEC OK bit, signatures are added in the answer section, or the corresponding DS RRSet and signatures are added in the authority section. Details are defined in [RFC4035] and [RFC5155].¶
(This section may be removed by the RFC editor.)¶
This section records the status of known implementations of these best practices defined by this specification at the time of publication, and any deviation from the specification.¶
Please note that the listing of any individual implementation here does not imply endorsement by the IETF. Furthermore, no effort has been spent to verify the information presented here that was supplied by IETF contributors.¶
BIND 9 does not implement the recommendations 1 and 2 in Section 3.1.¶
BIND 9 on Linux sets IP_MTU_DISCOVER to IP_PMTUDISC_OMIT with a fallback to IP_PMTUDISC_DONT.¶
BIND 9 on systems with IP_DONTFRAG (such as FreeBSD), IP_DONTFRAG is disabled.¶
Accepting PATH MTU Discovery for UDP is considered harmful and dangerous. BIND 9's settings avoid attacks to path MTU discovery.¶
For recommendation 3, BIND 9 will honor the requestor's size up to the
configured limit (max-udp-size). The UDP response packet is bound to be
between 512 and 4096 bytes, with the default set to 1232. BIND 9 supports the
requestor's size up to the configured limit (max-udp-size).¶
In the case of recommendation 4, and the send fails with EMSGSIZE, BIND 9 set the TC bit and try to send a minimal answer again.¶
In the first recommendation of Section 3.2, BIND 9 uses the edns-buf-size
option, with the default of 1232.¶
BIND 9 does implement recommendation 2 of Section 3.2.¶
For recommendation 3, after two UDP timeouts, BIND 9 will fallback to TCP.¶
Both Knot servers set IP_PMTUDISC_OMIT to avoid path MTU spoofing. UDP size limit is 1232 by default.¶
Fragments are ignored if they arrive over an XDP interface.¶
TCP is attempted after repeated UDP timeouts.¶
Minimal responses are returned and are currently not configurable.¶
Smaller signatures are used, with ecdsap256sha256 as the default.¶
Unbound sets IP_MTU_DISCOVER to IP_PMTUDISC_OMIT with fallback to IP_PMTUDISC_DONT. It also disables IP_DONTFRAG on systems that have it, but not on Apple systems. On systems that support it Unbound sets IPV6_USE_MIN_MTU, with a fallback to IPV6_MTU at 1280, with a fallback to IPV6_USER_MTU. It also sets IPV6_MTU_DISCOVER to IPV6_PMTUDISC_OMIT with a fallback to IPV6_PMTUDISC_DONT.¶
Unbound requests UDP size 1232 from peers, by default. The requestors size is limited to a max of 1232.¶
After some timeouts, Unbound retries with a smaller size, if that is smaller, at size 1232 for IPv6 and 1472 for IPv4. This does not do anything since the flag day change to 1232.¶
Unbound has minimal responses as an option, default on.¶