Network Working Group | S. Cheshire |
Internet-Draft | Apple Inc. |
Updates: 7050 (if approved) | D. Schinazi |
Intended status: Standards Track | Google LLC |
Expires: May 8, 2019 | November 4, 2018 |
Special Use Domain Name 'ipv4only.arpa'
draft-cheshire-sudn-ipv4only-dot-arpa-14
The specification for how a client discovers its local network's NAT64 prefix [RFC7050] defines the special name 'ipv4only.arpa' for this purpose, but in its Domain Name Reservation Considerations section that specification indicates that the name actually has no particularly special properties would require special handling, and does not request IANA to record the name in the Special-Use Domain Names registry.
Consequently, despite the well articulated special purpose of the name, 'ipv4only.arpa' was not recorded in the Special-Use Domain Names registry as a name with special properties.
As a result of this omission, in cases where software needs to give this name special treatment in order for it to work correctly, there was no clear mandate authorizing software authors to implement that special treatment. Software implementers were left with the choice between not implementing the special behavior necessary for the name queries to work correctly, or implementing the special behavior and being accused of being noncompliant with some RFC.
This document describes the special treatment required, formally declares the special properties of the name, and adds similar declarations for the corresponding reverse mapping names.
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 May 8, 2019.
Copyright (c) 2018 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.
The specification for how a client discovers its local network's NAT64 prefix defines the special name 'ipv4only.arpa' for this purpose, but in its Domain Name Reservation Considerations section that specification indicates that the name actually has no particularly special properties would require special handling, and does not request IANA to record the name in the Special-Use Domain Names registry.
Consequently, despite the well articulated special purpose of the name, 'ipv4only.arpa' was not recorded in the Special-Use Domain Names registry as a name with special properties.
This omission was discussed in the Special-Use Domain Names Problem Statement.
As a result of this omission, in cases where software needs to give this name special treatment in order for it to work correctly, there was no clear mandate authorizing software authors to implement that special treatment. Software implementers were left with the choice between not implementing the special behavior necessary for the name queries to work correctly, or implementing the special behavior and being accused of being noncompliant with some RFC.
This document describes the special treatment required, formally declares the special properties of the name, and adds similar declarations for the corresponding reverse mapping names.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY",
and "OPTIONAL" in this section are to be interpreted as described
in "Key words for use in RFCs to Indicate Requirement Levels",
when, and only when, they appear in all capitals, as shown here
[RFC2119] [RFC8174].
The hostname 'ipv4only.arpa' is peculiar in that it was never intended to be treated like a normal hostname.
A typical client never looks up the IPv4 address records for 'ipv4only.arpa', because it is already known, by specification, to have exactly two IPv4 address records, 192.0.0.170 and 192.0.0.171. No client ever has to look up the name in order to learn those two addresses.
In contrast, clients often look up the IPv6 AAAA address records for 'ipv4only.arpa', which is contrary to general DNS expectations, given that it is already known, by specification, that 'ipv4only.arpa' is an IPv4-only name, which has no IPv6 AAAA address records. And yet, clients expect to receive, and do in fact receive, positive answers for these IPv6 AAAA address records that apparently should not exist.
This is clearly not a typical DNS name. In normal operation, clients never query for the two records that do in fact exist; instead clients query for records that are known to not exist, and then get positive answers to those abnormal queries. Clients are using DNS to perform queries for this name, but they are certainly not using DNS to learn legitimate answers from the name's legitimate authoritative server. Instead, the DNS protocol has, in effect, been co-opted as an impromptu client-to-middlebox communication protocol, to communicate with the NAT64/DNS64 [RFC6146] [RFC6147] gateway, if present, and request that it disclose the prefix it is using for IPv6 address synthesis.
This use of specially-crafted DNS queries as an impromptu client-to-middlebox communication protocol has a number of specific consequences, outlined below, which client software needs to take into account if the queries are to produce the desired results, particularly when used on a multi-homed host, or when a VPN tunnel is in use. The name 'ipv4only.arpa' is most definitely a special name, and needs to be listed in IANA's registry along with other DNS names that have special uses.
As a result of the original specification not formally declaring 'ipv4only.arpa' to have special properties, there was no clear mandate for DNS software to treat this name specially. In particular, this lack of mandate for special treatment is relevant (a) to the name resolution APIs and libraries on client devices, and (b) to DNS64 [RFC6147] implementations. These two aspects are discussed in more detail below.
A serious problem can occur with NAT64/DNS64 when a device is configured to use a recursive resolver other than the one it learned from the network.
Typically a device joining a NAT64 network will learn the recursive resolver recommended for that network either via IPv6 Router Advertisement Options for DNS Configuration or via DNS Configuration options for DHCPv6. On a NAT64 network it is essential that the client use the DNS64 recursive resolver recommended for that network, since only that recursive resolver can be relied upon to know the appropriate prefix(es) to use for synthesizing IPv6 addresses that will be acceptable to the NAT64 gateway.
However, it is becoming increasingly common for users to manually override their default DNS configuration because they wish to use some other public recursive resolver on the Internet, which may offer better speed, better reliability, or better privacy than the local network's default recursive resolver. At the time of writing, examples of widely known public recursive resolver services include 1.1.1.1, 8.8.8.8, and 9.9.9.9.
Another common scenario is the use of corporate VPN client software. The local network's recursive resolver will typically be unable to provide answers for the company's private internal host names, so VPN client software overrides the local network's default configuration, to divert some or all DNS requests to the company's own private internal recursive resolver, reached through the VPN tunnel. As with the case described above of public recursive resolver services, the company's private internal recursive resolver cannot be expected to be able to synthesize IPv6 addresses correctly for use with the local network's NAT64 gateway, because the company's private internal recursive resolver is unlikely to be aware of the NAT64 prefix in use on the NAT64 network to which the client device is currently attached. It is clear that a single recursive resolver cannot meet both needs. The local network's recursive resolver cannot give answers for some company's private internal host names, and some company's private internal recursive resolver cannot give correctly synthesized IPv6 addresses suitable for the local network's NAT64 gateway.
The conflict here arises because DNS is being used for two unrelated purposes. The first purpose is retrieving data from a (nominally) global database -- generally retrieving the IP address(es) associated with a hostname. The second purpose is using the DNS protocol as a middlebox communication protocol, to interrogate the local network infrastructure to discover the IPv6 prefix(es) in use by the local NAT64 gateway for address synthesis.
As a result of there being no mandate for special treatment, queries for 'ipv4only.arpa' had to be handled normally, resulting in DNS64 gateways performing unnecessary IPv6 address record queries (DNS qtype "AAAA", always returning negative responses) and IPv4 address record queries (DNS qtype "A", always returning the same positive responses) to the authoritative 'arpa' name servers.
Having DNS64 gateways around the world issue these queries generated additional load on the authoritative 'arpa' name servers, which was redundant when the name 'ipv4only.arpa' is defined, by Internet Standard, to have exactly two IPv4 address records, 192.0.0.170 and 192.0.0.171, and no other IPv4 or IPv6 address records.
Also, at times, for reasons that remain unclear, the authoritative 'arpa' name servers have been observed to be slow or unresponsive. The failures of these 'ipv4only.arpa' queries result in unnecessary failures of the DNS64 gateways and of the client devices that depend on them for DNS64 address synthesis.
Even when the authoritative 'arpa' name servers are operating correctly, having to perform an unnecessary query to obtain an answer that is already known in advance can add precious milliseconds of delay, affecting user experience on the client devices waiting for those synthesized replies.
This document leverages operational experience to update the Domain Name Reservation Considerations section of the earlier specification with one that more accurately lists the actual special properties of the name 'ipv4only.arpa', so that software can legitimately implement the correct behavior necessary for better performance, better reliability, and correct operation.
These changes affect two bodies of software, (a) the name resolution APIs and libraries on client devices, and (b) DNS64 implementations.
The new special rules specified in this document for name resolution APIs and libraries state how they should select which recursive resolver to query to learn the IPv6 address synthesis prefix in use on a particular physical or virtual interface. Specifically: When querying for the name 'ipv4only.arpa', name resolution APIs and libraries should use the recursive resolver recommended by the network for the interface in question, rather than a recursive resolver configured manually, a recursive resolver configured by VPN software, or a full-service recursive resolver running on the local host.
The new special rules specified in this document for DNS64 implementations recommend that they avoid performing run-time network queries for values that are known to be fixed by specification.
A useful property of the way NAT64 Prefix Discovery was originally specified was that it allowed for incremental deployment. Even if existing DNS64 gateways, that were unaware of the special 'ipv4only.arpa' name, were already deployed, once IANA created the appropriate 'ipv4only.arpa' records, clients could begin to use the new facility immediately. Clients could send their special queries for 'ipv4only.arpa' to an ipv4only-unaware DNS64 gateway, and (after a query to IANA's servers) the DNS64 gateway would then generate the correct synthesized response.
While this was a useful transition strategy to enable rapid adoption, it is not the ideal end situation. For better performance, better reliability, and lower load in IANA's servers, it is preferable for DNS64 gateways to be aware of the special 'ipv4only.arpa' name so that they can avoid issuing unnecessary queries. Network operators who wish to provide reliable, high performance service to their customers are strongly motivated to prefer DNS64 gateways that recognize the special 'ipv4only.arpa' name and apply the appropriate optimizations.
One of the known concerns with DNS64 is that it conflicts with DNSSEC. If DNSSEC is used to assert cryptographically that a name has no IPv6 AAAA records, then this interferes with using DNS64 address synthesis to assert that those nonexistent IPv6 AAAA records do exist.
... DNS64 receives a query with the DO bit set and the CD bit set. In this case, the DNS64 is supposed to pass on all the data it gets to the query initiator. This case will not work with DNS64, unless the validating resolver is prepared to do DNS64 itself.
Section 3 of the DNS64 specification discusses this:
The NAT64 Prefix Discovery specification provides the mechanism for the query initiator to learn the NAT64 prefix so that it can do its own validation and DNS64 synthesis as described above. With this mechanism the client can (i) interrogate the local NAT64/DNS64 gateway with an 'ipv4only.arpa' query to learn the IPv6 address synthesis prefix, (ii) query for the (signed) IPv4 address records itself, and validate the response, and then (iii) perform its own IPv6 address synthesis locally, combining the IPv6 address synthesis prefix learned from the local NAT64/DNS64 gateway with the validated DNSSEC-signed data learned from the global Domain Name System.
It is conceivable that over time, if DNSSEC adoption continues to grow, the majority of clients could move to this validate-and-synthesize-locally model, which reduces the DNS64 machinery to the vestigial role of simply responding to the 'ipv4only.arpa' query to report the local IPv6 address synthesis prefix. In no case does the client care what answer(s) the authoritative 'arpa' name servers might give for that query. The 'ipv4only.arpa' query is being used purely as a local client-to-middlebox communication message.
This approach is even more attractive if it does not create an additional dependency on the authoritative 'arpa' name servers to answer a query that is unnecessary because the NAT64/DNS64 gateway already knows the answer before it even issues the query. Avoiding this unnecessary query improves performance and reliability for the client, and reduces unnecessary load for the authoritative 'arpa' name servers.
Hard-coding the known answers for 'ipv4only.arpa' IPv4 address record queries (DNS qtype "A") in recursive resolvers also reduces the risk of malicious devices intercepting those queries and returning incorrect answers. Because the 'ipv4only.arpa' zone has to be an insecure delegation (see below) DNSSEC cannot be used to protect these answers from tampering by malicious devices on the path.
With respect to the question of whether 'ipv4only.arpa' should be a secure or insecure delegation, we need to consider two paths of information flow through the network: The path from the authoritative server to the DNS64 recursive resolver, and the path from the DNS64 recursive resolver to the ultimate client. On either or both of these paths there may be one or more DNS64-unaware recursive resolvers.
The path from the authoritative server to the DNS64 recursive resolver (queries for IPv4 address records) need not be protected by DNSSEC, because the DNS64 recursive resolver already knows, by specification, what the answers are. In principle, if this were a secure delegation, and 'ipv4only.arpa' were a signed zone, then the path from the authoritative server to the DNS64 recursive resolver would still work, but DNSSEC is not necessary here. Run-time cryptographic signatures are not needed to verify compile-time constants.
The path from the DNS64 recursive resolver to the ultimate client (queries for IPv6 address records) *cannot* be protected by DNSSEC, because the DNS64 recursive resolver is synthesizing IPv6 address answers, and does not possess the secret key required to sign those answers.
Consequently, the 'ipv4only.arpa' zone MUST be an insecure delegation, to give NAT64/DNS64 gateways the freedom to synthesize answers to those queries at will, without the answers being rejected by DNSSEC-capable resolvers. DNSSEC-capable resolvers that follow this specification MUST NOT attempt to validate answers received in response to queries for the IPv6 AAAA address records for 'ipv4only.arpa'.
A signed "ipv4only.arpa." allows validating DNS64 servers (see [RFC6147] Section 3, Case 5, for example) to detect malicious AAAA resource records. Therefore, the zone serving the well-known name has to be protected with DNSSEC.
The original NAT64 Prefix Discovery specification stated, incorrectly:
This document updates the previous specification to correct that error. The 'ipv4only.arpa' zone MUST be an insecure delegation.
[Once published] IANA has recorded the following names in the
Special-Use Domain Names registry:
IANA has recorded the following IPv4 addresses in the
IPv4 Special-Purpose Address Registry:
The name 'ipv4only.arpa' is defined, by Internet Standard, to have two IPv4 address records with rdata 192.0.0.170 and 192.0.0.171.
When queried via a DNS64 recursive resolver, the name 'ipv4only.arpa' is also defined to have IPv6 AAAA records, with rdata synthesized from a combination of the NAT64 IPv6 prefix(es) and the IPv4 addresses 192.0.0.170 and 192.0.0.171. This can return more than one pair of IPv6 addresses if there are multiple NAT64 prefixes.
The name 'ipv4only.arpa' has no other IPv4 or IPv6 address records.
There are no subdomains of 'ipv4only.arpa'. All names falling below 'ipv4only.arpa' are defined to be nonexistent (NXDOMAIN).
The name 'ipv4only.arpa' is special to
(a) client software wishing to perform DNS64 address synthesis,
(b) APIs responsible for retrieving the correct information, and
(c) the DNS64 recursive resolver responding to such requests.
These three considerations are listed in items 2, 3 and 4 below:
Since the IPv4 addresses 192.0.0.170 and 192.0.0.171 are defined to be special, and are listed in the IPv4 Special-Purpose Address Registry, the corresponding reverse mapping names in the in‑addr.arpa domain are similarly special.
The name '170.0.0.192.in‑addr.arpa' is defined, by Internet Standard, to have only one DNS record, type PTR, with rdata 'ipv4only.arpa'.
The name '171.0.0.192.in‑addr.arpa' is defined, by Internet Standard, to have only one DNS record, type PTR, with rdata 'ipv4only.arpa'.
There are no subdomains of '170.0.0.192.in‑addr.arpa' or '171.0.0.192.in‑addr.arpa'. All names falling below these names are defined to be nonexistent (NXDOMAIN).
Practically speaking these two names are rarely used, but to the extent that they may be, they are special only to resolver APIs and libraries, as described in item 3 below:
For all IPv6 addresses synthesized by a DNS64 recursive resolver, the DNS64 recursive resolver is responsible for synthesizing the appropriate 'ip6.arpa' reverse mapping PTR records too, if it chooses to provide reverse mapping PTR records. The same applies to the synthesized IPv6 addresses corresponding to the IPv4 addresses 192.0.0.170 and 192.0.0.171.
Generally a DNS64 recursive resolver synthesizes appropriate 'ip6.arpa' reverse mapping PTR records by extracting the embedded IPv4 address from the encoded IPv6 address, performing a reverse mapping PTR query for that IPv4 address, and then synthesizing a corresponding 'ip6.arpa' reverse mapping PTR record containing the same rdata.
In the case of synthesized IPv6 addresses corresponding to the IPv4 addresses 192.0.0.170 and 192.0.0.171, the DNS64 recursive resolver does not issue reverse mapping queries for those IPv4 addresses, but instead, according to rule 3 above, immediately returns the answer 'ipv4only.arpa'.
In the case of a client that uses the 'ipv4only.arpa' query to discover the IPv6 prefixes in use by the local NAT64 gateway, and then proceeds to perform its own address synthesis locally (which has benefits such as allowing DNSSEC validation), that client MUST also synthesize 'ip6.arpa' reverse mapping PTR records for those discovered prefix(es), according to the rules above: When a client's name resolution APIs and libraries receive a request to look up an 'ip6.arpa' reverse mapping PTR record for an address that falls within one of the discovered NAT64 address synthesis prefixes, the software extracts the embedded IPv4 address and then, for IPv4 addresses 192.0.0.170 and 192.0.0.171, returns the fixed answer 'ipv4only.arpa', and for all other IPv4 addresses performs a reverse mapping PTR query for the IPv4 address, and then synthesizes a corresponding 'ip6.arpa' reverse mapping PTR record containing the same rdata.
Thanks to Jouni Korhonen, Teemu Savolainen, and Dan Wing, for devising the NAT64 Prefix Discovery mechanism, and for their feedback on this document.
Thanks to Geoff Huston for his feedback on this document.
Thanks to Erik Kline for pointing out that the in‑addr.arpa names are special too.
Thanks to Mark Andrews for pointing out the reasons why the 'ipv4only.arpa' zone MUST be an insecure delegation in order for the NAT64 Prefix Discovery mechanism to work.
Thanks particularly to Lorenzo Colitti for an especially spirited hallway discussion at IETF 96 in Berlin, which lead directly to significant improvements in how this document presents the issues.
Thanks to Dave Thaler and Warren Kumari for generously helping shepherd this document through the publication process.
[RFC6303] | Andrews, M., "Locally Served DNS Zones", BCP 163, RFC 6303, DOI 10.17487/RFC6303, July 2011. |
[RFC8244] | Lemon, T., Droms, R. and W. Kumari, "Special-Use Domain Names Problem Statement", RFC 8244, DOI 10.17487/RFC8244, October 2017. |
[SUDN] | "Special-Use Domain Names Registry" |
[SUv4] | "IANA IPv4 Special-Purpose Address Registry" |
A BIND 9 recursive resolver can be configured to act as authoritative for the necessary DNS64 names as described below.
zone "ipv4only.arpa" { type master; file "ipv4only"; };
In /etc/named.conf the following line is added:
$TTL 86400 ; Default TTL 24 hours @ IN SOA nameserver.example. admin.nameserver.example. ( 2016052400 ; Serial 7200 ; Refresh ( 7200 = 2 hours) 3600 ; Retry ( 3600 = 1 hour) 15724800 ; Expire (15724800 = 6 months) 60 ; Minimum ) @ IN NS nameserver.example. @ IN A 192.0.0.170 @ IN A 192.0.0.171 @ IN AAAA 64:ff9b::192.0.0.170 ; If not using Well-Known Prefix @ IN AAAA 64:ff9b::192.0.0.171 ; place chosen NAT64 prefix here
The file /var/named/ipv4only is created with the following content: