Internet Engineering Task Force (IETF) P. Koch Request for Comments: 8109 DENIC eG BCP: 209 M. Larson Category: Best Current Practice P. Hoffman ISSN: 2070-1721 ICANN March 2017
Initializing a DNS Resolver with Priming Queries
Abstract
This document describes the queries that a DNS resolver should emit to initialize its cache. The result is that the resolver gets both a current NS Resource Record Set (RRset) for the root zone and the necessary address information for reaching the root servers.
Status of This Memo
This memo documents an Internet Best Current Practice.
This document is a product of the Internet Engineering Task Force (IETF). It represents the consensus of the IETF community. It has received public review and has been approved for publication by the Internet Engineering Steering Group (IESG). Further information on BCPs is available in Section 2 of RFC 7841.
Information about the current status of this document, any errata, and how to provide feedback on it may be obtained at http://www.rfc-editor.org/info/rfc8109.
Copyright Notice
Copyright (c) 2017 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 (http://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.
Recursive DNS resolvers need a starting point to resolve queries. [RFC1034] describes a common scenario for recursive resolvers: they begin with an empty cache and some configuration for finding the names and addresses of the DNS root servers. [RFC1034] describes that configuration as a list of servers that will give authoritative answers to queries about the root. This has become a common implementation choice for recursive resolvers, and is the topic of this document.
This document describes the steps needed for this common implementation choice. Note that this is not the only way to start a recursive name server with an empty cache, but it is the only one described in [RFC1034]. Some implementers have chosen other directions, some of which work well and others of which fail (sometimes disastrously) under different conditions. For example, an implementation that only gets the addresses of the root name servers from configuration, not from the DNS as described in this document, will have stale data that could cause slower resolution.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [RFC2119].
This document only deals with recursive name servers (recursive resolvers, resolvers) for the IN class.
Priming is the act of finding the list of root servers from a configuration that lists some or all of the purported IP addresses of some or all of those root servers. A recursive resolver starts with no information about the root servers, and ends up with a list of their names and their addresses.
Priming is described in Sections 5.3.2 and 5.3.3 of [RFC1034]. The scenario used in that description, that of a recursive server that is also authoritative, is no longer as common.
The configured list of IP addresses for the root servers usually comes from the vendor or distributor of the recursive server software. This list is usually correct and complete when shipped, but may become out of date over time.
The list of root server operators and the domain name associated with each one has been stable since 1997. However, there are address changes for the root server domain names, both for IPv4 and IPv6 addresses. However, research shows that after those addresses change, some resolvers never get the new addresses. Therefore, it is important that resolvers be able to cope with change, even without relying upon configuration updates to be applied by their operator. Root server change is the main reason that resolvers need to do priming instead of just going from a configured list to get a full and accurate list of root servers.
A priming query is a DNS query used to get the root server information in a resolver. It has a QNAME of "." and a QTYPE of NS, and is sent to one of the addresses in the configuration for the recursive resolver. The priming query can be sent over either UDP or TCP. If the query is sent over UDP, the source port SHOULD be randomly selected (see [RFC5452]). The Recursion Desired (RD) bit MAY be set to 0 or 1, although the meaning of it being set to 1 is undefined for priming queries.
The recursive resolver SHOULD use EDNS(0) [RFC6891] for priming queries and SHOULD announce and handle a reassembly size of at least 1024 octets [RFC3226]. Doing so allows responses that cover the size of a full priming response (see Section 4.2) for the current set of root servers. See Section 3.3 for discussion of setting the DNSSEC OK (DO) bit (defined in [RFC4033]).
The recursive resolver SHOULD send a priming query only when it is needed, such as when the resolver starts with an empty cache and when the NS RRset for the root zone has expired. Because the NS records for the root are not special, the recursive resolver expires those NS records according to their TTL values. (Note that a recursive resolver MAY pre-fetch the NS RRset before it expires.)
If a priming query does not get a response, the recursive resolver needs to retry the query with a different target address from the configuration.
In order to spread the load across all the root server domain names, the recursive resolver SHOULD select the target for a priming query randomly from the list of addresses. The recursive resolver might choose either IPv4 or IPv6 addresses based on its knowledge of whether the system on which it is running has adequate connectivity on either type of address.
Note that this recommended method is not the only way to choose from the list in a recursive resolver's configuration. Two other common methods include picking the first from the list, and remembering which address in the list gave the fastest response earlier and using that one. There are probably other methods in use today. However, the random method listed above SHOULD be used for priming.
The resolver MAY set the DNSSEC OK (DO) bit. At the time of publication, there is little use to performing DNSSEC validation on the priming query. Currently, all root name server names end in "root-servers.net" and the AAAA and A RRsets for the root server names reside in the "root-servers.net" zone. All root servers are also authoritative for this zone, allowing priming responses to include the appropriate root name server A and AAAA RRsets. But, because the "root-servers.net" zone is not currently signed, these RRsets cannot be validated.
A man-in-the-middle attack on the priming query could direct a resolver to a rogue root name server. Note, however, that a validating resolver will not accept responses from rogue root name servers if they are different from the real responses because the
Koch et al. Best Current Practice [Page 4]
RFC 8109 DNS Priming Queries March 2017
resolver has a trust anchor for the root and the answers from the root are signed. Thus, if there is a man-in-the-middle attack on the priming query, the only result for a validating resolver will be a denial of service, not the resolver's accepting the bad responses.
If the "root-servers.net" zone is later signed, or if the root servers are named in a different zone and that zone is signed, having DNSSEC validation for the priming queries might be valuable.
A priming query is a normal DNS query. Thus, a root name server cannot distinguish a priming query from any other query for the root NS RRset. Thus, the root server's response will also be a normal DNS response.
The priming response is expected to have an RCODE of NOERROR, and to have the Authoritative Answer (AA) bit set. Also, it is expected to have an NS RRset in the Answer section (because the NS RRset originates from the root zone), and an empty Authority section (because the NS RRset already appears in the Answer section). There will also be an Additional section with A and/or AAAA RRsets for the root name servers pointed at by the NS RRset.
Resolver software SHOULD treat the response to the priming query as a normal DNS response, just as it would use any other data fed to its cache. Resolver software SHOULD NOT expect exactly 13 NS RRs because, historically, some root servers have returned fewer.
There are currently 13 root servers. All have one IPv4 address and one IPv6 address. Not even counting the NS RRset, the combined size of all the A and AAAA RRsets exceeds the original 512-octet payload limit from [RFC1035].
In the event of a response where the Additional section omits certain root server address information, re-issuing of the priming query does not help with those root name servers that respond with a fixed order of addresses in the Additional section. Instead, the recursive resolver needs to issue direct queries for A and AAAA RRsets for the remaining names. Currently, these RRsets would be authoritatively available from the root name servers.
Spoofing a response to a priming query can be used to redirect all of the queries originating from a victim recursive resolver to one or more servers for the attacker. Until the responses to priming queries are protected with DNSSEC, there is no definitive way to prevent such redirection.
An on-path attacker who sees a priming query coming from a resolver can inject false answers before a root server can give correct answers. If the attacker's answers are accepted, this can set up the ability to give further false answers for future queries to the resolver. False answers for root servers are more dangerous than, say, false answers for Top-Level Domains (TLDs), because the root is the highest node of the DNS. See Section 3.3 for more discussion.
In both of the scenarios above, a validating resolver will be able to detect the attack if its chain of queries comes to a zone that is signed, but not for those that are unsigned.