| Names and Identifiers Program | B. Trammell |
| Internet-Draft | ETH Zurich |
| Intended status: Informational | March 11, 2016 |
| Expires: September 12, 2016 |
Properties of an Ideal Naming Service
draft-trammell-inip-pins-01
This document specifies a set of necessary functions and desirable properties of an ideal system for resolving names to addresses and associated information for establishing communication associations in the Internet. For each property, it briefly explains the rationale behind it, and how the property is or could be met with the present Domain Name System. It is intended to start a discussion within the IAB’s Names and Identifiers program about gaps between the present reality of DNS and the naming service the Internet needs by returning to first principles.
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 http://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 September 12, 2016.
Copyright (c) 2016 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.
The Internet’s Domain Name System (DNS) [RFC1035] is an excellent illustration of the advantages of the decentralized architecture that have made the Internet able to scale to its present size. However, the choices made in the evolution of the DNS since its initial design are only one path through the design space of Internet-scale naming services. Many other naming services have been proposed, though none has been remotely as successful for general- purpose use in the Internet.
This document returns to first principles, to determine the dimensions of the design space of desirable properties of an Internet-scale naming service. It is a work in progress, intended to start a discussion within the IAB’s Names and Identifiers program about gaps between the present reality of DNS and the naming service the Internet needs.
Section 3 and Section 4 define the set of operations a naming service should provide for queriers and authorities, Section 5 defines a set of desirable properties of the provision of this service, and Section 6 examines implications of these properties.
The following capitalized terms are defined and used in this document:
[EDITOR’S NOTE: need to make a terminology unification pass]
At its core, a naming service must provide a few basic functions for queriers, associating a Subject of a query with information about that subject. The information available from a naming service is that which is necessary for a querier to establish a connection with some other entity in the Internet, given a name identifying it.
Given a Subject name, the naming service returns a set of addresses associated with that name, if such an association exists, where the association is determined by the authority for that name. Names may be associated with addresses in one or more address families (e.g. IP version 4, IP version 6). A querier may specify which address families it is interested in receiving addresses for, and the naming system treats all address families equally.
This mapping is implemented in the DNS protocol via the A and AAAA RRTYPES.
Given an Subject address, the naming service returns a set of names associated with that address, if such an association exists, where the association is determined by the authority for that address.
This mapping is implemented in the DNS protocol via the PTR RRTYPE. IPv4 mappings exist within the in-addr.arpa. zone, and IPv6 mappings in the ip6.arpa. zone. This mechanism has the disadvantage that delegations in IPv4 only happen on octet (8-bit) boundaries, and in IPv6 only happen on hex digit (4-bit) boundaries, which make delegations on other prefixes operationally difficult. [EDITOR’S NOTE: is there a citation for practical workarounds here?]
Given a Subject name, the naming service returns a set of object names associated with that name, if such an association exists, where the association is determined by the authority for the subject name.
This mapping is implemented in the DNS protocol via the CNAME RRTYPE. CNAME does not allow the association of multiple object names with a single subject, and CNAME may not combine with other RRTYPEs (e.g. NS, MX) arbitrarily.
Given a Subject name, the naming service returns other auxiliary information associated with that name that is useful for establishing communication over the Internet with the entities associated with that name.
Most of the other RRTYPES in the DNS protocol implement these sort of mappings.
As a name might be associated with more than one address, auxiliary information as above may be associated with a name/address pair, as opposed to just with a name.
This mapping is not presently supported by the DNS protocol.
The query interface is not the only interface to the naming service: the interface a naming service presents to an Authority allows updates to the set of Associations and Delegations in that Authority’s namespace. Updates consist of additions of, changes to, and deletions of Associations and Delegations. In the present DNS, this interface consists of the publication of a new zone file with an incremented version number, but other authority interfaces are possible.
The following properties are desirable in a naming service providing the functions in Section 3 and Section 4.
Every Association among names, addresses, and auxiliary data is subject to some Authority: an entity which has the right to determine which Associations and Subjects exist in its namespace. The following are properties of Authorities in our ideal naming service:
An Authority can delegate some part of its namespace to some other subordinate Authority. This property allows the naming service to scale to the size of the Internet, and leads to a tree-structured namespace, where each Delegation is itself identified with a Subject at a given level in the namespace.
In the DNS protocol, this federation of authority is implemented using the NS RRTYPE, redirecting queries to subordinate authorities recursively to the final authority.
For a given Subject, there is a single Authority that has the right to determine the Associations and/or Delegations for that subject. The unitary authority for the root of the namespace tree may be special, though; see Section 5.1.5.
In the DNS protocol as deployed, unitary authority is approximated by the entity identified by the SOA RRTYPE. The existence of registrars, which use the Extensible Provisioning Protocol (EPP) [RFC5730] to modify entries in the zones under the authority of a top-level domain registry, complicates this somewhat.
A querier can determine the identity of the Authority for a given Association. An Authority cannot delegate its rights or responsibilities with respect to a subject without that Delegation being exposed to the querier.
In DNS, the authoritative name server(s) to which a query is delegated via the NS RRTYPE are known. However, we note that in the case of authorities which delegate the ability to write to the zone to other entities (i.e., the registry-registrar relationship), the current DNS provides no facility for a querier to understand on whose behalf an authoritative assertion is being made; this information is instead available via WHOIS. To our knowledge, no present DNS name servers use WHOIS information retrieved out of band to make policy decisions.
An ideal naming service allows the revocation and replacement of an authority at any level in the namespace, and supports the revocation and replacement of authorities with minimal operational disruption.
The current DNS allows the replacement of any level of delegation except the root through changes to the appropriate NS and DS records. Authority revocation in this case is as consistent as any other change to the DNS.
Authority at the top level of the namespace tree is delegated according to a process such that there is universal agreement throughout the Internet as to the subordinates of those Delegations.
[EDITOR’S NOTE: Today, this is the root zone. But note that this property does not necessarily imply a single authority at the root as with the present arrangement, only that the process by which the root is changed and operated leads to a universally consistent result.]
A querier must be able to verify that the answers that it gets from the naming service are authentic.
Given a Delegation from a superordinate to a subordinate Authority, a querier can verify that the superordinate Authority authorized the Delegation.
Authenticity of delegation in DNS is provided by DNSSEC [RFC4033].
The authenticity of every answer is verifiable by the querier. The querier can confirm that the Association returned in the answer is correct according to the Authority for the Subject of the query.
Authenticity of response in DNS is provided by DNSSEC.
Some queries will yield no answer, because no such Association exists. In this case, the querier can confirm that the Authority for the Subject of the query asserts this lack of Association.
Authenticity of negative response in DNS is provided by DNSSEC.
Consistency in a naming service is important. The naming service should provide the most globally consistent view possible of the set of associations that exist at a given point in time, within the limits of latency and bandwidth tradeoffs.
When an Authority makes changes to an Association, every query for a given Subject returns either the new valid result or a previously valid result, with known and predictable bounds on “how previously”. Given that additions of, changes to, and deletions of associations may have different operational causes, different bounds may apply to different operations.
The time-to-live (TTL) on a resource record in DNS provides a mechanism for expiring old resource records. We note that this mechanism makes additions to the system propagate faster than changes and deletions, which may not be a desirable property.
Some techniques require giving different answers to different queries, even in the absence of changes: the stable state of the namespace is not globally consistent. This inconsistency should be explicit: a querier can know that an answer might be dependent on its identity, network location, or other factors.
One example of such desirable inconsistency is the common practice of “split horizon” DNS, where an organization makes internal names available on its own network, but only the names of externally-visible subjects available to the Internet at large.
Another is the common practice of DNS-based content distribution, in which an authoritative name server gives different answers for the same query depending on the network location from which the query was received, or depending on the subnet in which the end client originating a query is located (via the EDNS Client Subnet extension [I-D.ietf-dnsop-edns-client-subnet]). Such inconsistency based on client identity or network address may increase query linkability (see Section 5.4.4).
We note that while DNS can be deployed to allow essentially unlimited kinds of inconsistency in its responses, there is no protocol support for a query to express the kind of consistency it desires, or for a response to explicitly note that it is inconsistent. [I-D.ietf-dnsop-edns-client-subnet] does allow a querier to note that it would specifically like the view of the state of the namespace offered to a certain part of the network, and as such can be seen as inchoate support for this property.
A naming service must provide appropriate performance guarantees to its clients. As these properties deal with the operational parameters of the naming service, interesting tradeoffs are available among them, both at design time as well as at run time (on which see Section 5.4.5).
The naming service as a whole is resilient to failures of individual nodes providing the naming service, as well as to failures of links among them. Intentional prevention of successful, authenticated query by an adversary should be as hard as practical.
The DNS protocol was designed to be highly available through the use of secondary nameservers. Operational practices (e.g. anycast deployment) also increase the availability of DNS as currently deployed.
The time for the entire process of looking up a name and other necessary associated data from the point of view of the querier, amortized over all queries for all connections, should not significantly impact connection setup or resumption latency.
The bandwidth cost for looking up a name and other associated data necessary for establishing communication with a given Subject, from the point of view of the querier, amortized over all queries for all connections, should significantly impact total bandwidth demand for an application.
It should be costly for an adversary to monitor the infrastructure in order to link specific queries to specific queriers.
The DPRIVE working group is currently working on approaches to improve confidentiality of stub- to recursive-resolver communications in order to reduce query linkability; see e.g. [I-D.ietf-dprive-dns-over-tls], [I-D.ietf-dprive-dnsodtls].
A querier should be able to indicate the desire for a benefit with respect to one performance property by accepting a tradeoff in another, including:
There is no support for explicit tradeoffs in performance properties available to clients in the present DNS.
On a cursory examination, many of the properties of our ideal name service can be met, or could be met, by the present DNS protocol or extensions thereto. We note that there are further possibilities for the future evolution of naming services meeting these properties. This section contains random observations that might inform future work.
Any system which can provide the authenticity properties in Section 5.2 is freed from one of the design characteristics of the present domain name system: the requirement to bind a zone of authority to a specific set of authoritative servers. Since the authenticity of delegation must be a protected by a chain of signatures back to the root of authority, the location within the infrastructure where an authoritative mapping “lives” is no longer bound to a specific name server. While the present design of DNS does have its own scalability advantages, this implication allows a much larger design space to be explored for future name service work, as a Delegation need not always be implemented via redirection to another name server.
Much of the difficulty with explicit inconsistency (Section 5.3.2) derives from the fact that assertions and queries about subjects exist within a context: .local names on the local network (whether link or site local), split-DNS names within the context of the “inside” side of the recursive resolver, DNS geographic load balancing within the geographic context of the client. Because DNS provides no protocol-level support for expressing these contexts, they remain implicit.
We note that protocol-level support for this context explicit could point toward solutions for a variety of problems in currently deployed naming services, from generalized solutions with privacy/efficiency tradeoffs to the ([I-D.ietf-dnsop-edns-client-subnet] aside), to explicit redirection to alternate naming resolution for “special” names [RFC6761].
This document has no actions for IANA
[EDITOR’S NOTE: todo]
This document is, in part, an output of design work on naming services at the Network Security Group at ETH Zurich. Thanks to the group, including Daniele Asoni, Steve Matsumoto, and Stephen Shirley, for discussions leading to this document. Thanks as well to Ted Hardie, Wendy Selzter, Andrew Sullivan, and Suzanne Woolf for input and feedback.
| [I-D.ietf-dnsop-edns-client-subnet] | Contavalli, C., Gaast, W., tale, t. and W. Kumari, "Client Subnet in DNS Queries", Internet-Draft draft-ietf-dnsop-edns-client-subnet-06, December 2015. |
| [I-D.ietf-dprive-dns-over-tls] | Zi, Z., Zhu, L., Heidemann, J., Mankin, A., Wessels, D. and P. Hoffman, "DNS over TLS: Initiation and Performance Considerations", Internet-Draft draft-ietf-dprive-dns-over-tls-05, January 2016. |
| [I-D.ietf-dprive-dnsodtls] | Reddy, T., Wing, D. and P. Patil, "DNS over DTLS (DNSoD)", Internet-Draft draft-ietf-dprive-dnsodtls-04, January 2016. |
| [RFC1035] | Mockapetris, P., "Domain names - implementation and specification", STD 13, RFC 1035, DOI 10.17487/RFC1035, November 1987. |
| [RFC4033] | Arends, R., Austein, R., Larson, M., Massey, D. and S. Rose, "DNS Security Introduction and Requirements", RFC 4033, DOI 10.17487/RFC4033, March 2005. |
| [RFC5730] | Hollenbeck, S., "Extensible Provisioning Protocol (EPP)", STD 69, RFC 5730, DOI 10.17487/RFC5730, August 2009. |
| [RFC6761] | Cheshire, S. and M. Krochmal, "Special-Use Domain Names", RFC 6761, DOI 10.17487/RFC6761, February 2013. |