]> AFNIC

1, rue Stephenson78180Montigny-le-BretonneuxFrance +33 1 39 30 83 46bortzmeyer+ietf@nic.frhttp://www.afnic.fr/

DNS Privacy (dprive) Working Group This document proposes a mechanism for securing (privacy-wise) the communication between the DNS resolver and the authoritative name server. REMOVE BEFORE PUBLICATION: this document should be discussed in the IETF DPRIVE group, through its mailing list. The source of the document, as well as a list of open issues, is currently kept at Github.

To improve the privacy of the DNS user (), the standard solution is to encrypt the requests with TLS (). We use this DNS-over-TLS solution as well here, since it is standardized, already implemented in many programs, and relies on a well-known security protocol (inventing a new security protocol is quite dangerous). But just encrypting, without authenticating the remote server, leaves the user's privacy vulnerable to active man-in-the-middle attacks. and describe how to authenticate the DNS resolver, in the stub-to-resolver link. We describe here authentication of the authoritative name server, in the resolver-to-authoritative link. A stub DNS resolver has only a few resolvers, and there is typically a pre-existing relationship. But a resolver speaks to many authoritative name servers, without any prior relationship. This means that, for instance, having a static key for the resolver makes sense while it would be clearly unrealistic for the authoritative server. Instead, we rely on DANE (). Authoritative name servers are known by name (obtained from zone delegation). The manager of the ns1.example.net name server adds a TLSA record under example.net. The client establishes the TLS session, then authenticate in the normal DANE way. The original charter of the DPRIVE working group, in force at the time of this draft, says "The primary focus of this Working Group is to develop mechanisms that provide confidentiality between DNS Clients and Iterative Resolvers" and adds "but it may also later consider mechanisms that provide confidentiality between Iterative Resolvers and Authoritative Servers". This document is here for this second step, "between Iterative Resolvers and Authoritative Servers". It will probably require a rechartering of the group.

A DNS full-service resolver who needs to query an authoritative name server establishes a TLS-over-TCP session with this authoritative name server. If the DNS material to perform DANE authentication is sent in the TLS session (), it uses it. Otherwise, the resolver queries TLSA records () for this name server and authenticates the key or certificate of the server this way. If the name server is ns1.example.net, the TLSA record to query is _853._tcp.ns1.example.net. Note that the server MAY use raw public keys () and so there is not always a certificate. If the server uses raw public keys, the TLSA record's Selector field must be 1 (SPKI, SubjectPublicKeyInfo). The recommended order is to try TLS before querying the TLSA records. True, DANE signals if the server is willing to make DNS-over-TLS (and can therefore save a TLS attempt) but cannot guarantee that it will work (for instance if a middlebox blocks port 853). Also, the DANE records may be transferred in the TLS session, not through the DNS. If the TLS session establishement fails, or if the DANE authentication fails, the result depends on whether the resolver runs in strict or opportunistic mode (). In strict mode, the resolver MUST stop using this authoritative name server, and MUST try other servers of the DNS zone. In opportunistic mode, the resolver MUST use the authoritative name server despite the failure. It MAY try other name servers of the zone before, in the hope they will accept TLS and be authenticated. To avoid a chicken-and-egg problem, the resolver, even in strict mode, MAY use unsecure servers for the meta-queries (getting the TLSA records). More specifically: (0)The resolver remembers the keys of the authoritative name servers (in the same way it remembers the lowest RTT among an NS RRset), (1)When the resolver needs to talk to a server (say ns2.example.net) for which it does not know the key, it does a TLSA request for _853._tcp.ns2.example.net, (2)If the resolution of this request requires that we talk to the same server for which we’re searching for the TLSA record, the resolver connects to this server with TLS to port 853, does not bother to authenticate, and sends the query. This step offers no authentication. (See also , section 5.) A resolver MAY use the knowledge of TLS authentication it has to choose an authoritative name server among a NS RRset. As of this revision, we do not expect resolvers to use strict mode, since the encryption and authentication modes described in this document are not yet supported in authoritative name servers.

DNS-over-TLS depends on TCP, and the resolver and the authoritative name server must therefore support persistent TCP connections (, specially section 6.2.1). A resolver may have a lot of client-side state, when managing hundreds of connections to remote authoritative servers (). The latency when connecting to a authoritative name server is certainly an issue. TLS 1.3 and TCP Fast Open () may help. Open question: do we require a minimum TLS version of 1.3? () Because the resolver cannot know in advance if the TLS connection will work (even if there is a DANE record), using parallel attempts ("happy eyeballs", ) is important. A resolver working in opportunistic mode should try ports 53 and 853 in parallel. An authoritative name server cannot know if the resolver authenticated it, nor how. In the future, it may be interesting to have an EDNS option to signal a successful authentication, or a failure, but this is out of scope currently. If it is a concern that the same authoritative name servers are used for ordinary DNS and for encrypted DNS, there are several ways to address this concern. A server operator may use front-end systems dispatching requests to ports 53 and 853 to different servers. A resolver must be configurable to operate in strict or opportunistic modes. Until the features described herein are widely supported, opportunistic mode should not be the default since strict mode would yield frequent failures. A resolver may have a configuration mechanism to be in strict mode only for some domains.

No action for IANA. This section can be deleted.

The state to be kept in both the client and the server may make some denial-of-service attacks easier. Following the advice contained in section 10 of is recommended. In opportunistic mode, there is no guarantee to have a secure use of the DNS, or even a guarantee to be informed of a problem. Opportunistic mode is a "best effort" privacy service. Even in strict mode, some leaks may occur, through the DANE meta-queries, and through SNI indication () in the TLS session. Neither transport encryption nor authentication protect DNS users from authentic servers which nonetheless abuse users’ privacy once they’ve received their queries. These techniques must therefore be combined with data minimization techniques ().

&rfc6698; &rfc7858; &I-D.ietf-dprive-dtls-and-tls-profiles; &rfc7250; &rfc7413; &rfc7626; &rfc7766; &rfc7816; &rfc8305; &I-D.ietf-tls-tls13; &I-D.ietf-tls-dnssec-chain-extension; &I-D.ietf-tls-sni-encryption; &I-D.bortzmeyer-dprive-step-2;

Thanks to Bill Woodcock for a detailed review.

A number of other possible solutions to this problem may be found in in .