| dnsop | J. Dickinson |
| Internet-Draft | S. Dickinson |
| Obsoletes: 5966 (if approved) | Sinodun |
| Intended status: Standards Track | R. Bellis |
| Expires: May 5, 2016 | ISC |
| A. Mankin | |
| D. Wessels | |
| Verisign Labs | |
| November 2, 2015 |
DNS Transport over TCP - Implementation Requirements
draft-ietf-dnsop-5966bis-04
This document specifies the requirement for support of TCP as a transport protocol for DNS implementations and provides guidelines towards DNS-over-TCP performance on par with that of DNS-over-UDP. This document obsoletes RFC5966.
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 May 5, 2016.
Copyright (c) 2015 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.
Most DNS [RFC1034] transactions take place over UDP [RFC0768]. TCP [RFC0793] is always used for full zone transfers (AXFR) and is often used for messages whose sizes exceed the DNS protocol's original 512-byte limit. The growing deployment of DNSSEC and IPv6 has increased response sizes and therefore the use of TCP. The need for increased TCP use has also been driven by the protection it provides against address spoofing and therefore exploitation of DNS in reflection/amplification attacks. It is now widely used in Response Rate Limiting [RRL].
Section 6.1.3.2 of [RFC1123] states:
However, some implementors have taken the text quoted above to mean that TCP support is an optional feature of the DNS protocol.
The majority of DNS server operators already support TCP and the default configuration for most software implementations is to support TCP. The primary audience for this document is those implementors whose limited support for TCP restricts interoperability and hinders deployment of new DNS features.
This document therefore updates the core DNS protocol specifications such that support for TCP is henceforth a REQUIRED part of a full DNS protocol implementation.
There are several advantages and disadvantages to the increased use of TCP (see Appendix A) as well as implementation details that need to be considered. This document addresses these issues and presents TCP as a valid transport alternative for DNS. It extends the content of [RFC5966], with additional considerations and lessons learned from research, developments and implementation of TCP in DNS and in other internet protocols.
Whilst this document makes no specific requirements for operators of DNS servers to meet, it does offer some suggestions to operators to help ensure that support for TCP on their servers and network is optimal. It should be noted that failure to support TCP (or the blocking of DNS over TCP at the network layer) may result in resolution failure and/or application-level timeouts.
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].
In the absence of EDNS0 (Extension Mechanisms for DNS 0 [RFC6891]) (see below), the normal behaviour of any DNS server needing to send a UDP response that would exceed the 512-byte limit is for the server to truncate the response so that it fits within that limit and then set the TC flag in the response header. When the client receives such a response, it takes the TC flag as an indication that it should retry over TCP instead.
RFC 1123 also says:
Existing deployments of DNS Security (DNSSEC) [RFC4033] have shown that truncation at the 512-byte boundary is now commonplace. For example, a Non-Existent Domain (NXDOMAIN) (RCODE == 3) response from a DNSSEC-signed zone using NextSECure 3 (NSEC3) [RFC5155] is almost invariably larger than 512 bytes.
Since the original core specifications for DNS were written, the Extension Mechanisms for DNS have been introduced. These extensions can be used to indicate that the client is prepared to receive UDP responses larger than 512 bytes. An EDNS0-compatible server receiving a request from an EDNS0-compatible client may send UDP packets up to that client's announced buffer size without truncation.
However, transport of UDP packets that exceed the size of the path MTU causes IP packet fragmentation, which has been found to be unreliable in many circumstances. Many firewalls routinely block fragmented IP packets, and some do not implement the algorithms necessary to reassemble fragmented packets. Worse still, some network devices deliberately refuse to handle DNS packets containing EDNS0 options. Other issues relating to UDP transport and packet size are discussed in [RFC5625].
The MTU most commonly found in the core of the Internet is around 1500 bytes, and even that limit is routinely exceeded by DNSSEC-signed responses.
The future that was anticipated in RFC 1123 has arrived, and the only standardised UDP-based mechanism that may have resolved the packet size issue has been found inadequate.
All general-purpose DNS implementations MUST support both UDP and TCP transport.
Regarding the choice of when to use UDP or TCP, Section 6.1.3.2 of RFC 1123 also says:
This requirement is hereby relaxed. Stub resolvers and recursive resolvers MAY elect to send either TCP or UDP queries depending on local operational reasons. TCP MAY be used before sending any UDP queries. If it already has an open TCP connection to the server it SHOULD reuse this connection. In essence, TCP ought to be considered a valid alternative transport to UDP, not purely a fallback option.
In addition it is noted that all Recursive and Authoritative servers MUST send responses using the same transport as the query arrived on. In the case of TCP this MUST also be the same connection.
Section 4.2.2 of [RFC1035] says:
Other more modern protocols (e.g., HTTP/1.1 [RFC7230]) have support by default for persistent TCP connections for all requests. Connections are then normally closed via a 'connection close' signal from one party.
The description in [RFC1035] is clear that servers should view connections as persistent (particularly after receiving an SOA), but unfortunately does not provide enough detail for an unambiguous interpretation of client behaviour for queries other than a SOA. Additionally, DNS does not yet have a signalling mechanism for connection timeout or close, although some have been proposed.
There is no clear guidance today in any RFC as to when a DNS client should close a TCP connection, and there are no specific recommendations with regard to DNS client idle timeouts. However it is common practice for clients to close the TCP connection after sending a single request (apart from the SOA/AXFR case).
Many DNS server implementations use a long fixed idle timeout and default to a small number of TCP connections. They also offer little by the way of TCP connection management options. The disadvantages of this include:
Note that this denial-of-service is only on the TCP service. However, in these cases it affects not only clients wishing to use TCP for their queries for operational reasons, but all clients who choose to fall back to TCP from UDP after receiving a TC=1 flag.
The following sections include recommendations that are intended to result in more consistent and scalable implementations of DNS-over-TCP.
One perceived disadvantage to DNS over TCP is the added connection setup latency, generally equal to one RTT. To amortize connection setup costs, both clients and servers SHOULD support connection reuse by sending multiple queries and responses over a single persistent TCP connection.
When sending multiple queries over a TCP connection clients MUST take care to avoid Message ID collisions. In other words, they MUST NOT re-use the DNS Message ID of an in-flight query on the same TCP connection. This is especially important if the server could be performing out-of-order processing (see Section 7).
Due to the historical use of TCP primarily for zone transfer and truncated responses, no existing RFC discusses the idea of pipelining DNS queries over a TCP connection.
In order to achieve performance on par with UDP DNS clients SHOULD pipeline their queries. When a DNS client sends multiple queries to a server, it SHOULD NOT wait for an outstanding reply before sending the next query. Clients SHOULD treat TCP and UDP equivalently when considering the time at which to send a particular query.
It is likely that DNS servers need to process pipelined queries concurrently and also send out-of-order responses over TCP in order to provide the level of performance possible with UDP transport. If TCP performance is of importance, clients might find it useful to use server processing times as input to server and transport selection algorithms.
DNS servers (especially recursive) SHOULD expect to receive pipelined queries. The server SHOULD process TCP queries concurrently, just as it would for UDP. The server SHOULD answer all pipelined queries, even if they are sent in quick succession. The handling of responses to pipelined queries is covered in Section 7.
To mitigate the risk of unintentional server overload, DNS clients MUST take care to minimize the number of concurrent TCP connections made to any individual server. It is RECOMMENDED that for any given client/server interaction there SHOULD be no more than one connection for regular queries, one for zone transfers and one for each protocol that is being used on top of TCP, for example, if the resolver was using TLS. It is however noted that certain primary/secondary configurations with many busy zones might need to use more than one TCP connection for zone transfers for operational reasons.
Similarly, servers MAY impose limits on the number of concurrent TCP connections being handled for any particular client IP address or subnet. These limits SHOULD be much looser than the client guidelines above, because the server does not know, for example, if a client IP address belongs to a single client or is multiple resolvers on a single machine, or multiple clients behind a device performing Network Address Translation (NAT).
To mitigate the risk of unintentional server overload, DNS clients MUST take care to minimize the idle time of established DNS-over-TCP sessions made to any individual server. DNS clients SHOULD close the TCP connection of an idle session, unless an idle timeout has been established using some other signalling mechanism, for example, [edns-tcp-keepalive].
To mitigate the risk of unintentional server overload it is RECOMMENDED that the default server application-level idle period be of the order of seconds, but no particular value is specified. In practice, the idle period can vary dynamically, and servers MAY allow idle connections to remain open for longer periods as resources permit. A timeout of at least a few seconds is advisable for normal operations to support those clients that expect the SOA and AXFR request sequence to be made on a single connection as originally specified in [RFC1035]. Servers MAY use zero timeouts when experiencing heavy load or are under attack.
DNS messages delivered over TCP might arrive in multiple segments. A DNS server that resets its idle timeout after receiving a single segment might be vulnerable to a "slow read attack." For this reason, servers SHOULD apply the idle timeout to the receipt of a full DNS message, rather than to receipt of any part of a DNS message.
Under normal operation clients typically initiate connection closing on idle connections however servers can close the connection if their local idle timeout policy is exceeded. Connections can be also closed by either end under unusual conditions such as defending against an attack or system failure/reboot.
Clients SHOULD retry unanswered queries if the connection closes before receiving all outstanding responses. No specific retry algorithm is specified in this document.
If a server finds that a client has closed a TCP session, or if the session has been otherwise interrupted, before all pending responses have been sent then the server MUST NOT attempt to send those responses. Of course the server MAY cache those responses.
RFC 1035 is ambiguous on the question of whether TCP responses may be reordered -- the only relevant text is in Section 4.2.1, which relates to UDP:
For the avoidance of future doubt, this requirement is clarified. Authoritative servers and recursive resolvers are RECOMMENDED to support the preparing of responses in parallel and sending them out-of-order, regardless of the transport protocol in use. Stub and recursive resolvers MUST be able to process responses that arrive in a different order to that in which the requests were sent, regardless of the transport protocol in use.
In order to achieve performance on par with UDP, recursive resolvers SHOULD process TCP queries in parallel and return individual responses as soon as they are available, possibly out-of-order.
Since pipelined responses can arrive out-of-order, clients MUST match responses to outstanding queries on the same TCP connection using the Message ID. If the response contains a question section the client MUST match the QNAME, QCLASS and QTYPE fields. Failure by clients to properly match responses to outstanding queries can have serious consequences for interoperability.
For reasons of efficiency, DNS clients and servers SHOULD pass the two-octet length field, and the message described by that length field, to the TCP layer at the same time (e.g., in a single "write" system call) to make it more likely that all the data will be transmitted in a single TCP segment.
This additionally avoids problems due to some DNS servers being very sensitive to timeout conditions on receiving messages (they might abort a TCP session if the first TCP segment does not contain both the length field and the entire message). Such behavior is certainly undesirable. As described in Section 6.2.3, servers SHOULD apply connection timeouts to the receipt of a full message and MUST NOT close a connection simply because the first "read" from the TCP layer does not contain the entire message.
This section is non-normative.
TCP Fast Open [RFC7413] (TFO) allows data to be carried in the SYN packet, reducing the cost of re-opening TCP connections. It also saves up to one RTT compared to standard TCP.
TFO mitigates the security vulnerabilities inherent in sending data in the SYN, especially on a system like DNS where amplification attacks are possible, by use of a server-supplied cookie. TFO clients request a server cookie in the initial SYN packet at the start of a new connection. The server returns a cookie in its SYN-ACK. The client caches the cookie and reuses it when opening subsequent connections to the same server.
The cookie is stored by the client's TCP stack (kernel) and persists if either the client or server processes are restarted. TFO also falls back to a regular TCP handshake gracefully.
DNS services taking advantage of IP anycast [RFC4786] might need to take additional steps when enabling TFO. From [RFC7413]:
This memo includes no request to IANA.
Some DNS server operators have expressed concern that wider promotion and use of DNS over TCP will expose them to a higher risk of denial-of-service (DoS) attacks on TCP (both accidental and deliberate).
Although there is a higher risk of some specific attacks against TCP-enabled servers, techniques for the mitigation of DoS attacks at the network level have improved substantially since DNS was first designed.
Readers are advised to familiarise themselves with [CPNI-TCP], a security assessment of TCP detailing known TCP attacks and countermeasures which references most of the relevant RFCs on this topic.
To mitigate the risk of DoS attacks, DNS servers are advised to engage in TCP connection management. This could include maintaining state on existing connections, re-using existing connections and controlling request queues to enable fair use. It is likely to be advantageous to provide configurable connection management options, for example:
No specific values are recommended for these parameters.
Operators are advised to familiarise themselves with the configuration and tuning parameters available in the operating system TCP stack. However detailed advice on this is outside the scope of this document.
Operators of recursive servers are advised to ensure that they only accept connections from expected clients (for example by the use of an ACL), and do not accept them from unknown sources. In the case of UDP traffic, this will help protect against reflection attacks [RFC5358] and in the case of TCP traffic it will prevent an unknown client from exhausting the server's limits on the number of concurrent connections.
The authors would like to thank Francis Dupont and Paul Vixie for detailed review, Andrew Sullivan, Tony Finch, Stephane Bortzmeyer, Joe Abley, Tatuya Jinmei and the many others who contributed to the mailing list discussion. Also Liang Zhu, Zi Hu, and John Heidemann for extensive DNS-over-TCP discussions and code. Lucie Guiraud and Danny McPherson for reviewing early versions of this document. We would also like to thank all those who contributed to RFC5966.
| [RFC0768] | Postel, J., "User Datagram Protocol", STD 6, RFC 768, DOI 10.17487/RFC0768, August 1980. |
| [RFC0793] | Postel, J., "Transmission Control Protocol", STD 7, RFC 793, DOI 10.17487/RFC0793, September 1981. |
| [RFC1034] | Mockapetris, P., "Domain names - concepts and facilities", STD 13, RFC 1034, DOI 10.17487/RFC1034, November 1987. |
| [RFC1035] | Mockapetris, P., "Domain names - implementation and specification", STD 13, RFC 1035, DOI 10.17487/RFC1035, November 1987. |
| [RFC1123] | Braden, R., "Requirements for Internet Hosts - Application and Support", STD 3, RFC 1123, DOI 10.17487/RFC1123, October 1989. |
| [RFC2119] | Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997. |
| [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. |
| [RFC4786] | Abley, J. and K. Lindqvist, "Operation of Anycast Services", BCP 126, RFC 4786, DOI 10.17487/RFC4786, December 2006. |
| [RFC5155] | Laurie, B., Sisson, G., Arends, R. and D. Blacka, "DNS Security (DNSSEC) Hashed Authenticated Denial of Existence", RFC 5155, DOI 10.17487/RFC5155, March 2008. |
| [RFC5358] | Damas, J. and F. Neves, "Preventing Use of Recursive Nameservers in Reflector Attacks", BCP 140, RFC 5358, DOI 10.17487/RFC5358, October 2008. |
| [RFC5625] | Bellis, R., "DNS Proxy Implementation Guidelines", BCP 152, RFC 5625, DOI 10.17487/RFC5625, August 2009. |
| [RFC5966] | Bellis, R., "DNS Transport over TCP - Implementation Requirements", RFC 5966, DOI 10.17487/RFC5966, August 2010. |
| [RFC6891] | Damas, J., Graff, M. and P. Vixie, "Extension Mechanisms for DNS (EDNS(0))", STD 75, RFC 6891, DOI 10.17487/RFC6891, April 2013. |
| [RFC7230] | Fielding, R. and J. Reschke, "Hypertext Transfer Protocol (HTTP/1.1): Message Syntax and Routing", RFC 7230, DOI 10.17487/RFC7230, June 2014. |
The TCP handshake generally prevents address spoofing and, therefore, the reflection/amplification attacks which plague UDP.
IP fragmentation is less of a problem for TCP than it is for UDP. TCP stacks generally implement Path MTU Discovery so they can avoid IP fragmentation of TCP segments. UDP, on the other hand, does not provide reassembly, which means datagrams that exceed the path MTU size must experience fragmentation [RFC5405]. Middleboxes are known to block IP fragments, leading to timeouts and forcing client implementations to "hunt" for EDNS0 reply size values supported by the network path. Additionally, fragmentation may lead to cache poisoning [fragmentation-considered-poisonous].
TCP setup costs an additional RTT compared to UDP queries. Setup costs can be amortized by reusing connections, pipelining queries, and enabling TCP Fast Open.
TCP imposes additional state-keeping requirements on clients and servers. The use of TCP Fast Open reduces the cost of closing and re-opening TCP connections.
Long-lived TCP connections to anycast servers might be disrupted due to routing changes. Clients utilizing TCP for DNS need to always be prepared to re-establish connections or otherwise retry outstanding queries. It might also be possible for TCP Multipath [RFC6824] to allow a server to hand a connection over from the anycast address to a unicast address.
There are many "Middleboxes" in use today that interfere with TCP over port 53 [RFC5625]. This document does not propose any solutions, other than to make it absolutely clear that TCP is a valid transport for DNS and support for it is a requirement for all implementations.
A more in-depth discussion of connection orientated DNS can be found elsewhere [Connection-Oriented-DNS].
[Note to RFC Editor: please remove this section prior to publication.]
This document differs from RFC 5966 in four additions: