RFC 7828






Internet Engineering Task Force (IETF)                        P. Wouters
Request for Comments: 7828                                       Red Hat
Category: Standards Track                                       J. Abley
ISSN: 2070-1721                                                Dyn, Inc.
                                                            S. Dickinson
                                                                 Sinodun
                                                               R. Bellis
                                                                     ISC
                                                              April 2016


                  The edns-tcp-keepalive EDNS0 Option

Abstract



   DNS messages between clients and servers may be received over either
   UDP or TCP.  UDP transport involves keeping less state on a busy
   server, but can cause truncation and retries over TCP.  Additionally,
   UDP can be exploited for reflection attacks.  Using TCP would reduce
   retransmits and amplification.  However, clients commonly use TCP
   only for retries and servers typically use idle timeouts on the order
   of seconds.

   This document defines an EDNS0 option ("edns-tcp-keepalive") that
   allows DNS servers to signal a variable idle timeout.  This
   signalling encourages the use of long-lived TCP connections by
   allowing the state associated with TCP transport to be managed
   effectively with minimal impact on the DNS transaction time.

Status of This Memo



   This is an Internet Standards Track document.

   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
   Internet Standards is available in Section 2 of RFC 5741.

   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/rfc7828.









Wouters, et al.              Standards Track                    [Page 1]

RFC 7828           The edns-tcp-keepalive EDNS0 Option        April 2016


Copyright Notice



   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.

Table of Contents



   1.  Introduction  . . . . . . . . . . . . . . . . . . . . . . . .   3
   2.  Requirements Notation . . . . . . . . . . . . . . . . . . . .   4
   3.  The edns-tcp-keepalive Option . . . . . . . . . . . . . . . .   5
     3.1.  Option Format . . . . . . . . . . . . . . . . . . . . . .   5
     3.2.  Use by DNS Clients  . . . . . . . . . . . . . . . . . . .   5
       3.2.1.  Sending Queries . . . . . . . . . . . . . . . . . . .   5
       3.2.2.  Receiving Responses . . . . . . . . . . . . . . . . .   6
     3.3.  Use by DNS Servers  . . . . . . . . . . . . . . . . . . .   6
       3.3.1.  Receiving Queries . . . . . . . . . . . . . . . . . .   6
       3.3.2.  Sending Responses . . . . . . . . . . . . . . . . . .   6
     3.4.  TCP Session Management  . . . . . . . . . . . . . . . . .   7
     3.5.  Non-clean Paths . . . . . . . . . . . . . . . . . . . . .   8
     3.6.  Anycast Considerations  . . . . . . . . . . . . . . . . .   8
   4.  Intermediary Considerations . . . . . . . . . . . . . . . . .   8
   5.  Security Considerations . . . . . . . . . . . . . . . . . . .   9
   6.  IANA Considerations . . . . . . . . . . . . . . . . . . . . .   9
   7.  References  . . . . . . . . . . . . . . . . . . . . . . . . .   9
     7.1.  Normative References  . . . . . . . . . . . . . . . . . .   9
     7.2.  Informative References  . . . . . . . . . . . . . . . . .  10
   Acknowledgements  . . . . . . . . . . . . . . . . . . . . . . . .  11
   Authors' Addresses  . . . . . . . . . . . . . . . . . . . . . . .  11













Wouters, et al.              Standards Track                    [Page 2]

RFC 7828           The edns-tcp-keepalive EDNS0 Option        April 2016


1.  Introduction



   DNS messages between clients and servers may be received over either
   UDP or TCP [RFC1035].  Historically, DNS clients used APIs that only
   facilitated sending and receiving a single query over either UDP or
   TCP.  New APIs and deployment of DNSSEC validating resolvers on hosts
   that in the past were using stub resolving only is increasing the DNS
   client base that prefer using long-lived TCP connections.  Long-lived
   TCP connections can result in lower request latency than the case
   where UDP transport is used and truncated responses are received.
   This is because clients that retry over TCP following a truncated UDP
   response typically only use the TCP session for a single (request,
   response) pair, continuing with UDP transport for subsequent queries.

   The use of TCP transport requires state to be retained on DNS
   servers.  If a server is to perform adequately with a significant
   query load received over TCP, it must manage its available resources
   to ensure that all established TCP sessions are well-used, and idle
   connections are closed after an appropriate amount of time.

   UDP transport is stateless, and hence presents a much lower resource
   burden on a busy DNS server than TCP.  An exchange of DNS messages
   over UDP can also be completed in a single round trip between
   communicating hosts, resulting in optimally short transaction times.
   UDP transport is not without its risks, however.

   A single-datagram exchange over UDP between two hosts can be
   exploited to enable a reflection attack on a third party.  Response
   Rate Limiting [RRL] is designed to help mitigate such attacks against
   authoritative-only servers.  One feature of RRL is to let some amount
   of responses "slip" through the rate limiter.  These are returned
   with the TC (truncation) bit set, which causes legitimate clients to
   resend the same query using TCP transport.

   [RFC1035] specified a maximum DNS message size over UDP transport of
   512 bytes.  Deployment of DNSSEC [RFC4033] and other protocols
   subsequently increased the observed frequency at which responses
   exceed this limit.  EDNS0 [RFC6891] allows DNS messages larger than
   512 bytes to be exchanged over UDP, with a corresponding increased
   incidence of fragmentation.  Fragmentation is known to be problematic
   in general, and has also been implicated in increasing the risk of
   cache poisoning attacks [fragmentation-considered-poisonous].

   TCP transport is less susceptible to the risks of fragmentation and
   reflection attacks.  However, TCP transport for DNS as currently
   deployed has expensive setup overhead, compared to using UDP (when no
   retry is required).




Wouters, et al.              Standards Track                    [Page 3]

RFC 7828           The edns-tcp-keepalive EDNS0 Option        April 2016


   The overhead of the three-way TCP handshake for a single DNS
   transaction is substantial, increasing the transaction time for a
   single (request, response) pair of DNS messages from 1x RTT to 2x
   RTT.  There is no such overhead for a session that is already
   established; therefore, the overhead of the initial TCP handshake is
   minimised when the resulting session is used to exchange multiple DNS
   message pairs over a single session.  The extra RTT time for session
   setup can be represented as the equation (1 + N)/N, where N
   represents the number of DNS message pairs that utilize the session
   and the result approaches unity as N increases.

   With increased deployment of DNSSEC and new RR types containing
   application-specific cryptographic material, there is an increase in
   the prevalence of truncated responses received over UDP with retries
   over TCP.  The overhead for a DNS transaction over UDP truncated due
   to RRL is 3x RTT higher than the overhead imposed on the same
   transaction initiated over TCP.

   This document proposes a signalling mechanism between DNS clients and
   servers that encourages the use of long-lived TCP connections by
   allowing the state associated with TCP transport to be managed
   effectively with minimal impact on the DNS transaction time.

   This mechanism will be of benefit for both stub-resolver and
   resolver-authoritative TCP connections.  In the latter case, the
   persistent nature of the TCP connection can provide improved defence
   against attacks including DDoS.

   The reduced overhead of this extension adds up significantly when
   combined with other EDNS0 extensions, such as [CHAIN-QUERY] and
   [DNS-over-TLS].  For example, the combination of these EDNS0
   extensions make it possible for hosts on high-latency mobile networks
   to natively and efficiently perform DNSSEC validation and encrypt
   queries.

2.  Requirements Notation



   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].











Wouters, et al.              Standards Track                    [Page 4]

RFC 7828           The edns-tcp-keepalive EDNS0 Option        April 2016


3.  The edns-tcp-keepalive Option



   This document specifies a new EDNS0 [RFC6891] option, edns-tcp-
   keepalive, which can be used by DNS clients and servers to signal a
   willingness to keep an idle TCP session open to conduct future DNS
   transactions, with the idle timeout being specified by the server.
   This specification does not distinguish between different types of
   DNS client and server in the use of this option.

   [RFC7766] defines an 'idle DNS-over-TCP session' from both the client
   and server perspective.  The idle timeout described here begins when
   the idle condition is met per that definition and should be reset
   when that condition is lifted, i.e., when a client sends a message or
   when a server receives a message on an idle connection.

3.1.  Option Format



   The edns-tcp-keepalive option is encoded as follows:

                        1                   2                   3
    0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
   +-------------------------------+-------------------------------+
   !         OPTION-CODE           !         OPTION-LENGTH         !
   +-------------------------------+-------------------------------+
   |           TIMEOUT             !
   +-------------------------------+

   where:

   OPTION-CODE:   the EDNS0 option code assigned to edns-tcp-keepalive,
      11

   OPTION-LENGTH:   the value 0 if the TIMEOUT is omitted, the value 2
      if it is present;

   TIMEOUT:   an idle timeout value for the TCP connection, specified in
      units of 100 milliseconds, encoded in network byte order.

3.2.  Use by DNS Clients



3.2.1.  Sending Queries



   DNS clients MUST NOT include the edns-tcp-keepalive option in queries
   sent using UDP transport.

   DNS clients MAY include the edns-tcp-keepalive option in the first
   query sent to a server using TCP transport to signal their desire to
   keep the connection open when idle.



Wouters, et al.              Standards Track                    [Page 5]

RFC 7828           The edns-tcp-keepalive EDNS0 Option        April 2016


   DNS clients MAY include the edns-tcp-keepalive option in subsequent
   queries sent to a server using TCP transport to signal their
   continued desire to keep the connection open when idle.

   Clients MUST specify an OPTION-LENGTH of 0 and omit the TIMEOUT
   value.

3.2.2.  Receiving Responses



   A DNS client that receives a response using UDP transport that
   includes the edns-tcp-keepalive option MUST ignore the option.

   A DNS client that receives a response using TCP transport that
   includes the edns-tcp-keepalive option MAY keep the existing TCP
   session open when it is idle.  It SHOULD honour the timeout received
   in that response (overriding any previous timeout) and initiate close
   of the connection before the timeout expires.

   A DNS client that receives a response that includes the edns-tcp-
   keepalive option with a TIMEOUT value of 0 SHOULD send no more
   queries on that connection and initiate closing the connection as
   soon as it has received all outstanding responses.

   A DNS client that sent a query containing the edns-keepalive-option
   but receives a response that does not contain the edns-keepalive-
   option SHOULD assume the server does not support keepalive and behave
   following the guidance in [RFC7766].  This holds true even if a
   previous edns-keepalive-option exchange occurred on the existing TCP
   connection.

3.3.  Use by DNS Servers



3.3.1.  Receiving Queries



   A DNS server that receives a query using UDP transport that includes
   the edns-tcp-keepalive option MUST ignore the option.

   A DNS server that receives a query using TCP transport that includes
   the edns-tcp-keepalive option MAY modify the local idle timeout
   associated with that TCP session if resources permit.

3.3.2.  Sending Responses



   A DNS server that receives a query sent using TCP transport that
   includes an OPT RR (with or without the edns-tcp-keepalive option)
   MAY include the edns-tcp-keepalive option in the response to signal
   the expected idle timeout on a connection.  Servers MUST specify the
   TIMEOUT value that is currently associated with the TCP session.  It



Wouters, et al.              Standards Track                    [Page 6]

RFC 7828           The edns-tcp-keepalive EDNS0 Option        April 2016


   is reasonable for this value to change according to local resource
   constraints.  The DNS server SHOULD send an edns-tcp-keepalive option
   with a timeout of 0 if it deems its local resources are too low to
   service more TCP keepalive sessions or if it wants clients to close
   currently open connections.

3.4.  TCP Session Management



   Both DNS clients and servers are subject to resource constraints that
   will limit the extent to which TCP sessions can persist.  Effective
   limits for the number of active sessions that can be maintained on
   individual clients and servers should be established, either as
   configuration options or by interrogation of process limits imposed
   by the operating system.  Servers that implement edns-tcp-keepalive
   should also engage in TCP connection management by recycling existing
   connections when appropriate, closing connections gracefully, and
   managing request queues to enable fair use.

   In the event that there is greater demand for TCP sessions than can
   be accommodated, servers may reduce the TIMEOUT value signalled in
   successive DNS messages to minimise idle time on existing sessions.
   This also allows, for example, clients with other candidate servers
   to query to establish new TCP sessions with different servers in
   expectation that an existing session is likely to be closed or to
   fall back to UDP.

   Based on TCP session resources, servers may signal a TIMEOUT value of
   0 to request clients to close connections as soon as possible.  This
   is useful when server resources become very low or a denial-of-
   service attack is detected and further maximises the shifting of
   TIME_WAIT state to well-behaved clients.

   However, it should be noted that RFC 6891 states:

      Lack of presence of an OPT record in a request MUST be taken as an
      indication that the requestor does not implement any part of this
      specification and that the responder MUST NOT include an OPT
      record in its response.

   Since servers must be faithful to this specification even on a
   persistent TCP connection, it means that (following the initial
   exchange of timeouts) a server may not be presented with the
   opportunity to signal a change in the idle timeout associated with a
   connection if the client does not send any further requests
   containing EDNS0 OPT RRs.  This limitation makes persistent
   connection handling via an initial idle timeout signal more





Wouters, et al.              Standards Track                    [Page 7]

RFC 7828           The edns-tcp-keepalive EDNS0 Option        April 2016


   attractive than a mechanism that establishes default persistence and
   then uses a connection close signal (in a similar manner to HTTP 1.1
   [RFC7230]).

   If a client includes the edns-tcp-keepalive option in the first
   query, it SHOULD include an EDNS0 OPT RR periodically in any further
   messages it sends during the TCP session.  This will increase the
   chance of the client being notified should the server modify the
   timeout associated with a session.  The algorithm for choosing when
   to do this is out of scope of this document and is left up to the
   implementor and/or operator.

   DNS clients and servers MAY close a TCP session at any time in order
   to manage local resource constraints.  The algorithm by which clients
   and servers rank active TCP sessions in order to determine which to
   close is not specified in this document.

3.5.  Non-clean Paths



   Many paths between DNS clients and servers suffer from poor hygiene,
   limiting the free flow of DNS messages that include particular EDNS0
   options or messages that exceed a particular size.  A fallback
   strategy similar to that described in [RFC6891], Section 6.2.2 SHOULD
   be employed to avoid persistent interference due to non-clean paths.

3.6.  Anycast Considerations



   DNS servers of various types are commonly deployed using anycast
   [RFC4786].

   Changes in network topology between clients and anycast servers may
   cause disruption to TCP sessions making use of edns-tcp-keepalive
   more often than with TCP sessions that omit it, since the TCP
   sessions are expected to be longer lived.  It might be possible for
   anycast servers to avoid disruption due to topology changes by making
   use of TCP multipath [RFC6824] to anchor the server side of the TCP
   connection to an unambiguously unicast address.

4.  Intermediary Considerations



   It is RECOMMENDED that DNS intermediaries that terminate TCP
   connections implement edns-tcp-keepalive.  An intermediary that does
   not implement edns-tcp-keepalive but sits between a client and server
   that both support edns-tcp-keepalive might close idle connections
   unnecessarily.






Wouters, et al.              Standards Track                    [Page 8]

RFC 7828           The edns-tcp-keepalive EDNS0 Option        April 2016


5.  Security Considerations



   The edns-tcp-keepalive option can potentially be abused to request
   large numbers of long-lived sessions in a quick burst.  When a DNS
   server detects abusive behaviour, it SHOULD immediately close the TCP
   connection and free the resources used.

   Servers could choose to monitor client behaviour with respect to the
   edns-tcp-keepalive option to build up profiles of clients that do not
   honour the specified timeout.

   Readers are advised to familiarise themselves with the security
   considerations outlined in [RFC7766]

6.  IANA Considerations



   IANA has assigned an EDNS0 option code for the edns-tcp-keepalive
   option from the "DNS EDNS0 Option Codes (OPT)" registry as follows:

           +-------+--------------------+----------+-----------+
           | Value | Name               | Status   | Reference |
           +-------+--------------------+----------+-----------+
           | 11    | edns-tcp-keepalive | Standard | RFC 7828  |
           +-------+--------------------+----------+-----------+

7.  References



7.1.  Normative References



   [RFC1035]  Mockapetris, P., "Domain names - implementation and
              specification", STD 13, RFC 1035, DOI 10.17487/RFC1035,
              November 1987, <http://www.rfc-editor.org/info/rfc1035>.

   [RFC2119]  Bradner, S., "Key words for use in RFCs to Indicate
              Requirement Levels", BCP 14, RFC 2119,
              DOI 10.17487/RFC2119, March 1997,
              <http://www.rfc-editor.org/info/rfc2119>.

   [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,
              <http://www.rfc-editor.org/info/rfc4033>.

   [RFC4786]  Abley, J. and K. Lindqvist, "Operation of Anycast
              Services", BCP 126, RFC 4786, DOI 10.17487/RFC4786,
              December 2006, <http://www.rfc-editor.org/info/rfc4786>.





Wouters, et al.              Standards Track                    [Page 9]

RFC 7828           The edns-tcp-keepalive EDNS0 Option        April 2016


   [RFC6891]  Damas, J., Graff, M., and P. Vixie, "Extension Mechanisms
              for DNS (EDNS(0))", STD 75, RFC 6891,
              DOI 10.17487/RFC6891, April 2013,
              <http://www.rfc-editor.org/info/rfc6891>.

   [RFC7230]  Fielding, R., Ed. and J. Reschke, Ed., "Hypertext Transfer
              Protocol (HTTP/1.1): Message Syntax and Routing",
              RFC 7230, DOI 10.17487/RFC7230, June 2014,
              <http://www.rfc-editor.org/info/rfc7230>.

   [RFC7766]  Dickinson, J., Dickinson, S., Bellis, R., Mankin, A., and
              D. Wessels, "DNS Transport over TCP - Implementation
              Requirements", RFC 7766, DOI 10.17487/RFC7766, March 2016,
              <http://www.rfc-editor.org/info/rfc7766>.

7.2.  Informative References



   [CHAIN-QUERY]
              Wouters, P., "Chain Query requests in DNS", Work in
              Progress, draft-ietf-dnsop-edns-chain-query-07, February
              2016.

   [DNS-over-TLS]
              Hu, Z., Zhu, L., Heidemann, J., Mankin, A., Wessels, D.,
              and P. Hoffman, "Specification for DNS over TLS", Work in
              Progress, draft-ietf-dprive-dns-over-tls-09, March 2016.

   [fragmentation-considered-poisonous]
              Herzberg, A. and H. Shulman, "Fragmentation Considered
              Poisonous", arXiv: 1205.4011, May 2012,
              <http://arxiv.org/abs/1205.4011>.

   [RFC6824]  Ford, A., Raiciu, C., Handley, M., and O. Bonaventure,
              "TCP Extensions for Multipath Operation with Multiple
              Addresses", RFC 6824, DOI 10.17487/RFC6824, January 2013,
              <http://www.rfc-editor.org/info/rfc6824>.

   [RRL]      Vixie, P. and V. Schryver, "DNS Response Rate Limiting
              (DNS RRL)", ISC-TN 2012-1-Draft1, April 2012,
              <https://ftp.isc.org/isc/pubs/tn/isc-tn-2012-1.txt>.











Wouters, et al.              Standards Track                   [Page 10]

RFC 7828           The edns-tcp-keepalive EDNS0 Option        April 2016


Acknowledgements



   The authors acknowledge the contributions of Jinmei TATUYA and Mark
   Andrews.  Thanks to Duane Wessels for detailed review and the many
   others who contributed to the mailing list discussion.

Authors' Addresses



   Paul Wouters
   Red Hat

   Email: pwouters@redhat.com


   Joe Abley
   Dyn, Inc.
   103-186 Albert Street
   London, ON  N6A 1M1
   Canada

   Phone: +1 519 670 9327
   Email: jabley@dyn.com


   Sara Dickinson
   Sinodun Internet Technologies
   Magdalen Centre
   Oxford Science Park
   Oxford  OX4 4GA
   United Kingdom

   Email: sara@sinodun.com
   URI:   http://sinodun.com


   Ray Bellis
   Internet Systems Consortium, Inc
   950 Charter Street
   Redwood City, CA  94063
   United States

   Phone: +1 650 423 1200
   Email: ray@isc.org
   URI:   http://www.isc.org







Wouters, et al.              Standards Track                   [Page 11]