RFC 4074 INFORMATIONAL DNS · IPV6 · OPERATIONS

AAAA Query Misbehavior: Common Misbehavior Against DNS Queries for IPv6 Addresses

How some DNS servers answer IPv6 address queries wrongly, and why that can break IPv4 connections.

On rfc-editor.org ↗
PUBLISHED
June 2005
AUTHORS
Y. Morishita, T. Jinmei
STREAM
IETF
CORRECT REPLY
RCODE 0, empty answer
NAME ERROR CODE
RCODE 3
SERVER FAILURE CODE
RCODE 2
BROKEN AAAA DATA LENGTH
4 bytes
DOI
10.17487/RFC4074

RFC text

Contents
Original text · RFC Editor View on rfc-editor.org ↗
Network Working Group                                       Y. Morishita
Request for Comments: 4074                                          JPRS
Category: Informational                                        T. Jinmei
                                                                 Toshiba
                                                                May 2005


       Common Misbehavior Against DNS Queries for IPv6 Addresses

Status of This Memo

   This memo provides information for the Internet community.  It does
   not specify an Internet standard of any kind.  Distribution of this
   memo is unlimited.

Copyright Notice

   Copyright (C) The Internet Society (2005).

Abstract

   There is some known misbehavior of DNS authoritative servers when
   they are queried for AAAA resource records.  Such behavior can block
   IPv4 communication that should actually be available, cause a
   significant delay in name resolution, or even make a denial of
   service attack.  This memo describes details of known cases and
   discusses their effects.

1.  Introduction

   Many existing DNS clients (resolvers) that support IPv6 first search
   for AAAA Resource Records (RRs) of a target host name, and then for A
   RRs of the same name.  This fallback mechanism is based on the DNS
   specifications, which if not obeyed by authoritative servers, can
   produce unpleasant results.  In some cases, for example, a web
   browser fails to connect to a web server it could otherwise reach.
   In the following sections, this memo describes some typical cases of
   such misbehavior and its (bad) effects.

   Note that the misbehavior is not specific to AAAA RRs.  In fact, all
   known examples also apply to the cases of queries for MX, NS, and SOA
   RRs.  The authors believe this can be generalized for all types of
   queries other than those for A RRs.  In this memo, however, we
   concentrate on the case for AAAA queries, since the problem is
   particularly severe for resolvers that support IPv6, which thus
   affects many end users.  Resolvers at end users normally send A
   and/or AAAA queries only, so the problem for the other cases is
   relatively minor.

2.  Network Model

   In this memo, we assume a typical network model of name resolution
   environment using DNS.  It consists of three components: stub
   resolvers, caching servers, and authoritative servers.  A stub
   resolver issues a recursive query to a caching server, which then
   handles the entire name resolution procedure recursively.  The
   caching server caches the result of the query and sends the result to
   the stub resolver.  The authoritative servers respond to queries for
   names for which they have the authority, normally in a non-recursive
   manner.

3.  Expected Behavior

   Suppose that an authoritative server has an A RR but has no AAAA RR
   for a host name.  Then, the server should return a response to a
   query for an AAAA RR of the name with the response code (RCODE) being
   0 (indicating no error) and with an empty answer section (see
   Sections 4.3.2 and 6.2.4 of [1]).  Such a response indicates that
   there is at least one RR of a different type than AAAA for the
   queried name, and the stub resolver can then look for A RRs.

   This way, the caching server can cache the fact that the queried name
   has no AAAA RR (but may have other types of RRs), and thus improve
   the response time to further queries for an AAAA RR of the name.

4.  Problematic Behaviors

   There are some known cases at authoritative servers that do not
   conform to the expected behavior.  This section describes those
   problematic cases.

4.1.  Ignore Queries for AAAA

   Some authoritative servers seem to ignore queries for an AAAA RR,
   causing a delay at the stub resolver to fall back to a query for an A
   RR.  This behavior may cause a fatal timeout at the resolver or at
   the application that calls the resolver.  Even if the resolver
   eventually falls back, the result can be an unacceptable delay for
   the application user, especially with interactive applications like
   web browsing.

4.2.  Return "Name Error"

   This type of server returns a response with RCODE 3 ("Name Error") to
   a query for an AAAA RR, indicating that it does not have any RRs of
   any type for the queried name.

   With this response, the stub resolver may immediately give up and
   never fall back.  Even if the resolver retries with a query for an A
   RR, the negative response for the name has been cached in the caching
   server, and the caching server will simply return the negative
   response.  As a result, the stub resolver considers this to be a
   fatal error in name resolution.

   Several examples of this behavior are known to the authors.  As of
   this writing, all have been fixed.

4.3.  Return Other Erroneous Codes

   Other authoritative servers return a response with erroneous response
   codes other than RCODE 3 ("Name Error").  One such RCODE is 4 ("Not
   Implemented"), indicating that the servers do not support the
   requested type of query.

   These cases are less harmful than the previous one; if the stub
   resolver falls back to querying for an A RR, the caching server will
   process the query correctly and return an appropriate response.

   However, these can still cause a serious effect.  There was an
   authoritative server implementation that returned RCODE 2 ("Server
   failure") to queries for AAAA RRs.  One widely deployed mail server
   implementation with a certain type of resolver library interpreted
   this result as an indication of retry and did not fall back to
   queries for A RRs, causing message delivery failure.

   If the caching server receives a response with these response codes,
   it does not cache the fact that the queried name has no AAAA RR,
   resulting in redundant queries for AAAA RRs in the future.  The
   behavior will waste network bandwidth and increase the load of the
   authoritative server.

   Using RCODE 1 ("Format error") would cause a similar effect, though
   the authors have not seen such implementations yet.

4.4.  Return a Broken Response

   Another type of authoritative servers returns broken responses to
   AAAA queries.  Returning a response whose RR type is AAAA with the
   length of the RDATA being 4 bytes is a known behavior of this
   category.  The 4-byte data looks like the IPv4 address of the queried
   host name.

   That is, the RR in the answer section would be described as follows:

     www.bad.example. 600 IN AAAA 192.0.2.1

   which is, of course, bogus (or at least meaningless).

   A widely deployed caching server implementation transparently returns
   the broken response (and caches it) to the stub resolver.  Another
   known server implementation parses the response by itself, and sends
   a separate response with RCODE 2 ("Server failure").

   In either case, the broken response does not affect queries for an A
   RR of the same name.  If the stub resolver falls back to A queries,
   it will get an appropriate response.

   The latter case, however, causes the same bad effect as that
   described in the previous section: redundant queries for AAAA RRs.

4.5.  Make Lame Delegation

   Some authoritative servers respond to AAAA queries in a way that
   causes lame delegation.  In this case, the parent zone specifies that
   the authoritative server should have the authority of a zone, but the
   server should not return an authoritative response for AAAA queries
   within the zone (i.e., the AA bit in the response is not set).  On
   the other hand, the authoritative server returns an authoritative
   response for A queries.

   When a caching server asks the server for AAAA RRs in the zone, it
   recognizes the delegation is lame, and returns a response with RCODE
   2 ("Server failure") to the stub resolver.

   Furthermore, some caching servers record the authoritative server as
   lame for the zone and will not use it for a certain period of time.
   With this type of caching server, even if the stub resolver falls
   back to querying for an A RR, the caching server will simply return a
   response with RCODE 2, since all the servers are known to be "lame."

   There is also an implementation that relaxes the behavior a little
   bit.  It tries to avoid using the lame server, but continues to try
   it as a last resort.  With this type of caching server, the stub
   resolver will get a correct response if it falls back after Server
   failure.  However, this still causes redundant AAAA queries, as
   explained in the previous sections.

5.  Security Considerations

   The CERT/CC pointed out that the response with RCODE 3 ("Name
   Error"), described in Section 4.2, can be used for a denial of
   service attack [2].  The same argument applies to the case of "lame
   delegation", described in Section 4.5, with a certain type of caching
   server.

6.  Acknowledgements

   Erik Nordmark encouraged the authors to publish this document as an
   RFC.  Akira Kato and Paul Vixie reviewed a preliminary version of
   this document.  Pekka Savola carefully reviewed a previous version
   and provided detailed comments.  Bill Fenner, Scott Hollenbeck,
   Thomas Narten, and Alex Zinin reviewed and helped improve the
   document at the last stage for publication.

7.  Informative References

   [1]  Mockapetris, P., "Domain names - concepts and facilities", STD
        13, RFC 1034, November 1987.

   [2]  The CERT Coordination Center, "Incorrect NXDOMAIN responses from
        AAAA queries could cause denial-of-service conditions",
        March 2003, <http://www.kb.cert.org/vuls/id/714121>.

Authors' Addresses

   MORISHITA Orange Yasuhiro
   Research and Development Department, Japan Registry Services Co.,Ltd.
   Chiyoda First Bldg. East 13F, 3-8-1 Nishi-Kanda
   Chiyoda-ku, Tokyo  101-0065
   Japan

   EMail: yasuhiro@jprs.co.jp


   JINMEI Tatuya
   Corporate Research & Development Center, Toshiba Corporation
   1 Komukai Toshiba-cho, Saiwai-ku
   Kawasaki-shi, Kanagawa  212-8582
   Japan

   EMail: jinmei@isl.rdc.toshiba.co.jp

Full Copyright Statement

   Copyright (C) The Internet Society (2005).

   This document is subject to the rights, licenses and restrictions
   contained in BCP 78, and except as set forth therein, the authors
   retain all their rights.

   This document and the information contained herein are provided on an
   "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS
   OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY AND THE INTERNET
   ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED,
   INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE
   INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED
   WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.

Intellectual Property

   The IETF takes no position regarding the validity or scope of any
   Intellectual Property Rights or other rights that might be claimed to
   pertain to the implementation or use of the technology described in
   this document or the extent to which any license under such rights
   might or might not be available; nor does it represent that it has
   made any independent effort to identify any such rights.  Information
   on the procedures with respect to rights in RFC documents can be
   found in BCP 78 and BCP 79.

   Copies of IPR disclosures made to the IETF Secretariat and any
   assurances of licenses to be made available, or the result of an
   attempt made to obtain a general license or permission for the use of
   such proprietary rights by implementers or users of this
   specification can be obtained from the IETF on-line IPR repository at
   http://www.ietf.org/ipr.

   The IETF invites any interested party to bring to its attention any
   copyrights, patents or patent applications, or other proprietary
   rights that may cover technology that may be required to implement
   this standard.  Please address the information to the IETF at ietf-
   ipr@ietf.org.

Acknowledgement

   Funding for the RFC Editor function is currently provided by the
   Internet Society.

AI ELI5

On this page
PROBLEMIPv6-capable resolvers ask for AAAA records first and then fall back to A records, but some authoritative servers answer AAAA queries incorrectly.
MECHANISMThe memo lists five known misbehaviors: ignoring the query, returning Name Error, returning other error codes, returning a broken answer, and creating a lame delegation. It explains the effect of each.
WHY IT MATTERSThe results include blocked IPv4 connectivity, long delays, wasted queries and a possible denial of service. The correct behavior is an empty answer with RCODE 0.

§1Why AAAA Queries Matter

Many resolvers that support IPv6 first look up AAAA resource records (IPv6 addresses) for a host name and then look up A records (IPv4 addresses) for the same name. This fallback follows the DNS specifications. If authoritative servers do not obey them, a web browser may fail to connect to a server it could otherwise reach.

The memo assumes three components: stub resolvers, caching servers and authoritative servers. A stub resolver sends a recursive query to a caching server, which does the full resolution and caches the result. Authoritative servers answer for their own names, normally non-recursively. The authors say the misbehavior is not specific to AAAA: all known examples also apply to MX, NS and SOA queries. They focus on AAAA because IPv6-capable resolvers at end users make the problem particularly severe.

Your computer first asks for the new kind of address, and if there is none it asks for the old kind. That only works if the answer to the first question is honest.

§3The Expected Answer

If a server has an A record but no AAAA record for a name, it should answer an AAAA query with RCODE 0 (no error) and an empty answer section. That tells the stub resolver the name exists with at least one record of another type, so it can look for A records.

It also lets the caching server remember that this name has no AAAA record while other types may exist, which speeds up later queries.

The right answer is 'I have no new-style address for that name, but the name is real.' Then everyone can move on and try the old kind.

§4Five Kinds of Misbehavior

Ignoring the query. The stub resolver waits and may hit a fatal timeout, or the user suffers a long delay, which is worst for interactive uses like web browsing.

Returning Name Error (RCODE 3). This says no records of any type exist for the name. The resolver may give up and never fall back, and the caching server caches the negative response and returns it for A queries too. The authors say all the examples known to them have been fixed.

Other error codes. RCODE 4 (Not Implemented) is less harmful because a fallback to A works. But one server returned RCODE 2 (Server failure), and a widely deployed mail server with a certain resolver library treated that as a reason to retry and never fell back, so mail delivery failed. Caching servers do not cache these errors, which causes redundant AAAA queries, wasted bandwidth and more load on the authoritative server. RCODE 1 (Format error) would have a similar effect, but the authors had not seen it.

Broken responses. Some servers return an AAAA record whose data is only 4 bytes, like an IPv4 address (the memo shows www.bad.example. 600 IN AAAA 192.0.2.1). One widely deployed caching server passes and caches it. Another replies with RCODE 2, with the redundant-query effect.

Lame delegation. The parent zone names a server as authoritative, but that server does not give an authoritative answer (the AA bit is not set) to AAAA queries while it does for A queries. The caching server treats the delegation as lame and returns RCODE 2. Some caching servers then mark the server lame for a period, so even the A fallback fails. Others still try it as a last resort, which only costs redundant AAAA queries.

Bad servers might stay silent, say the name does not exist, shout an error, hand back nonsense, or act like they are not in charge. Each one can leave your computer stuck or doing useless work.

RESPONSE CODES DISCUSSED IN RFC 4074
  • 0No error — Correct reply to an AAAA query for a name with no AAAA record, with an empty answer.
  • 1Format error — Would have a similar effect to other error codes; not seen in implementations by the authors.
  • 2Server failure — Seen from a server and from caching servers after bad or lame answers; may stop the fallback.
  • 3Name Error — Says the name has no records of any type; can stop the fallback and be cached.
  • 4Not Implemented — The server does not support the requested query type.

§5Denial of Service

The CERT/CC pointed out that the Name Error (RCODE 3) response described above can be used for a denial of service attack. The memo says the same argument applies to lame delegation with a certain type of caching server.

If a server wrongly says a real name does not exist, people cannot reach that place, and a bad actor could use that on purpose.

§5References

[1]RFC 1034Mockapetris, P., "Domain names - concepts and facilities", STD 13, RFC 1034, November 1987.November 1987
[2]The CERT Coordination Center, "Incorrect NXDOMAIN responses from AAAA queries could cause denial-of-service conditions", March 2003.