[Docs] [txt|pdf] [Tracker] [Email] [Diff1] [Diff2] [Nits]

Versions: 00 01 02 03

Network Working Group                                        Keith Moore
Internet-Draft                                   University of Tennessee
2 March 2001
Expires: 2 September 2001

                              6to4 and DNS


     This document is an Internet-Draft and is in full conformance with
all provisions of Section 10 of RFC2026.

     Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF), its areas, and its working groups.  Note that other
groups may also distribute working documents as Internet-Drafts.

     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."

     The list of current Internet-Drafts can be accessed at

     The list of Internet-Draft Shadow Directories can be accessed at

     Comments regarding this internet-draft should be sent to the
mailing list of the IETF ngtrans working group.  Refer to the IETF web
site at http://www.ietf.org/ for current contact information for IETF
working groups.  Please include the document identifier "draft-
moore-6to4-dns-02.txt" in any comments regarding this document.


     This memo discusses several potential mechanisms for locating the
DNS servers which provide "reverse lookup" of 6to4 addresses.

     Please note that this is a preliminary draft which only attempts to
outline possible means of solving the problem, for purpose of
discussion.  This version of the proposal is NOT rigorously specified,
and the author claims to have significant expertise in neither DNS nor
anycast.  Nevertheless, it is hoped that these proposals are
sufficiently detailed to allow reviewers to make a first-order
assessment of their viability.  The assistance of appropriate experts in
drafting future revisions of these proposals would be most welcome.

Moore                   Expires 2 September 2001                [Page 1]

6to4 and DNS                 INTERNET-DRAFT                 2 March 2001

1. Introduction

     6to4 [1] defines a mechanism for allowing sites to communicate
using IPv6 over the public IPv4 Internet.  It does so by assigning a
block of IPv6 addresses corresponding to any "public" (globally-scoped)
IPv4 address, and a means of tunneling IPv6 traffic destined for such
addresses over the IPv4 Internet.  In this way, any site which is
connected to the IPv4 Internet and which has at least one global IPv4
address assigned to it, can communicate with IPv6.

     The advantage of 6to4 is that it decouples deployment of IPv6 by
the core of the network (e.g. Internet Service Providers or ISPs) from
deployment of IPv6 at the edges (e.g. customer sites), allowing each
site or ISP to deploy IPv6 support in its own time frame according to
its own priorities.  With 6to4, the edges may communicate with one
another using IPv6 even if one or more of their ISPs do not yet provide
native IPv6 service.  In addition, the principal cost of the 6to4
transition mechanism is borne by those who benefit from it.

     However, the ability to perform so-called "reverse lookups" (IP
address to domain name lookups) in DNS requires that there be a
delegation path corresponding to the IP address being queried, from the
DNS root to the DNS server which provides PTR records for that IP
address.  Under normal circumstances, the necessary DNS servers and
records for IPv6 reverse lookups would be maintained by the each
organization to which an address block is delegated; the delegation path
of DNS records reflects the delegation of address blocks themselves.
For IPv6 addresses beginning with the 6to4 address prefix, the DNS
records would need to reflect IPv4 address delegation, and might not be
present for a 6to4 site whose ISP did not yet support IPv6 in any form.

     This memo discusses several potential mechanisms for locating the
DNS servers which are assumed to provide "reverse lookup" of 6to4

     The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
document are to be interpreted as described in RFC 2119.

     The characters { and } are used to indicate protocol elements where
literal DNS labels or addresses would appear in actual use; neither
these delimiters nor the text appearing within are to be interpreted

2. Design Goals

     An ideal solution to this problem would have several

Moore                   Expires 2 September 2001                [Page 2]

6to4 and DNS                 INTERNET-DRAFT                 2 March 2001

-    Minimal impact on existing software and operations.

-    Reasonable efficiency for lookup of names corresponding to 6to4

-    Minimal effort in deployment of DNS support.

-    Costs borne primarily by those who immediately benefit.

-    Does not adversely affect security of DNS queries.

-    Any assumptions made by client or server software as to the
     location of authoritative DNS server(s) for reverse lookup of a
     6to4 address, are made only if no explicit referral information is

     However, no attempt has yet been made to establish relative impor-
tance of these goals.

3. Methods of Inferring Delegation Paths

     The author has identified two methods of inferring delegation paths
in the absence of explicit delegation information (NS or DNAME records)
for reverse lookups of IPv6 addresses in the DNS.  The first is to
assume that the default DNS servers for reverse lookup of a 6to4 address
are the  same servers that are responsible for reverse lookup of the
corresponding IPv4 address.  The second is to assume that the default
DNS servers for reverse lookup of a 6to4 address are reachable via some
well-known anycast address which is derivable from the 6to4 prefix.
While it might be possible to employ both of these methods, or use them
in some combination, at first glance it seems better to choose one
method or the other.

     In both methods, the actual PTR records for 6to4 addresses are
explicitly maintained by the site to which that portion of 6to4 space is
assigned (i.e. the site to whom the corresponding portion of IPv4 space
is assigned).  This proposal never makes assumptions about the mapping
between specific 6to4 addresses and specific host names.

     Note also that both methods infer NS records, rather than DNAME
records [2], for query referral.  This is because it seems undesirable
for any automatically generated resource record, or a resource record
which is assumed by a third party, to make assumptions about a different
organization's domain name space.  In other words, while it might seem
fairly safe to say "If there are PTR resource records for an address in
this portion of 6to4 space, they will be found on the same servers as
the PTR records for the corresponding portions of IPv4 space" but not to
say "If there are PTR records for an address in this portion of 6to4

Moore                   Expires 2 September 2001                [Page 3]

6to4 and DNS                 INTERNET-DRAFT                 2 March 2001

space, those records will be named after the DNS name(s) of the
server(s) used for the same portion of IPv4 space".

3.1  6to4 NS records derived from IPv4 NS records

     This method assumes that the default DNS servers for reverse lookup
of a 6to4 address are the same servers that are responsible for reverse
lookup of the corresponding IPv4 address.  In effect, if there is a NS
resource record that refers reverse queries for a portion of IPv4
address space to some set of DNS servers, we want to behave (in the
absence of explicit records to the contrary) as if there is a similar NS
record for the portion of IPv6 address space corresponding to those IPv4

     More formally, for every resource record of the form:

{address-bits}.IN-ADDR.ARPA.  NS   some-domain.example.com.

we want to have the effect of also having a resource record of the form:

{address-bits}.\[x2002].IP6.ARPA. NS    some-domain.example.com.

unless the lookup for the IPv6 address can be fulfilled by explicit (NS
or DNAME) resource records.  The following sections discuss various ways
of producing the effect.  The NS records so generated or assumed (by
whatever means) are termed "pseudo-records" to distinguish them from
explicitly-supplied NS records.

     Note that due to the different ways of representing {address-bits}
in DNS labels between IPv4 and IPv6, and the different ways of referring
queries in each address space, a transformation (to be specified later)
will be required.  The TTLs of the NS pseudo-records so generated should
be no larger than those of the NS records from which they were derived;
in some cases it may be desirable to make them smaller.

     This method has the advantage that 6to4 sites do not need to estab-
lish new DNS servers, nor to get those servers to answer to new
addresses, in order to implement reverse lookup service for 6to4
addresses.  It need only add the appropriate resource records to its
existing DNS servers which perform those functions for IPv4.  However,
this method only works for sites that already operate their own DNS
servers that provide reverse lookup for IPv4 addresses.  Specifically,
sites that have only a single IPv4 address are unlikely to operate their
own DNS servers for IPv4 addresses, but they may be significant users of

Moore                   Expires 2 September 2001                [Page 4]

6to4 and DNS                 INTERNET-DRAFT                 2 March 2001

3.2  6to4 NS records inferred from 6to4 prefix

     This method assumes that the default DNS servers for reverse lookup
of a 6to4 address are reachable at a local (to the destination network)
anycast address which is derived from the 6to4 prefix.

     More formally, in the absence of any explicit DNAME or NS resource
records for the suffix {IPv4-address}.\[x2002].IP6.ARPA, a resource
record of the form

{IPv4-address}.\[x2002].IP6.ARPA. NS    {label}
{label}                     AAAA 2002:{IPv4-address}:{suffix}

where {IPv4-address} is 32 bits of address represented as a bit-string
label; {label} is a domain-name which is created for the purpose of
associating a 6to4 address with its DNS servers (since NS records must
refer to a DNS name rather than an IPv6 address), and {suffix} is a
well-known constant bit pattern (to be determined) which is treated as
an anycast address by the 6to4 network.  The 6to4 network then estab-
lishes one or more DNS servers to listen to that anycast address which
will answer reverse lookup queries.

     Note that if a site uses more than one 6to4 prefix (because it has
more than one IPv4 address assigned to it), its DNS servers which are
responsible for reverse lookups will be required to accept queries at
multiple addresses.

     A variant of this method would be to define a set of suffixes for
this purpose, rather than a single suffix, and to infer a set of AAAA
records (one for each of the suffixes) rather than a single AAAA record.
This would allow a 6to4 site to establish multiple servers for reverse
lookups without having to arrange for anycast access.  (The difficulty
is in arranging for the hosts to respond to NS queries at those
addresses only when the DNS servers on those hosts are correctly
operating; without such a mechanism, client-based fail-over would be
more reliable, if slower, than server selection by anycast.)

3.3  DNS-based Multihoming

     With either of these methods, sites which have multiple IPv4
address blocks and which wish to run "multihomed 6to4" may still do so
by installing their own DNAME records.  That is, if an organization is
assigned {IPv4-prefix-1} and {IPv4-prefix-2}, it may still maintain the
address-to-name mappings of its 6to4 hosts in a single DNS zone, by
creating DNAME records of the form:

Moore                   Expires 2 September 2001                [Page 5]

6to4 and DNS                 INTERNET-DRAFT                 2 March 2001

{IPv4-prefix-1}.\[x2002].IP6.ARPA       DNAME     ip6dns.example.com.
{IPv4-prefix-2}.\[x2002].IP6.ARPA       DNAME     ip6dns.example.com.

on the appropriate DNS servers.  A similar technique can be used to
allow a site to share a single set of PTR records between 6to4 and
native prefixes (and thus ease the transition from 6to4 to native), pro-
vided that the "locally assigned" bits of each 6to4 address will also
fit within the space remaining after each of the "native" prefixes.

4  Methods of adapting existing software to infer delegation paths

     The following paragraphs detail several possible techniques which
might allow existing platforms to infer these delegation paths with
varying degrees of disruption.  They are not mutually-exclusive; it is
possible to employ more than of these techniques.  Some of them are less
attractive than others.  At present the purpose of this document is to
outline several possible approaches, and serve as a focal point of
discussion, rather than to categorically recommend any particular

     Most of these implementation methods can be used with either method
of inferring NS records - either deriving them from v4 NS records (sec-
tion 3.1) or using an anycast address (section 3.2).

4.1. No Change To Any Server Software

     This implementation method makes no changes to any DNS client or
server software.  Rather, it expects that the root servers, ISPs DNS
servers, and the DNS servers of other organizations which delegate IPv4
address space, will be populated with similar NS records which refer
reverse lookup queries from 6to4 space.

     Unless and until the assignee of the IPv4 address requested that
IPv6 queries be referred to different servers (i.e. that new DNAME or NS
RRs be installed), any changes made to the NS records for IPv4 addresses
would also need to be reflected in the corresponding NS records for IPv6
addresses in 6to4 space.

     As stated above, this technique requires no software changes to
either DNS server or client software.  However, it would certainly
require changes to the software used by registries, ISPs, and other
networks, to maintain the DNS records needed to provide reverse lookups.

     This implementation method may be used with either method of
inferring NS records.   In other words, either new NS records could
either be derived from existing NS records for IPv4 addresses, or new NS
and AAAA records could be created assuming that servers would be
established at one or more suffixes within a 6to4 subnet prefix.  In

Moore                   Expires 2 September 2001                [Page 6]

6to4 and DNS                 INTERNET-DRAFT                 2 March 2001

either case a site must be allowed to change the records associated with
its 6to4 prefix after they are established.

     This implementation method avoids kludges to DNS software but is
assumed to be difficult to deploy, as it requires several different
organizations to explicitly support 6to4.

4.2. Pseudo-records generated by DNS servers for the IPv4 zones

     In this technique, the authoritative DNS servers for IN-ADDR.ARPA
and its subdomains would be modified to return "psuedo-records" for any
query of type PTR or NS which matched a name of the form

     In particular,

-    if the server had explicit records matching the label of a PTR
     query, those records would be returned and no pseudo-records would
     be returned;

-    if the server had explicit NS records matching the label or a
     suffix of the label of an NS or PTR query, those records would be
     returned and no pseudo-records would be returned;

-    (if the method in section 3.1 were used) otherwise, if the server
     had NS records matching {something}.IN-ADDR.ARPA, or matching any
     IPv4 address prefix of {something}.IN-ADDR.ARPA, NS pseudo-records
     corresponding to the longest matching prefixes would be returned.
     The pseudo-records so returned would be marked authoritative, and
     their TTLs would be no larger than the TTLs of the explicit records
     from which the pseudo-records were derived.

-    (if the method in section 3.2 is used) otherwise, the server would
     return a NS pseudo-record corresponding to the 6to4 prefix, which
     pointed to a label for which one or more AAAA pseudo-records
     containing the well-known address(es) for reverse lookups at that
     prefix.   The AAAA addresses would be returned as additional
     information in response to the NS or PRT query, but would
     necesarily also be obtainable from a separate AAAA or A6 query for
     any {label} returned in an NS pseudo-record.

     This technique is assumed to be somewhat easier to deploy than the
previous one, because it automates the generation of the pseudo-records
and avoids the need for each organization that delegates IPv4 space to
change its DNS maintenance procedures.  However, it still requires
changes to DNS servers, and it requires those organizations to upgrade
their DNS servers to include those changes, before the change will be
useful.  It also requires cooperation on behalf of the owner of the DNS

Moore                   Expires 2 September 2001                [Page 7]

6to4 and DNS                 INTERNET-DRAFT                 2 March 2001

servers providing lookup for an IPv4 address, which might not be the
same party that is using the corresponding 6to4 addresses.

4.3. Pseudo-records generated by DNS resolvers

     In this technique, DNS servers which act as resolvers behave as if
pseudo-records had been returned to them when some kinds of queries
fail.  In some cases they may return pseudo-records when a query fails.

     When such a resolver received a PTR or NS query for a label that
had a \[x2002].IP6.ARPA suffix, it would first attempt to satisfy that
query from its cache, or failing that, by forwarding the query to an
upstream server.  If that query failed due to a "no such domain" error,
the resolver would then attempt to find the server for the
{something}.\[x2002].IP6.ARPA label by (if the method in section 3.1 is
used) issuing an NS query for {something}.IN-ADDR.ARPA, or (if the
method in section 3.2 is used) inferring NS and AAAA records based on
the 6to4 prefix of the address.

     If the method in section 3.1 were used, and the original query was
for PTR records, and one or more NS records were found for
{something}.IN-ADDR.ARPA, the resolver would then forward the original
query for {something}.\[x2002].IP6.ARPA to one or more of those servers,
and return the results from one of the forwarded queries if any were
successful.  If the original query was for NS records, and one or more
NS records were found for {something}.IN-ADDR.ARPA, the resolver would
then return the pseudo-records corresponding to the IN-ADDR.ARPA
domains.  Those pseudo-records would NOT be marked as authoritative, and
the resolver would NOT cache those records.

     Similarly, if the method in section 3.2 were used, the resolver
would return NS and AAAA pseudo-records derived from the IPv6 address
being queried.

     Note that while the DNS resolver effectively behaves as if pseudo-
records had been returned to it by other servers, it MUST NOT cache
those pseudo-records.  However, it MAY cache the actual NS or PTR
records returned by those servers and use such cached data to generate
additional pseudo-records.

     This technique requires changes to DNS resolver software, and
requires that sites using IPv6 and wishing to communicate with 6to4
sites, upgrade their DNS resolvers to include this change.  However it
does not require changes of IPv6 hosts.

Moore                   Expires 2 September 2001                [Page 8]

6to4 and DNS                 INTERNET-DRAFT                 2 March 2001

4.4 Pseudo-records generated by DNS query libraries

     In this technique, the run-time library used on a host by
applications is modified to process DNS queries in the following manner:

     If the query is of type PTR or NS, and the label queried has a
suffix of \[x2002].IP6.ARPA, or if the query is otherwise intended to
perform an reverse lookup, and the address being looked up is a 6to4
address, an attempt is first made to look up the address via normal
means.  If this attempt failed due to the lack of any delegation of the
6to4 prefix, NS and perhaps AAAA pseudo-records are inferred according
to sections 3.1 and/or 3.2 (whichever ends up being chosen).

     If the method in section 3.1 is chosen, a query is made
(internally) for NS records corresponding to the embedded IPv4 address.
If this secondary query is successful, the original DNS query for the
6to4 address is re-issued to the servers which are authoritative for
that IPv4 address; the result is determined from the response to that

     This technique requires changes to DNS query libraries (or
applications), and requires that hosts and/or applications using IPv6,
and which wish to communicate with hosts and/or applications at 6to4
sites, upgrade their DNS libraries to include this change.

5. Author's Recommendations

     For the purpose of facilitating discussion, the author tentatively
recommends that the following combination of methods be used:

     Locations of DNS servers to be used for reverse lookups should be
obtained in the following manner:

-    First, attempt to perform the lookup in the normal way used for any
     IPv6 address, by issuing a query for {address}.ip6.arpa.  If the
     result of this query is one or more PTR records, these results are
     used and the lookup is complete.

-    Else, if the result of this query indicates that lookups for a pre-
     fix of the queried IPv6 address, greater than or equal to 48 bits
     in length, have explicitly been delegated, but the query could not
     be completed due to some error, the error is returned and the
     lookup is complete.

-    Else, the method of inferring NS and AAAA records described in sec-
     tion 3.2 is used, with two or three well-known suffixes chosen
     rather than a single anycast address.  Assigning two or three well-
     known suffixes rather than a single suffix allows a small site to

Moore                   Expires 2 September 2001                [Page 9]

6to4 and DNS                 INTERNET-DRAFT                 2 March 2001

     provide redundant servers for reverse lookup without having to
     implement anycast.

     This method is recommended both in preference to, and instead of,
     the method in section 3.1 because it is anticipated that many 6to4
     sites will be using a single IPv4 address and will not have reverse
     lookup for that IPv4 address delegated to their name servers.  (In
     other words, NS records delegating the reverse lookup of 32-bit
     IPv4 prefixes are assumed to be rare.)

     Implementation of the above algorithm should be provided by both
host-based DNS query libraries and (as a configuration option) by
resolver servers.  Thus, if either the host-based query library (for
dynamically-linked applications) or the local resolver server has been
upgraded to infer delegation of 6to4 addresses, applications on that
host will be able to perform lookups of 6to4 addresses in the absence of
explicit delegation.

     This compromise largely preserves the favorable deployment charac-
teristics of 6to4 - namely, that hosts and networks can use 6to4 without
explicit support from the existing IPv4 network infrastructure.  Imple-
menting the algorithm in both query libraries in resolvers allows exist-
ing IPv6 hosts and applications to lookup 6to4 addresses without having
to upgrade all of their hosts, while still allowing lookups for single
hosts and small sites which cannot reconfigure their DNS resolver
servers.  However it does require that all IPv6 sites - not just those
on 6to4 networks - upgrade their query libraries and/or resolvers if
they wish to perform reverse lookups on 6to4 addresses.

6. Security Considerations

     The use of well-known address suffixes for DNS servers would allow
hosts that could choose their own addresses to provide inverse name
lookups in the absence of explicit delegation by the network
administrators.  For this reason, it is necessary to check for explicit
delegation of reverse lookup service before using results obtained from
queries to well-known addresses.

     In addition, sites running 6to4 which do not provide reverse lookup
service at each of the well-known address suffixes, should take measures
to prevent ordinary hosts from assuming the role of DNS servers.  For
example, a site might make a decision to disallow those addresses being
used by ordinary hosts and to filter any traffic originating from those
addresses which were not assigned to DNS servers.

     Pseudo-records that are automatically derived from other DNS
records cannot be signed using DNSSEC, even if the explicit records from
which the pseudo-records are derived are signed.  Since explicit records

Moore                   Expires 2 September 2001               [Page 10]

6to4 and DNS                 INTERNET-DRAFT                 2 March 2001

take precedence over pseudo-records, a host or application SHOULD NOT
trust a signed NS record referring a query for some portion of IPv4
space as evidence of authoritative referral to the corresponding portion
of 6to4 space unless it has evidence that there are no explicit records
present for that portion of 6to4 space.

7. Author's Address

Keith Moore
University of Tennessee, Knoxville
1122 Volunteer Blvd, Suite 203
Knoxville TN, 37996-3450
email: moore@cs.utk.edu

8. References

[1]. Carpenter, B., Moore, K.  Connection of IPv6 Domains via IPv4
     Clouds.  RFC 3056, February 2001.

[2]. Crawford, M., Huitema, C., Thomson, S.  DNS Extensions to Support
     IPv6 Address Aggregation and Renumbering. RFC 2874, July 2000.

Moore                   Expires 2 September 2001               [Page 11]

Html markup produced by rfcmarkup 1.129d, available from https://tools.ietf.org/tools/rfcmarkup/