One document matched: draft-dempsky-dnscurve-00.xml

<?xml version='1.0'?>
<!DOCTYPE rfc SYSTEM 'rfc2629.dtd' [
 <!ENTITY RFC2119 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.2119.xml">
 <!ENTITY rfc4648 PUBLIC '' 'http://xml.resource.org/public/rfc/bibxml/reference.RFC.4648.xml'>
]>

<?rfc toc='yes'?>
<?rfc symrefs='yes'?>
<?rfc compact='yes'?>
<?rfc subcompact='no'?>

<rfc category='std' ipr='trust200902' docName='draft-dempsky-dnscurve-00'>
<front>
<title abbrev='DNSCurve: Link-level security for DNS'>DNSCurve: Link-level security for the Domain Name System</title>

<author initials='M.' surname='Dempsky' fullname='Matthew Dempsky'>
<organization>OpenDNS, Inc.</organization>
<address>
<postal>
<street>199 Fremont St, Fl 12</street>
<city>San Francisco</city> <region>CA</region> <code>94105-6629</code>
<country>US</country>
</postal>
<phone>+1 415 680 3742</phone>
<email>matthew@dempsky.org</email>
</address>
</author>

<date month='August' year='2009' />

<abstract>
<t>
This document describes DNSCurve,
a protocol extension that adds link-level security
to the Domain Name System (DNS).
</t>
</abstract>
</front>

<middle>
<section title='Introduction'>
<t>
DNSCurve adds link-level security to the Domain Name System (DNS). It
includes a key distribution mechanism compatible with today's name
server software and registry services, and two packet formats: a
simple streamlined format requiring minimal packet overhead and a
mostly backwards-compatible format intended for use with strict
firewalls and DNS proxies.
</t>
<t>
DNSCurve packets include a cryptographic MAC to provide integrity and
availability. Clients can be confident that verified responses came
from the appropriate server and were not forged by a blind or even
sniffing attacker, while servers can be confident that responses will
not be replayed against other unintended clients. Additionally,
DNSCurve packets are encrypted to provide some confidentiality.
</t>
<t>
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 RFC 2119
<xref target='RFC2119' />.
</t>
</section>

<section title='Overview'>
<t>
DNSCurve uses Curve25519XSalsa20Poly1305, a particular combination of
the Curve25519, Salsa20, and Poly1305 primitives as described in <xref
target='naclcrypto' />. In particular, it is a cryptosystem featuring
256-bit public and secret keys, 192-bit nonces, and 128-bit
authenticators.
</t>
<t>
Each DNSCurve client and server has a secret key and a corresponding
public key. DNSCurve servers distribute their public keys by encoding
them in name server names embedded in standard DNS NS records.
DNSCurve clients distribute their public keys by including them in
their query packets.
</t>
<t>
When a DNSCurve client is about to send a DNS query to a name server,
if the name contains a DNSCurve public key, it can instead use this
public key along with its own secret key and a nonce to protect its
query in a "cryptographic box" as described in <xref
target='naclcrypto' />. The client then encodes this cryptographic box
along with the nonce and its own public key as an expanded DNSCurve
query packet, which it sends to the DNSCurve server.
</t>
<t>
Upon receiving a DNS query packet, a DNSCurve name server first checks
if it appears to be an expanded DNSCurve query packet. If not, then it
responds normally. Otherwise, it extracts the client's DNSCurve public
key, nonce, and boxed query, and attempts to open the box using the
public key and nonce and its own secret key. If this fails (i.e., the
authenticator is invalid), then the packet is not an expanded DNSCurve
query packet, and the server responds as a normal DNS query.
</t>
<t>
Otherwise, if the unboxing succeeds, then the server discovers the
client's original query packet. To send a response, the server chooses
a nonce extension to append to the client-chosen nonce, and protects
its response packet in a cryptographic box using the same keys and the
extended nonce. The server then encodes this cryptographic box as an
expanded DNSCurve response packet, which it sends to the DNSCurve
client.
</t>
<t>
Meanwhile, the DNSCurve client waits for an expanded DNSCurve response
packet. If it receives a non-DNSCurve response packet, an expanded
DNSCurve response packet with an invalid nonce (i.e., not an extension
of its original nonce) or an invalid cryptographic box (i.e., cannot
be opened using the same keys and the extended nonce), then it discards
the packet and continues waiting. Once it receives a valid expanded
DNSCurve response packet, it opens the cryptographic box to discover
the server's original DNS response.
</t>
</section>

<section title='Base-32 encoding'>
<t>
Sometimes DNSCurve communicates arbitrary byte strings inside domain
names. While the DNS protocol is 8-bit safe for names and labels
(except for case-insensitive handling of ASCII alphabetic characters),
many tools have trouble with arbitrary characters in domain names, in
particular domain registrar software. To cope with this limitation,
DNSCurve encodes byte strings using a set of safe alphanumeric
characters.
</t>
<t>
In DNSCurve's base-32 encoding, a byte string is interpreted as a
number in little-endian form. Each 5-bit sequence of this number, from
least significant to most significant, is encoded as one of the
standard "digits" "0123456789bcdfghjklmnpqrstuvwxyz". A final sequence
of fewer than 5 bits is zero-extended before encoding. Decoders MUST
accept "BCDFGHJKLMNPQRSTUVWXYZ" as synonyms for
"bcdfghjklmnpqrstuvwxyz".
</t>
<t>
For example, the two-byte string with bytes {0x64,0x88} (i.e., {100,136}
decimal) is interpreted as the integer 0x8864 (i.e., 34916). The bits
1000100001100100 of this integer are divided into 5-bit parts 00100,
00011, 00010, 00001, which in turn are encoded as "4", "3", "2", "1".
The original string is therefore encoded as the string "4321".
</t>
<t>
N.B., this is not the same encoding as defined in <xref
target='RFC4648' />.  In particular, the byte string is chunked into
5-bit sequences differently, and a different alphabet is used.  The
first allows DNSCurve public keys to be encoded slightly more
compactly (see <xref target='pubkeys' />), and the second helps to
further prevent false positives when searching for base-32 encoded
strings in domain names.
</t>

<section title='Examples'>

<texttable>
<ttcol>Byte string</ttcol>
<ttcol>Base-32 encoding</ttcol>

<c>{}</c>
<c>""</c>

<c>{0x88}</c>
<c>"84"</c>


<c>{0x9f,0x0b}</c>
<c>"zw20"</c>


<c>{0x17,0xa3,0xd4}</c>
<c>"rs89f"</c>


<c>{0x2a,0xa9,0x13,0x7e}</c>
<c>"b9b71z1"</c>


<c>{0x7e,0x69,0xa3,0xef,0xac}</c>
<c>"ycu6urmp"</c>


<c>{0xe5,0x3b,0x60,0xe8,0x15,0x62}</c>
<c>"5zg06nr223"</c>

<c>{0x72,0x3c,0xef,0x3a,0x43,0x2c,0x8f}</c>
<c>"l3hygxd8dt31"</c>


<c>{0x17,0xf7,0x35,0x09,0x41,0xe4,0xdc,0x01}</c>
<c>"rsxcm44847r30"</c>

</texttable>

</section>
</section>

<section anchor='pubkeys' title='Encoding public keys in name server names'>
<t>
DNSCurve public keys are encoded in name server names as a 54-byte
label consisting of the magic string "uz5" followed by the first 51
bytes of the base-32 encoding of the public key. (Curve25519 public
keys are actually 255-bit integers in little-endian, so the 52nd byte
of the base-32 encoding will always be "0".)
</t>
<t>
When a DNSCurve client is searching a name server name for a DNSCurve
public key, it MUST check every label for an encoded public key. If
multiple public keys are found, the left-most label MUST be chosen.
String comparison with "uz5" MUST be performed case-insensitively.
</t>

<!--
Bother with this, or just provide reference code?

<section title='Examples'>

<t>
A few examples of domain names 
and their respective DNSCurve public keys:
<list style='symbols'>
<t>ns1.example.com: none</t>
<t>uz5foo.example.com: none</t>
<t>uz5vowels
<t>uz5[bcd...].example.com: decode([bcd...])</t>
<t>x.uz5[mnp...].example.com: decode([mnp...])</t>
<t>uz5[fgh...].uz5[jkl...].example.com: decode([fgh...])</t>
<t>uz5.uz5[mnp...].example.com: decode([mnp...])</t>
<t>uz5[aaa...].uz5[qrs...].example.com: decode([qrs...])</t>
</list>
</t>

</section>
-->

</section>

<section title='Nonce generation'>
<t>
For every request, DNSCurve clients generate a 96-bit nonce, and for
every response, DNSCurve servers generate a 96-bit nonce extension.
Nonces MUST be unique for distinct packets for the same client-server
key pair. A simple way to achieve this is to choose a unique nonce
for each packet and for each retransmission. Additionally, servers
MUST use a non-zero nonce extension (because nonces are zero extended
in query packets). Clients and servers may otherwise generate nonces
however they choose.
</t>
<t>
Two recommended ways to generate a 96-bit nonce or nonce extension are
<list style='numbers'>
<t>a 64-bit counter (starting at 1) followed by a 32-bit random number
and</t>
<t>a 64-bit timestamp (e.g., nanoseconds since 1970) followed by a
32-bit random number.</t>
</list>
In either case the 64-bit value MUST NOT decrease even if the software
restarts or the system clock jumps backwards.
</t>
<t>
If multiple clients or multiple servers share a DNSCurve secret key,
then they MUST make sure no two separate clients or servers generate
the same nonce. A simple way to achieve this is to use nonce
separation; e.g., one server uses only even nonces and the other uses
only odd nonces.
</t>
</section>

<section title='DNSCurve expanded formats'>
<t>
DNSCurve defines two expanded formats: "streamlined" and "TXT". Each
includes a format for expanded queries and a format for expanded
responses. DNSCurve clients may send DNSCurve expanded queries using
whichever format it chooses, but they are encouraged to use the
streamlined format when possible. A DNSCurve server MUST support
DNSCurve expanded queries in either format and MUST send expanded
responses using the corresponding format.
</t>

<section title='Streamlined format'>
<t>
An expanded query packet in streamlined format has the following bytes:
<list style='symbols'>
<t>8 bytes: the magic string "Q6fnvWj8".</t>
<t>32 bytes: the client's DNSCurve public key.</t>
<t>12 bytes: a client-selected nonce for this packet.</t>
<t>A cryptographic box containing the original DNS query packet.</t>
</list>
</t>
<t>
An expanded response packet in streamlined format has the following bytes:
<list style='symbols'>
<t>8 bytes: the magic string "R6fnvWJ8".</t>
<t>12 bytes: the client's nonce.</t>
<t>12 bytes: a server-selected nonce extension.</t>
<t>A cryptographic box containing the original DNS response packet.</t>
</list>
</t>
<t>
Note that this streamlined response format does not repeat the
client's query name, and in particular does not repeat the client's
public key. However, it does repeat the client's nonce.
</t>
</section>

<section title='TXT format'>
<t>
The "TXT" format receives its name from the fact that expanded query
and response packets in this format appear to casual inspection to be
standard DNS packets with two possible exceptions: 1) the query name
exceed 255 bytes and 2) the total packet may exceed 512 bytes.
</t>
<t>
When encoding an expanded query packet in TXT format, a DNSCurve
client MUST create a DNS standard query packet with the AA, TC, RD,
RA, Z, and RCODE bits cleared, a single entry in the question section,
and no records in the answer, authority records, or additional records
sections. The one question MUST be an Internet class question for TXT
records for the query name constructed from the concatenation of the
following labels:
<list style='symbols'>
<t>
One or more labels, each label before the last being exactly 50 bytes,
the last label being at most 50 bytes. The concatenation of these
labels is the base-32 encoding of a 96-bit client-selected nonce for
this packet followed by a cryptographic box containing the original
DNS query packet.
</t>
<t>
One 54-byte label: the client's DNSCurve public key, encoded as
described in <xref target='pubkeys' />, except with the magic string
"x1a" instead of "uz5".
</t>
<t>
Zero or more additional labels specifying the name of the zone served
by this server; i.e., the owner name of the relevant NS record.
</t>
</list>
</t>
<t>
A DNSCurve server SHOULD be lenient in decoding expanded query packets
in TXT format. In particular, it MUST allow the RD bit to either be
set or clear, MUST allow records in the answer, authority records, and
additional records sections, and MUST allow any labels to follow the
DNSCurve public key in the query name. However, it MUST discard
packets with the QR bit set.
</t>
<t>
When encoding an expanded response packet in TXT format, a DNSCurve
server MUST create a DNS standard response packet copying the ID, RD
bit, and questions section from the expanded query packet, setting the
AA bit, leaving the TC and RA bits cleared and Z and RCODE values set
to 0, containing one record in the answer section, and no records in
the authority records or additional records section. The record in the
answer section MUST be an Internet-class TXT record for the query name
from the questions section with a TTL of 0. The RDATA of this record
is the 96-bit server-selected nonce extension followed by a
cryptographic box containing the original DNS response packet, encoded
as a sequence of one or more strings of at most 255 bytes in standard
DNS TXT RDATA format.
</t>
<t>
Similarly, a DNSCurve client SHOULD be lenient in decoding expanded
response packets in TXT format. In particular, it MUST allow the
server to alter the case of the query name when repeating it in the
questions section.
</t>
</section>

</section> <!-- formats -->

<section title='UDP and TCP'>
<t>
If a normal DNS response packet is larger than 512 bytes then the
server replaces it by an explicitly truncated packet. The client then
tries again through TCP. Servers are not required to support TCP if no
responses are above 512 bytes; clients are permitted to try TCP only
if the server has explicitly indicated truncation.
</t>
<t>
DNSCurve does not require TCP support from servers that were not
already supporting TCP. If the original DNS response packet is at most
512 bytes then the server is permitted to send the expanded response
packet as a UDP packet. DNSCurve clients are required to set aside a
4096-byte buffer for receiving a UDP response packet.
</t>
<t>
If the original DNS response packet is above 512 bytes then it is
replaced by an explicitly truncated packet and the truncated packet is
protected by DNSCurve. In this case the client tries again by TCP,
sending its DNSCurve query packet through TCP and receiving the
DNSCurve response through TCP.
</t>
<t>
TCP is considerably more expensive for clients and servers than UDP
is, and TCP has no protection against denial of service, so server
administrators are advised to stay below 512 bytes if possible.
DNSCurve adds some denial-of-service protection for UDP but cannot do
anything to help TCP.
</t>
<t>
If a protected DNS query includes an EDNS0 OPT record, then the payload
size field refers to how large the original DNS response packet can be
before encoding as a DNSCurve response packet.
Clients MUST reduce the payload size they advertise to account for overhead
from encoding the response as an expanded response packet.
If a server builds a response within the payload size limit,
but cannot fit the encoded response in 4096 bytes,
then it MAY silently discard the response.
</t>
</section>

<section title='Security considerations'>
<t>
The security of the Curve25519XSalsa20Poly1305 cryptosystem and its
underlying cryptographic primitives is discussed in <xref
target='naclcrypto' />.  In summary, it is designed to meet the
standard notions of privacy and third-party unforgeability for a
public-key authenticated-encryption scheme using nonces.
</t>
<t>
DNSCurve only provides link-level security between a client-server
pair.  It does not attempt to ensure end-to-end security for queries
and responses relayed by untrusted DNS proxies and caches.
</t>
</section>

<section title='IANA considerations'>
<t>
This document has no actions for IANA.
</t>
</section>

</middle>

<back>

<references title='Normative references'>
&RFC2119;
&rfc4648;

<reference anchor='naclcrypto'>
<front>
  <title>Cryptography in NaCl</title>
  <author initials='D. J.' surname='Bernstein' fullname='Daniel J. Bernstein'>
    <organization abbrev='UIC'>
      University of Illinois at Chicago
    </organization>
  </author>
  <date month='March' year='2009' />
</front>
<format type='html' target='http://cr.yp.to/papers.html#naclcrypto' />
</reference>
</references>

</back>
</rfc>