INTERNET-DRAFT DNS Protocol Security Extensions
23 February 1994
Expires 22 August 1994
Domain Name System Protocol Security Extensions
------ ---- ------ -------- -------- ----------
Donald E. Eastlake 3rd & Charles W. Kaufman
Status of This Document
This draft, file name draft-ietf-dnssec-secext-00.txt, is intended to be
become a standards track RFC. Distribution of this document is
unlimited. Comments should be sent to the DNS Security Working Group
mailing list <dns-security@tis.com> or to the authors.
This document is an Internet Draft. Internet Drafts are working
documents of the Internet Engineering Task Force (IETF), its Areas,
its Working Groupsd and other organizations or individuals.
Internet Drafts are draft documents valid for a maximum of six
months. Internet Drafts may be updated, replaced, or obsoleted by
other documents at any time. It is not appropriate to use Internet
Drafts as reference material or to cite them other than as a
``working draft'' or ``work in progress.'' Please check the 1id-
abstracts.txt listing contained in the internet-drafts Shadow
Directories on ds.internic.net, nic.nordu.net, ftp.nisc.sri.com, or
munnari.oz.au to learn the current status of any Internet Draft.
Abstract
The Domain Name System has become a critical operational part of the
Internet infrastructure yet it has no security mechanisms to assure
data integrity or authentication. Extensions to the DNS protocol are
proposed to provide these services to security aware resolvers or
applications through the use of cryptographic digital signatures.
These digital signatures are added to secured zones as resource
records. They can be most efficiently handled by security aware
servers but security can still be provided to security aware
resolvers or applications even by non-security aware primary,
secondary, and caching servers.
In addition, the extensions provide for the storage of authenticated
keys in the DNS, so that a security aware resolver can learn the
authenticating key of zones in addition to those for which it is
initially configured, and for other purposes.
Finally, the extensions optionally provide for authentication of DNS
protocol messages for additional security.
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 1]
INTERNET-DRAFT DNS Protocol Security Extensions
Table of Contents
Status of This Document....................................1
Abstract...................................................1
Acknowledgements...........................................2
Table of Contents..........................................2
1. Introduction............................................4
2. Overview of the Protocol...............................4
2.1 Data Origin Authentication.............................4
2.1.1 Security Provided...................................4
2.1.2 The SIG Resource Record..............................5
2.1.3 The RSA Resource Record..............................6
2.1.4 Signers Other Than The Zone..........................6
2.1.5 Special Problems With Time-to-Live...................6
2.1.6 Improved Performance At The Expense Of Compatibility.7
2.2 DNS Message Authentication.............................8
3. Services, Requirement, and Non-Requirements.............9
3.1 Requirements...........................................9
3.2 Non-Requirements......................................11
4. The Security Desired & Security Available Bits.........12
5. The SIG Resource Record................................13
5.1 SIG RDATA Format......................................13
5.1.1 Signature Format....................................14
5.1.2 Signet Format.......................................15
5.1.2.1 Direct Resource Record Signets....................16
5.1.2.2 Direct Glue Record Signet.........................17
5.1.2.3 Hashed Resource Record(s) Signet..................17
5.1.2.4 Partial RR Set Flag Signet........................18
5.1.2.5 Partial SIG Set Flag Signet.......................19
5.1.2.6 AXFR Signets......................................19
5.1.2.7 Reserved Signet Prefixes..........................20
5.2 SIG RRs in the Construction of Responses..............20
5.3 Processing Responses with SIG RRs.....................21
5.4 File Representation of SIG RRs........................22
5.4.1 Size of Data........................................22
5.4.2 RR Numbering........................................22
5.4.3 SIG RR Scope........................................23
5.4.4 RRs Surpressed by a SIG RR..........................23
6. The RSA Resource Record................................25
6.1 RSA RDATA format......................................25
6.2 Types of DNS Names and Keys...........................26
6.3 RSA RR Flag Bits......................................26
6.4 RSA RRs in the Construction of Responses..............27
6.5 File Representation of RSA RRs........................28
7. How to Resolve Securely................................29
7.1 Boot File Format......................................29
7.2 Chaining Through Zones................................29
7.3 Secure Time...........................................30
8. Operational Considerations.............................32
8.1 Modulus Size Considerations...........................32
8.2 Key Storage...........................................32
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 2]
INTERNET-DRAFT DNS Protocol Security Extensions
8.3 Key Generation........................................33
8.4 Key Lifetimes.........................................33
8.5 Key Revocation........................................33
8.6 Root..................................................34
9. Conformance............................................35
10. Security Considerations...............................36
References................................................36
Authors Addresses.........................................37
Expiration and File Name..................................37
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 3]
INTERNET-DRAFT DNS Protocol Security Extensions
1. Introduction
[To be written]
2. Overview of the Protocol
These DNS protocol extensions provide two distinct services: data
origin authentication, described in section 2.1 below, and message
authentication, described in section 2.2 below. In addition, the
resource records added to support these authentication services
permit the association of keys with DNS names. These keys could be
used in support of other security services such as IP level security.
2.1 Data Origin Authentication
There are two distinct aspects to the data origin authentication
service. The purpose of the first is to add security; the purpose of
the second is to improve performance when the first is used. Adding
security requires no changes to the "on-the-wire" DNS protocol beyond
the addition of two new resource types: signatures and keys. This
service could be supported by existing resolver and server
implementations so long as they could support the additional resource
types.
If signatures are always retrieved and verified when retrieving the
information they authenticate, there will be more trips to the server
and performance will suffer. The revisions to the DNS wire protocol
for security aware servers are an attempt to mitigate that
degradation by automatically sending exactly the signatures needed
and by skipping the sending of the data if it can be derived from the
signature.
2.1.1 Security Provided
Security is provided by associating with each item of information in
DNS a cryptographically generated digital signature. Commonly, there
will be a single RSA key that signs for an entire zone. If the
resolver reliably learns the public RSA key of the zone, it can
verify that all the data read was properly authorized and is
reasonably current. The expected implementation is for the zone
private key to be kept off-line and used to re-sign all of the
records in the zone periodically.
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 4]
INTERNET-DRAFT DNS Protocol Security Extensions
The data origin authentication key belongs to the zone and not to the
servers that store copies of the data. That means compromise of a
secondary or caching server will not affect the degree of assurance
that a resolver has that the data is genuine. However, such a server
can (except in the case of a zone transfer) claim that a name does
not exist and a resolver may not be able to determine otherwise.
A resolver can learn the public key of a zone either by having it
manually configured or by reading it from DNS. To reliably learn the
public key by reading it from DNS, the key itself must be signed.
Thus to provide any reasonable degree of security, the resolver must
be configured with at least the public key of one zone. From that,
it can securely read the public keys of other zones. It is in
principle more secure to have the resolver manually configured with
the public keys of multiple zones, since then the compromise of a
single zone would not permit the faking of information from other
zones. It is also more administratively cumbersome, however,
particularly when public keys change.
2.1.2 The SIG Resource Record
The syntax of a SIG resource record (signature) is described in
Section 5. It includes the name of the signer, the time at which the
signature was created, the time it expires (when it is no longer to
be believed), its original time to live (which may be longer than its
current time to live but cannot be shorter), and an RSA signature.
There are a number of unusual aspects to the construction of the RSA
signature that are intended to maximize performance in this
application. Unlike some other digital signature schemes like El
Gamal or DSS, RSA signatures have the property that when a signature
is verified, it produces a message that is almost the size of the
signature itself. In most uses, for example Privacy Enhanced Mail,
the message to be signed is much larger than the signature (which is
generally 64-256 bytes long), so a message digest of the message is
computed (a 16-20 byte "fingerprint") and that quantity is signed.
For DNS, however, it will be common that the messages being signed,
will be very short - sometimes shorter than 16-20 bytes -
particularly with the abbreviation techniques used herein.
Further, there are commonly multiple resource records associated with
a DNS name, and it should be efficient to verify a signature on a
single one of those records or any subset of them. If a 64-256 byte
signature record were created for every resource record, there would
be an unacceptable explosion of data.
The SIG Resource record syntax proposed therefore has two unusual
properties: (1) when it signs a resource record, it may contain
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 5]
INTERNET-DRAFT DNS Protocol Security Extensions
either the resource record itself or a message digest of the resource
record; and (2) a single signature may sign multiple multiple
resource records associated with a single name.
Every name in a zone supporting signed data will have associated with
it one or more SIG resource records - as many as required to sign all
of the non-SIG resource records. A security aware server supporting
the performance enhanced version of the DNS protocol security
extensions will return with all records retrieved the corresponding
SIG records. If a server does not support the protocol, the resolver
must retrieve all the SIG records for a name, verify them all, and
find the one or ones that sign the resource records that resolver is
interested in. As a further optimization, a server supporting the
performance enhanced version of the protocol will return only the
signature - and skip the requested data - in the case where the
signature contains enough information to reconstruct the data in
full. Because of this, in some cases the authenticated data being
sent via SIG records can be shorter than the plain data would have
been.
2.1.3 The RSA Resource Record
The syntax of an RSA resource record (key) is described in Section 6.
It is present for two reasons: to support the DNS infrastructure
itself so that a resolver that is manually configured with the public
keys of one or more zones can securely learn the public keys of other
zone; and to allow the storing of RSA public keys of DNS-named
entities other than zones for applications like IP-Security.
2.1.4 Signers Other Than The Zone
There are two cases where a signature is generated by other than the
zone private key. One is for future support of dynamic update where
an entity is permitted authenticate/update its own record. The
public key of the entity must be present in the DNS and signed with
the zone key, but the other RRs may be signed with the entity's key.
The other is for support of message authentication as described in
2.2 below.
2.1.5 Special Problems With Time-to-Live
A digital signature will fail to verify if any change has occurred to
the data between the time it was originally signed and the time the
signature is verified. This conflicts with our desire to have the
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 6]
INTERNET-DRAFT DNS Protocol Security Extensions
time-to-live field tick down when resource records are cached.
This could be avoided by leaving the time-to-live out of the digital
signature, but that would allow unscrupulous secondaries to set
arbitrarily long time to live values undetected. Instead, we include
the "original" time-to-live in the signature and communicate that
data in addition to the current time-to-live. Unscrupulous servers
under this scheme can fail to decrement time to live like they are
supposed to, but they cannot increase it beyond its original value.
Separately, signatures include a time signed and an expiration time.
A resolver that knows an absolute time can determine securely whether
a signature has expired.
In order to keep the data as compressed as possible, we don't want to
have to include the original TTL for every resource record included
in a SIG when usually they are all the same. We therefore assume
that the original TTL is equal to the original TTL of the SIG
resource record (which is sent for every SIG resource record), and in
the rare case where the TTL on the other resource record differs we
permit it to be explicitly included.
2.1.6 Improved Performance At The Expense Of Compatibility
To run the high performance version of the protocol, the server
should remember for each resource record: (1) which SIG record
includes the signature for that record and (2) whether the SIG record
contains the resource record in full or in digested form.
When the server is responding to a request, it should for each record
requested return the corresponding SIG (removing duplicates) and also
it should suppress the sending of the record itself if it is present
in the signature in undigested form. Since a resolver running the
secure protocol will not believe any record that is not signed, there
would be no point in returning the record without the SIG. And if
the resolver is going to see the RR in full in the course of
verifying the signature there is no point in wasting bandwidth by
sending the RR being authenticated.
The high performance version of the protocol can only be used if both
resolver and server understand it. Negotiation is done via some DNS
message header bits we believe that existing servers will ignore and
existing resolvers will not set.
If signature verification is not done by the DNS resolver code but
rather by some application that is retrieving resource records
through that resolver, the standard protocol must be used.
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 7]
INTERNET-DRAFT DNS Protocol Security Extensions
2.2 DNS Message Authentication
The data origin authentication service described above protects
resource records but provides no protection for message headers and
limited protection against resource record addition to or deletion
from a message. If header bits or, in some cases, the resource
record set in a response, are falsely set by a server, there is
little that can be done. However, it is at least possible to add
message authentication. Such authentication means that a resolver
can be sure it is getting messages from the server it thinks it is, a
server can be sure it is getting requests from the resolver it thinks
it is, and that in both cases these messages have not been diddled in
transit.
This is accomplished by adding a SIG resource record to end of the
message which digitally signs the message by the server or resolver.
The private key used belongs to the host composing the message, not
to the zone being queried. The corresponding public key is stored in
and retrieved from the DNS. Because messages are highly variable,
message authentication SIGs can not be precalculated. Thus it will
be necessary to keep the key on-line, for example in software or in a
directly connected piece of hardware.
The best way to get the security provided by the message
authentication service would be to use a good IP level security
protocol. The authors of this draft decry the every growing number
of IP application level security protocols such as Telnet, NTP, FTP,
etc., etc. when a single IP-security protocol could secure most of
these applications.
Unfortunately, an IP-security standard has not yet been adopted. And
even if it had, there will be many systems for many years where it
will be hard to add IP security but relatively easy to replace the
DNS components. Furthermore, the data original authentication
service requires the implementation of essentially all the mechanisms
needed for a rudamentary message authentication service. Thus a
simple message authentication service using mechanisms already
required by DNS security is included as a strictly optional part of
these extensions.
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 8]
INTERNET-DRAFT DNS Protocol Security Extensions
3. Services, Requirement, and Non-Requirements
This section is based on the 19 November 1993 email message from
James M. Galvin summarizing the Houston IETF meeting of the DNS
Security Working Group. At this meeting a list of requirements,
shown in 3.1 below, and non-requirement shown in 3.2 below, were
arrived at.
The security services desired were set at the following:
data integrity, and
data origin authentication.
It was noted at that meeting that these services can be provided by
digital signatures.
3.1 Requirements
The numbered items below were determined to be "requirements", not in
the sense of being mandatory but in the sense of all being highly
desirable. A comment appears after each requirement as to if/how it
is met by this proposal.
<1> Sites must be able to support at least their internal users in
the presence of external network failures.
Security aware resolvers can be configured with the public key of
their local apex zone. The needed SIG RRs can be added to that zone
and any desired lower level zones either off-line or on-line. Thus
this requirement is met.
<2> it should be possible for a site to pre-configure other
authoritative servers without having to query the "system" to find
the server.
Security aware resolvers can be configured with the public key of any
other zones and the IP address of their servers.
<3> It should be possible to request services only from security
enhanced servers, only from non-security enhanced servers, or a
indicate that either is acceptable.
These proposed protocol extensions do not provide any enhancement to
the NS RR or otherwise to indicate whether or not a server is
security aware without actually querying it. It is believed that
this additional complexity is not warranted. Non-security aware
servers can still support security aware resolvers, although less
efficiently. It is possible to tell if a server is security aware by
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 9]
INTERNET-DRAFT DNS Protocol Security Extensions
the SA (Security Available) bit in the header of its responses.
<4> It should be possible to recognize security enhanced responses.
Security enhanced responses can be recognized by the presence of SIG
RRs.
<5> It should be possible to assign cryptographic keys (make use of
the security services) to leaf nodes in the DNS tree, i.e., fully
qualified domain names.
The proposed extensions allow RSA key RRs to be associated with any
node in the DNS tree. Indeed, more than one key can be associated
with a node which serves multiple functions.
<6> It should be possible to not trust secondary servers.
The proposed extensions can be implemented so that the zone private
key is never on line on the network. Thus, ignoring denial of
service threats, it is possible to have untrusted secondaries and
even untrusted primaries.
<7> A mechanism must exist for revoking cryptographic keys that works
within the DNS time-to-live framework.
A bit is defined in the RSA RR to indicate if it is a key assertion
or key revocation and key revocations are opportunisticly flooded as
additional information on every query to their zone if the resolver
and server are security aware.
However, an additional mechanism may be necessary to notify
secondaries, caching servers, etc. to assure that a revocation is
noticed within the TTL. This is really just a special case of a
change in zone information and a general mechanism such as the NOTIFY
operation described in draft-ietf-dns-ixfr-01.txt could be used.
However, guaranteed revocation is not possible (for example in a
partitioned network) without introducing unacceptable denial of
service risks (such as having to wait "forever" to get a current "key
revocation list" for a zone if the network is partitioned).
<8> Security services should be supported with no additional DNS
queries beyond what would be required if security was not supported.
No additional queries are required, barring possible UDP truncation
problems, to obtain authentication from or to securely learn the
public key for a zone when its names servers are being obtain if a
security enhanced server is being used by a security aware resolver.
<9> It must be possible to ensure that cached data expires according
to its TTL.
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 10]
INTERNET-DRAFT DNS Protocol Security Extensions
A security aware resolver has both the TTL for an RR and the
expiration time of the SIG covering the RR. Cached data is always
invalid after the SIG expiration time plus the original TTL.
3.2 Non-Requirements
It was explicitly decided by the DNS Security Working Group at the
Houston IETF meeting that the following were not requirements:
<1> Confidentiality: DNS data is "public" and no effort need be made
to provide encryption of queries/responses. (This service may be
available via an IP level security protocol which is currently being
worked on by the IP Security Working Group of the IETF <ipsec-
request@ans.net>.)
<2> Access control: DNS data is "public" and no effort need be made
to provide access control lists, or similar mechanisms, as part of
this DNS security effort.
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 11]
INTERNET-DRAFT DNS Protocol Security Extensions
4. The Security Desired & Security Available Bits
The header section of DNS messages is modified to define bits 9 and
10 in the second word (fourth octet). These were formerly the top
two bits of the Z field defined as "Reserved for future use."
[RFC1035]
These bits are defined as security desired, labeled SD, and security
available, labeled SA. With the definition of these bits, the header
looks like the following:
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
| ID |
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
| QR| Opcode | AA| TC| RD| RA| SD| SA| Z | RCODE |
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
| QDCOUNT |
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
| ANCOUNT |
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
| NSCOUNT |
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
| ARCOUNT |
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
All fields are as before except that the Z field is reduced to one
bit and SD and SA are defined as follows:
SD
Security Desired - this bit may be set in a query. If the
server to which the query is sent does not support the DNS security
protocol, the bit should be ignored except that it may be copied into
the response. If the server does support this protocol, the bit MUST
copied into the response and, if the bit is set, the server MUST
provide any SIG and KEY RRs as described in the sections below
concerning these RRs.
SA
Security Available - this is to be set or cleared in a
response. If set, it indicates that the server supports this
security protocol.
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 12]
INTERNET-DRAFT DNS Protocol Security Extensions
5. The SIG Resource Record
The SIG or "signature" resource record (RR) is the fundamental way
that resource records (and optionally message) are authenticated.
The SIG RR unforgably binds one or more other RRs (or a DNS message)
to a time and the signer's fully qualified domain name using
cryptographic techniques and the signer's private key. The signer is
the owner of the zone the RR originated from (or the composer of the
authenticated DNS message).
5.1 SIG RDATA Format
The RDATA portion of a SIG RR is as shown below. The integrity of
the RDATA information and that of the SIG RRs owner, type, and class
are protected by the signature field.
1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 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
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| signer's name |
/ /
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| original TTL |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| time signed |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| signature expiration |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| sig length | /
+-+-+-+-+-+-+-+-+ signature -+
/ /
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
The value of the SIG type is xxx.
The "signer's name" field is the fully qualified domain name of the
signer of node generating the SIG RR. This is the zone which
contained the RR(s) being authenticated (or the host which is the
source of a DNS message that is being authenticated).
The "original TTL" field is included in the RDATA portion to avoid
authentication problems that caching servers would otherwise cause by
decrementing the real TTL field and security problems that
unscrupulous servers could otherwise cause by manipulating the real
TTL field. This original TTL is protected by the signature while the
real TTL field is not.
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 13]
INTERNET-DRAFT DNS Protocol Security Extensions
The "time signed" field is an unsigned number of seconds since the
start of 1 January 1970.
The SIG is valid until the signature expiration time which is a field
of the same format as the time signed.
The "sig length" field is an unsigned 8 bit count of the number of
octets in the signature field.
The structured of the "signature" field is described below.
[It would be possible to allow additional optional fields after the
above in the SIG RR as described in draft-ietf-dns-ixfr-01.txt]
5.1.1 Signature Format
The actual signature portion of the SIG RR binds the owner, signer,
class, original TTL, time signed, and expiration time of the RR to
one or more RRs being authenticated (or to the entire DNS message in
which it occurs). To accomplish this, it contains at least one
"signet", as defined in the following section, and a "self-hash"
field covering the above items. The structure of signets is
described in section 5.1.2 below.
The signature is then calculated using the RSA public key system as
follows
s = ( 01 | FF* | 00 | signets | self-hash ) ** e (mod n)
where "|" is concatenation. "signets" is the concatenation of the
signets included in this signature. "self-hash" is the 20 octet hash
using the Secure Hash Standard [SHS] of the SIG RR name, class,
signer name, original TTL, time signed, and expiration time. "e" is
the secret key exponent of the signer, and "n" is the public modulus
that is the signer's public key. 01, FF, and 00 are fixed octets of
the corresponding hexadecimal value. The FF octet is repeated the
maximum number of times such that the value of the quantity being
exponentiated is less than the value of n. No FF octets need occur
if "signets" is long enough. The order of the signets is not
significant.
The size of n, including most and least significant bits (which will
be 1) SHALL be at least 641 and not more than 2040. n and e MUST be
chosen such that the public exponent is less than 2**24 - 1 and
SHOULD be chosen such that the public exponent is small.
The above specifications are a profiling of PKCS #1 [PKCS1] except
that, under most circumstances, one additional byte of data is
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 14]
INTERNET-DRAFT DNS Protocol Security Extensions
allowed.
(A public exponent of 3 minimizes the effort needed to decode a
signature. Use of 3 as the public exponent may be weak for
confidentiality uses since, if the same data can be collected
encrypted under three different keys with an exponent of 3 then,
using the Chinese Remainder Theorem [ref], the original plain text
can be easily recovered. This weakness is not significant for DNS
because we seek only authentication, not confidentiality.)
5.1.2 Signet Format
Each signet consists of a prefix octet, with which the length of the
signet can be unambiguously determined, followed by the signet data.
Signet's are of three basic types: hashed, direct, and flag.
A hashed signet consists of the prefix octet, some additional data,
and a 20 bit hash of the data covered using the Secure Hash Standard
[SHS]. Since this hash algorithm is not invertible, the data, such
as RRs, covered by a hashed signet must also be included in the reply
to a query if it was requested.
A direct signet includes all of the data it covers. Some of the data
may be implicit or compressed but it is all unambiguously
recoverable. If a direct signet is present covering an RR, then that
RR SHOULD be surpressed from the reply message if the SD bit is on in
the query and the server is security aware.
Flag signets occur with direct signets and multiple SIG RRs. They
are used to determine if a complete set of RRs of a particular
variety are present.
Signet prefix octets are as follows:
0000000* Illegal
0LLLLLLL Direct Resource Record - type & rdata
10LLLLLL Direct Glue Record - name, type, & rdata
110***** (reserved)
1110**** (reserved)
11110NHT Hashed RR(s) - N, type, T, & hash
111110** (reserved
11111100 Partial SIG Set Flag
11111101 Partial RR Set Flag
1111111* Illegal
In the above table, "*" represents 0 or 1 and L, N, H, and T are bits
whose meaning is defined in the signet descriptions below. The
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 15]
INTERNET-DRAFT DNS Protocol Security Extensions
illegal prefixes should never occur. Any signature that appears to
include one should be considered invalid.
5.1.2.1 Direct Resource Record Signets
This signet is an actual RR but with some fields surpressed.
In order to avoid inconsistencies, all RRs of the same type and class
have to have the same TTL, at least for currently defined RRs. SIG
RRs need not survive beyond the RRs they authenticate but must live
as long as the covered RRs do. Thus SIG RRs may be constrained to
having the same TTL as the RRs they cover in most cases. The SIG RR
will always have the same class and name as the RRs it covers(except
for glue RRs as described in 5.1.2.2 and 5.1.2.3 below). Finally,
the signet data length in the prefix octet can be used to calculate
the RDSIZE. Thus RRs directly represented by this variety of signet
are compressed by omitting their name, class, TTL, and RDLENGTH
fields. Only the type, and RDATA are present. For example, the
direct signet for a type A record in the IN (Internet) class would be
seven octets as follows:
06 00 01 xx yy zz ww
where 06 is the prefix indicating a direct RR signet with six data
octets, 00 01 is hex for the type code for type A, and xx yy zz ww is
the 32 bit IP address from the RR. These 7 octets in a SIG RR
completely represent the 15+ (1+ name, 2 type, 2 class, 4 TTL, 2
rdlength, 4 ip address) octets of the original type A RR. The class,
name, and TTL are all recoverable from the same fields of the SIG RR
whose signature field includes this signet. Thus the original type A
RR can be surpressed from the answer if given to a security aware
resolver.
Note that any names in the RDATA area of this type of signet can not
be abbreviated by pointers outside of the reconstructed RR. That is
because the SIG RR and its signets are frozen at the time the
signature is encrypted under the signer's private key and this frozen
SIG may then be used in a variety of DNS messages. The RDATA area
may, however, if it has multiple names, abbreviate them by references
to earlier names in the reconstructed RR.
The direct representation of RRs makes maximum use of the relatively
large size of RSA digital signatures for common cases. The direct
representation also avoids the computational effort of calculating a
hash code. Because the original RRs type field must always be
present, the minimum length of the data after this type of signet
prefix is 2, thus prefixes of 00 and 01 hex are illegal. The maximum
size direct RR signet is 128 octets.
All RRs need not be included within a signature using this direct
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 16]
INTERNET-DRAFT DNS Protocol Security Extensions
signet. If the data portion of a direct RR signet exceeds 22 octets
(i.e., a total signet size of 23 octets including the prefix count
octet), space inside the signature can be saved by going to the
hashed RR signet described below. If an RR compressed for a direct
signet exceeds 127 bytes or the amount of space available for signets
in the signature part of a SIG RR, then it must appear separately and
be authenticated by a hashed RR signet.
5.1.2.2 Direct Glue Record Signet
Glue records must be handled a little differently. These are
currently always A records with a name which usually isn't even in
the zone being handled but are associated with an RR in the zone.
This type of signet allows such glue RRs to be included within a SIG.
The direct glue record signet is just like the direct resource record
signet described above except that the name is included right after
the prefix and before the type and RDATA.
If the Glue Record signet won't fit, a hashed RR signet, described
below, must be used. Note that names in the DNS can be up to 256
bytes long which would not fit inside a SIG RR signature field.
5.1.2.3 Hashed Resource Record(s) Signet
RRs can also be protected by a signet with a hash code. If a hashed
signet is used, all RRs of the same name, type, class, and TTL MUST
be hashed into a single signet.
To avoid inconsistencies, RRs of the same name, type, class, and TTL
must either all be present or all be absent. Thus a single hash code
covering such multiple RRs is all that is required. The signet is
then formed by a single octet 11110NHT (binary) prefix followed by a
possible name field depending on "N" and "H" as described below, the
two octet type for the RRs covered, a possible TTL field depending on
the values of "T" described below, and then the 20 octet hash code.
The hash is calculated by concatenating the full RRs, with all names
fully expanded and any required RDLENGTH adjustments made, in a
canonical order, and applying the Secure Hash Standard [SHS]. The
canonical order for RRs is to sort them in ascending order as left
justified unsigned octet sequences where a missing octet sorts before
a zero octet. Thus the hex sequences FE 01 02, 07, FF 03, FE FD, FF,
and 01 01 01 01 would sort and be concatenated into the sequence 01
01 01 01 07 FE 01 02 FE FD FF FF 03.
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 17]
INTERNET-DRAFT DNS Protocol Security Extensions
The "N" bit in the prefix stands for Name. It is normally zero
because almost all RRs associated with a name in a zone have that
name. The exception is glue records. These are currently A records
with owner names outside the zone. To be able to cover these in a
SIG for a different name where they cannot be included as a direct
glue record signet as discussed in the section above, the N bit makes
it possible to include the different name in the signet immediately
after the prefix. If the N bit is a one but the H bit is zero, the
name is included in full. Because names in DNS can be up to 256
bytes long, the 20 byte hash of the full name, using the Secure Hash
Standard, can be used instead of the actual name and is indicated by
turning on the H (or "hash") bit. To match this against RRs in the
reply, their names must be hashed and compared.
The "T" bit in the prefix stands for TTL. It is normally zero. For
the presently defined RRs, all RRs of the same type, class, and name
should have the same TTL. Future RRs may be defined for which it is
useful for such RRs to have different TTL's. In that case the T bit
must be one in the signet prefix octet, and the TTL of the hashed
RR(s) included after the two octet type and before the hash code.
If both the N and T bits are on, the name appears immediately after
the prefix byte followed by the type, then the TTL, then the 20 byte
hash code. This is the standard order for these fields in an RR.
5.1.2.4 Partial RR Set Flag Signet
Verification of a hashed RR signet against the RR(s) included in the
hash provides a guarantee that none have been omitted. The same
assurance is not provided by the direct RR signet unless all of the
direct RR signets of the same type and class are included in one SIG
RR. If direct RR signets are split over more than one RR, they MUST
be covered by a partial RR set flag signet in each RR.
The partial RR set flag signet is indicated by a hex FC prefix
followed a two octet type and then by a count octet. The most
significant 4 bits (nibble) of this count octet indicates one less
than the total number of SIG RRs that include all of the direct
signets for the variety of RRs in question. The least significant
nibble is used to distinguish the different SIG RRs required and
varies from zero through the value of the first nibble. It is
permissible, but unnecessary, to include a partial RR set flag signet
prefix followed by the type and a zero byte (i.e., 1 of 1) in the SIG
RR containing direct signets for all RRs of a particular varient.
For example, an signet of FC000131 (hex) means the SIG RR is the 2nd
of the 4 SIG RRs covering A type RRs.
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 18]
INTERNET-DRAFT DNS Protocol Security Extensions
Should there be a case where more than 16 SIG RRs could be required
to hold the direct signets for a particular variety of RR, direct
signets may not be used. The RRs must appear directly in the DNS
answer and a hashed signet must be used for authentication.
5.1.2.5 Partial SIG Set Flag Signet
Since the SIG RRs for an owner name are not in themselves covered by
yet another SIG (except for the case of zone transfers), a malicious
server might choose to provide only some of them in response to a
query for SIG RRs. The partial RR set flag signet defined above is
not guaranteed to help here.
The signet prefix of hex FD is followed by an unsigned byte which is
one less than the total number of SIG RRs associated with the name
and a second unsigned byte which varies from zero through the value
of the first byte. It is permissible, but unnecessary, to include a
partial SIG set flag signet prefix followed by two zero bytes (i.e, 1
of 1) if only one SIG RR is associated with a name.
More than 256 SIGs many not be associated with the same name and
class.
5.1.2.6 AXFR Signets
To secure zone transfers, a SIG under the zone name will have a
hashed RR signet with the AXFR type. It will be calculated by
hashing together all other static zone RRs, including SIGs. The RRs
are ordered and concatenated for hashing as described in Section
5.1.2.3. This SIG, other than having to be calculated last of all
zone key signed SIGs in the zone, is the same as any other SIG. It
can contain non-AXFR signets, be numbered with the partial SIG set
flag signet along with other zone level SIGs, if any, etc.
Dynamic zone RRs which might be added by some future dynamic zone
update protocol and signed by an end entity key rather than a zone
key (see Section 6.2) are not included. They originate in the
network and will not, in general, be migrated to the recommended off
line zone signing procedure (see Section 8.2). Thus such dynamic RRs
are not directly signed by the zone and are not generally protected
against omission during zone transfers.
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 19]
INTERNET-DRAFT DNS Protocol Security Extensions
5.1.2.7 Reserved Signet Prefixes
A number of signet prefixes are reserved for future allocation by the
Internet Assigned Numbers Authority (IANA). All such signets will
have an unsigned length octet immediately following their prefix
code. This will be the length of the signet data not including the
prefix or length octets. Thus their length can be unambiguously
determined.
Such signets are not to be generated by any implementation except for
a use for which they have been allocated by IANA. Such signets are
to be ignored on receipt by any implementation which does not
understand them.
5.2 SIG RRs in the Construction of Responses
SIG RRs MUST NOT be sent in response to a query where the SD header
bit is clear unless the query specifically requests the SIG type.
When the SD header bit is set, the DNS server MUST, for every RR the
query will return, attempt to send the available SIG RRs which
authenticates the requested RR. If multiple such SIGs are available,
there may be insufficient space in the response to include them all.
In this case, SIGs whose signer is the zone containing the RR MUST be
given highest priority and retained even if SIGs with other signers
must be dropped.
Furthermore, this automatic inclusion of SIGs in a response is NOT
additional information RR processing. To minimize possible
truncation problems, if a SIG covers any RR that would be in the
answer section of the response, it MUST appear in the answer section.
If it covers an RR that would appear in the authority section and
does not cover any answer section RR, it MUST appear in the authority
section. If it covers an RR that would appear in the additional
information section and does not cover any answer or authority
section RR, it MUST appear in the additional information section.
In many cases, as described below, the full authenticated RR will be
included inside the SIG RR. In such cases, the DNS server SHOULD
send only the SIG and surpress the directly requested RR.
The server should assume that the inquirer has the necessary public
key to authenticate RRs with the SIG and that, in general, an RR not
covered by a SIG may be considered worthless by the inquirer.
However, if a SIG including a full RR or an RR and its authenticating
SIG will not fit in a response, but the RR alone will, a server MAY
send the unauthenticated RR notwithstanding the set SD bit in the
query header.
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 20]
INTERNET-DRAFT DNS Protocol Security Extensions
Optionally, DNS messages may be authenticated by a SIG RR at the end
of the message in the additional information section. Such SIG RRs
are signed by the DNS server originating the message.
5.3 Processing Responses with SIG RRs
If SIG RRs are received in response to a query specifying the SIG
type, no special processing is required but a security aware client
may wish to authenticate them by decoding the signature and applying
consistency checks.
If SIG RRs are received in any other response, a security aware
client should decoded them using the public key of the signer and the
data coded into the signature should be carefully examined.
It should start with a 01 FF* 00 octet sequence (a 01 octet, zero or
more FF octets, and a 00 octet) followed immediately by one or more
concatenated signets and ending with the 20 byte self hash field
matching the hash of the relevant SIG RR fields outside of the
signature. If the decoded signature can not be parsed or the self
hash check fails, the SIG RR is invalid and should be ignored. The
time of receipt of the SIG RR must be in the inclusive range of the
time signed and the signature expiration but the SIG can be retained
and remains locally valid until the expiration time plus the range
authenticated TTL. Next, the contents of the one or more direct RR,
hashed RR, or flag signets present should be examined.
For all direct RR signets, the original RR should be reconstructed if
they are of a type that should have been retrieved by the query. If
they are of another type, they can be optionally reconstructed or
ignored. For all reconstructed RRs, there must be a complete set of
partial set flag signets or all must be included in one SIG. For
hashed RR signets, the hash should be computed from RRs present in
the response and compared for authentication. Hashed RR signets for
a type not requested in the query must be ignored.
If the SIG RR is the last RR in a response in the additional
information section, it may contain a hashed message signet covering
the preceding data in the response. This should be checked and the
message rejected if the check fails but it does NOT authenticate any
RRs in the message. Only proper direct or hashed RR signets signed
by the originating zone can authenticate RRs. The hashed message
signet merely protects from tampering between the DNS server and the
resolver making the query.
If all reasonable checks indicate that the SIG RR is valid then RRs
reconstructed or verified by hash should be considered authenticated
and all other RRs in the response should be considered with
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 21]
INTERNET-DRAFT DNS Protocol Security Extensions
suspicion. The probability that a SIG RR that has been tampered with
(without knowledge of the secret key) will pass reasonable checks is
vanishingly small (less than 1 in 2**150).
If a SIG RR is received in the additional information section of a
query, rather than a response, it can be optionally used to
authenticate the query. Warning: many current implementations of the
DNS ignore queries with a non-zero additional information count. A
message authenticating SIG RR should NOT be included in a query
unless you have outside knowledge that the queried system will permit
it or have received a DNS message from the system with the SA bit on
in the header.
5.4 File Representation of SIG RRs
A SIG RR covering RRs can be represented as a single logical line in
a zone data file [RFC1033] but there are some special problems as
described below. (It does not make sense to include a message
authenticating SIG RR in a file as it is a transient authentication
that must be calculated in real time by the message composing DNS
host.)
5.4.1 Size of Data
There is no particular problem with the signer and times. The time
fields appears in the form YYYYMMDDHHMMSS where YYYY is the year, the
first MM is the month number (01-12), DD is the day of the month
(01-31), HH is the hour in 24 hours notation (00-23), the second MM
is the minute (00-59), and SS is the second (00-59).
The original TTL appears as an unsigned integer.
However, the signature itself can be up to 255 octets. It is the
last data field and is represented in hex and may be divided up into
any number of white space separated substrings, down to single hex
digits, which are concatenated to obtain the full signature. These
hex substrings can be split between lines using the standard
parenthesis.
5.4.2 RR Numbering
A SIG RR stored in a zone file covers and in some cases surpresses a
number of resource records via one or more direct or hashed signets.
In order to represent this, RRs can be numbered by an integer field
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 22]
INTERNET-DRAFT DNS Protocol Security Extensions
enclosed in curly braces "{}". For compatibility with earlier DNS
zone file implementations, this field can occur after all data fields
but before any comments, and can be optionally preceded by a single
pound sign ("#") immediately before the open curly brace ("{"). This
"#" will cause the RR number to be treated as a comment by non-
security aware servers but the "#{ number }" will be recognized by
security aware servers.
[give example]
Actually, the RR number is the first subfield of three within the
curly braces. As described below additional fields can occur
separated by semi-colons (";").
5.4.3 SIG RR Scope
The RRs that are covered by a SIG RR are represented by a second
sub-field inside the curly brace field which must be present for SIG
RRs. This subfield consists of a comma and/or white space separated
list of RR numbers, or ranges of the form n-m to indicates all
integers from n through m inclusive. As an abbreviation, an asterisk
("*") appearing in this subfield means all RRs appearing before the
SIG RR in the zone file back to but not include the immediately
previous SIG RR or the beginning of the file, whether or not such
covered RRs are numbered.
The SIG RR should also be considered to cover any RRs that it
surpresses as explained in the section below.
The SIG RR itself need not be numbered unless it needs to be referred
to.
For example
[give examples here]
5.4.4 RRs Surpressed by a SIG RR
Where a SIG RR includes direct RR signets, the RRs being
authenticated should normally be surpressed when the SIG RR appears
in a response. This is indicated in the zone data file by a third
sub-field inside the curly brace field that may be present with SIG
RRs and must be present if they have direct RR signets.
As with the scope subfield, this subfield consists of a comma and/or
white space separated list of RR numbers or ranges of the form n-m to
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 23]
INTERNET-DRAFT DNS Protocol Security Extensions
indicate all integers from n through m inclusive. As an
abbreviation, an asterisk ("*") appearing in this subfield means all
RRs appearing before the SIG RR in the zone file back to but not
include the immediately previous SIG RR or the beginning of the file,
whether or not such surpressed RRs are numbered. An RR that is
surpressed is implicitly covered and may, but need not, also be
listed in the scope sub-field described in 5.4.2.
[give example]
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 24]
INTERNET-DRAFT DNS Protocol Security Extensions
6. The RSA Resource Record
The RSA RR is used to document a public key that is associated with a
DNS name. This can be a zone owner, the name of a DNS host for
message authentication, or any DNS name. An RSA RR is, like any
other RR, authenticated by a SIG RR. Security aware DNS
implementations should be designed to handle at least two
simultaneously valid keys associated with a name and try both for
decoding relevant SIG RRs to handle key roll over.
The type number for the RSA RR is xxx.
6.1 RSA RDATA format
The RDATA for a RSA RR consists of a start and end time, an octet of
flags, the public exponent, and the public modulus. The format is as
follows:
1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 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
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| start time |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| end time |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| flags | public key exponent |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| modulus length| |
+---------------+ public key modulus -+
/ /
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
The time fields are unsigned in seconds since the start of 1 January
1970.
The public key exponent is an unsigned 24 bit integer.
The modulus length is an unsigned octet. This limits keys to a
maximum of 255 bytes. A zero modulus length is special and indicates
the specific assertion that there is no key associated with the
owner.
The public key modulus field is a multiprecision unsigned integer.
The bits in the flag octet are described in Section 6.3 belows.
[It would be possible to allow additional optional fields after the
above in the SIG RR as described in draft-ietf-dns-ixfr-01.txt]
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 25]
INTERNET-DRAFT DNS Protocol Security Extensions
6.2 Types of DNS Names and Keys
The same DNS name may refer to up to three things at present. For
example, dee.lkg.dec.com could be (1) a zone, (2) a host, and (3) the
mapping into a DNS name of the user dee@lkg.dec.com (although at
present it is only the last of these three). Thus, the flag byte in
the RSA RR has bits to indicate with which of these roles a public
key is associated as described below. It is possible to use the same
key for these different things with the same DNS name but this is
discouraged.
The case of the host or other end entity is further subdivided. One
bit indicates that a key is generally associated with that entity. A
second indicates that the key belongs to the end entity and is
authorized to authenticate RRs for the end entity. In this case, the
thing represented by the key is the same and it would be reasonable
to use the same key for both the general and DNS updating /
authenticating roles but the freedom is provided to use different
keys.
It would be desirable for the growth of DNS to be managed so that
additional possible simultaneous uses for names are NOT added. New
uses should be distinguished by exclusive domains. For example, all
telephone numbers in the world have been mapped into the tpc.int
domain of the operational DNS system. This is preferable to having
the same name possibly be a telephone number and a host as well as a
zone and a user, depending on the RRs present.
6.3 RSA RR Flag Bits
Bit 0 (the most significant bit) a zero means the RSA RR asserts
the validity of the public key from the start to the end time,
inclusive. If it is a one, the RSA RR is a revocation of the key as
above. The strength of these assertions depends on the SIG RR(s), if
any, authenticating the RSA RR.
Bits 1 is the "mandatory" bit. Keys may be associated with zone
or entities for experimental, trial, or optional use, in which case
this bit will be zero. If this bit is a one, it means that the use
or availability of security based on the key is "mandatory". Thus,
if this bit is on for a zone, the zone should be assumed secured by
SIG RRs and any responses indicating the zone is not secured should
be considered bogus. Similarly, if this bit were on for a host key
and attempts to negotiate IP-security with the host produced
indications that IP-security was not supported, it should be assumed
that the host has been compromised or communications with it are
being spoofed. On the other hand, if this bit were a zero, the host
might very well sometimes operate in a secure mode and at other times
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 26]
INTERNET-DRAFT DNS Protocol Security Extensions
operate without the availability of IP-security. (This bit is
meaningless in a revocation RSA RR.)
Bits 2-3 are reserved and must be zero. If they are found non-
zero, they should be ignored and the RSA RR used as indicated by the
other flags.
Bit 4 on indicates that this is a zone key for the zone whose
name is the RSA RR owner name. This is the fundamental type of data
origin authentication public key.
Bit 5 on indicates that this is a key associated with the end
entity whose name is the RR owner name by which that entity is
authorized to authenticate DNS entries for itself. This assertion
is, of course, only valid if the asserting RSA RR is signed by a
valid zone key. This is intended to support certain types of dynamic
update.
Bit 6 on indicates that this is a key associated with the end
entity whose name is the RR owner name. This will commonly be a host
but could, in some part of the DNS tree, be some other type of
entity. This is the public key used in connection with the optional
message authentication service defined in this draft. It could also
be used in an IP-security protocol where authentication of a host was
desired. This would be useful in IP or other security for host level
services such as DNS, NTP, routing, etc.
Bit 7 on indicates that this is a key associated with a "user"
or "account" at an end entity, usually a host. The coding of the
owner name is that used for the responsible individual in the SOA
record: The owner name is the user name as the name of a node under
the entity name. For example, "j.random_user" on
host.subdomain.domain could have a public key associated then through
an RSA RR with name j\.random_user.host.subdomain.domain. It could
be used in an IP-security protocol where authentication of a user was
desired. This key would be useful in IP or other security for a user
level service such a telnet, ftp, rlogin, etc.
6.4 RSA RRs in the Construction of Responses
A request for RSA RRs does not cause any additional information
process because of these RSA RRs except, of course, for SIG RRs if
security is requested and available.
Security aware DNS servers will include RSA RRs as additional
information in responses where security is requested under
appropriate conditions as follows:
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 27]
INTERNET-DRAFT DNS Protocol Security Extensions
On the retrieval of NS RRs, the zone key RSA RR for the zone
served by these name servers will be included. If not all additional
info will fit, the RSA RR has higher priority than type A RRs.
On retrieval of type A RRs, the end entity RSA RR for the host
named will be included. On inclusion of A RRs as additional
information, they will also be included but with lower priority than
the relevant A RRs.
On any retrieval with security requested and available from a
zone, any revocation RSAs that fit will be included as additional
information with low priority and the relevant SIGs for such
revocation RSAs will also be included with lower priority than the
RSA RRs they sign.
6.5 File Representation of RSA RRs
RSA RRs may appear as lines in a zone data file.
The time fields appear in the form YYYYMMDDHHMMSS where YYYY is the
year, the first MM is the month number (01-12), DD is the day of the
month (01-31), HH is the hour in 24 hours notation (00-23), the
second MM is the minute (00-59), and SS is the second (00-59).
The flags field is represented as an unsigned integer.
The public key exponent appears as an unsigned integer from 3 to
16777215.
The public key modulus can be quite large, up to 255 octets. It is
the last data field and is represented in hex and may be divided up
into any number of white space separated substrings, down to single
hex digits, which are concatenated to obtain the full signature.
These hex substrings can span lines using the standard parenthesis.
The special case of a null key is indicated by a single zero digit.
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 28]
INTERNET-DRAFT DNS Protocol Security Extensions
7. How to Resolve Securely
Retrieving or resolving secure data from the DNS involves starting
with one or more trusted public keys and, in general, progressing
securely from them through the DNS structure to the zone of interest.
Such trusted public keys would normally be configured in a manner
similar to that described in section 7.1. However, as a practical
matter, a security aware resolver would still gain some confidence in
the results it returns even if it was not configured with any keys
but trusted what it got from a local well known server as a starting
point.
7.1 Boot File Format
The recommended format for a boot file line to configure starting
keys is as follows:
zonekey f.q.d.n exponent modulus
for a zone public key or
hostkey f.q.d.n exponent modulus
for a host public key. f.q.d.n is the domain name of the zone or
host, exponent is the public key exponent between 3 and 16777215, and
modulus is the public key modulus in hex. Appropriate "start" and
"end" times should be synthesized when the boot file is read.
While it might seem logical for everyone to start with the key for
the root zone, this has problems. The logistics of updating every
DNS resolver in the world when the root key changes would be
excessive. It may be some time before there even is a root key. And
furthermore, some organizations may explicitly wish their "interior"
DNS implementations to trust only their own zone. These interior
resolvers can then go through the organization's zone server to
access data outsize the organization's domain.
If desired, the IP address for the f.q.d.n's with configured keys can
generally also be configured via an /etc/hosts or similar local file.
7.2 Chaining Through Zones
Starting with one trusted zone key, it is possible to retrieve signed
keys for subzones which have them. Every secure zone (except root)
should also include the RSA record for its super-zone signed by the
secure zone. This makes it possible to climb the tree of zones if
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 29]
INTERNET-DRAFT DNS Protocol Security Extensions
one starts below root.
A resolver should keep track of the number of successive secure zones
traversed from a starting point to any secure zone it can reach. In
general, the lower such a distance number is, the greater the
confidence in the data and data configured via a boot file should be
given a distance number of zero. Should a query encounter different
data with different distance values, that with a larger value should
be ignored.
A security conscious resolver should completely refuse to step from a
secure zone into a non-secure zone unless the non-secure zone is
certified to be non-secure or only optionally secure by the present
of an authenticated RSA RR for the non-secure zone with a zero length
modulus or the presence of a non-zero length modulus RSA RR without
the mandatory bit set. Otherwise the resolver could be getting
completely bogus or spoofed replies.
If legitimate non-secure zones are encountered in traversing the DNS
tree, then no zone can be trusted as secure that can be reached only
via information from such non-secure zones. Since the non-secure zone
data could have been spoofed, the "secure" zone reach via it could be
counterfeit. The "distance" to data in such zones or zones reached
via such zones could be set to 512 or more as this exceeds the
largest possible distance through secure zones in the DNS. Never the
less, continuing to apply secure checks withing "secure" zones
reached via non-secure zones will, as a practical matter, provide
some increase in security.
7.3 Secure Time
Coordinated interpretation of the time fields in SIG and RSA RRs
requires that secure consistent time be available to the hosts
implementing the DNS security extensions. The Network Time Protocol
(NTP) [ref] provides an excellent means for coordinating consistent
time. It also includes strong security but has no key management
provisions. It just assumes that symmetric keying material will be
on each pair of communicating nodes.
In the absence of security, NTP or other time synchronization
protocols could be spoofed. A solution to this would be to do NTP
over IP-security. This may seem circular, where the security system
is used to protect the synchronization of time needed for the
security system. In practice, manual set up of approximate time
would be adequate to bootstrap the system which could then securely
synchronize itself more accurately.
To accommodate the secure time requirement, all DNS servers should
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 30]
INTERNET-DRAFT DNS Protocol Security Extensions
also be NTP servers.
NTP assumes that time servers are organized into numbered strata
where the servers at each strata are clients to a lower numbered
strata and servers to higher numbered strata. This can be
accomplished in the DNS context by have each primary or secondary DNS
server be an NTP client to the servers up the DNS tree from those
zones they provide (ignoring zones that are subzones of other zones
the server carries). Stub resolvers or the like could look to their
default server for NTP service. Special arrangements would have to
been made for the root primary and its secondaries to either have
reliable hardware time sources or be secure clients to machines with
such sources.
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 31]
INTERNET-DRAFT DNS Protocol Security Extensions
8. Operational Considerations
8.1 Modulus Size Considerations
There are a number of factors that effect modulus size choice for use
in the DNS security extension. Unfortunately, these factors do not
all point in the same direction. Choice of a modulus size should be
made by the zone administrator depending on their local conditions.
Larger moduluses are more secure but slower. The recommended minimum
modulus size, 641 bit, is believed by the authors to be secure at
this time and for some years but high level nodes in the DNS tree may
wish to set a higher minimum, perhaps 1000 bits, for security
reasons. (Since the United States National Security Agency generally
permits export of encryption systems using an RSA modulus of up to
512 bits, use of that small a modulus, i.e. n, must be considered
weak.)
Because this protocol packs information inside an RSA signature,
larger moduluses also increase the efficiency of use of space with
SIG RRs. There is a 22 byte overhead (prefix and self hash) within
the signature plus all the SIG RR fields outside of the signature.
However, larger moduluses also lead to larger SIG RRs which may lead
to lower packing density of SIG RRs in a maximum length DNS UDP
packet.
With the minimum modulus size required by this protocol, the
signature before RSA encoding is 80 octets (usually resulting in 81
octets after encoding). After deducting 2 octets for the minimum 01
00 signature prefix and 20 octets for a hashed self signet, 56 octets
would be available for other signets. With the maximum modulus size
permitted by this protocol, the signature is usually 255 octets which
leaves 233 bytes for other signets.
Zones may wish to adopt policies on the size of host, user, or even
subzone keys within them such that the RSA RRs for these keys will
fit within a zone signed SIG for efficiency.
8.2 Key Storage
It is strongly recommended that zone private keys and the zone file
master copy be kept and used in off-line non-network connected
machines. Periodically an application can be run to re-sign the RRs
in a zone by adding SIG RRs and then the signed file transferred,
perhaps by sneaker-net, to the networked server machine.
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 32]
INTERNET-DRAFT DNS Protocol Security Extensions
The idea is to have a one way information flow to the network to
avoid the possibility of tampering from the network. Keeping the
zone master file on-line on the network and simply cycling it through
an off-line signer does not do this. The on-line version could still
be tampered with if the host it resides on is compromised. The
master copy of the zone file should also be off net and should not be
updated based on a solely network mediated communication.
8.3 Key Generation
Careful key generation is a sometimes over looked but absolutely
essential element in any cryptographically secure system. The
strongest algorithms used with the longest keys are still of no use
if an adversary can guess enough to lower the size of the likely key
space so that it can be exhaustively searched. Suggestions will be
found in draft-ietf-security-randomness-01.txt.
8.4 Key Lifetimes
No key should be used forever. The longer a key is in use, the
greater the probability that it will have been compromised through
carelessness, accident, espionage, or cryptanalysis.
No DNS security extensions key should have a lifetime significantly
over five years. The recommended lifetime for zone keys that are
kept off-line and carefully guarded is 13 months with the intent that
they be replaced every year. The recommended lifetime for host keys
that are used for IP-security or the like and are kept on line is 35
days with the intent that they be replaced monthly.
8.5 Key Revocation
In cases where an erroneous signed key exists in the DNS system, it
is useful to be able to propagate a revocation. In most cases, the
natural refresh processes of DNS will eventually obtain valid up to
date key information for secondaries. However, there could be stale
information out in caching servers or the like for a long time,
particularly since accident or malicious action could have cause RRs,
including SIGs, RSAs, etc., with very long TTLs and/or distant end-
times to be distributed.
There are limits to how much can be done about this problem. The DNS
security extensions provide for revocation of keys. The revocation
information is provided when current key information is retrieved.
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 33]
INTERNET-DRAFT DNS Protocol Security Extensions
In addition, security aware servers opportunisticly flood revocation
information by including it with low priority in the additional
information section of any retrieval from the zone containing the
revoked key. It is a matter of server policy how to choose which
extra revocations to include as additional information. Some
revocations may be "more important" than others or it may be best to
simply pick as much as will fit at random.
On receipt of an authenticated revocation, a resolver should expunge
all RRs authenticated by the revoked key. It is a matter of resolver
policy what revocations, if any, to cache and for how long. If a
relavent revocation is received by a resolver but is not accompanied
by an authenticating SIG, the resolver should normally attempt to
retrieve such a SIG.
Resolvers can not keep an indefinite number of revocations for an
indefinite time. The possibility of denial of service attacks based
on fabricating many revocations must be considered.
It should be noted that like assertions, revocations have start and
end times. Thus, for example, a valid key validly generated but
accidentally given an excessive lifetime can be revoked for just the
later part of that lifetime by setting appropriate times in the
revocation RSA RR.
8.6 Root
The root zone RSA key is self-signed.
It should also be noted that in DNS the root is a zone unto itself.
Thus the root key should only be see signing itself or signing RRs
with names one level below root, such as .aq, edu, or .arpa.
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 34]
INTERNET-DRAFT DNS Protocol Security Extensions
9. Conformance
[this section needs work ...]
- Security aware server needs to respond normally to requests that do
not have the Security Desired bit set. [should response still have
the Security Available bit set if SD wasn't?]
- Minimal server compliance is ability to handle SIG and RSA RRs in
zone files, etc.
- Full server compliance is ability to handle SD and SA bits,
automatically include SIG and RSA RRs in responses as appropriate,
etc.
- Resolver compliance...
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 35]
INTERNET-DRAFT DNS Protocol Security Extensions
10. Security Considerations
The entirety of this document concerns extensions to the Domain Name
System (DNS) protocol to provide data origin authentication, DNS
message authentication, and key storage.
References
[PKCS1] - PKCS #1: RSA Encryption Standard, RSA Data Security, Inc.,
3 June 1991, Version 1.4.
[RFC1032] - Domain Administrators Guide, M. Stahl, November 1987
[RFC1033] - Domain Administrators Operations Guide, M. Lottor,
November 1987
[RFC1034] - Domain Names - Concepts and Facilities, P. Mockapetris,
November 1987
[RFC1035] - Domain Names - Implementation and Specifications
[SHS] - NIST FIPS PUB 180, Secure Hash Standard, National Institute
of Science and Technology, U.S. Department of Commerce, April 1993.
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 36]
INTERNET-DRAFT DNS Protocol Security Extensions
Authors Addresses
Donald E. Eastlake 3rd
Digital Equipment Corporation
550 King Street, LKG2-1/BB3
Littleton, MA 01460
Telephone: +1 508 486 6577(w) +1 508 287 4877(h)
EMail: dee@lkg.dec.com
Charles W. Kaufman
Digital Equipment Corporation
110 Spit Brook Road, ZKO3-3/U14
Nashua, NH 03062
Telephone: +1 603-881-1495
EMail: kaufman@zk3.dec.com
Expiration and File Name
This draft expires 22 August 1994
Its file name is draft-ietf-dnssec-secext-00.txt.
Donald E. Eastlake 3rd & Charles W. Kaufman [Page 37]
