[Search] [txt|pdf|bibtex] [Tracker] [WG] [Email] [Diff1] [Diff2] [Nits]

Versions: 00 01 02 rfc2726                                              
IETF RPS WG + RIPE DBSEC Task Force                             J. Zsako
INTERNET-DRAFT                                                   BankNet
<draft-ietf-rps-dbsec-pgp-authent-01.txt>                     April 1999

              PGP authentication for RIPE database updates

Status of this Memo

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

   Copyright (C) The Internet Society (1999).  All Rights Reserved.

   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-

   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


   This document presents the proposal for a stronger authentication
   method of the updates of the RIPE database based on digital
   signatures. The proposal tries to be as general as possible as far as
   digital signing methods are concerned, however, it concentrates
   mainly on PGP, as the first method to be implemented.  The proposal
   is the result of the discussions within the RIPE DBSEC Task Force.

1. Rationale

   An increasing need has been identified for a stronger authentication
   of the database maintainer upon database updates (addition,
   modification and deletion of objects). The existing authentication
   methods have serious security problems: the MAIL-FROM has the
   drawback that a mail header is very easy to forge whereas CRYPT-PW is
   exposed to message interception, since the password is sent
   unencrypted in the update mail message.

   The goal was to implement a digital signature mechanism based on a
   widely available and deployed technology. The first choice was PGP,
   other methods may follow at a later date. PGP is presently quite
   widely used within the Internet community and is available both in
   and outside the US.

   The current aim is for an improved authentication method and nothing
   more (in particular, this paper does not try to cover authorization
   issues other than those related to authentication).

2. Changes to the RIPE database

   In order to make the database as much self consistent as possible,
   the key certificates are stored in the RIPE database. For efficiency
   reasons a local keyring of public keys will also be maintained,
   however, the local keyring will only contain a copy of the key
   certificates present in the database. The synchronization of the
   database with the local keyring will be made as often as possible.
   The database objects will be created only via the current e-mail
   mechanism (auto-dbm@ripe.net), in particular no public key
   certificate will be retrieved from a key server by the database

   The presence of the key certificates in the database will allow the
   users of the database to check the "identity" of the maintainer, in
   the sense that they can query the database for the certificate of the
   key the database software uses for authenticating the maintainer.
   This key certificate can then be checked for existing signatures and
   can possibly be compared with the key certificate obtained by other
   means for the same user (e.g. from the owner himself of from a public
   key server). Although the key certificates can be stored in the RIPE
   database with any number of signatures, since the RIPE database is
   not communicating directly with the public key servers, it is a good
   practice to add the key certificate with the minimum number of
   signatures possible (preferably with just one signature: the one of
   itself).  See also section 4. for more details.

2.1. The key-cert object

   A new object type is defined below for the purpose of storing the key
   certificates of the maintainers:

   key-cert:  [mandatory]  [single]     [primary/look-up key]
   method:    [generated]  [single]
   owner:     [generated]  [multiple]   [ ]
   fingerpr:  [generated]  [single]     [ ]
   certif:    [mandatory]  [single]   [ ]
   remarks:   [optional]   [multiple]   [ ]
   notify:    [optional]   [multiple]   [inverse key]
   mnt-by:    [mandatory]  [multiple]   [inverse key]
   changed:   [mandatory]  [multiple]   [ ]
   source:    [mandatory]  [single]     [ ]

   The syntax and the semantics of the different attributes are
   described below.

   PGPKEY-hhhhhhhh, where hhhhhhhh stands for the hex representation of
   the four bytes ID of the PGP key.  The key certificate detailed in
   the certif attribute belongs to the PGP key with the id hhhhhhhh. The
   reason for having PGPKEY- as a prefix is to allow for other types of
   key certificates at a later date, and at the same time to be able to
   clearly differentiate at query time between a person query and a key
   certificate query.  at the time of the creation/modification of the
   key-cert object, the database software checks whether the key
   certificate in the certif attribute indeed belongs to the PGP id
   specified here. The creation/modification is authorized only upon the
   match of these two ids.

   method:  line containing the name of the signing method.  this is the
   name of the digital signature method. The present certificate belongs
   to a key for digitally signing messages using the specified method.
   the method attribute is generated automatically by the database
   software upon creation of the key-cert object.  Any method attribute
   present in the object at the time of the submission for creation is
   ignored.  The method has to be consistent with both the prefix of the
   id in the key-cert attribute and with the certificate contained in
   the certif attributes. If these latter two (i.e. prefix and
   certificate) are not consistent, the key-cert object creation is
   refused. For the PGP method this will be the string "PGP" (without
   the quotes).

   owner:  line containing a description of the owner of the key.  for a
   PGP key, the owners are the user ids associated with the key.  for
   each user id present in the key certificate, an owner attribute is
   generated automatically by the database software upon creation of the
   key-cert object.  Any owner attribute present in the object at the
   time of the submission for creation is ignored.

   fingerpr:  a given number of hex encoded bytes, separated for better
   readability by spaces.  it represents the fingerprint of the key
   associated with the present certificate.  this is also a field
   generated upon creation of the object instance. Any fingerpr
   attribute submitted to the robot is ignored.  The reason for having
   this attribute (and the owner attribute) is to allow for an easy
   check of the key certificate upon a query of the database.  The
   querier gets the owner and fingerprint information without having to
   add the certificate to his/her own public keyring.  Also, since these
   two attributes are _generated_ by the database software from the
   certificate, one can trust them (as much as one can trust the
   database itself).

   certif:  line containing a line of the ASCII armoured key
   certificate.  the certif attribute lines contain the key certificate.
   In the case of PGP, they also contain the delimiting lines (BEGIN/END
   PGP PUBLIC KEY BLOCK).  Obviously the order of the lines is
   essential, therefore the certif attribute lines are presented at
   query time in the same order as they have been submitted at creation.
   A database client application could contain a script that strips the
   certif attribute lines (returned as a result of a query) from the
   leading "certif:" string and the following white spaces and import
   the remainder in the local keyring.

   mnt-by:  the usual syntax the usual semantics this attribute is
   _mandatory_ for this object. Therefore, the existence of a mntner
   object is a prerequisite for the creation of a key-cert object. The
   mntner referenced in the mnt-by attribute may not have the auth
   attribute set to NONE.

   source:  the usual syntax the usual semantics.

   In the case of PGP, when a key-cert object is created, the associated
   key is also added to a local keyring of public keys. When a key-cert
   object is deleted, the corresponding public key is deleted from the
   local keyring as well. Whenever a key-cert object is modified, the
   key is deleted from the local keyring and the key associated with the
   new certificate is added to the keyring (obviously this is performed
   only when the database update is authorized, in particular if the new
   key certificate does belong to the id specified in the attribute
   key-cert, see section 2.1).

2.2. Changes to the mntner object

   The only change is that there is a new possible value for the auth
   attribute.  This value is of the form PGPKEY-<id>, where <id> is the
   hex representation of the four bytes id of the PGP public key used
   for authentication.

   The semantics of this new value is that the PGP key associated with
   the key certificate stored in the key-cert object identified by
   PGPKEY-id is used to check the signature of any
   creation/modification/deletion message sent to auto-dbm@ripe.net
   affecting an object maintained by this mntner.

   Just as with other values, the auth attribute can be multiple. It
   does not make much sense to have two auth attributes with different
   methods (e.g.  PGPKEY-<id> and NONE :)) ), just as it didn't earlier

   If there are several auth methods with a PGPKEY-<id> value, the
   semantics is the already known one, namely that _either_ signature is

3. The PGP signed creation/modification/deletion

   The whole message has to be signed. This means, that the database
   software first checks whether the message is a PGP signed message. If
   it is, it checks for a valid signature and associates this signature
   with the objects submitted in the message. A message may contain only
   one PGP signature.

   If an object present in a message has a mnt-by attribute, and the
   respective mntner has auth attribute(s) with PGPKEY-<id> value, the
   database software checks whether the object has a signature
   associated with it (i.e. whether the message being processed had been
   signed) and whether the type of the signature (PGP in this
   implementation phase) and the id of the key used for signing the
   message is the same as the one in (one of) the auth attribute(s). The
   creation/modification/deletion of the object is performed only if
   this authentication succeeds.

   This approach allows for a simplification of the message parsing
   process.  A different approach would be necessary if one would sign
   the _objects_, rather then the update messages. In case the objects
   would be signed, the parser would have to identify which objects were
   signed, check the signature(s) on each object individually and
   permit/refuse the update at an object level, depending on (amongst
   others) whether the signature is valid and whether it belongs to (one
   of) the maintainer(s). This approach would allow for mixing in the
   same e-mail message objects signed by different maintainers (which
   would probably not be typical), and it would also allow for storing
   the signature in the database (in order to allow for the verification
   of the signature at query time). This latter (i.e. storing the
   signatures in the database) is beyond the scope of the first phase of
   the implementation. It may become a goal at a later date.

   It is recommended to check that the mailer program does not make any
   transformations on the text of the e-mail message (and possibly
   configure it not to do any). Such common transformations are line-
   wrapping after a given number of characters, transforming of tabs in
   spaces, etc. Also check that you only use ASCII characters in the

4. Requirements the PGP key certificates must meet

   There is no limitation imposed with respect to the version of the PGP
   software that is/was used for the creation of the key. Key of both
   version 2.x and 5.0 are supported, although the keys generated with
   version 5.0 are recommended.

   The key certificates submitted for creating a key-cert object must
   contain a signature of the key itself. Although the certificate may
   contain other signatures as well, it is recommended to create the
   key-cert object with as few signatures as possible in the
   certificate. Anyone concerned about the trustfulness of the key
   should retrieve a copy of the key certificate from a public key
   server (or by any other appropriate means and check the signatures
   present in _that_ certificate. If such a check is performed one
   should take care to check both the key fingerprint and the key type
   and length in order to make sure the two certificates (the one
   retrieved from the RIPE database and the one retrieved from the
   public key server or collected by other means) belong to the same

   Although it is highly unlikely, it may happen that a key-cert with
   the id identical to the id of the key of a maintainer already exists
   in the RIPE database.  In case this latter key had been used for a
   while and it had been registered at public key servers for some time,
   the given person should contact the RIPE NCC and report this to
   ripe-dbm@ripe.net. Anyway, he/she may have to create a new key and
   register _its_ certificate into the RIPE database. Such a procedure,
   although highly unlikely to happen, should not create serious
   problems to the respective maintainer.

5. Short overview of the tasks to be performed in order to use PGP

   You must have a mntner object in the RIPE database with auth: other
   than NONE.  The mntner object has to be created in the traditional

   You must get a certificate of your own key and prepare a key-cert
   object from it. The object has to reference in mnt-by the mntner
   mentioned above (a.).

   Create the key-cert object in the RIPE database, by sending the
   object prepared above (b.) to auto-dbm@ripe.net. Obviously you must
   pass the authentication checks required by the mntner object (i.e.
   mail from a predefined address or send the correct password).

   Change the mntner object to have the auth: attribute value of
   PGPKEY-<id>, where <id> is the hex id of your PGP key.

   Check all objects maintained by the given mntner (preferably with the
   command This is the only way to benefit from the stronger
   authentication method in order to assign more trustfulness to the
   database. Remember that you are the only person who can check for and
   correct possible inconsistencies.

   >From now on always sign the (the whole) update messages that refer
   to objects maintained by you, with the key you submitted to the RIPE

6. Example of objects using the new feature

mntner:      AS3244-MNT
descr:       BankNet, Budapest HU
descr:       Eastern European Internet Provider via own VSAT network
admin-c:     JZ38
tech-c:      JZ38
tech-c:      IR2-RIPE
upd-to:      ncc@banknet.net
mnt-nfy:     ncc@banknet.net
auth:        PGPKEY-23F5CE35
remarks:     This is the maintainer of all BankNet related objects
notify:      ncc@banknet.net
mnt-by:      AS3244-MNT
changed:     zsako@banknet.net 19980525
source:      RIPE

key-cert: PGPKEY-23F5CE35
method:   PGP
owner:    Janos Zsako <zsako@banknet.net>
fingerpr: B5 D0 96 D0 D0 D3 2B B2  B8 C2 5D 22 D4 F5 78 92
certif: -----BEGIN PGP PUBLIC KEY BLOCK-----
 Version: 2.6.2i
remarks: This is an example of PGP key certificate
mnt-by:  AS3244-MNT
changed: zsako@banknet.net 19980525
source:  RIPE


   The present proposal is the result of the discussions within the RIPE
   DBSEC Task Force, which was set up at RIPE 27 in Dublin at the
   initiative of Joachim Schmitz and Wilfried Woeber. The list of
   participants who have contributed to the discussions at different
   ocasions (TF meetings and via e-mail) is (in alphabetical order):
   Cengiz Allaettinoglu, Joao Luis Silva Damas, Havard Eidnes, Chris
   Fletcher, Daniel Karrenberg, David Kessens, Jake Khuon, Craig
   Labovitz, Carl Malamud, Dave Meyer, Maldwyn Morris, Sandy Murphy,
   Mike Norris, Carol Orange, Joachim Schmitz, Tom Spindler, Don
   Stikvoort, Curtis Villamizar, Gerald Winters, Wilfried Woeber, Janos

Author's Address

   Janos Zsako
   1121 Budapest
   Konkoly-Thege ut 29-33.

   Phone: +36 1 395 90 28
   Fax:   +36 1 395 90 32
   EMail: zsako@banknet.net