INTERNET-DRAFTClifford Neuman draft-ietf-cat-kerberos-pk-init-03.txtBrian Tung draft-ietf-cat-kerberos-pk-init-04.txt Clifford Neuman Updates: RFC 1510 ISI expiresSeptember 30, 1997January 31, 1998 John Wray Digital Equipment Corporation Ari Medvinsky Matthew Hur CyberSafe Corporation Jonathan Trostle Novell Public Key Cryptography for Initial Authentication in Kerberos 0. Status OfthisThis Memo This document is an Internet-Draft. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." To learn the current status of any Internet-Draft, please check the "1id-abstracts.txt" listing contained in the Internet-Drafts Shadow Directories on ds.internic.net (US East Coast), nic.nordu.net (Europe), ftp.isi.edu (US West Coast), or munnari.oz.au (Pacific Rim). The distribution of this memo is unlimited. It is filed asdraft-ietf-cat-kerberos-pk-init-03.txt,draft-ietf-cat-kerberos-pk-init-04.txt, and expiresSeptember 30, 1997.January 31, 1998. Please send comments to the authors. 1. Abstract This document defines extensions (PKINIT) to the Kerberos protocol specification (RFC 1510 [1]) to provide a method for using public key cryptography during initial authentication. The methods defined specify the ways in which preauthentication data fields and error data fields in Kerberos messages are to be used to transport public key data. 2. Introduction The popularity of public key cryptography has produced a desire for its support in Kerberos [2]. The advantages provided by public key cryptography include simplified key management (from the Kerberos perspective) and the ability to leverage existing and developing public key certification infrastructures. Public key cryptography can be integrated into Kerberos in a number of ways. One is totoassociate a key pair with each realm, which can then be used to facilitate cross-realm authentication; this is the topic of another draft proposal. Another way is to allow users with public key certificates to use them in initial authentication. This is the concern of the current document. One of the guiding principles in the design of PKINIT is that changes should be as minimal as possible. As a result, the basic mechanism of PKINIT is as follows: The user sends a request to the KDC as before, except that if that user is to use public key cryptography in the initial authentication step, his certificate accompanies the initial request, in the preauthentication fields. Upon receipt of this request, the KDC verifies the certificate and issues a ticket granting ticket (TGT) as before, except that instead of being encrypted in the user's long-term key (which is derived from a password), it is encrypted in a randomly-generated key. This random key is in turn encrypted using the public key from the certificate that came with the request and signed using the KDC'ssignatureprivate key, and accompanies the reply, in the preauthentication fields. PKINIT also allows for users with only digital signature keys to authenticate using those keys, and for users to store and retrieve private keys on the KDC. The PKINIT specification may also be used for direct peer to peer authentication without contacting a central KDC. This application of PKINIT is described in PKTAPP [4] and is based on concepts introduced in [5, 6]. For direct client-to-server authentication, the client uses PKINIT to authenticate to the end server (instead of a central KDC), which then issues a ticket for itself. This approach has an advantage over SSL [7] in that the server does not need to save state (cache session keys). Furthermore, an additional benefit is that Kerberos tickets can facilitate delegation (see [8]). 3. Proposed Extensions This section describes extensions to RFC 1510 for supporting the use of public key cryptography in the initial request for a ticket granting ticket (TGT). In summary, the following changes to RFC 1510 are proposed: --> Users may authenticate using either a public key pair or a conventional (symmetric) key. If public key cryptography is used, public key data is transported in preauthentication data fields to help establish identity. --> Users may store private keys on the KDC for retrieval during Kerberos initial authentication. This proposal addresses two ways that users may use public key cryptography for initial authentication. Users may present public key certificates, or they may generate their own session key, signed by their digital signature key. In either case, the end result is that the user obtains an ordinary TGT that may be used for subsequent authentication, with such authentication using only conventional cryptography. Section 3.1 provides definitions to help specify message formats. Section 3.2 and 3.3 describe the extensions for the two initial authentication methods. Section3.33.4 describes a way for the user to store and retrieve his private key on theKDC.KDC, as an adjunct to the initial authentication. 3.1. DefinitionsHash andThe extensions involve new encryptiontypes will be specified using ENCTYPE tags;methods; we propose the addition of the following types:#define ENCTYPE_SIGN_DSA_GENERATE 0x0011 #define ENCTYPE_SIGN_DSA_VERIFY 0x0012 #define ENCTYPE_ENCRYPT_RSA_PRIV 0x0021 #define ENCTYPE_ENCRYPT_RSA_PUB 0x0022 allowing further signature types to be defined in the range 0x0011 through 0x001f, and furtherdsa-sign 8 rsa-priv 9 rsa-pub 10 rsa-pub-md5 11 rsa-pub-sha1 12 The proposal of these encryption typesto be defined innotwithstanding, we do not mandate therange 0x0021 through 0x002f.use of any particular public key encryption method. The extensions involve new preauthenticationfields. The preauthentication data types are infields; we propose therange 17 through 21. These values are also specified along with their corresponding ASN.1 definition. #defineaddition of the following types: PA-PK-AS-REQ17 #define14 PA-PK-AS-REP18 #define15 PA-PK-AS-SIGN19 #define16 PA-PK-KEY-REQ20 #define17 PA-PK-KEY-REP2118 The extensions also involve new errortypes. The new error types aretypes; we propose the addition of the following types: KDC_ERR_CLIENT_NOT_TRUSTED 62 KDC_ERR_KDC_NOT_TRUSTED 63 KDC_ERR_INVALID_SIG 64 KDC_ERR_KEY_TOO_WEAK 65 In many cases, PKINIT requires the encoding of an X.500 name as a Realm. In these cases, the realm will be represented using a different style, specified in RFC 1510 with therange 227 through 229. They are: #define KDC_ERROR_CLIENT_NOT_TRUSTED 227 #define KDC_ERROR_KDC_NOT_TRUSTED 228 #define KDC_ERROR_INVALID_SIG 229following example: NAMETYPE:rest/of.name=without-restrictions For a realm derived from an X.500 name, NAMETYPE will have the value X500-ASN1-BASE64. The full realm name will appear as follows: X500-ASN1-BASE64:Base64Encode(DistinguishedName) where Base64 is an ASCII encoding of binary data as per RFC 1521, and DistinguishedName is the ASN.1 encoding of the X.500 Distinguished Name from the X.509 certificate. Similarly, PKINIT may require the encoding of an X.500 name as a PrincipalName. In these cases, the name-type of the principal name shall be set to NT-X500-PRINCIPAL, and the name-string shall be set as follows: Base64Encode(DistinguishedName) as described above. [Similar description needed on how realm names and principal names are to be derived from PGP names.] 3.1.1. Encryption and Key Formats In the exposition below, we use thefollowing terms: encryption key, decryption key, signature key, verification key.terms public key and private key generically. It should be understood that the term "public key" may be used to refer to either a public encryptionandkey or a signature verificationkeys are essentially public keys,key, and that the term "private key" may be used to refer to either a private decryptionandkey or a signaturekeys are essentially private keys.generation key. The fact thattheythese are logically distinct does not preclude the assignment of bitwise identical keys. All additional symmetric keys specified in this draft shall use the same encryption type as the session key in the response from the KDC. These include the temporary keys used to encrypt the signed random key encrypting the response, as well as the key derived from Diffie-Hellman agreement. In the case of Diffie-Hellman, the key shall be produced from the agreed bit string as follows: * Truncate the bit string to the appropriate length. * Rectify parity in each byte (if necessary) to obtain the key. For instance, in the case of a DES key, we take the first eight bytes of the bit stream, and then adjust the least significant bit of each byte to ensure that each byte has odd parity. RFC 1510, Section 6.1, defines EncryptedData as follows: EncryptedData ::= SEQUENCE { etype [0] INTEGER, kvno [1] INTEGER OPTIONAL, cipher [2] OCTET STRING -- is CipherText } RFC 1510 suggests that ciphertext is coded as follows: CipherText ::= ENCRYPTED SEQUENCE { confounder [0] UNTAGGED OCTET STRING(conf_length) OPTIONAL, check [1] UNTAGGED OCTET STRING(checksum_length) OPTIONAL, msg-seq [2] MsgSequence, pad [3] UNTAGGED OCTET STRING(pad_length) OPTIONAL } The PKINIT protocol introduces several new types of encryption. Data that is encrypted with a public key will allow only the check optional field, as it is defined above. This type of the checksum will be specified in the etype field, together with the encryption type. In order to identify the checksum type, etype will have the following values: rsa-pub-MD5 rsa-pub-SHA1 In the case that etype is set to rsa-pub, the optional 'check' field will not be provided. Data that is encrypted with a private key will not use any optional fields. PKINIT uses private key encryption only for signatures, which are encrypted checksums. Therefore, the optional check field is not needed. 3.2. Standard Public Key Authentication Implementation of the changes in this section is REQUIRED for compliance withpk-init.PKINIT. It is assumed that all public keys are signed by some certification authority (CA). The initial authentication request is sent as per RFC 1510, except that a preauthentication field containing data signed by the user'ssignatureprivate key accompanies the request: PA-PK-AS-REQ::-::= SEQUENCE { -- PA TYPE17 signedPKAuth14 signedAuthPack [0]SignedPKAuthenticator,SignedAuthPack userCert [1] SEQUENCE OF Certificate OPTIONAL, -- the user's certificate-- optionally followed by that -- certificate's certifierchain trustedCertifiers [2] SEQUENCE OF PrincipalName OPTIONAL -- CAs that the client trusts }SignedPKAuthenticatorSignedAuthPack ::= SEQUENCE {pkAuthauthPack [0]PKAuthenticator, pkAuthSigAuthPack, authPackSig [1] Signature, -- ofpkAuthauthPack -- using user'ssignatureprivate key } AuthPack ::= SEQUENCE { pkAuthenticator [0] PKAuthenticator, clientPublicValue [1] SubjectPublicKeyInfo OPTIONAL -- if client is using Diffie-Hellman } PKAuthenticator ::= SEQUENCE {cuseckdcName [0] PrincipalName, kdcRealm [1] Realm, cusec [2] INTEGER, -- for replay prevention ctime[1][3] KerberosTime, -- for replay prevention nonce[2] INTEGER, -- binds response to this request kdcName [3] PrincipalName, clientPubValue[4]SubjectPublicKeyInfo OPTIONAL, -- for Diffie-Hellman algorithmINTEGER } Signature ::= SEQUENCE { signedHash [0] EncryptedData -- of type Checksum-- encrypted under signature key} Checksum ::= SEQUENCE { cksumtype [0] INTEGER, checksum [1] OCTET STRING } -- as specified by RFC 1510 SubjectPublicKeyInfo ::= SEQUENCE { algorithm [0]algorithmIdentifier,AlgorithmIdentifier, subjectPublicKey [1] BIT STRING -- for DH, equals -- public exponent (INTEGER encoded -- as payload of BIT STRING) } -- as specified by the X.509 recommendation [9] AlgorithmIdentifier ::= SEQUENCE { algorithm [0] ALGORITHM.&id, -- for DH, equals -- dhKeyAgreement -- ({iso(1) member-body(2) US(840) -- rsadsi(113549) pkcs(1) pkcs-3(3) -- 1}) parameters [1] ALGORITHM.&type -- for DH, is DHParameter } -- as specified by the X.509 recommendation [9] DHParameter ::= SEQUENCE { prime [0] INTEGER, -- p base [1] INTEGER, -- g privateValueLength [2] INTEGER OPTIONAL } Certificate ::= SEQUENCE {CertTypecertType [0] INTEGER, -- type of certificate -- 1 = X.509v3 (DER encoding) -- 2 = PGP (per PGPdraft) CertDataspecification) certData [1] OCTET STRING -- actual certificate -- type determined byCertTypecertType }Note: If the signature uses RSA keys, then it is to be performed as per PKCS #1.The PKAuthenticator carries information to foil replay attacks, to bind the request and response, and to optionally pass the client's Diffie-Hellman public value (i.e. for using DSA in combination with Diffie-Hellman). The PKAuthenticator is signed with the private key corresponding to the public key in the certificate found in userCert (or cached by the KDC). In the PKAuthenticator, the client may specify the KDC name in one of two ways:1) a* The Kerberos principalname, or 2)name krbtgt/<realm_name>@<realm_name>, where <realm_name> is replaced by the applicable realm name. * The name in the KDC's certificate (e.g., an X.500 name, or a PGP name). Note that the first case#1requires that the certificate name and the Kerberos principal name be bound together (e.g., via an X.509v3 extension). The userCert field is a sequence of certificates, the first of which must be the user's public key certificate. Any subsequent certificates will be certificates of the certifiers of the user's certificate. These cerificates may be used by the KDC to verify the user's public key. This fieldismay be left empty if the KDC already has the user'scertifcate.certificate. The trustedCertifiers field contains a list of certification authorities trusted by the client, in the case that the client does not possess the KDC's public key certificate. Upon receipt of the AS_REQ with PA-PK-AS-REQ pre-authentication type, the KDC attempts to verify the user's certificate chain (userCert), if one is provided in the request. This is done by verifying the certification path against the KDC's policy of legitimate certifiers. This may be based on a certification hierarchy, or it may be simply a list of recognized certifiers in a system like PGP. Ifthe certification path does not match oneverification of theKDC's trusted certifiers,user's certificate fails, the KDC sends back an error message of typeKDC_ERROR_CLIENT_NOT_TRUSTED, and it includes in the error dataKDC_ERR_CLIENT_NOT_TRUSTED. The e-data fielda list of its own trusted certifiers, upon which the client resendscontains additional information pertaining to this error, and is formatted as follows: METHOD-DATA ::= SEQUENCE { method-type [0] INTEGER, -- 1 = cannot verify public key -- 2 = invalid certificate -- 3 = revoked certificate -- 4 = invalid KDC name method-data [1] OCTET STRING OPTIONAL } -- syntax as for KRB_AP_ERR_METHOD (RFC 1510) The values for therequest.method-type and method-data fields are described in Section 3.2.1. If trustedCertifiers is provided in the PA-PK-AS-REQ, the KDC verifies that it has a certificate issued by one of the certifiers trusted by the client. If it does not have a suitable certificate, the KDC returns an error message of typeKDC_ERROR_KDC_NOT_TRUSTEDKDC_ERR_KDC_NOT_TRUSTED to the client. If a trust relationship exists, the KDC then verifies the client's signature on PKAuthenticator. If that fails, the KDC returns an error message of typeKDC_ERROR_INVALID_SIG.KDC_ERR_INVALID_SIG. Otherwise, the KDC uses the timestamp in the PKAuthenticator to assure that the request is not a replay. The KDC also verifies that its name is specified in the PKAuthenticator. If the clientPublicValue field is filled in, indicating that the client wishes to use Diffie-Hellman key agreement, then the KDC checks to see that the parameters satisfy its policy. If they do not (e.g., the prime size is insufficient for the expected encryption type), then the KDC sends back an error message of type KDC_ERR_KEY_TOO_WEAK. Otherwise, it generates its own public and private values for the response. The KDC also checks that the timestamp in the PKAuthenticator is within the allowable window. If the local (server) time and the client time in the authenticator differ by more than the allowable clock skew, then the KDC returns an error message of type KRB_AP_ERR_SKEW. Assuming no errors, the KDC replies as per RFC 1510, except as follows: The user's name in the ticket is as represented in the certificate, unless a Kerberos principal name is bound to the name in the certificate (e.g., via an X.509v3 extension). The user's realm in the ticket shall be the name of the Certification Authority thatitissued the user's public key certificate. The KDC encrypts the reply not with the user's long-term key, but with a random key generated only for this particular response. This random key is sealed in the preauthentication field: PA-PK-AS-REP ::= SEQUENCE { -- PA TYPE18 kdcCert15 encSignedReplyKeyPack [0]SEQUENCE OF Certificate OPTIONAL,EncryptedData, --the KDC's certificateof type SignedReplyKeyPack --optionally followed by thatusing the temporary key --certificate's certifier chain encPaReplyin encTmpKey encTmpKeyPack [1] EncryptedData, -- of typePaReplyTmpKeyPack -- using either the client public -- key or the Diffie-Hellman key -- specified by SignedDHPublicValuesignedDHPublicValuesignedKDCPublicValue [2]SignedDHPublicValueSignedKDCPublicValue OPTIONAL -- if one was passed in the request kdcCert [3] SEQUENCE OF Certificate OPTIONAL, -- the KDC's certificate chain }PaReplySignedReplyKeyPack ::= SEQUENCE {replyEncKeyPackreplyKeyPack [0]ReplyEncKeyPack, replyEncKeyPackSigReplyKeyPack, replyKeyPackSig [1] Signature, -- of replyEncKeyPack -- using KDC'ssignatureprivate key }ReplyEncKeyPackReplyKeyPack ::= SEQUENCE {replyEncKeyreplyKey [0] EncryptionKey, -- used to encrypt main reply nonce [1] INTEGER -- binds response to the request -- must be same as the nonce -- passed in the PKAuthenticator }SignedDHPublicValueTmpKeyPack ::= SEQUENCE { tmpKey [0] EncryptionKey, -- used to encrypt the -- SignedReplyKeyPack } SignedKDCPublicValue ::= SEQUENCE {dhPublicValuekdcPublicValue [0] SubjectPublicKeyInfo,dhPublicValueSig-- as described above kdcPublicValueSig [1] Signature -- ofdhPublicValuekdcPublicValue -- using KDC'ssignatureprivate key } The kdcCert field is a sequence of certificates, the first of which must be the KDC's public key certificate. Any subsequent certificates will be certificates of the certifiers of the KDC's certificate. The last of these must have as itsrootcertifier one of the certifiers sent to the KDC in the PA-PK-AS-REQ.Any subsequent certificates will be certificates of the certifiers of the KDC's certificate.These cerificates may be used by the client to verify the KDC's public key. This field is empty if the client did not send to the KDC a list of trusted certifiers (the trustedCertifiers field was empty). Since each certifier in the certification path of a user's certificate is essentially a separate realm, the name of each certifier shall be added to the transited field of the ticket. The format of these realm namesshall follow the naming constraints set forthis defined inRFC 1510 (sections 7.1 and 3.3.3.1). Note thatSection 3.1 of thiswill require new nametypes todocument. If applicable, the transit-policy-checked flag should bedefined for PGP certifiers and other types of realms as they arise.set in the issued ticket. The KDC's certificate must bind the public key to a name derivable from the name of the realm for that KDC. For an X.509 certificate, this is done as follows. The certificate will contain a distinguished X.500 name contains, in addition to other attributes, an extended attribute, called principalName, with the KDC's principal name as its value (as the text string krbtgt/<realm_name>@<realm_name>, where <realm_name> is the realm name of the KDC): principalName ATTRIBUTE ::= { WITH SYNTAX PrintableString(SIZE(1..ub-prinicipalName)) EQUALITY MATCHING RULE caseExactMatch ORDERING MATCHING RULE caseExactOrderingMatch SINGLE VALUE TRUE ID id-at-principalName } ub-principalName INTEGER ::= 1024 id-at OBJECT IDENTIFIER ::= {joint-iso-ccitt(2) ds(5) 4} id-at-principalName OBJECT IDENTIFIER ::= {id-at 60} where ATTRIBUTE is as defined in X.501, and the matching rules are as defined in X.520. [Still need to register principalName.] [Still need to discuss what is done for a PGP certificate.] The client then extracts the random key used to encrypt the main reply. This random key (in encPaReply) is encrypted with either the client's public key or with a key derived from the DH values exchanged between the client and the KDC. 3.2.1. Additional Information for Errors This section describes the interpretation of the method-type and method-data fields of the KDC_ERR_CLIENT_NOT_TRUSTED error. If method-type=1, the client's public key certificate chain does not contain a certificate that is signed by a certification authority trusted by the KDC. The format of the method-data field will be an ASN.1 encoding of a list of trusted certifiers, as defined above: TrustedCertifiers ::= SEQUENCE OF PrincipalName If method-type=2, the signature on one of the certificates in the chain cannot be verified. The format of the method-data field will be an ASN.1 encoding of the integer index of the certificate in question: CertificateIndex ::= INTEGER -- 0 = 1st certificate, -- 1 = 2nd certificate, etc If method-type=3, one of the certificates in the chain has been revoked. The format of the method-data field will be an ASN.1 encoding of the integer index of the certificate in question: CertificateIndex ::= INTEGER -- 0 = 1st certificate, -- 1 = 2nd certificate, etc If method-type=4, the KDC name or realm in the PKAuthenticator does not match the principal name of the KDC. There is no method-data field in this case. 3.3. Digital Signature Implementation of the changes in this section are OPTIONAL for compliance withpk-init.PKINIT. We offer this option with the warning that it requires the client to generate a random key; the client may not be able to guarantee the same level of randomness as the KDC. If the userregisteredregistered, or presents a certificate for, a digital signature key with the KDC instead of an encryption key, then a separate exchange must be used. The client sends a request for a TGT as usual, except that it (rather than the KDC) generates the random key that will be used to encrypt the KDC response. This key is sent to the KDC along with the request in a preauthenticationfield:field, encrypted with the KDC's public key: PA-PK-AS-SIGN ::= SEQUENCE { -- PA TYPE19 encSignedKeyPack16 encSignedRandomKeyPack [0]EncryptedDataEncryptedData, -- ofSignedKeyPacktype SignedRandomKeyPack -- using the key in encTmpKeyPack encTmpKeyPack [1] EncryptedData, -- of type TmpKeyPack -- using the KDC's public key userCert [2] SEQUENCE OF Certificate OPTIONAL -- the user's certificate chain }SignedKeyPackSignedRandomKeyPack ::= SEQUENCE {signedKeyrandomkeyPack [0]KeyPack, signedKeyAuthRandomKeyPack, randomkeyPackSig [1]PKAuthenticator, signedKeySig [2]Signature -- ofsignedKey.signedKeyAuthkeyPack -- using user'ssignatureprivate key }KeyPackRandomKeyPack ::= SEQUENCE { randomKey [0] EncryptionKey, -- will be used to encrypt replynoncerandomKeyAuth [1]INTEGER } where thePKAuthenticator -- nonceiscopied from AS-REQ } If therequest.KDC does not accept client-generated random keys as a matter of policy, then it sends back an error message of type KDC_ERR_KEY_TOO_WEAK. Otherwise, it extracts the random key as follows. Upon receipt of the PA-PK-AS-SIGN, the KDC decrypts then verifies the randomKey. It then replies as per RFC 1510, except that the reply is encrypted not with a password-derived user key, but with the randomKey sent in the request. Since the client already knows this key, there is no need to accompany the reply with an extra preauthentication field. The transited field of the ticket should specify the certification path as described in Section 3.2. 3.4. Retrieving the User's Private KeyFromfrom the KDC Implementation of the changes described in this sectionis RECOMMENDEDare OPTIONAL for compliance withpk-init.PKINIT. When the user's private key is not stored local to the user, he may choose to store the private key (normally encrypted using a password-derived key) on the KDC. In this case, the client makes a request as described above, except that instead of preauthenticating with his private key, he uses a symmetric key shared with the KDC. For simplicity's sake, this shared key is derived from the password- derived key used to encrypt the private key, in such a way that the KDC can authenticate the user with the shared key without being able to extract the private key. We provide this option to present the user with an alternative to storing the private key on local disk at each machine where he expects to authenticate himself usingpk-init.PKINIT. It should be noted that it replaces the added risk of long-term storage of the private key on possibly many workstations with the added risk of storing the private key on the KDC in a form vulnerable to brute-force attack.In orderDenote by K1 the symmetric key used toobtain aencrypt the private key. Then construct symmetric key K2 as follows: * Perform a hash on K1. * Truncate the digest to Length(K1) bytes. * Rectify parity in each byte (if necessary) to obtain K2. The KDC stores K2, the public key, and the encrypted private key. This key pair is designated as the "primary" key pair for that user. This primary key pair is the one used to perform initial authentication using the PA-PK-AS-REP preauthentication field. If he desires, he may also store additional key pairs on the KDC; these may be requested in addition to the primary. When the clientincludesrequests initial authentication using public key cryptography, it must then include in its request, instead of apreauthentication field withPA-PK-AS-REQ, theAS-REQ message:following preauthentication sequence: PA-PK-KEY-REQ ::= SEQUENCE { -- PA TYPE20 patimestamp17 signedPKAuth [0]KerberosTime OPTIONAL, -- used to address replay attacks. pausecSignedPKAuth, trustedCertifiers [1]INTEGERSEQUENCE OF PrincipalName OPTIONAL, --used to address replay attacks. nonce [2] INTEGER, -- bindsCAs that thereply to this request privkeyID [3]client trusts keyIDList [2] SEQUENCE OFKeyIDChecksum OPTIONAL --constructed as apayload is hash of--public key -- corresponding to--desired -- private key -- if absent, KDC will return all -- stored private keys }KeyIDSignedPKAuth ::= SEQUENCE {KeyIdentifierpkAuth [0]OCTET STRINGPKAuthenticator, pkAuthSig [1] Signature -- of pkAuth -- using the symmetric key K2 }The client may requestIf aspecifickeyIDList is present, the first identifier should indicate the primary private key. No public keyby sendingcertificate is required, since thecorresponding ID.KDC stores the public key along with the private key. Ifthis fieldthere isleft empty, thenno keyIDList, all the user's private keys are returned. Upon receipt, the KDC verifies the signature using K2. If the verification fails, the KDC sends back an error of type KDC_ERR_INVALID_SIG. If the signature verifies, but the requested keys are not found on the KDC, then the KDC sends back an error of type KDC_ERR_PREAUTH_FAILED. If all checks out, the KDC responds as described inthe above sections,Section 3.2, except thatan additional preauthentication field, containingin addition, theuser's private key, accompaniesKDC appends thereply:following preauthentication sequence: PA-PK-KEY-REP ::= SEQUENCE { -- PA TYPE21 nonce18 encKeyRep [0]INTEGER,EncryptedData --binds the reply toof type EncKeyReply -- using therequest KeyData [1]symmetric key K2 } EncKeyReply ::= SEQUENCE { keyPackList [0] SEQUENCE OF KeyPack, -- the first KeyPair is -- the primary key pair nonce [1] INTEGER -- binds reply to request -- must be identical to the nonce -- sent in the SignedAuthPack }KeyPairKeyPack ::= SEQUENCE {privKeyIDkeyID [0]OCTET STRING, -- corresponding to encPrivKeyChecksum, encPrivKey [1] OCTET STRING }3.4.1. Additional ProtectionUpon receipt ofRetrieved Private Keys We solicit discussion onthefollowing proposal: thatreply, the client extracts the encrypted private keys (and mayoptionally include in its request additional data to encryptstore them, at the client's option). The primary private key, whichis currently only protected bymust be theuser's password. One possibilityfirst private key in the keyPack SEQUENCE, isthatused to decrypt theclient might generate arandomstring of bits, encrypt it with the publickeyofin theKDC (asPA-PK-AS-REP; this key in turn is used to decrypt theSignedKeyPack, but with an ordinary OCTET STRINGmain reply as described inplace of an EncryptionKey),Section 3.2. 4. Logistics andinclude this withPolicy This section describes a way to define the policy on the use of PKINIT for each principal and request. The KDCthen XORs each returned key with this random bit string. (If the bit stringistoo short, the KDC couldnot required to contain a database record for users that use eitherreturn an error, or XORthereturned keyStandard Public Key Authentication or Public Key Authentication with arepetition ofDigital Signature. However, if these users are registered with thebit string.) In orderKDC, it is recommended that the database record for these users be modified tomake this work,include three additionalmeans of preauthentication need to be devisedflags inorder to prevent attackers from simply inserting their own bit string. One way to do this is to store a hash ofthepassword-derivedattributes field. The first flag, use_standard_pk_init, indicates that the user should authenticate using standard PKINIT as described in Section 3.2. The second flag, use_digital_signature, indicates that the user should authenticate using digital signature PKINIT as described in Section 3.3. The third flag, store_private_key, indicates that the user has stored his private key(theon the KDC and should retrieve it using the exchange described in Section 3.4. If oneused to encryptof theprivate key). This hashpreauthentication fields defined above is included in the request, thenusedthe KDC shall respond as described inturn to derive a second key (calledSections 3.2 through 3.4, ignoring thehash-key);aforementioned database flags. If more than one of thehash-keypreauthentication fields isused to encryptpresent, the KDC shall respond with anASN.1 structure containingerror of type KDC_ERR_PREAUTH_FAILED. In thegenerated bit string and a nonce valueevent thatbinds it tonone of therequest. Sincepreauthentication fields defined above are included in the request, the KDCpossesseschecks to see if any of thehash,above flags are set. If the first flag is set, then itcan generatesends back an error of type KDC_ERR_PREAUTH_REQUIRED indicating that a preauthentication field of type PA-PK-AS-REQ must be included in thehash-key and verify this (weaker) preauthentication, and yet cannot reproducerequest. Otherwise, if theprivate key itself, sincefirst flag is clear, but thehashsecond flag is set, then the KDC sends back an error of type KDC_ERR_PREAUTH_REQUIRED indicating that aone-way function. 4. Logistics and Policy Issues We solicit discussion on how clients and KDCs shouldpreauthentication field of type PA-PK-AS-SIGN must beconfiguredincluded inorder to determine which oftheoptions described above (if any) should be used. One possibility is to setrequest. Lastly, if theuser's database record to indicate that authenticationfirst two flags are clear, but the third flag isto use public key cryptography; this will not work, however, inset, then theeventKDC sends back an error of type KDC_ERR_PREAUTH_REQUIRED indicating that a preauthentication field of type PA-PK-KEY-REQ must be included in theclient needs to know before making the initialrequest. 5.Compatibility with One-Time Passcodes We solicit discussionDependence onhow the protocol changes proposed inTransport Mechanisms Certificate chains can potentially grow quite large and span several UDP packets; thisdraft will interact within turn increases theproposed use of one-time passcodes discussedprobability that a Kerberos message involving PKINIT extensions will be broken indraft-ietf-cat-kerberos-passwords-00.txt. 6. Strength of Cryptographic Schemestransit. In light ofrecent findings onthestrength of MD5 and DES,possibility that the Kerberos specification will allow TCP as a transport mechanism, we solicit discussion onwhich encryption types to incorporate intowhether using PKINIT should encourage theprotocol changes. 7.use of TCP. 6. Bibliography [1] J. Kohl, C. Neuman. The Kerberos Network Authentication Service (V5). Request forComments: 1510Comments 1510. [2] B.C. Neuman, Theodore Ts'o. Kerberos: An Authentication Service for Computer Networks, IEEE Communications, 32(9):33-38. September 1994. [3] A. Medvinsky, M. Hur. Addition of Kerberos Cipher Suites to Transport Layer Security (TLS). draft-ietf-tls-kerb-cipher-suites-00.txt [4] A. Medvinsky, M. Hur, B. Clifford Neuman. Public Key Utilizing Tickets for Application Servers (PKTAPP). draft-ietf-cat-pktapp-00.txt [5] M. Sirbu, J. Chuang. Distributed Authentication in Kerberos Using Public Key Cryptography. Symposium On Network and Distributed System Security, 1997. [6] B. Cox, J.D. Tygar, M. Sirbu. NetBill Security and Transaction Protocol. In Proceedings of the USENIX Workshop on Electronic Commerce, July 1995. [7] Alan O. Freier, Philip Karlton and Paul C. Kocher. The SSL Protocol, Version 3.0 - IETF Draft. [8] B.C. Neuman, Proxy-Based Authorization and Accounting for Distributed Systems. In Proceedings of the 13th International Conference on Distributed Computing Systems, May19931993. [9] ITU-T (formerly CCITT) Information technology - Open Systems Interconnection - The Directory: Authentication Framework Recommendation X.509 ISO/IEC 9594-88.7. Acknowledgements Sasha Medvinsky contributed several ideas to the protocol changes and specifications in this document. His additions have been most appreciated. Some of the ideas on which this proposal is based arose during discussions over several years between members of the SAAG, the IETF CAT working group, and the PSRG, regarding integration of Kerberos and SPX. Some ideas have also been drawn from the DASS system. These changes are by no means endorsed by these groups. This is an attempt to revive some of the goals of those groups, and this proposal approaches those goals primarily from the Kerberos perspective. Lastly, comments from groups working on similar ideas in DCE have been invaluable.9.8. Expiration Date This draft expiresSeptember 30,January 31, 1997.10.9. AuthorsClifford NeumanBrian Tung Clifford Neuman USC Information Sciences Institute 4676 Admiralty Way Suite 1001 Marina del Rey CA 90292-6695 Phone: +1 310 822 1511 E-mail:{bcn, brian}@isi.edu{brian, bcn}@isi.edu John Wray Digital Equipment Corporation 550 King Street, LKG2-2/Z7 Littleton, MA 01460 Phone: +1 508 486 5210 E-mail: wray@tuxedo.enet.dec.com Ari Medvinsky Matthew Hur CyberSafe Corporation 1605 NW Sammamish Road Suite 310 Issaquah WA 98027-5378 Phone: +1 206 391 6000 E-mail: {ari.medvinsky, matt.hur}@cybersafe.com Jonathan Trostle Novell Corporation Provo UT E-mail:jonathan.trostle@novell.comjtrostle@novell.com