Internet Engineering Task Force (IETF) D. Harkins Request for Comments: 8146 HP Enterprise Updates: 5931 April 2017 Category: Informational ISSN: 2070-1721
Adding Support for Salted Password Databases to EAP-pwd
EAP-pwd is an Extensible Authentication Protocol (EAP) method that utilizes a shared password for authentication using a technique that is resistant to dictionary attacks. It includes support for raw keys and double hashing of a password in the style of Microsoft Challenge Handshake Authentication Protocol version 2 (MSCHAPv2), but it does not include support for salted passwords. There are many existing databases of salted passwords, and it is desirable to allow their use with EAP-pwd.
Status of This Memo
This document is not an Internet Standards Track specification; it is published for informational purposes.
This document is a product of the Internet Engineering Task Force (IETF). It represents the consensus of the IETF community. It has received public review and has been approved for publication by the Internet Engineering Steering Group (IESG). Not all documents approved by the IESG are a candidate for any level of Internet Standard; see Section 2 of RFC 7841.
Copyright (c) 2017 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
Databases of stored passwords present an attractive target for attack -- get access to the database, learn the passwords. To confound such attacks, a random "salt" was hashed with the password and the resulting digest stored, along with the salt, instead of the raw password. This has the effect of randomizing the password; even if two, distinct users have chosen the same password, the stored, and salted, password will be different. It also requires an adversary who has compromised the security of the stored database to launch a dictionary attack per entry to recover passwords.
Dictionary attacks, especially using custom hardware, represent real- world attacks and merely salting a password is insufficient to protect a password database. To address these attacks, a sequential memory hard function, such as described in [RFC7914], is used.
While salting a password database is not sufficient to deal with many real-world attacks, the historic popularity of password salting means there are a large number of such databases deployed, and EAP-pwd needs to be able to support them. In addition, EAP-pwd needs to be able to support databases using more modern sequential memory hard functions for protection.
EAP-pwd imposes an additional security requirement on a database of salted passwords that otherwise would not exist, see Section 4.
EAP-pwd is based on the "dragonfly" Password-Authenticated Key Exchange (PAKE) -- see [RFC7664]. This is a balanced PAKE and requires that each party to the protocol obtain an identical representation of a processed password (see Section 4). Therefore, salting of a password is treated as an additional password preprocessing technique of EAP-pwd. The salt and digest to use are conveyed to the peer by the server, and the password is processed prior to fixing the password element (see Section 2.8.3 of [RFC5931]).
Harkins Informational [Page 3]
RFC 8146 NaCled EAP-pwd April 2017
This memo defines eight (8) new password preprocessing techniques for EAP-pwd:
o 0x03: a random salt with SHA-1
o 0x04: a random salt with SHA-256
o 0x05: a random salt with SHA-512
o 0x06: UNIX crypt()
o 0x07: scrypt
o 0x08: PBKDF2 with SHA-256
o 0x09: PBKDF2 with SHA-512
o 0x0A: SASLprep then a random salt with SHA-1
o 0x0B: SASLprep then a random salt with SHA-256
o 0x0C: SASLprep then a random salt with SHA-512
o 0x0D: SASLprep then UNIX crypt()
o 0x0E: OpaqueString then scrypt
o 0x0F: OpaqueString then PBKDF2 with SHA-256
o 0x10: OpaqueString then PBKDF2 with SHA-512
When passing salt, the size of the salt SHOULD be at least as long as the message digest of the hash algorithm used. There is no guarantee that deployed salted databases have followed this rule, and in the interest of interoperability, an EAP peer SHOULD NOT abort an EAP-pwd exchange if the length of the salt conveyed during the exchange is less than the message digest of the indicated hash algorithm.
UNIX crypt() ([CRY]), scrypt ([RFC7914]), and PBKDF2 ([RFC8018]) impose additional formatting requirements on the passed salt. See below.
Plain salting techniques using [SHS] are included for support of existing databases. scrypt and PBKDF2 techniques are RECOMMENDED for new password database deployments.
SASLprep has been deprecated, but databases treated with SASLprep exist; it is necessary to provide code points for them. When using
Harkins Informational [Page 4]
RFC 8146 NaCled EAP-pwd April 2017
SASLprep, a password SHALL be considered a "stored string" per [RFC3454]; therefore, unassigned code points are prohibited. The output of SASLprep SHALL be the binary representation of the processed UTF-8 character string. Prohibited output and unassigned code points encountered in SASLprep preprocessing SHALL cause a failure of preprocessing, and the output SHALL NOT be used with EAP- pwd.
When performing one of the preprocessing techniques (0x0E-0x10), the password SHALL be a UTF-8 string and SHALL be preprocessed by applying the Preparation and Enforcement steps of the OpaqueString profile in [RFC7613] to the password. The output of OpaqueString, also a UTF-8 string, becomes the EAP-pwd password and SHALL be hashed with the indicated algorithm.
There is a large number of deployed password databases that use salting and hashing in the style of [RFC7616], but these deployments require a nonce contribution by the client (as well as the server), and EAP-pwd does not have the capability to provide that information.
For both parties to derive the same salted password, there needs to be a canonical method of salting a password. When using EAP-pwd, a password SHALL be salted by hashing the password followed by the salt using the designated hash function:
salted-password = Hash(password | salt)
The server stores the salted-password, and the salt, in its database and the client derives the salted password on the fly.
Different algorithms are supported with the UNIX crypt() function. The particular algorithm used is indicated by prepending an encoding of "setting" to the passed salt. The specific algorithm used is opaque to EAP-pwd as the entire salt, including the encoded "setting", is passed as an opaque string for interpretation by crypt(). The salted password used for EAP-pwd SHALL be the output of crypt():
salted-password = crypt(password, salt)
The server stores the salted-password, and the encoded algorithm plus salt, in its database and the client derives the salted-password on- the-fly.
Harkins Informational [Page 5]
RFC 8146 NaCled EAP-pwd April 2017
If the server indicates a crypt() algorithm that is unsupported by the client, the exchange fails and the client MUST terminate the connection.
These parameters are encoded into the "salt" field of the modified EAP-pwd message. Parameters r and dkLen SHALL be 16-bit integers in network order. The other parameters SHALL each be 32-bit integers in network order. The "salt" field that gets transmitted in EAP-pwd SHALL therefore be:
N || r || p || dkLen || salt
where || represents concatenation.
The value of N represents the exponent taken to the power of two in order to determine the CPU/Memory cost of scrypt -- i.e., the value is 2^N. Per [RFC7914], the resulting CPU/Memory cost value SHALL be less than 2^(128 * r / 8), and the value p SHALL be less than or equal to ((2^32 - 1) * 32) / (128 * r).
Note: EAP-pwd uses the salted password directly as the authentication credential and will hash it with a counter in order to obtain a secret element in a finite field. Therefore, it makes little sense to use dkLen greater than the length of the digest produced by the underlying hash function, but the capability is provided to do so anyway.
These parameters are encoded into the "salt" field of the modified EAP-pwd message. The parameters SHALL be 16-bit integers in network order. The "salt" field that gets transmitted in EAP-pwd SHALL therefore be:
c || dkLen || salt
where || represents concatenation.
Note: EAP-pwd uses the salted password directly as the authentication credential and will hash it with a counter in order to obtain a secret element in a finite field. Therefore, it makes little sense to use a dkLen greater than the length of the digest produced by the underlying hash function, but the capability is provided to do so anyway.
Like all EAP methods, EAP-pwd is server initiated, and the initial identity supplied by the client is not useful for authentication purposes. Because of this, the server is required to indicate its intentions, including the password preprocessing it wishes to use, before it knows the true identity of the client. This prevents the server from supporting multiple salt digests simultaneously in a single password database. To support multiple salt digests simultaneously, it is necessary to maintain multiple password databases and use the routable portion of the client identity to select one when initiating EAP-pwd.
The server uses the EAP-pwd-ID/Request to indicate the password preprocessing technique. The client indicates its acceptance of the password preprocessing technique and identifies itself in the EAP- pwd-ID/Response. If the client does not accept any of the offered preprocessing techniques, it SHALL terminate the exchange. Upon receipt of the EAP-pwd-ID/Response, the server knows the identity of the client and can look up the client's salted password and the salt from the database. The server adds the length of the salt and the salt itself to the EAP-pwd-Commit/Request message (see Section 2.7).
The server can fix the password element (Section 2.8.3 of [RFC5931]) as soon as the salted password has been looked up in the database. The client, though, is required to wait until receipt of the server's EAP-pwd-Commit/Request before it begins fixing the password element.
When a salted password preprocessing technique is agreed upon during the EAP-pwd-ID exchange, the EAP-pwd-Commit payload is modified to include the salt and salt length (see Figure 1). The server passes the salt and salt length in the EAP-pwd-Commit/Request; the client's EAP-pwd-Commit/Response is unchanged, and it MUST NOT echo the salt length and salt in its EAP-pwd-Commit/Response.
The "salt-len" SHALL be non-zero, and it indicates the length, in octets, of the salt that follows. The "Salt" SHALL be a binary string. The "Element" and "Scalar" are encoded according to Section 3.3 of [RFC5931].
Note: when a non-salted password preprocessing method is used, for example, any of the methods from [RFC5931], the EAP-pwd-Commit payload MUST NOT be modified to include the salt and salt length.
EAP-pwd requires each side to produce an identical representation of the (processed) password before the password element can be fixed. This symmetry undercuts one of the benefits to salting a password database because the salted password from a compromised database can be used directly to impersonate the EAP-pwd client -- since the plaintext password need not be recovered, no dictionary attack is needed. While the immediate effect of such a compromise would be compromise of the server, the per-user salt would still prevent the adversary from recovering the password, barring a successful dictionary attack, to use for other purposes.
Salted password databases used with EAP-pwd MUST be afforded the same level of protection as databases of plaintext passwords.
Hashing a password with a salt increases the work factor for an attacker to obtain the cleartext password, but dedicated hardware makes this increased work factor increasingly negligible in real- world scenarios. Cleartext password databases SHOULD be protected with a scheme that uses a sequential memory hard function such as [RFC7914].
EAP-pwd sends the salt in the clear. If EAP-pwd is not tunneled in another, encrypting, EAP method, an adversary that can observe traffic from server to authenticator or from authenticator to client would learn the salt used for a particular user. While knowledge of a salt by an adversary may be of a somewhat dubious nature (pre-image resistance of the hash function used will protect the client's password and, as noted above, the database of salted passwords must still be protected from disclosure), it does represent potential additional meta-data in the hands of a untrusted third party.