| Internet-Draft | RADIUS over TLS | October 2022 |
| Rieckers | Expires 27 April 2023 | [Page] |
This document specifies a transport profile for RADIUS using Transport Layer Security (TLS) over TCP as the transport protocol. This enables dynamic trust relationships between RADIUS servers.¶
This note is to be removed before publishing as an RFC.¶
Status information for this document may be found at https://datatracker.ietf.org/doc/draft-rieckers-radext-rfc6614bis/.¶
Discussion of this document takes place on the RADIUS EXTensions Working Group mailing list (mailto:radext@ietf.org), which is archived at https://mailarchive.ietf.org/arch/browse/radext/.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
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."¶
This Internet-Draft will expire on 27 April 2023.¶
Copyright (c) 2022 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 (https://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 Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
The RADIUS protocol [RFC2865] is a widely deployed authentication and authorization protocol. The supplementary RADIUS Accounting specification [RFC2866] provides accounting mechanisms, thus delivering a full Authentication, Authorization, and Accounting (AAA) solution. However, RADIUS is experiencing several shortcomings, such as its dependency on the unreliable transport protocol UDP and the lack of security for large parts of its packet payload. RADIUS security is based on the MD5 algorithm, which has been proven to be insecure.¶
The main focus of RADIUS over TLS is to provide a means to secure the communication between RADIUS/TCP peers using TLS. The most important use of this specification lies in roaming environments where RADIUS packets need to be transferred through different administrative domains and untrusted, potentially hostile networks. An example for a worldwide roaming environment that uses RADIUS over TLS to secure communication is "eduroam", see [eduroam].¶
There are multiple known attacks on the MD5 algorithm that is used in RADIUS to provide integrity protection and a limited confidentiality protection (see [MD5-attacks]). RADIUS over TLS wraps the entire RADIUS packet payload into a TLS stream and thus mitigates the risk of attacks on MD5.¶
Because of the static trust establishment between RADIUS peers (IP address and shared secret), the only scalable way of creating a massive deployment of RADIUS servers under the control of different administrative entities is to introduce some form of a proxy chain to route the access requests to their home server. This creates a lot of overhead in terms of possible points of failure, longer transmission times, as well as middleboxes through which authentication traffic flows. These middleboxes may learn privacy-relevant data while forwarding requests. The new features in RADIUS over TLS obsolete the use of IP addresses and shared MD5 secrets to identify other peers and thus allow the use of more contemporary trust models, e.g., checking a certificate by inspecting the issuer and other certificate properties.¶
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶
a RADIUS-over-TLS client or server¶
a RADIUS-over-TLS instance that initiates a new connection.¶
a RADIUS-over-TLS instance that listens on a RADIUS-over-TLS port and accepts new connections¶
a classic RADIUS transport over UDP as defined in [RFC2865]¶
This document is an Experimental RFC.¶
It is one out of several approaches to address known cryptographic weaknesses of the RADIUS protocol (see also Section 4). The specification does not fulfill all recommendations on a AAA transport profile as per [RFC3539]; in particular, by being based on TCP as a transport layer, it does not prevent head-of-line blocking issues.¶
If this specification is indeed selected for advancement to Standards Track, certificate verification options (Section 2.3, point 2) need to be refined.¶
Another experimental characteristic of this specification is the question of key management between RADIUS/TLS peers. RADIUS/UDP only allowed for manual key management, i.e., distribution of a shared secret between a client and a server. RADIUS/TLS allows manual distribution of long-term proofs of peer identity as well (by using TLS-PSK ciphersuites, or identifying clients by a certificate fingerprint), but as a new feature enables use of X.509 certificates in a PKIX infrastructure. It remains to be seen if one of these methods will prevail or if both will find their place in real-life deployments. The authors can imagine pre-shared keys (PSK) to be popular in small-scale deployments (Small Office, Home Office (SOHO) or isolated enterprise deployments) where scalability is not an issue and the deployment of a Certification Authority (CA) is considered too much of a hassle; however, the authors can also imagine large roaming consortia to make use of PKIX. Readers of this specification are encouraged to read the discussion of key management issues within [RFC6421] as well as [RFC4107].¶
It has yet to be decided whether this approach is to be chosen for Standards Track. One key aspect to judge whether the approach is usable on a large scale is by observing the uptake, usability, and operational behavior of the protocol in large-scale, real-life deployments.¶
An example for a worldwide roaming environment that uses RADIUS over TLS to secure communication is "eduroam", see [eduroam].¶
The default destination port number for RADIUS over TLS is TCP/2083. There are no separate ports for authentication, accounting, and dynamic authorization changes. The source port is arbitrary. See Section 3.4 for considerations regarding the separation of authentication, accounting, and dynamic authorization traffic.¶
RADIUS/TLS has no notion of negotiating TLS in an established connection. Servers and clients need to be preconfigured to use RADIUS/TLS for a given endpoint.¶
RADIUS/TLS nodes¶
after completing the TCP handshake, immediately negotiate TLS sessions according to [RFC5246] or its predecessor TLS 1.1. The following restrictions apply:¶
Peer authentication can be performed in any of the following three operation models:¶
TLS with X.509 certificates using PKIX trust models (this model is mandatory to implement):¶
In RADIUS/UDP, clients are uniquely identified by their IP address. Since the shared secret is associated with the origin IP address, if more than one RADIUS client is associated with the same IP address, then those clients also must utilize the same shared secret, a practice that is inherently insecure, as noted in [RFC5247].¶
RADIUS/TLS supports multiple operation modes.¶
In TLS-PSK operation, a client is uniquely identified by its TLS identifier.¶
In TLS-X.509 mode using fingerprints, a client is uniquely identified by the fingerprint of the presented client certificate.¶
In TLS-X.509 mode using PKIX trust models, a client is uniquely identified by the tuple (serial number of presented client certificate;Issuer).¶
Note well: having identified a connecting entity does not mean the server necessarily wants to communicate with that client. For example, if the Issuer is not in a trusted set of Issuers, the server may decline to perform RADIUS transactions with this client.¶
There are numerous trust models in PKIX environments, and it is beyond the scope of this document to define how a particular deployment determines whether a client is trustworthy. Implementations that want to support a wide variety of trust models should expose as many details of the presented certificate to the administrator as possible so that the trust model can be implemented by the administrator. As a suggestion, at least the following parameters of the X.509 client certificate should be exposed:¶
In TLS-PSK operation, at least the following parameters of the TLS connection should be exposed:¶
Authentication, Authorization, and Accounting packets are sent according to the following rules:¶
RADIUS/TLS clients transmit the same packet types on the connection they initiated as a RADIUS/UDP client would (see Section 3.4 (3) and (4)). For example, they send¶
and they receive¶
RADIUS/TLS servers transmit the same packet types on connections they have accepted as a RADIUS/UDP server would. For example, they send¶
and they receive¶
Due to the use of one single TCP port for all packet types, it is required that a RADIUS/TLS server signal which types of packets are supported on a server to a connecting peer. See also Section 3.4 for a discussion of signaling.¶
This section explains the design decisions that led to the rules defined in the previous section.¶
One mechanism to discover RADIUS-over-TLS peers dynamically via DNS is specified in [RFC7585]. While this mechanism is still under development and therefore is not a normative dependency of RADIUS/TLS, the use of dynamic discovery has potential future implications that are important to understand.¶
Readers of this document who are considering the deployment of DNS-based dynamic discovery are thus encouraged to read [RFC7585] and follow its future development.¶
If a RADIUS/TLS client is in possession of multiple certificates from different CAs (i.e., is part of multiple roaming consortia) and dynamic discovery is used, the discovery mechanism possibly does not yield sufficient information to identify the consortium uniquely (e.g., DNS discovery). Subsequently, the client may not know by itself which client certificate to use for the TLS handshake. Then, it is necessary for the server to signal to which consortium it belongs and which certificates it expects. If there is no risk of confusing multiple roaming consortia, providing this information in the handshake is not crucial.¶
If a RADIUS/TLS server is in possession of multiple certificates from different CAs (i.e., is part of multiple roaming consortia), it will need to select one of its certificates to present to the RADIUS/TLS client. If the client sends the Trusted CA Indication, this hint can make the server select the appropriate certificate and prevent a handshake failure. Omitting this indication makes it impossible to deterministically select the right certificate in this case. If there is no risk of confusing multiple roaming consortia, providing this indication in the handshake is not crucial.¶
Not all TLS ciphersuites in [RFC5246] are supported by available TLS tool kits, and licenses may be required in some cases. The existing implementations of RADIUS/TLS use OpenSSL as a cryptographic backend, which supports all of the ciphersuites listed in the rules in the normative section.¶
The TLS ciphersuite TLS_RSA_WITH_3DES_EDE_CBC_SHA is mandatory to implement according to [RFC4346]; thus, it has to be supported by RADIUS/TLS nodes.¶
The two other ciphersuites in the normative section are widely implemented in TLS tool kits and are considered good practice to implement.¶
After the TLS session is established, RADIUS packet payloads are exchanged over the encrypted TLS tunnel. In RADIUS/UDP, the packet size can be determined by evaluating the size of the datagram that arrived. Due to the stream nature of TCP and TLS, this does not hold true for RADIUS/TLS packet exchange. Instead, packet boundaries of RADIUS packets that arrive in the stream are calculated by evaluating the packet's Length field. Special care needs to be taken on the packet sender side that the value of the Length field is indeed correct before sending it over the TLS tunnel, because incorrect packet lengths can no longer be detected by a differing datagram boundary. See Section 2.6.4 of [RFC6613] for more details.¶
Within RADIUS/UDP [RFC2865], a shared secret is used for hiding attributes such as User-Password, as well as in computation of the Response Authenticator. In RADIUS accounting [RFC2866], the shared secret is used in computation of both the Request Authenticator and the Response Authenticator. Since TLS provides integrity protection and encryption sufficient to substitute for RADIUS application-layer security, it is not necessary to configure a RADIUS shared secret. The use of a fixed string for the obsolete shared secret eliminates possible node misconfigurations.¶
RADIUS/UDP [RFC2865] uses different UDP ports for authentication, accounting, and dynamic authorization changes. RADIUS/TLS allocates a single port for all RADIUS packet types. Nevertheless, in RADIUS/TLS, the notion of a client that sends authentication requests and processes replies associated with its users' sessions and the notion of a server that receives requests, processes them, and sends the appropriate replies is to be preserved. The normative rules about acceptable packet types for clients and servers mirror the packet flow behavior from RADIUS/UDP.¶
RADIUS/UDP [RFC2865] uses negative ICMP responses to a newly allocated UDP port to signal that a peer RADIUS server does not support the reception and processing of the packet types in [RFC5176]. These packet types are listed as to be received in RADIUS/TLS implementations. Note well: it is not required for an implementation to actually process these packet types; it is only required that the NAK be sent as defined above.¶
RADIUS/UDP [RFC2865] uses negative ICMP responses to a newly allocated UDP port to signal that a peer RADIUS server does not support the reception and processing of RADIUS Accounting packets. There is no RADIUS datagram to signal an Accounting NAK. Clients may be misconfigured for sending Accounting packets to a RADIUS/TLS server that does not wish to process their Accounting packet. To prevent a regression of detectability of this situation, the Accounting-Response + Error-Cause signaling was introduced.¶
The IETF defines multiple alternative transports to the classic UDP transport model as defined in [RFC2865], namely RADIUS over TCP [RFC6613] and the present document on RADIUS over TLS. The IETF also proposed RADIUS over Datagram Transport Layer Security (DTLS) [RFC7360].¶
RADIUS/TLS does not specify any inherent backward compatibility to RADIUS/UDP or cross compatibility to the other transports, i.e., an implementation that utilizes RADIUS/TLS only will not be able to receive or send RADIUS packet payloads over other transports. An implementation wishing to be backward or cross compatible (i.e., wishes to serve clients using other transports than RADIUS/TLS) will need to implement these other transports along with the RADIUS/TLS transport and be prepared to send and receive on all implemented transports, which is called a "multi-stack implementation".¶
If a given IP device is able to receive RADIUS payloads on multiple transports, this may or may not be the same instance of software, and it may or may not serve the same purposes. It is not safe to assume that both ports are interchangeable. In particular, it cannot be assumed that state is maintained for the packet payloads between the transports. Two such instances MUST be considered separate RADIUS server entities.¶
Since RADIUS/TLS is only a new transport profile for RADIUS, the compatibility of RADIUS/TLS - Diameter [RFC3588] and RADIUS/UDP [RFC2865] - Diameter [RFC3588] is identical. The considerations regarding payload size in [RFC6613] apply.¶
The computational resources to establish a TLS tunnel are significantly higher than simply sending mostly unencrypted UDP datagrams. Therefore, clients connecting to a RADIUS/TLS node will more easily create high load conditions and a malicious client might create a Denial-of-Service attack more easily.¶
Some TLS ciphersuites only provide integrity validation of their payload, and provide no encryption. This specification forbids the use of such ciphersuites. Since the RADIUS payload's shared secret is fixed to the well-known term "radsec" (see Section 2.3 (4)), failure to comply with this requirement will expose the entire datagram payload in plaintext, including User-Password, to intermediate IP nodes.¶
By virtue of being based on TCP, there are several generic attack vectors to slow down or prevent the TCP connection from being established; see [RFC4953] for details. If a TCP connection is not up when a packet is to be processed, it gets re-established, so such attacks in general lead only to a minor performance degradation (the time it takes to re-establish the connection). There is one notable exception where an attacker might create a bidding-down attack though. If peer communication between two devices is configured for both RADIUS/TLS (i.e., TLS security over TCP as a transport, shared secret fixed to "radsec") and RADIUS/UDP (i.e., shared secret security with a secret manually configured by the administrator), and the RADIUS/UDP transport is the failover option if the TLS session cannot be established, a bidding-down attack can occur if an adversary can maliciously close the TCP connection or prevent it from being established. Situations where clients are configured in such a way are likely to occur during a migration phase from RADIUS/UDP to RADIUS/TLS. By preventing the TLS session setup, the attacker can reduce the security of the packet payload from the selected TLS ciphersuite packet encryption to the classic MD5 per-attribute encryption. The situation should be avoided by disabling the weaker RADIUS/UDP transport as soon as the new RADIUS/TLS connection is established and tested. Disabling can happen at either the RADIUS client or server side:¶
RADIUS/TLS provides authentication and encryption between RADIUS peers. In the presence of proxies, the intermediate proxies can still inspect the individual RADIUS packets, i.e., "end-to-end" encryption is not provided. Where intermediate proxies are untrusted, it is desirable to use other RADIUS mechanisms to prevent RADIUS packet payload from inspection by such proxies. One common method to protect passwords is the use of the Extensible Authentication Protocol (EAP) and EAP methods that utilize TLS.¶
When using certificate fingerprints to identify RADIUS/TLS peers, any two certificates that produce the same hash value (i.e., that have a hash collision) will be considered the same client. Therefore, it is important to make sure that the hash function used is cryptographically uncompromised so that an attacker is very unlikely to be able to produce a hash collision with a certificate of his choice. While this specification mandates support for SHA-1, a later revision will likely demand support for more contemporary hash functions because as of issuance of this document, there are already attacks on SHA-1.¶
No new RADIUS attributes or packet codes are defined. IANA has updated the already assigned TCP port number 2083 to reflect the following: * Reference: [RFC6614] * Assignment Notes: The TCP port 2083 was already previously assigned by IANA for "RadSec", an early implementation of RADIUS/ TLS, prior to issuance of this RFC. This early implementation can be configured to be compatible to RADIUS/TLS as specified by the IETF. See RFC 6614, Appendix A for details.¶
RADIUS/TLS was first implemented as "RADSec" by Open Systems Consultants, Currumbin Waters, Australia, for their "Radiator" RADIUS server product (see [radsec-whitepaper]).¶
Funding and input for the development of this document was provided by the European Commission co-funded project "GEANT2" [TODO: outdated reference] and further feedback was provided by the TERENA Task Force on Mobility and Network Middleware [TODO: outdated reference].¶
Radiator implements the RadSec protocol for proxying requests with the <Authby RADSEC> and <ServerRADSEC> clauses in the Radiator configuration file.¶
The <AuthBy RADSEC> clause defines a RadSec client, and causes Radiator to send RADIUS requests to the configured RadSec server using the RadSec protocol.¶
The <ServerRADSEC> clause defines a RadSec server, and causes Radiator to listen on the configured port and address(es) for connections from <Authby RADSEC> clients. When an <Authby RADSEC> client connects to a <ServerRADSEC> server, the client sends RADIUS requests through the stream to the server. The server then handles the request in the same way as if the request had been received from a conventional UDP RADIUS client.¶
Radiator is compliant to RADIUS/TLS if the following options are used:¶
<AuthBy RADSEC>¶
<ServerRADSEC>¶
As of Radiator 3.15, the default shared secret for RadSec connections is configurable and defaults to "mysecret" (without quotes). For compliance with this document, this setting needs to be configured for the shared secret "radsec". The implementation uses TCP keepalive socket options, but does not send Status-Server packets. Once established, TLS connections are kept open throughout the server instance lifetime.¶
The RADIUS proxy named radsecproxy was written in order to allow use of RadSec in current RADIUS deployments. This is a generic proxy that supports any number and combination of clients and servers, supporting RADIUS over UDP and RadSec. The main idea is that it can be used on the same host as a non-RadSec client or server to ensure RadSec is used on the wire; however, as a generic proxy, it can be used in other circumstances as well.¶
The configuration file consists of client and server clauses, where there is one such clause for each client or server. In such a clause, one specifies either "type tls" or "type udp" for TLS or UDP transport. Versions prior to 1.6 used "mysecret" as a default shared secret for RADIUS/TLS; version 1.6 and onwards uses "radsec". For backwards compatibility with older versions, the secret can be changed (which makes the configuration not compliant with this specification).¶
In order to use TLS for clients and/or servers, one must also specify where to locate CA certificates, as well as certificate and key for the client or server. This is done in a TLS clause. There may be one or several TLS clauses. A client or server clause may reference a particular TLS clause, or just use a default one. One use for multiple TLS clauses may be to present one certificate to clients and another to servers.¶
If any RadSec (TLS) clients are configured, the proxy will, at startup, listen on port 2083, as assigned by IANA for the OSC RadSec implementation. An alternative port may be specified. When a client connects, the client certificate will be verified, including checking that the configured Fully Qualified Domain Name (FQDN) or IP address matches what is in the certificate. Requests coming from a RadSec client are treated exactly like requests from UDP clients.¶
At startup, the proxy will try to establish a TLS connection to each (if any) of the configured RadSec (TLS) servers. If it fails to connect to a server, it will retry regularly. There is some back-off where it will retry quickly at first, and with longer intervals later. If a connection to a server goes down, it will also start retrying regularly. When setting up the TLS connection, the server certificate will be verified, including checking that the configured FQDN or IP address matches what is in the certificate. Requests are sent to a RadSec server, just like they would be to a UDP server.¶
The proxy supports Status-Server messages. They are only sent to a server if enabled for that particular server. Status-Server requests are always responded to.¶
This RadSec implementation has been successfully tested together with Radiator. It is a freely available, open-source implementation. For source code and documentation, see [TODO outdated reference].¶