Internet-Draft | DNS-Bound Identities Architecture | July 2023 |
Wilson, et al. | Expires 24 January 2024 | [Page] |
This architecture document defines terminology, interaction, and authentication patterns, related to the use of DANE DNS records for TLS client and messaging peer identity, within the context of existing object security and TLS-based protocols.¶
This note is to be removed before publishing as an RFC.¶
Discussion of this document takes place on the DANE Authentication for Network Clients Everywhere Working Group mailing list (dance@ietf.org), which is archived at https://mailarchive.ietf.org/arch/browse/dance/.¶
Source for this draft and an issue tracker can be found at https://github.com/ashdwilson/draft-dance-architecture.¶
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 24 January 2024.¶
Copyright (c) 2023 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.¶
A digital identity, in an abstract sense, possesses at least two features: an identifier (or name), and a means of proving ownership of the identifier. One of the most resilient mechanisms for tying an identifier to a method for proving ownership of the identifier is the digital certificate, issued by a well-run Certification Authority (CA). The CA acts as a mutually trusted third party, a root of trust.¶
Certificate-based identities are limited in scope by the issuing CA, or by the namespace of the application responsible for issuing or validating the identity.¶
An example of this limitation is well-illustrated by organizational Public Key Infrastructure (PKI). Organizational PKI is very often coupled with email and LDAP systems, and can be used for associating a human or machine identity identifier with a public key. Within the organization, authentication systems already agree on the roots of trust for validating entity certificates issued by organizational PKI.¶
Attempting to use organizational PKI outside the organization can be challenging. In order to authenticate a certificate, the certificate’s CA must be trusted. CAs have no way of controlling identifiers in certificates issued by other CAs. Consequently, trusting multiple CAs at the same time can enable entity identifier collisions. Asking an entity to trust your CA implies trust in anything that your CA signs. This is why many organizations operate a private CA, and require devices connecting to the organization’s networks or applications to possess certificates issued by the organization’s CA.¶
These limitations make the implementation and ongoing maintenance of a PKI costly, and have a chilling effect on the broader adoption of certificate-based IoT device identity. If certificate-based device identity were easier to manage, more broadly trusted, and less operationally expensive, more organizations and applications would be able to use it.¶
The lack of trust between PKI domains has lead to a lack of simple and globally scalable solutions for secure end-to-end inter-domain communication between entities, such as SIP phones, email and chat accounts and IoT devices belonging to different organizations.¶
DANCE seeks to make PKI-based IoT device identity universally discoverable, more broadly recognized, and less expensive to maintain by using DNS as the constraining namespace and lookup mechanism. DANCE builds on patterns established by the original DANE RFCs to enable client and sending entity certificate, public key, and trust anchor discovery. DANCE allows entities to possess a first-class identity, which, thanks to DNS, may be trusted by any application also trusting the DNS. A first-class identity is an application-independent identity.¶
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.¶
This section will be interesting to define. We have great examples of identity terminology in the https://datatracker.ietf.org/doc/html/draft-sarikaya-t2trg-sbootstrapping-06 document, but this document also admits that there is semantic drift on terms like "bootstrapping", depending on who's talking.¶
How to Dance with ENTITY: This architecture document delegates many details of how DANCE can be used with some specific protocol to a document with the names "How to Dance with entity".¶
Identity provisioning: This refers to the set of tasks required to securely provision an asymmetric key pair for the device, sign the certificate (if the public credential is not simply a raw public key), and publish the public key or certificate in DNS. Under some circumstances, these steps are not all performed by the same party or organization. A manufacturer may instantiate the key pair, and a systems integrator may be responsible for issuing (and publishing) the device certificate in DNS. In some circumstances, a manufacturer may also publish device identity records in DNS. In this case, the system integrator needs to perform network and application access configuration, since the identity already exists in DNS.¶
DANCEr: A DANCEr is the term which is used to describe a protocol that has been taught to use DANE, usually through a How to Dance with document.¶
Identity provisioning: This refers to the set of tasks required to securely provision an asymmetric key pair for the device, sign the certificate (if the public credential is not simply a raw public key), and publish the public key or certificate in DNS. Under some circumstances, these steps are not all performed by the same party or organization. A manufacturer may instantiate the key pair, and a systems integrator may be responsible for issuing (and publishing) the device certificate in DNS. In some circumstances, a manufacturer may also publish device identity records in DNS. In this case, the system integrator needs to perform network and application access configuration, since the identity already exists in DNS.¶
Security Domain: DNS-bound client identity allows the device to establish secure communications with any server with a DNS-bound identity, as long as a network path exists, the entity is configured to trust its communicating peer by name, and agreement on protocols can be achieved. The act of joining a security domain, in the past, may have involved certificate provisioning. Now, it can be as simple as using a manufacturer-provisioned identity to join the device to the network and application. [Is the security domain defined by how broadly the identity is recognized, or by the breadth of the application or network access policy?]¶
Client: This architecture document adopts the definition of "Client" from RFC 8446: "The endpoint initiating the TLS connection"¶
Server: This architecture document adopts the definition of "Server" from RFC 8446: "The endpoint that did not initiate the TLS connection"¶
Sending agent: Software which encodes and transmits messages. A sending agent may perform tasks related to generating cryptographic signatures and/or encrypting messages before transmission.¶
Receiving agent: Software which interprets and processes messages. A receiving agent may perform tasks related to the decryption of messages, and verification of message signatures.¶
Store-and-forward system: A message handling system in-path between the sending agent and the receiving agent.¶
Hardware supplier role: The entity which manufactures or assembles the physical device. In many situations, multiple hardware suppliers are involved in producing a given device. In some cases, the hardware supplier may provision an asymmetric key pair for the device and establish the device identity in DNS. In some cases, the hardware supplier may ship a device with software pre-installed.¶
Systems integrator: The party responsible for configuration and deployment of application components. In some cases, the systems integrator also installs the software onto the device, and may provision the device identity in DNS.¶
Consumer: The entity or organization which pays for the value provided by the application, and defines the success criteria for the output of the application.¶
Client/server communication patterns imply a direct connection between an entity which provides a service (the server), and an entity which initiates a connection to the server, called a client. A secure implementation of this pattern includes a TLS-protected session directly between the client and the server. A secure implementation may also include public key-based mutual authentication.¶
Extending DANE to include client identity allows the server to authenticate clients independent of the private PKI used to issue the client certificate. This reduces the complexity of managing the CA certificate collection, and mitigates the possibility of client identifier collision.¶
The extension also allows an application to find an application identity and set up a secure communication channel directly. This pattern can be used in mesh networking, IoT and in many communication protocols for multimedia sessions, chat and messaging.¶
Decoupled architecture, frequently incorporating store-and-forward systems, provides no direct connection between the producer and consumer of information. The producer (or sending agent) and consumer (or receiving agent) are typically separated by at least one layer of messaging-oriented middleware. The Messaging-oriented middleware components may act as a server for the purpose of establishing TLS sessions for the producer and consumer. This allows the assertion of identity between the middleware and sending agent, and the middleware and receiving agent. The trust relationship between the sending agent and receiving agent is based on the presumed trustworthiness of the middleware, unless an identity can be attached to the message itself, independent of transport and middleware components.¶
Within many existing store-and-forward protocols, certificates may be transmitted within the signed message itself. An example of this is S/MIME. Within IoT applications, we find that networks may be more constrained. Including certificates in message payloads can present an unnecessary overhead on constrained network links. Decoupled applications benefit from an out-of-band public key discovery mechanism, which may enable the retrieval of certificates only when needed, and sometimes using a less expensive network connection.¶
The client sets up a TLS connection to a server, attaches a client certificate with a subjectAltName dNSName indicating the name of the client. In the TLS connection the DANE-client-id extension is used to tell the server to use the certificate dNSName to find a DANE record including the public key of the certificate to be able to validate. If the server can validate the DNSSEC response, the server validates the certificate and completes the TLS connection setup.¶
Using DNS to convey certificate information for authenticating TLS clients gives a not-yet-authenticated client the ability to trigger a DNS lookup on the server side of the TLS connection. An opportunity for DDOS may exist when malicious clients can trigger arbitrary DNS lookups. For instance, an authoritative DNS server which has been configured to respond slowly, may cause a high concurrency of in-flight TLS authentication processes as well as open connections to upstream resolvers. This sort of attack (of type slowloris) could have a performance or availability impact on the TLS server.¶
Direct device-to-cloud communication is common in simple IoT applications. Authentication in these applications is usually accomplished using shared credentials like API keys, or using client certificates. Client certificate authentication frequently requires the consumer to maintain a CA. The CA trust anchor certificate is installed into the cloud application, and used in the TLS authentication process.¶
Using DANE for device identity can allow parties other than the implementer to operate the CA. A hardware manufacturer can provide a pre-established identity, with the certificate or public key already published in DNS. This makes PKI-based identity more approachable for small organizations which currently lack the resources to operate an organizational CA.¶
For the end-device onboarding in LoRaWAN, the "network server" and the "join server" [RFC8376] needs to establish mutual TLS authentication in order to exchange configuration parameters. Certificate Authority based mutual TLS authentication doesn't work in LoRaWAN due to the non availability of the CA trust store in the LoRaWAN network stack. Self-signed certificate based mutual-TLS authentication method is the alternative solution.¶
DANE based client identity allows the server to authenticate clients during the TLS handhsake. Thus, independent of the private PKI used to issue the client's self-signed certificate, the "network server" and the "join server" could be mutually authenticated.¶
https://datatracker.ietf.org/doc/html/draft-hong-t2trg-iot-edge-computing-01 may require devices to mutually authenticate in the field. A practical example of this pattern is the edge computing in construction use case [https://datatracker.ietf.org/doc/html/draft-hong-t2trg-iot-edge-computing-01#section-6.2.1]. Using traditional certificate-based identity, the sensor and the gateway may have certificates issued by the same organizational PKI. By using DANE for client and sender identity, the sensor and the gateway may have identities represented by the equipment supplier, and still be able to mutually authenticate. Important sensor measurements forwarded by the gateway to the cloud may bear the DNS name and signature of the originating sensor, and the cloud application may authenticate the measurement independent of the gateway which forwarded the information to the application.¶
End to end security in SIP is currently based on a classical S/MIME model which has not received much implementation. There are also SIP standards that build upon a trust chained anchored on the HTTP trust chain (SIP identity, STIR). WebRTC has a trust model between the web browser and the servers using TLS, but no inter-domain trust infrastructure. WebRTC lacks a definition of namespace to map to DNS, where SIP is based on an email-style addressing scheme. For WebRTC the application developer needs to define the name space and mapping to DNS.¶
By using DNS as a shared root of trust SIP and WebRTC end points can anchor the keys used for DTLS/SRTP media channel setup. In addition, SIP devices can establish security in the SIP messaging by using DNS to find the callee’s and the callers digital identity.¶
[I-D.johansson-sipcore-dane-sip](SIPDANE)¶
Issue #8¶
SSH servers have for some time been able to put their host keys into DNS using [RFC4255].¶
In many SSH server implementations the list of users that is authorized to login to an account is given by listing their public keys in a per-user file ("authorized_keys"). The file provides both authorization (who may login), and authentication (how they prove their identity). While this is an implementation detail, doing both in one place has been one of Secure Shell's major reason for success.¶
However, there are downsides to this: a user can not easily replace their key without visiting every host they are authorized to access and update the key on that host. Separation of authorization and authentication in this case would involve putting the key material in a third place, such as in a DANE record in DNS, and then listing only the DNS name in the authorization file:¶
The DNS record used could be TLSA, but it is possible with some protocol work that it could instead be SSHFP.¶
Network access refers to an authentication process by which a node is admitted securely onto network infrastructure. This is most common for wireless networks (wifi, 802.15.4), but has also routine been done for wired infrastructure using 802.1X mechanisms with EAPOL.¶
While there are EAP protocols that do not involve certificates, such as EAPSIM ([RFC4186], the use of symmetric key mechanisms as the "network key" is common in many homes. The use of certificate based mechanisms are expected to increase, due to challenges, such as Randomized and Changing MAC addresses (RCM), as described in [I-D.ietf-madinas-use-cases].¶
Enterprise EAP methods use a version of TLS to form a secure transport. Client and server-side certificates are used as credentials. EAP-TLS does not run over TCP, but rather over a reliable transport provided by EAP. To keep it simple the EAP "window" is always one, and there are various amounts of overhead that needs to be accounted for, and the EAP segment size is often noticeably smaller than the normal ethernet 1500 bytes. [RFC3748] does guarantee a minimum payload of 1020 bytes.¶
The client side certificates are often larger than 1500 bytes and can take two or three round trip times to transport from the supplicant to the authenticator. In worst case scenarios, which are common with eduroam [RFC7593], the EAP packets are transported some distance, easily across the entire planet. The authenticating system (the "authentication server" in EAP terms) is a system at the institute that issued the client side certificate, and so already has access to the entire client certificate. Transferring the client certificate is redundant. That is, the authenticator already has access to the entire certificate, but the client does not know this to tbe case, so it sends the entire certificate anyway.¶
The use of DANE Client IDs in TLS as described in [I-D.ietf-dance-tls-clientid] reduces the redundant bytes of certificate sent.¶
Supplicant: The entity which acts as the TLS client in the EAP-TLS authentication protocol. This term is defined in IEEE 802.1x. The suppliant acts as a client in the EAPOL (EAP over LAN) protocol, which is terminated at the authenticator (defined below).¶
Authentication server: The entity which acts as the TLS server in the EAP-TLS protocol. RADIUS (RFC 2865) is a frequently-used authentication server protocol.¶
Authenticator: The authenticator is the device which acts as a server the EAPOL (EAP over LAN) protocol, and is a client of the authentication server. The authenticator is responsible for passing EAP messages between the supplicant and the authentication server, and for ensuring that only authenticated supplicants gain access to the network.¶
https://datatracker.ietf.org/doc/html/rfc5216 is a mature and widely-used protocol for network authentication, for IoT and IT equipment. IEEE 802.1x defines the encapsulation of EAP over LAN access technologies, like IEEE 802.11 wireless and IEEE 802.3 ethernet. RADIUS is a protocol and server technology frequently used for supporting the server side of EAP-TLS authentication. Guidance for implementing RADIUS strongly encourages the use of a single common CA for all supplicants, to mitigate the possibility of identifier collisions across PKIs. The use of DANE for client identity can allow the safe use of any number of CAs. DNS acts as a constraining namespace, which prevents two unrelated CAs from issuing valid certificates bearing the same identifier. Certificates represented in DNS are valid, and all others are un-trusted.¶
The RADIUS protocol has a few recognized security problems. https://datatracker.ietf.org/doc/html/rfc6614 addresses the challenges related to the weakness of MD5-based authentication and confidentiality over untrusted networks by establishing a TLS session between the RADIUS protocol client and the RADIUS protocol server. RADIUS datagrams are then transmitted between the authenticator and authentication server within the TLS session. Updating the RADSEC standard to include the use of DANE for client and server identity would allow a RADIUS server and client to mutually authenticate, independent of the client’s and server’s issuing CAs. The benefit for this use case is that a hosted RADIUS service may mutually authenticate any client device, like a WiFi access point or ethernet switch, via RADSEC, without requiring the distribution of CA certificates.¶
Issue #13¶
JOSE and COSE provide formats for exchanging authenticated and encrypted structured data. JOSE defines the x5u field in [RFC7515], Section 4.1.5, and COSE defines a field of the same name in [I-D.ietf-cose-x509], Section 2.¶
However, this URL field points to where the key can be found. There is, as yet, no URI scheme which says that the key can be found via the DNS lookup itself.¶
In order to make use of x5u, a DANCEr would have to define a new URI scheme that explained how to get the right key from DNS. (Open Issue #22, about [RFC4501])¶
Issue #14¶
DNS clients should use DNS over TLS with trusted DNS resolvers to protect the identity of authenticating peers.¶
The integrity of public keys represented in DNS is most important. An altered public key can enable device impersonation, and the denial of existence for a valid identity can cause devices to become un-trusted by the network or the application. DNS records should be validated by the DNS stub resolver, using the DNSSEC protocol.¶
Compartmentalizing failure domains within an application is a well-known architectural best practice. Within the context of protecting DNS-based identities, this compartmentalization may manifest by hosting an identity zone on a DNS server which only supports the resource record types essential for representing device identities. This can prevent a compromised identity zone DNS server from presenting records essential for impersonating web sites under the organization’s domain name.¶
The naming pattern suggested in https://datatracker.ietf.org/doc/html/draft-huque-dane-client-cert includes an underscore label (_device) which also prevents the issuance of Web PKI-validating certificates in the event a DNS server hosting a client identity zone, which is capable of presenting A and AAAA records, is compromised.¶
One of the advantages of DNS is that it has more than fourty years of demonstrated scaling. It is a distributed database with a caching mechanism, and properly configured, it has proven resilient to many kinds of outages and attacks.¶
A key part of this availability is the proper use of Time To Live (TTL) values for resource records. A cache is allowed to hang on to the data for a set time, the TTL, after which it must do a new query to find out if the data has changed, or perhaps been deleted.¶
There is therefore a tension between resilience (higher TTL values), and agility (lower TTL values). A lower TTL value allows for revocation or replacement of a key to become known much faster. This allows for a more agile security posture.¶
On the other hand, lower TTLs cause the queries to occur more often, which may reveal more information to an observer about which devices are active. Encrypted transports like DoT/DoH/DoQ make these queries far less visible. In addition to the on-path observer being able to see more, the resolver logs also may be a source of information. It also allows for more opportunities for an attacker to affect the response time of the queries.¶
If the name of the identity proven by a certificate is directly or indirectly relatable to a person, privacy needs to be considered when forming the name of the DNS resource record for the certificate. When creating the name of the RR, effects of DNS zone walking and possible harvesting of identities in the DNS zone will have to be considered. The name of the RR may note have to have a direct relation to the name of the subject of the certificate.¶
Further work has do be done in this area.¶
AW: Consider if an approach like the email local-part hashing used in SMIMEA https://datatracker.ietf.org/doc/html/rfc8162 might work for this. If the identifier/local-part is hashed and the certificate association is a SHA256 or SHA512 hash, the effort required to walk a zone would not produce much useful information.¶
In the use case for IoT an implementation must be scalable to a large amount of devices. In many cases, identities may also be very short lived as revocation is performed by simply removing a DNS record. A zone will have to manage a large amount of changes as devices are constantly added and de-activated.¶
In these cases it is important to consider the architecture of the DNS zone and when possible use a tree-like structure with many subdomain parts, much like reverse DNS records or how telephone numbers are represented in the ENUM standard (RFC 6116).¶
If an authoritative resolver were configured to respond quite slowly (think slow loris [XXXrefereceXXX]), is it possible to cause a DoS on the TLS server via complete exhaustion of TCP connections?¶
The availability of a client identity zone is essential to permitting clients to authenticate. If the DNS infrastructure hosting client identities becomes unavailable, then the clients represented by that zone cannot be authenticated.¶
OEJ: We may want to have a discussion with the IETF DNS directorate. The scalability section above is from a discussion with one of the members...¶
One of the significant use cases is where the devices are identified by their manufacturer assigned identities. A significant savings was that enterprises would not have to run their own (private) PKI systems, sometimes even one system per device type. But, with this usage style for DANCE there is no private PKI to run, and as a result there is no change of ownership required. The device continues to use the manufacturer assigned identity.¶
The device OwnerOperator is therefore at risk if the device's manufacturer goes out of business, or decides that they no longer wish to manufacturer that device. Should that happen then the OwnerOperator of the device may be in trouble, and may find themselves having to replace the devices.¶
[RFC8995], Section 10.4 (BRSKI) deals with concerns about manufacturers influence on devices. In the case of BRSKI, the concern was limited to when the device ownership transfer was performed (the BRSKI transaction itself). There was no concern once the OwnerOperator had taken control over the device through an [RFC8366] voucher.¶
In the case of DANCE, the manufacturer is continuously involved with the day to day operation of the device.¶
If this is of concern, then the OwnerOperator should perform some kind of transfer of ownership, such as using DPP, [RFC8995](BRSKI), [RFC9140](EAP-NOOB), and others yet to come.¶
The DANCE method of using manufacturer assigned identities would therefore seem to be best used for devices which have a short lifetime: one much smaller than the uncertainty about the anticipated lifespan of the manufacturer. For instance, some kind of battery operated sensor which might be used in a large quantity at a construction site, and which can not be recharged.¶
This document has no IANA actions.¶
TODO acknowledge.¶