Ray time: Validating token expiry on an unbounded local time interval

1.1. Terminology

This document uses terminology from the ACE framework ([ACE]) to describe the typical participants, even in general parts that may apply beyond ACE. In particular:¶

• The Resource Server (RS) is a constrained device, also sometimes called “the device”.¶

  In the context of this document, it has no regular access to the Internet, and its ability to maintain a concept of local time is limited by intermittent power supplies.¶

• The Authorization Server (AS) is a party trusted by the RS.¶

  In the context of this document, it is connected to the Internet. (When more private networks are used, the term Internet can be replaced with the partition of the private networks that contains the AS). It is considered the source of trusted time in this document, as distributed time sources are not relevant to its mechanisms.¶

• The Client is a device that has obtained time- and scope limited credentials for the RS from the AS.¶

  In the context of this document, it acquires these credentials ahead of time before it moves out of communication range of the Internet and into communication range of the RS.¶

  The client is untrusted in the sense that no action of the RS should allow it to exceed the permissions encoded in the credentials it holds.¶

1.2. Motivating example

Devices that use Internet style connectivity and security have been deployed far beyond the range of easily accessible connectivity, for example in remote forests and in space.¶

For concrete example consider a scientific instrument set up by students. The instrument is under the administrative control of the university, which is issuing time limited access tokens that allow configuring experiments and reading data, but not altering inventory designations or handing off ownership.¶

The instrument is set up in a far-off valley without cellular reception, and visited by students once per week to collect data, verify that it is still operational, and perform adjustments or repairs on site as needed. Any repairs require powering down the device completely.¶

When following the mechanisms and guidance of this document, the instrument will be usable by the students even after they had to power it down completely, and (once used by the new students) justifiedly reject the tokens of students that used the device earlier. It will stay usable to malicious students if they disassemble it and use it after their allowed time on the instrument has expired, but can only be operated for a total time of about the validity time span of those users’ tokens.¶

1.3. Security and availability

When a device has no current time, no means of acquiring time, and receives a time limited token to authorize an action, two outcomes are possible: the action is rejected or allowed. Rejection is the safe alternative in terms of security, but reduces the availability of the device, possibly to the point of not providing its function at all. Conversely, allowing the action maintains availability at the risk of violating security requirements.¶

The fundamental trade-off between security and availability is common around time limited access, and usually manifested in a choice of expiry times: A token with a validity of one week (or, equivalently, a revocation list with a maximum age of one week) will be usable across a day-long outage of the authorization infrastructure. On the downside, provides technical authorization to an administratively deauthorized user for several days after revocation of permissions.¶

This document concerns itself with getting the most security out of the case that favors availability after a period of no operational clock.¶

1.4. Threat model

Three main threats are considered here:¶

• Using old tokens.¶

  A user who has obtained a token may attempt to use it after the token’s life time (and the user’s permission) has expired.¶

• Manipulation of time sources.¶

  An attacker controlling a time source which the device trusts may make false statements on the current time. Indicating an earlier time can be used to extend the usability of old tokens. Indicating a later time can be used to deny service to users of valid tokens.¶

• Manipulation of the clock.¶

  An attacker with physical access (possibly only to the device’s environment) may cut power to the device’s clock.¶

Other attacks, such as physical attacks on the device, or exploiting weakenesses of the used cryptographic systems, are relevant to the system, but have no different impact on a system following this document than a system that requires an operational clock before it allows access.¶

1.5. Applicability constraints

The mechanisms of this document are intended only for scenarios matching the listed prerequisites. If either of them are violated, there are better solutions available, with suggestens listed along the condition.¶

• There is no network connectivity, not even for the client.¶

  If there is, the RS may use communication with the AS to obtain a current time value (as described in [ace-time]), perform token validation ([ACE]). Alternatively, it may use other Internet services such as [roughtime].¶

  Note that due to CoAP’s good proxy support, the connectivity between the RS and the AS does not need to be in a continuous IP network.¶

• The life time of tokens is limited.¶

  When tokens of unlimited life time are used, there is no need to employ the methods described in this document.¶

• Devices have no means of reliably maintaining time throughout their whole lifetime.¶

  If the power supply of the device’s clock is reliable enough to be sure to outlive the device, regular evaluation of time bounds can be done, possibly accounting for an upper bound of clock skew.¶

• Devices need to stay accessible after local power interruptions, that is, favor availability over security after a loss of local time.¶

  Otherwise, the device can issue an AS Request Creation Hint containing a cnonce (described in [ACE]). The client then needs to travel back to into communication distance to the AS, and obtain a token confirming that cnonce. Alternatively, one of the time synchronization mechanisms mentioned above can be used.¶

The applicability is also limited when it comes to requirements of a certain date being in the past, such as “nbf” (“not before”) claims described in [JWT]. These claims are notoriously hard to verify after a loss of time (as discussed in [imperfect]), but fortunately also rarely needed. The use of such claims is not fundamentally prohibitive of this document’s ray time mechanism, but its approach favoring availability over security is inapplicable to them (as a device would accept any nbf value after power-up).¶

While data sources such as DCT77 or GNSSs such as GPS are frequently offered as a solution, neither is reliable enough for use with security systems: DCT77 is effectively unencrypted, and GPS signals can be forged with cheap hardware [gps].¶

3.1. Open questions

• Should the decision on whether ray time is OK to use be encoded in the token or in the application?¶

  Options are to use “exp” and say that the action will tell whether or not ray time is good enough to evaluate the credentials’ validity (some actions may require more assurances), or to use an “exp-besteffort” or similar indicator to describe the whole token. The latter approach may be limiting in that then not all permissions can be expressed in a single token, but then again, that’s also true with different validity times.¶

  More generally (also for interval time), can a token indicate whether the clock skew upper bound is used in favor or against the user? [JWT] talks of “some small leeway”, but is not overly precise.¶

• If the device is used online, can it set a cnonce in its creation hints and the application will still try to use a cnonce-less token? Or should optional cnonces use a different key? (Setting a cnonce in online situations does have the advantage that the device can convert from ray time to interval time. But is that even useful, unless the tokens used have nbf claims?)¶