CAProck provides an implementation of a distributed authorization scheme with cryptographic guarantees. It is essentially a distributed Object-capability model using a cryptographic variant of a Power of Attorney model for ensuring it can be used in a fully distributed manner.
Centralized authorization schemes query a centralized service whether an authenticated entity (a subject) currently holds a privilege or permission (a predicate) for a given object1.
Distributing this information means serializing it into a distributable form. However, to verify that it has not been tampered with, one needs to additionally add a cryptographic signature. In order to verify the signature, an issuer identifier is also helpful.
Distributing such serialized tokens or capabilities is now possible, and the recipient can verify it. This implies that the signer and recipient do not need to communicate directly. Instead, the signer may pass the token to e.g. the subject of the capability, which then hands it to the recipient at some later date. The recipient can validate the token, and grant access to the object without the signer getting involved.
The use cases for this kind of authorization scheme always involve scenarios in which the issuer of a token cannot be reached at the time the authorization needs to be enforced.
In IoT device provisioning, it may be necessary to configure a device via some physical medium before it being able to register on a network. For example, the device may need to send a token to a network server to join a system of IoT nodes.
When collaborating on a shared resource, it is necessary for a service at which the resource lives to grant access to all collaborators, whether the owner is currently reachable on the network.
This can also be addressed in a centralized system, and is often the cited reason for centralization - but a distributed authorization scheme makes such centralization unnecessary.
Other than the basic principle of signing and verifying authorization triples, care must be taken to not grant authorization in perpetuity.
- Typically, tokens include a valid time range to limit damage that may be done.
- Explicit revocations may still be necessary for emergencies.
- Wildcards may be useful to limit the number of tokens issued, e.g. by issuing one for all objects belonging to an issuer, etc.
CAProck addresses these concenrs; see the linked specifications for details. The quick version is that CAProck can produce and consume an ordered stream of tokens, and act upon the final authorization state that results from processing the stream.
In a distributed systems, a stream of tokens may not be fully synchronized. CAPRock’s approach is to embrace eventual consistency, and produce results based on the synchronized sub-set of the stream – on the assumption that in many cases, that will be entirely sufficient.
In principle that makes attacks possible where monitoring a stream and suppressing the delivery of selected tokens will manipulate the authorization results. In practice, the solution to this is to encrypt the stream, either in transport or by encapsulating it in a vessel resource.
The data queried therefore conceptually represents an RDF triple, even though IRIs may not be used. However, it is theoretically possible to use IRIs for these fields. ↩︎