NOT PRODUCTION READY!

Data Integrity algorithms for RDF Datasets — Proof of concepts implementation

This is a proof-of-concept implementation (in Typescript) of the Verifiable Credentials Data Integrity (DI) specification of the W3C, adapted to RDF Datasets.

The DI specification is primarily aimed at Verifiable Credentials (i.e., JSON-LD based RDF Datasets to express credentials), but the approach is general enough for any kind of RDF Datasets. This implementation is an attempt to implement that.

It is proof-of-concepts, meaning that it is not production ready, and there are also minor discrepancies with the official specification. These are:

• Primarily at validation time, it doesn't do all the checks that the DI specification defines.
• In contrast with the DI specification, the Verification Method (ie, the public key) is expected to be be present in the input. In other words, the package does not retrieve the keys through a URL, it looks for the respective quads in the input dataset.
• The management of proof chains is a bit restricted compared to the specification: proof chains and sets are not mixed. In other words, either all proofs are part of a set or form a chain. The case when a previous proof reference points at a set of proofs has not been implemented.
• It has not (yet) been cross-checked with other DI implementations and, in general, should be much more thoroughly tested.

There is also a missing feature in the DI specification regarding the usage for Datasets in general. For a Verifiable Credential there is a natural "anchor" Resource used to "connect" the input dataset with its proof. This is generally not true (see, e.g. separate discussion) and, in this implementation, it must be provided explicitly to embed the proof into the dataset.

What the implementation proves, however, is that the DI specification may indeed be used, with minor adjustment on the "anchor", to provide proofs for an RDF Dataset in the form of separate "Proof Graphs", i.e., RDF Graphs containing a signature and its metadata that can be separately verified.

Some details

The steps for signature follow the "usual" approach for signing data, namely:

1. The input RDF Dataset is canonicalized, using the RDF Dataset Canonicalization.
2. The resulting canonical N-Quads are sorted, and hashed to yield a canonical hash of the Dataset. By default, the hash is done using SHA-256, except if the key is ECDSA with a P-384 curve (in which case SHA-384 is used).
3. A "proof option graph" is created, which includes crypto keys and some metadata. The key is stored in JWK or in Multikey formats: the former is used for RSA keys (for which no Multikey encoding has been specified) and the latter is used for ECDSA and EdDSA, as required by the respective cryptosuite specifications. This separate graph is also canonicalized, sorted, and hashed.
4. The the two hash values are concatenated (in the order of the proof option graph and the original dataset), and signed using a secret key. The signature value is stored as a base64url value following the Multibase format, and its value is added to the proof option graph (turning it into a "proof graph").

The package has API entries to generate, and validate, such proof graphs. The API gives the possibility to use a set of keys, yielding a set of proof graphs, which can also be validated in one step.

It is also possible, following the DI spec, to create "embedded" proofs, i.e., a new dataset, containing the original data, as well as the proof graph(s), each as a separate graph within an RDF dataset. If a separate "anchor" resource is provided, then this new dataset will also contain additional RDF triples connecting the anchor to the proof graphs.

When embedding proof graphs, this can come in two flavors: proof sets and proof chains. Semantically, a proof set is just a collection of proofs. A proof chain implies an order of proofs: the specification requires that the previous proof in the chain is also "signed over" by the current proof, i.e., the dataset is expanded to include, for the purpose of a signature, the previous proof graph in its entirety. The different behaviors is reflected in the API by the type of the crypto key collection: if it is a Typescript Array, it is considered to be a chain of keys (and of embedded proofs), and a set otherwise (e.g., if a Typescript Set is used)

The crypto layer for the package relies on the Web Crypto API specification, and its implementation in node.js or deno. The following crypto algorithms are available:

• EDDSA, a.k.a. Ed25519. It is not official in the WebCrypto specification, but implemented both in node.js and deno. See also the EdDSA cryptosuite specification.
• ECDSA. See also the ECDSA cryptosuite specification.
• RSA-PSS. No DI cryptosuite specification exists.
• RSASSA-PKCS1-v1_5. No DI cryptosuite specification exists.

Although not strictly necessary for this package, a separate method is available as part of the API to generate cryptography keys for one of these four algorithms. The first two algorithms are identified by cryptosuite names, namely eddsa-rdfc-2022 and ecdsa-rdfc-2019, respectively. The other two are non-standard, and are identified with the temporary cryptosuite names of rsa-pss-rdfc-ih and rsa-ssa-rdfc-ih, respectively. Note that there are no Multikey encodings for RSA keys, so the keys are stored in the proof graphs in JWK format as a literal with an rdf:JSON datatype.

The user facing APIs accept WebCrypto CryptoKeyPair instances. To facilitate the application developer, the interface exports a function to generate keys that are usable with this package (see generateKey) and there is also an extra function to convert JWK instances into CryptoKeys (see jwkToCrypto). (Converting from WebCrypto to JWK can be done directly with the crypto.subtle.exportKey function). Note also the companion package multikey-webcrypto that performs conversions between Multikey and JWK, respectively WebCrypto.

For more details, see:

• Separate document for the API
• A small RDF graph and its "verifiable" version with embedded proof graphs.

(Note that the API works on an RDF Data model level, and does not include a Turtle/TriG parser or serializer; that should be done separately.)

Examples

import * as rdf from '@rdfjs/types';
import { 
    KeyData, 
    generateProofGraph, verifyProofGraph, 
    embedProofGraph, verifyEmbeddedProofGraph,
    VerificationResult, 
    generateKey 
} from 'rdfjs-di';

const dataset: rdf.DatasetCore = generateYourDataset();

// Generate an ECDSA keyPair, with the P-256 curve
const keyPair: KeyData = await generateKey(Cryptosuites.ecdsa);

// 'proof' is a separate RDF graph with the keys, metadata, and the signature
const proof: rdf.DatasetCore = await generateProofGraph(dataset, keyPair)

// You can verify the information
const result: VerificationResult = await verifyProofGraph(dataset, proof);

// If everything is fine, this should be true
console.log(result.verified);

// The proof can also be embedded into the result
const embeddedProof: rdf.DatasetCore = await embedProofGraph(dataset, keyPair, anchorResource);

// This can be verified as before
const embeddedResult: VerificationResult = await verifyEmbeddedProofGraph(proof, anchor);

// There may be several keys, in which case an array of proofs are created:
const keypairs: KeyData[] = generateYourWebCryptoKeyPairs();

// The function interfaces are all overloaded, so the call format does not really change:
const proofs: rdf.DatasetCore[] = await generateProofGraph(dataset, keyPairs);

// etc.
Copy