- JWT-Simple
A new JWT (JSON Web Tokens) implementation for Rust that focuses on simplicity, while avoiding common JWT security pitfalls.
jwt-simple
is unopinionated and supports all commonly deployed authentication and signature algorithms:
JWT algorithm name | Description |
---|---|
HS256 |
HMAC-SHA-256 |
HS384 |
HMAC-SHA-384 |
HS512 |
HMAC-SHA-512 |
RS256 |
RSA with PKCS#1v1.5 padding / SHA-256 |
RS384 |
RSA with PKCS#1v1.5 padding / SHA-384 |
RS512 |
RSA with PKCS#1v1.5 padding / SHA-512 |
PS256 |
RSA with PSS padding / SHA-256 |
PS384 |
RSA with PSS padding / SHA-384 |
PS512 |
RSA with PSS padding / SHA-512 |
ES256 |
ECDSA over p256 / SHA-256 |
ES384 |
ECDSA over p384 / SHA-384 |
ES256K |
ECDSA over secp256k1 / SHA-256 |
EdDSA |
Ed25519 |
jwt-simple
uses only pure Rust implementations, and can be compiled out of the box to WebAssembly/WASI. It is fully compatible with Fastly's Compute@Edge service.
Important: JWT's purpose is to verify that data has been created by a party knowing a secret key. It does not provide any kind of confidentiality: JWT data is simply encoded as BASE64, and is not encrypted.
cargo.toml
:
[dependencies]
jwt-simple = "0.10"
Rust:
use jwt_simple::prelude::*;
Errors are returned as jwt_simple::Error
values (alias for the Error
type of the thiserror
crate).
Authentication schemes use the same key for creating and verifying tokens. In other words, both parties need to ultimately trust each other, or else the verifier could also create arbitrary tokens.
Key creation:
use jwt_simple::prelude::*;
// create a new key for the `HS256` JWT algorithm
let key = HS256Key::generate();
A key can be exported as bytes with key.to_bytes()
, and restored with HS256Key::from_bytes()
.
Token creation:
/// create claims valid for 2 hours
let claims = Claims::create(Duration::from_hours(2));
let token = key.authenticate(claims)?;
-> Done!
let claims = key.verify_token::<NoCustomClaims>(&token, None)?;
-> Done! No additional steps required.
Key expiration, start time, authentication tags, etc. are automatically verified. The function fails with JWTError::InvalidAuthenticationTag
if the authentication tag is invalid for the given key.
The full set of claims can be inspected in the claims
object if necessary. NoCustomClaims
means that only the standard set of claims is used by the application, but application-defined claims can also be supported.
Extra verification steps can optionally be enabled via the ValidationOptions
structure:
let mut options = VerificationOptions::default();
// Accept tokens that will only be valid in the future
options.accept_future = true;
// accept tokens even if they have expired up to 15 minutes after the deadline
options.time_tolerance = Some(Duration::from_mins(15));
// reject tokens if they were issued more than 1 hour ago
options.max_validity = Some(Duration::from_hours(1));
// reject tokens if they don't include an issuer from that set
options.allowed_issuers = Some(HashSet::from_strings(&["example app"]));
// see the documentation for the full list of available options
let claims = key.verify_token::<NoCustomClaims>(&token, Some(options))?;
Note that allowed_issuers
and allowed_audiences
are not strings, but sets of strings (using the HashSet
type from the Rust standard library), as the application can allow multiple return values.
A signature requires a key pair: a secret key used to create tokens, and a public key, that can only verify them.
Always use a signature scheme if both parties do not ultimately trust each other, such as tokens exchanged between clients and API providers.
Key creation:
use jwt_simple::prelude::*;
// create a new key pair for the `ES256` JWT algorithm
let key_pair = ES256KeyPair::generate();
// a public key can be extracted from a key pair:
let public_key = key_pair.public_key();
use jwt_simple::prelude::*;
// create a new key pair for the `ES384` JWT algorithm
let key_pair = ES384KeyPair::generate();
// a public key can be extracted from a key pair:
let public_key = key_pair.public_key();
Keys can be exported as bytes for later reuse, and imported from bytes or, for RSA, from individual parameters, DER-encoded data or PEM-encoded data.
RSA key pair creation, using OpenSSL and PEM importation of the secret key:
openssl genrsa -out private.pem 2048
openssl rsa -in private.pem -outform PEM -pubout -out public.pem
let key_pair = RS384KeyPair::from_pem(private_pem_file_content)?;
let public_key = RS384PublicKey::from_pem(public_pem_file_content)?;
Token creation and verification work the same way as with HS*
algorithms, except that tokens are created with a key pair, and verified using the corresponding public key.
Token creation:
/// create claims valid for 2 hours
let claims = Claims::create(Duration::from_hours(2));
let token = key_pair.sign(claims)?;
Token verification:
let claims = public_key.verify_token::<NoCustomClaims>(&token, None)?;
Available verification options are identical to the ones used with symmetric algorithms.
Claim objects support all the standard claims by default, and they can be set directly or via convenient helpers:
let claims = Claims::create(Duration::from_hours(2)).
with_issuer("Example issuer").with_subject("Example subject");
But application-defined claims can also be defined. These simply have to be present in a serializable type (this requires the serde
crate):
#[derive(Serialize, Deserialize)]
struct MyAdditionalData {
user_is_admin: bool,
user_country: String,
}
let my_additional_data = MyAdditionalData {
user_is_admin: false,
user_country: "FR".to_string(),
};
Claim creation with custom data:
let claims = Claims::with_custom_claims(my_additional_data, Duration::from_secs(30));
Claim verification with custom data. Note the presence of the custom data type:
let claims = public_key.verify_token::<MyAdditionalData>(&token, None)?;
let user_is_admin = claims.custom.user_is_admin;
Properties such as the key identifier can be useful prior to tag or signature verification in order to pick the right key out of a set.
let metadata = Token::decode_metadata(&token)?;
let key_id = metadata.key_id();
let algorithm = metadata.algorithm();
// all other standard properties are also accessible
IMPORTANT: neither the key ID nor the algorithm can be trusted. This is an unfixable design flaw of the JWT standard.
As a result, algorithm
should be used only for debugging purposes, and never to select a key type.
Similarly, key_id
should be used only to select a key in a set of keys made for the same algorithm.
At the bare minimum, verification using HS*
must be prohibited if a signature scheme was originally used to create the token.
Key identifiers indicate to verifiers what public key (or shared key) should be used for verification. They can be attached at any time to existing shared keys, key pairs and public keys:
let public_key_with_id = public_key.with_key_id(&"unique key identifier");
Instead of delegating this to applications, jwt-simple
can also create such an identifier for an existing key:
let key_id = public_key.create_key_id();
This creates an text-encoded identifier for the key, attaches it, and returns it.
If an identifier has been attached to a shared key or a key pair, tokens created with them will include it.
jwt-simple
includes mechanisms to mitigate replay attacks:
- Nonces can be created and attached to new tokens using the
create_nonce()
claim function. The verification procedure can later reject any token that doesn't include the expected nonce (required_nonce
verification option). - The verification procedure can reject tokens created too long ago, no matter what their expiration date is. This prevents tokens from malicious (or compromised) signers from being used for too long.
- The verification procedure can reject tokens created before a date. For a given user, the date of the last successful authentication can be stored in a database, and used later along with this option to reject older (replayed) tokens.
The development code includes a cwt
cargo feature that enables experimental parsing and validation of CWT tokens.
Please note that CWT doesn't support custom claims. The required identifiers haven't been standardized yet.
Also, the existing Rust crates for JSON and CBOR deserialization are not safe. An untrusted party can send a serialized object that requires a lot of memory and CPU to deserialize. Band-aids have been added for JSON, but with the current Rust tooling, it would be tricky to do for CBOR.
As a mitigation, we highly recommend rejecting tokens that would be too large in the context of your application. That can be done by with the max_token_length
verification option.
This crate is not an endorsement of JWT. JWT is an awful design, and one of the many examples that "but this is a standard" doesn't necessarily mean that it is good.
I would highly recommend PASETO or Biscuit instead if you control both token creation and verification.
However, JWT is still widely used in the industry, and remains absolutely mandatory to communicate with popular APIs.
This crate was designed to:
- Be simple to use, even to people who are new to Rust
- Avoid common JWT API pitfalls
- Support features widely in use. I'd love to limit the algorithm choices to Ed25519, but other methods are required to connect to existing APIs, so just provide them (with the exception of the
None
signature method for obvious reasons). - Minimize code complexity and external dependencies
- Automatically perform common tasks to prevent misuse. Signature verification and claims validation happen automatically instead of relying on applications.
- Still allow power users to access everything JWT tokens include if they really need to
- Be as portable as possible by using only Rust implementations of cryptographic primitives
- Have no dependency on OpenSSL
- Work out of the box in a WebAssembly environment, so that it can be used in function-as-a-service platforms.