noble-ed25519
Fastest JS implementation of ed25519,
an elliptic curve that could be used for EDDSA signature scheme and X25519 ECDH key agreement.
Conforms to RFC7748, RFC8032 and ZIP215. Includes support for ristretto255: a technique for constructing prime order elliptic curve groups with non-malleable encodings.
Audited by an independent security firm: no vulnerabilities have been found. Check out the online demo.
This library belongs to noble crypto
noble-crypto — high-security, easily auditable set of contained cryptographic libraries and tools.
- No dependencies, one small file
- Easily auditable TypeScript/JS code
- Supported in all major browsers and stable node.js versions
- All releases are signed with PGP keys
- Check out homepage & all libraries:
secp256k1,
ed25519,
bls12-381,
hashes
Usage
Use NPM in node.js / browser, or include single file from
GitHub's releases page:
npm install @noble/ed25519
import * as ed from '@noble/ed25519';
(async () => {
const privateKey = ed.utils.randomPrivateKey();
const message = Uint8Array.from([0xab, 0xbc, 0xcd, 0xde]);
const publicKey = await ed.getPublicKey(privateKey);
const signature = await ed.sign(message, privateKey);
const isValid = await ed.verify(signature, message, publicKey);
})();
To use the module with Deno,
you will need import map:
deno run --import-map=imports.json app.ts
- app.ts:
import * as ed from "https://deno.land/x/ed25519/mod.ts";
- imports.json:
{"imports": {"crypto": "https://deno.land/std@0.125.0/node/crypto.ts"}}
API
getPublicKey(privateKey)
function getPublicKey(privateKey: Uint8Array | string | bigint): Promise<Uint8Array>;
privateKey: Uint8Array | string | bigint
will be used to generate public key. If you want to pass bigints,
ensure they are Big-Endian.- Returns
Promise<Uint8Array>
. Uses promises, because ed25519 uses SHA internally; and we're using built-in browser window.crypto
, which returns Promise
.
To generate ed25519 public key:
- private key is hashed with sha512, then first 32 bytes are taken from the hash
- 3 least significant bits of the first byte are cleared
- Use
Point.fromPrivateKey(privateKey)
if you want Point
instance instead - Use
Point.fromHex(publicKey)
if you want to convert hex / bytes into Point.
It will use decompression algorithm 5.1.3 of RFC 8032. - Use
utils.getExtendedPublicKey
if you need full SHA512 hash of seed
sign(message, privateKey)
function sign(message: Uint8Array | string, privateKey: Uint8Array | string): Promise<Uint8Array>;
message: Uint8Array | string
- message (not message hash) which would be signedprivateKey: Uint8Array | string
- private key which will sign the hash- Returns EdDSA signature. You can consume it with
Signature.fromHex()
method:
Signature.fromHex(ed25519.sign(hash, privateKey))
verify(signature, message, publicKey)
function verify(
signature: Uint8Array | string | Signature,
message: Uint8Array | string,
publicKey: Uint8Array | string | Point
): Promise<boolean>
signature: Uint8Array | string | Signature
- returned by the sign
functionmessage: Uint8Array | string
- message that needs to be verifiedpublicKey: Uint8Array | string | Point
- e.g. that was generated from privateKey
by getPublicKey
- Returns
Promise<boolean>
Verifies signature. Compatible with ZIP215, accepts:
0 <= sig.R/publicKey < 2**256
(can be >= curve.P
aka non-canonical encoding)0 <= sig.s < l
Not compatible with RFC8032 because rfc encorces canonical encoding of R/publicKey. There is no security risk in ZIP behavior, and there is no effect on honestly generated signatures. For additional info about verification strictness, check out It’s 255:19AM.
getSharedSecret(privateKey, publicKey)
function getSharedSecret(privateKey: Uint8Array | string | bigint, publicKey: Uint8Array | string): Promise<Uint8Array>;
Converts ed25519 private / public keys to Curve25519 and calculates
Elliptic Curve Diffie Hellman (ECDH) with X25519.
Conforms to RFC7748.
X25519 and curve25519
const pub = ed25519.curve25519.scalarMultBase(privateKey);
const shared = ed25519.curve25519.scalarMult(privateKeyA, publicKeyB);
The library includes namespace curve25519
that you could use to calculate
Curve25519 keys. It uses Montgomery Ladder specified in RFC7748.
You cannot use ed25519 keys, because they are hashed with sha512. However, you can use
Point#toX25519()
method on ed25519 public keys. See implementation of ed25519.getSharedSecret
for details.
Ristretto255
Each Point in ed25519 has 8 different equivalent points. This can be a great pain for some algorithms
e.g. ring signatures. In Tor, Ed25519 public key malleability would mean that every v3 onion service
has eight different addresses, causing mismatches with user expectations and potential gotchas for service operators.
Fixing this required expensive runtime checks in the v3 onion services protocol, requiring a full scalar multiplication,
point compression, and equality check. This check must be called in several places to validate that the onion service's
key does not contain a small torsion component.
No matter which one of these 8 equivalent points you give the Ristretto algorithm,
it will give you exactly the same one. The other 7 points are no longer representable. Two caveats:
- Always use
RistrettoPoint.fromHex()
and RistrettoPoint#toHex()
- Never mix
ExtendedPoint
& RistrettoPoint
: ristretto is not a subgroup of ed25519.
ExtendedPoint
you are mixing with, may not be the representative for the set of possible points.
import { RistrettoPoint } from '@noble/ed25519';
RistrettoPoint.fromHex(hex: Uint8Array | string): RistrettoPoint;
RistrettoPoint#toHex(): Uint8Array;
RistrettoPoint.hashToCurve(hash: Uint8Array | string): RistrettoPoint;
It extends Mike Hamburg's Decaf approach to cofactor elimination to support cofactor-8 curves such as Curve25519.
In particular, this allows an existing Curve25519 library to implement a prime-order group with only a thin abstraction layer, and makes it possible for systems using Ed25519 signatures to be safely extended with zero-knowledge protocols, with no additional cryptographic assumptions and minimal code changes.
For more information on the topic, check out:
Utilities
We provide a bunch of useful utils and expose some internal classes.
utils.randomPrivateKey();
utils.sha512(message: Uint8Array): Promise<Uint8Array>;
utils.mod(number: bigint, modulo = CURVE.P): bigint;
utils.invert(number: bigint, modulo = CURVE.P): bigint;
utils.bytesToHex(bytes: Uint8Array): string;
utils.getExtendedPublicKey(privateKey);
utils.precompute(W = 8, point = Point.BASE)
class Point {
constructor(x: bigint, y: bigint);
static fromHex(hash: string);
static fromPrivateKey(privateKey: string | Uint8Array);
toX25519(): Uint8Array;
toRawBytes(): Uint8Array;
toHex(): string;
isTorsionFree(): boolean;
equals(other: Point): boolean;
negate(): Point;
add(other: Point): Point;
subtract(other: Point): Point;
multiply(scalar: bigint): Point;
}
class ExtendedPoint {
constructor(x: bigint, y: bigint, z: bigint, t: bigint);
static fromAffine(point: Point): ExtendedPoint;
toAffine(): Point;
equals(other: ExtendedPoint): boolean;
add(other: ExtendedPoint): ExtendedPoint;
subtract(other: ExtendedPoint): ExtendedPoint;
multiply(scalar: bigint): ExtendedPoint;
multiplyUnsafe(scalar: bigint): ExtendedPoint;
}
class RistrettoPoint {
static hashToCurve(hex: Hex): RistrettoPoint;
static fromHex(hex: Hex): RistrettoPoint;
toRawBytes(): Uint8Array;
toHex(): string;
equals(other: RistrettoPoint): boolean;
add(other: RistrettoPoint): RistrettoPoint;
subtract(other: RistrettoPoint): RistrettoPoint;
multiply(scalar: number | bigint): RistrettoPoint;
}
class Signature {
constructor(r: bigint, s: bigint);
static fromHex(hex: Hex): Signature;
toRawBytes(): Uint8Array;
toHex(): string;
}
ed25519.CURVE.P
ed25519.CURVE.l
ed25519.Point.BASE
ed25519.utils.TORSION_SUBGROUP;
Security
Noble is production-ready.
- The library has been audited by an independent security firm cure53: PDF. No vulnerabilities have been found. See changes since audit.
- The library has also been fuzzed by Guido Vranken's cryptofuzz. You can run the fuzzer by yourself to check it.
We're using built-in JS BigInt
, which is "unsuitable for use in cryptography" as per official spec. This means that the lib is potentially vulnerable to timing attacks. But, JIT-compiler and Garbage Collector make "constant time" extremely hard to achieve in a scripting language. Which means any other JS library doesn't use constant-time bigints. Including bn.js or anything else. Even statically typed Rust, a language without GC, makes it harder to achieve constant-time for some cases. If your goal is absolute security, don't use any JS lib — including bindings to native ones. Use low-level libraries & languages. Nonetheless we've hardened implementation of ec curve multiplication to be algorithmically constant time.
We however consider infrastructure attacks like rogue NPM modules very important; that's why it's crucial to minimize the amount of 3rd-party dependencies & native bindings. If your app uses 500 dependencies, any dep could get hacked and you'll be downloading malware with every npm install
. Our goal is to minimize this attack vector.
Speed
Benchmarks done with Apple M1 on macOS 12.
getPublicKey(utils.randomPrivateKey()) x 7,600 ops/sec @ 131μs/op
sign x 3,819 ops/sec @ 261μs/op
verify x 762 ops/sec @ 1ms/op
Point.fromHex decompression x 12,055 ops/sec @ 82μs/op
ristretto255#hashToCurve x 5,617 ops/sec @ 178μs/op
ristretto255 round x 5,734 ops/sec @ 174μs/op
curve25519.scalarMultBase x 1,173 ops/sec @ 852μs/op
ed25519.getSharedSecret x 883 ops/sec @ 1ms/op
Compare to alternative implementations:
# tweetnacl-fast@1.0.3
getPublicKey x 920 ops/sec @ 1ms/op # aka scalarMultBase
sign x 519 ops/sec @ 2ms/op
# ristretto255@0.1.1
getPublicKey x 877 ops/sec @ 1ms/op # aka scalarMultBase
# sodium-native@3.2.1, native bindings to libsodium, node.js-only
sodium-native#sign x 58,661 ops/sec @ 17μs/op
Contributing
- Clone the repository
npm install
to install build dependencies like TypeScriptnpm run build
to compile TypeScript codenpm run test
to run jest on test/index.ts
License
MIT (c) 2019 Paul Miller (https://paulmillr.com), see LICENSE file.