Audited & minimal JS implementation of SHA, RIPEMD, BLAKE, HMAC, HKDF, PBKDF, Scrypt & Argon2.
- ๐ Audited by an independent security firm
- ๐ป Tree-shakeable: unused code is excluded from your builds
- ๐ Fast: hand-optimized for caveats of JS engines
- ๐ Reliable: chained / sliding window / DoS tests and fuzzing ensure correctness
- ๐ No unrolled loops: makes it easier to verify and reduces source code size up to 5x
- ๐ข Scrypt supports
N: 2**22
, while other implementations are limited to2**20
- ๐ฆ SHA3 supports Keccak, cSHAKE, KangarooTwelve, MarsupilamiFourteen and TurboSHAKE
- ๐ชถ 89KB (17KB gzipped) for everything, 10KB (2.5KB gzipped) for single-hash build
The library's initial development was funded by Ethereum Foundation.
For discussions, questions and support, visit GitHub Discussions section of the repository.
noble cryptography โ high-security, easily auditable set of contained cryptographic libraries and tools.
- Zero or minimal dependencies
- Highly readable TypeScript / JS code
- PGP-signed releases and transparent NPM builds
- All libraries: ciphers, curves, hashes, post-quantum, 4kb secp256k1 / ed25519
- Check out homepage for reading resources, documentation and apps built with noble
npm install @noble/hashes
We support all major platforms and runtimes. For Deno, ensure to use npm specifier. For React Native, you may need a polyfill for getRandomValues. A standalone file noble-hashes.js is also available.
// import * from '@noble/hashes'; // Error: use sub-imports, to ensure small app size
import { sha256 } from '@noble/hashes/sha2'; // ECMAScript modules (ESM) and Common.js
// import { sha256 } from 'npm:@noble/[email protected]/sha2'; // Deno
console.log(sha256(new Uint8Array([1, 2, 3]))); // Uint8Array(32) [3, 144, 88, 198, 242...]
// you could also pass strings that will be UTF8-encoded to Uint8Array
console.log(sha256('abc')); // == sha256(new TextEncoder().encode('abc'))
All hash functions:
- receive
Uint8Array
and returnUint8Array
- may receive
string
, which is automatically converted toUint8Array
via utf8 encoding (not hex) - support little-endian and big-endian architectures
- can hash up to 4GB per chunk, with any amount of chunks
function hash(message: Uint8Array | string): Uint8Array;
hash(new Uint8Array([1, 3]));
hash('string') == hash(new TextEncoder().encode('string'));
All hash functions can be constructed via hash.create()
method:
- the result is
Hash
subclass instance, which hasupdate()
anddigest()
methods digest()
finalizes the hash and makes it no longer usable
hash
.create()
.update(new Uint8Array([1, 3]))
.digest();
Some hash functions can also receive options
object, which can be either passed as a:
- second argument to hash function:
blake3('abc', { key: 'd', dkLen: 32 })
- first argument to class initializer:
blake3.create({ context: 'e', dkLen: 32 })
import { sha256, sha384, sha512, sha224, sha512_256, sha512_384 } from '@noble/hashes/sha2';
// also available as aliases:
// import ... from '@noble/hashes/sha256'
// import ... from '@noble/hashes/sha512'
// Variant A:
const h1a = sha256('abc');
// Variant B:
const h1b = sha256
.create()
.update(Uint8Array.from([1, 2, 3]))
.digest();
for (let hash of [sha384, sha512, sha224, sha512_256, sha512_384]) {
const res1 = hash('abc');
const res2 = hash.create().update('def').update(Uint8Array.from([1, 2, 3])).digest();
};
See RFC 4634 and the paper on truncated SHA512/256.
import {
sha3_224,
sha3_256,
sha3_384,
sha3_512,
keccak_224,
keccak_256,
keccak_384,
keccak_512,
shake128,
shake256,
} from '@noble/hashes/sha3';
const h5a = sha3_256('abc');
const h5b = sha3_256
.create()
.update(Uint8Array.from([1, 2, 3]))
.digest();
const h6a = keccak_256('abc');
const h7a = shake128('abc', { dkLen: 512 });
const h7b = shake256('abc', { dkLen: 512 });
See FIPS PUB 202, Website.
Check out the differences between SHA-3 and Keccak
import {
cshake128,
cshake256,
kmac128,
kmac256,
k12,
m14,
turboshake128,
turboshake256,
tuplehash128,
tuplehash256,
parallelhash128,
parallelhash256,
keccakprg,
} from '@noble/hashes/sha3-addons';
const h7c = cshake128('abc', { personalization: 'def' });
const h7d = cshake256('abc', { personalization: 'def' });
const h7e = kmac128('key', 'message');
const h7f = kmac256('key', 'message');
const h7h = k12('abc');
const h7g = m14('abc');
const h7t1 = turboshake128('abc');
const h7t2 = turboshake256('def', { D: 0x05 });
const h7i = tuplehash128(['ab', 'c']); // tuplehash(['ab', 'c']) !== tuplehash(['a', 'bc']) !== tuplehash(['abc'])
// Same as k12/blake3, but without reduced number of rounds. Doesn't speedup anything due lack of SIMD and threading,
// added for compatibility.
const h7j = parallelhash128('abc', { blockLen: 8 });
// pseudo-random generator, first argument is capacity. XKCP recommends 254 bits capacity for 128-bit security strength.
// * with a capacity of 254 bits.
const p = keccakprg(254);
p.feed('test');
const rand1b = p.fetch(1);
- Full NIST SP 800-185: cSHAKE, KMAC, TupleHash, ParallelHash + XOF variants
- Reduced-round Keccak:
- ๐ฆ K12 aka KangarooTwelve
- M14 aka MarsupilamiFourteen
- TurboSHAKE
- KeccakPRG: Pseudo-random generator based on Keccak
import { ripemd160 } from '@noble/hashes/ripemd160';
// function ripemd160(data: Uint8Array): Uint8Array;
const hash8 = ripemd160('abc');
const hash9 = ripemd160
.create()
.update(Uint8Array.from([1, 2, 3]))
.digest();
import { blake2b } from '@noble/hashes/blake2b';
import { blake2s } from '@noble/hashes/blake2s';
import { blake3 } from '@noble/hashes/blake3';
const h10a = blake2s('abc');
const b2params = { key: new Uint8Array([1]), personalization: t, salt: t, dkLen: 32 };
const h10b = blake2s('abc', b2params);
const h10c = blake2s
.create(b2params)
.update(Uint8Array.from([1, 2, 3]))
.digest();
// All params are optional
const h11 = blake3('abc', { dkLen: 256, key: 'def', context: 'fji' });
SHA1 was cryptographically broken, however, it was not broken for cases like HMAC.
See RFC4226 B.2.
Don't use it for a new protocol.
import { sha1 } from '@noble/hashes/sha1';
const h12 = sha1('def');
import { hmac } from '@noble/hashes/hmac';
import { sha256 } from '@noble/hashes/sha2';
const mac1 = hmac(sha256, 'key', 'message');
const mac2 = hmac
.create(sha256, Uint8Array.from([1, 2, 3]))
.update(Uint8Array.from([4, 5, 6]))
.digest();
Matches RFC 2104.
import { hkdf } from '@noble/hashes/hkdf';
import { sha256 } from '@noble/hashes/sha';
import { randomBytes } from '@noble/hashes/utils';
const inputKey = randomBytes(32);
const salt = randomBytes(32);
const info = 'abc';
const dkLen = 32;
const hk1 = hkdf(sha256, inputKey, salt, info, dkLen);
// == same as
import * as hkdf from '@noble/hashes/hkdf';
import { sha256 } from '@noble/hashes/sha2';
const prk = hkdf.extract(sha256, inputKey, salt);
const hk2 = hkdf.expand(sha256, prk, info, dkLen);
Matches RFC 5869.
import { pbkdf2, pbkdf2Async } from '@noble/hashes/pbkdf2';
import { sha256 } from '@noble/hashes/sha2';
const pbkey1 = pbkdf2(sha256, 'password', 'salt', { c: 32, dkLen: 32 });
const pbkey2 = await pbkdf2Async(sha256, 'password', 'salt', { c: 32, dkLen: 32 });
const pbkey3 = await pbkdf2Async(sha256, Uint8Array.from([1, 2, 3]), Uint8Array.from([4, 5, 6]), {
c: 32,
dkLen: 32,
});
Matches RFC 2898.
import { scrypt, scryptAsync } from '@noble/hashes/scrypt';
const scr1 = scrypt('password', 'salt', { N: 2 ** 16, r: 8, p: 1, dkLen: 32 });
const scr2 = await scryptAsync('password', 'salt', { N: 2 ** 16, r: 8, p: 1, dkLen: 32 });
const scr3 = await scryptAsync(Uint8Array.from([1, 2, 3]), Uint8Array.from([4, 5, 6]), {
N: 2 ** 22,
r: 8,
p: 1,
dkLen: 32,
onProgress(percentage) {
console.log('progress', percentage);
},
maxmem: 2 ** 32 + 128 * 8 * 1, // N * r * p * 128 + (128*r*p)
});
N, r, p
are work factors. To understand them, see the blog post.dkLen
is the length of output bytes- It is common to use N from
2**10
to2**22
and{r: 8, p: 1, dkLen: 32}
onProgress
can be used with async version of the function to report progress to a user.
Memory usage of scrypt is calculated with the formula N * r * p * 128 + (128 * r * p)
,
which means {N: 2 ** 22, r: 8, p: 1}
will use 4GB + 1KB of memory. To prevent
DoS, we limit scrypt to 1GB + 1KB
of RAM used, which corresponds to
{N: 2 ** 20, r: 8, p: 1}
. If you want to use higher values, increase
maxmem
using the formula above.
Note: noble supports 2**22
(4GB RAM) which is the highest amount amongst JS
libs. Many other implementations don't support it. We cannot support 2**23
,
because there is a limitation in JS engines that makes allocating
arrays bigger than 4GB impossible, but we're looking into other possible solutions.
Warning
Experimental: it may be removed at any time.
Argon2 RFC 9106 implementation.
import { argon2d, argon2i, argon2id } from '@noble/hashes/argon2';
const result = argon2id('password', 'salt', { t: 2, m: 65536, p: 1 });
import { bytesToHex as toHex, randomBytes } from '@noble/hashes/utils';
console.log(toHex(randomBytes(32)));
bytesToHex
will convertUint8Array
to a hex stringrandomBytes(bytes)
will produce cryptographically secure randomUint8Array
of lengthbytes
import { sha256, sha384, sha512, sha224, sha512_256, sha512_384 } from '@noble/hashes/sha2';
// prettier-ignore
import {
sha3_224, sha3_256, sha3_384, sha3_512,
keccak_224, keccak_256, keccak_384, keccak_512,
shake128, shake256
} from '@noble/hashes/sha3';
// prettier-ignore
import {
cshake128, cshake256,
turboshake128, turboshake256,
kmac128, kmac256,
tuplehash256, parallelhash256,
k12, m14, keccakprg
} from '@noble/hashes/sha3-addons';
import { ripemd160 } from '@noble/hashes/ripemd160';
import { blake3 } from '@noble/hashes/blake3';
import { blake2b } from '@noble/hashes/blake2b';
import { blake2s } from '@noble/hashes/blake2s';
import { hmac } from '@noble/hashes/hmac';
import { hkdf } from '@noble/hashes/hkdf';
import { pbkdf2, pbkdf2Async } from '@noble/hashes/pbkdf2';
import { scrypt, scryptAsync } from '@noble/hashes/scrypt';
import { sha1 } from '@noble/hashes/sha1'; // legacy
// small utility method that converts bytes to hex
import { bytesToHex as toHex } from '@noble/hashes/utils';
console.log(toHex(sha256('abc'))); // ba7816bf8f01cfea414140de5dae2223b00361a396177a9cb410ff61f20015ad
The library has been independently audited:
- at version 1.0.0, in Jan 2022, by cure53
- PDFs: online, offline, in-repo
- Changes since audit.
- Scope: everything, besides
blake3
,sha3-addons
,sha1
andargon2
, which have not been audited - The audit has been funded by Ethereum Foundation with help of Nomic Labs
It is tested against property-based, cross-library and Wycheproof vectors, and has fuzzing by Guido Vranken's cryptofuzz.
If you see anything unusual: investigate and report.
JIT-compiler and Garbage Collector make "constant time" extremely hard to achieve timing attack resistance in a scripting language. Which means any other JS library can't have constant-timeness. 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 โ including bindings to native ones. Use low-level libraries & languages. Nonetheless we're targetting algorithmic constant time.
The library shares state buffers between hash function calls. The buffers are zeroed-out after each call. However, if an attacker can read application memory, you are doomed in any case:
- At some point, input will be a string and strings are immutable in JS:
there is no way to overwrite them with zeros. For example: deriving
key from
scrypt(password, salt)
where password and salt are strings - Input from a file will stay in file buffers
- Input / output will be re-used multiple times in application which means it could stay in memory
await anything()
will always write all internal variables (including numbers) to memory. With async functions / Promises there are no guarantees when the code chunk would be executed. Which means attacker can have plenty of time to read data from memory- There is no way to guarantee anything about zeroing sensitive data without complex tests-suite which will dump process memory and verify that there is no sensitive data left. For JS it means testing all browsers (incl. mobile), which is complex. And of course it will be useless without using the same test-suite in the actual application that consumes the library
- Commits are signed with PGP keys, to prevent forgery. Make sure to verify commit signatures.
- Releases are transparent and built on GitHub CI. Make sure to verify provenance logs
- Rare releasing is followed to ensure less re-audit need for end-users
- Dependencies are minimized and locked-down:
- If your app has 500 dependencies, any dep could get hacked and you'll be downloading malware with every install. We make sure to use as few dependencies as possible
- We prevent automatic dependency updates by locking-down version ranges. Every update is checked with
npm-diff
- Dev Dependencies are only used if you want to contribute to the repo. They are disabled for end-users:
- scure-base, scure-bip32, scure-bip39, micro-bmark and micro-should are developed by the same author and follow identical security practices
- prettier (linter), fast-check (property-based testing) and typescript are used for code quality, vector generation and ts compilation. The packages are big, which makes it hard to audit their source code thoroughly and fully
We're deferring to built-in crypto.getRandomValues which is considered cryptographically secure (CSPRNG).
In the past, browsers had bugs that made it weak: it may happen again. Implementing a userspace CSPRNG to get resilient to the weakness is even worse: there is no reliable userspace source of quality entropy.
Benchmarks measured on Apple M1 with macOS 12.
Note that PBKDF2 and Scrypt are tested with extremely high work factor.
To run benchmarks, execute npm run bench:install
and then npm run bench
SHA256 32B x 1,219,512 ops/sec @ 820ns/op ยฑ 2.58% (min: 625ns, max: 4ms)
SHA384 32B x 512,032 ops/sec @ 1ฮผs/op
SHA512 32B x 509,943 ops/sec @ 1ฮผs/op
SHA3-256, keccak256, shake256 32B x 199,600 ops/sec @ 5ฮผs/op
Kangaroo12 32B x 336,360 ops/sec @ 2ฮผs/op
Marsupilami14 32B x 298,418 ops/sec @ 3ฮผs/op
BLAKE2b 32B x 379,794 ops/sec @ 2ฮผs/op
BLAKE2s 32B x 515,995 ops/sec @ 1ฮผs/op ยฑ 1.07% (min: 1ฮผs, max: 4ms)
BLAKE3 32B x 588,235 ops/sec @ 1ฮผs/op ยฑ 1.36% (min: 1ฮผs, max: 5ms)
RIPEMD160 32B x 1,140,250 ops/sec @ 877ns/op ยฑ 3.12% (min: 708ns, max: 6ms)
HMAC-SHA256 32B x 377,358 ops/sec @ 2ฮผs/op
HKDF-SHA256 32B x 108,377 ops/sec @ 9ฮผs/op
PBKDF2-HMAC-SHA256 262144 x 3 ops/sec @ 326ms/op
PBKDF2-HMAC-SHA512 262144 x 1 ops/sec @ 970ms/op
Scrypt r: 8, p: 1, n: 262144 x 1 ops/sec @ 616ms/op
Compare to native node.js implementation that uses C bindings instead of pure-js code:
SHA256 32B node x 1,302,083 ops/sec @ 768ns/op ยฑ 10.54% (min: 416ns, max: 7ms)
SHA384 32B node x 975,609 ops/sec @ 1ฮผs/op ยฑ 11.32% (min: 625ns, max: 8ms)
SHA512 32B node x 983,284 ops/sec @ 1ฮผs/op ยฑ 11.24% (min: 625ns, max: 8ms)
SHA3-256 32B node x 910,746 ops/sec @ 1ฮผs/op ยฑ 12.19% (min: 666ns, max: 10ms)
keccak, k12, m14 are not implemented
BLAKE2b 32B node x 967,117 ops/sec @ 1ฮผs/op ยฑ 11.26% (min: 625ns, max: 9ms)
BLAKE2s 32B node x 1,055,966 ops/sec @ 947ns/op ยฑ 11.07% (min: 583ns, max: 7ms)
BLAKE3 is not implemented
RIPEMD160 32B node x 1,002,004 ops/sec @ 998ns/op ยฑ 10.66% (min: 625ns, max: 7ms)
HMAC-SHA256 32B node x 919,963 ops/sec @ 1ฮผs/op ยฑ 6.13% (min: 833ns, max: 5ms)
HKDF-SHA256 32 node x 369,276 ops/sec @ 2ฮผs/op ยฑ 13.59% (min: 1ฮผs, max: 9ms)
PBKDF2-HMAC-SHA256 262144 node x 25 ops/sec @ 39ms/op
PBKDF2-HMAC-SHA512 262144 node x 7 ops/sec @ 132ms/op
Scrypt r: 8, p: 1, n: 262144 node x 1 ops/sec @ 523ms/op
It is possible to make this library 4x+ faster by doing code generation of full loop unrolls. We've decided against it. Reasons:
- the library must be auditable, with minimum amount of code, and zero dependencies
- most method invocations with the lib are going to be something like hashing 32b to 64kb of data
- hashing big inputs is 10x faster with low-level languages, which means you should probably pick 'em instead
The current performance is good enough when compared to other projects; SHA256 takes only 900 nanoseconds to run.
- Clone the repository
npm install
to install build dependencies like TypeScriptnpm run build
to compile TypeScript codenpm run test
will execute all main tests. See our approach to testingnpm run test:dos
will test against DoS; by measuring function complexity. Takes ~20 minutesnpm run test:big
will execute hashing on 4GB inputs, scrypt with 1024 differentN, r, p
combinations, etc. Takes several hours. Using 8-32+ core CPU helps.
Check out paulmillr.com/noble for useful resources, articles, documentation and demos related to the library.
The MIT License (MIT)
Copyright (c) 2022 Paul Miller (https://paulmillr.com)
See LICENSE file.