Skip to content

Keats/rust-bcrypt

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

bcrypt

Safety Dance Build Status Documentation

Installation

Add the following to Cargo.toml:

bcrypt = "0.16"

The minimum Rust version is 1.63.0.

Usage

The crate makes 3 things public: DEFAULT_COST, hash, verify.

extern crate bcrypt;

use bcrypt::{DEFAULT_COST, hash, verify};

let hashed = hash("hunter2", DEFAULT_COST)?;
let valid = verify("hunter2", &hashed)?;

The cost needs to be an integer between 4 and 31 (see benchmarks to have an idea of the speed for each), the DEFAULT_COST is 12.

Error on truncation

Most if not all bcrypt implementation truncates the password after 72 bytes. In specific use cases this can break 2nd pre-image resistance. One can enforce the 72-bytes limit on input by using non_truncating_hash, non_truncating_hash_with_result, non_truncating_hash_with_salt, and non_truncating_verify. The non_truncating_* functions behave identically to their truncating counterparts unless the input is longer than 72 bytes, in which case they will return BcryptError::Truncation.

If you are generating hashes from other libraries/languages, do not use the non_truncating_verify function.

no_std

bcrypt crate supports no_std platforms. When alloc feature is enabled, all crate functionality is available. When alloc is not enabled only the raw bcrypt() function is usable.

Benchmarks

Speed depends on the cost used: the highest the slowest. Here are some benchmarks on a 2019 Macbook Pro to give you some ideas on the cost/speed ratio. Note that I don't go above 14 as it takes too long.

test bench_cost_10      ... bench:  51,474,665 ns/iter (+/- 16,006,581)
test bench_cost_14      ... bench: 839,109,086 ns/iter (+/- 274,507,463)
test bench_cost_4       ... bench:     795,814 ns/iter (+/- 42,838)
test bench_cost_default ... bench: 195,344,338 ns/iter (+/- 8,329,675)

Acknowledgments

This gist for the hash splitting and the null termination.

Recommendations

While bcrypt works well as an algorithm, using something like Argon2 is recommended for new projects.

Changelog

  • 0.16.0: add non_truncating_* functions
  • 0.15.1: update base64 dependency
  • 0.15.0: add an alloc feature that can be disabled.
  • 0.14.0: use subtle crate for constant time comparison, update base64 and bump to 2021 edition
  • 0.13.0: make zeroize dep opt-out and use fixed salt length
  • 0.12.1: zero vec containing password in the hashing function before returning the hash
  • 0.12.0: allow null bytes in password
  • 0.11.0: update deps causing big bump in MSRV
  • 0.10.1: fix panic with invalid hashes and allow 2x
  • 0.10.0: update blowfish to 0.8 and minimum Rust version to 1.43.0.
  • 0.9.0: update base64 to 0.13 and getrandom to 0.2
  • 0.8.2: fix no-std build
  • 0.8.0: constant time verification for hash, remove custom base64 code from repo and add std feature
  • 0.7.0: add HashParts::from_str and remove Error::description impl, it's deprecated
  • 0.6.3: add hash_with_salt function and make Version::format_for_version public
  • 0.6.2: update base64 to 0.12
  • 0.6.1: update base64 to 0.11
  • 0.6.0: allow users to choose the bcrypt version and default to 2b instead of 2y
  • 0.5.0: expose the inner bcrypt function + edition 2018
  • 0.4.0: make DEFAULT_COST const instead of static
  • 0.3.0: forbid NULL bytes in passwords & update dependencies
  • 0.2.2: update rand
  • 0.2.1: update rand
  • 0.2.0: replace rust-crypto with blowfish, use some more modern Rust things like ? and handle more errors
  • 0.1.6: update rand and base64 deps
  • 0.1.5: update lazy-static to 1.0
  • 0.1.4: Replace rustc-serialize dependency with bcrypt
  • 0.1.3: Fix panic when password > 72 chars
  • 0.1.1: make BcryptResult, BcryptError public and update dependencies
  • 0.1.0: initial release