Smash out seed

This commit is contained in:
Luke Parker
2024-06-23 06:31:22 -04:00
parent 1e2e3bd5ce
commit 11dba9173f
22 changed files with 146 additions and 64 deletions

View File

@@ -0,0 +1,37 @@
[package]
name = "monero-seed"
version = "0.1.0"
description = "Rust implementation of Monero's seed algorithm"
license = "MIT"
repository = "https://github.com/serai-dex/serai/tree/develop/coins/monero/wallet/seed"
authors = ["Luke Parker <lukeparker5132@gmail.com>"]
edition = "2021"
rust-version = "1.79"
[package.metadata.docs.rs]
all-features = true
rustdoc-args = ["--cfg", "docsrs"]
[lints]
workspace = true
[dependencies]
std-shims = { path = "../../../../common/std-shims", version = "^0.1.1", default-features = false }
thiserror = { version = "1", default-features = false, optional = true }
zeroize = { version = "^1.5", default-features = false, features = ["zeroize_derive"] }
rand_core = { version = "0.6", default-features = false }
curve25519-dalek = { version = "4", default-features = false, features = ["alloc", "zeroize"] }
[features]
std = [
"std-shims/std",
"thiserror",
"zeroize/std",
"rand_core/std",
]
default = ["std"]

View File

@@ -1,6 +1,6 @@
# Monero Seeds
A Rust implementation of Monero's seed algorithm.
Rust implementation of Monero's seed algorithm.
This library is usable under no-std when the `std` feature (on by default) is
disabled.

View File

@@ -0,0 +1,339 @@
use core::ops::Deref;
use std_shims::{
sync::OnceLock,
vec::Vec,
string::{String, ToString},
collections::HashMap,
};
use zeroize::{Zeroize, Zeroizing};
use rand_core::{RngCore, CryptoRng};
use curve25519_dalek::scalar::Scalar;
const CLASSIC_SEED_LENGTH: usize = 24;
const CLASSIC_SEED_LENGTH_WITH_CHECKSUM: usize = 25;
/// An error when working with a seed.
#[derive(Clone, Copy, PartialEq, Eq, Debug)]
#[cfg_attr(feature = "std", derive(thiserror::Error))]
pub enum SeedError {
/// The deprecated English language option was used with a checksum.
///
/// The deprecated English language option did not include a checksum.
#[cfg_attr(feature = "std", error("deprecated English language option included a checksum"))]
DeprecatedEnglishWithChecksum,
#[cfg_attr(feature = "std", error("invalid seed"))]
/// The seed was invalid.
InvalidSeed,
/// The checksum did not match the data.
#[cfg_attr(feature = "std", error("invalid checksum"))]
InvalidChecksum,
}
/// Language options.
#[derive(Clone, Copy, PartialEq, Eq, Debug, Hash, Zeroize)]
pub enum Language {
/// Chinese language option.
Chinese,
/// English language option.
English,
/// Dutch language option.
Dutch,
/// French language option.
French,
/// Spanish language option.
Spanish,
/// German language option.
German,
/// Italian language option.
Italian,
/// Portuguese language option.
Portuguese,
/// Japanese language option.
Japanese,
/// Russian language option.
Russian,
/// Esperanto language option.
Esperanto,
/// Lojban language option.
Lojban,
/// The original, and deprecated, English language.
DeprecatedEnglish,
}
fn trim(word: &str, len: usize) -> Zeroizing<String> {
Zeroizing::new(word.chars().take(len).collect())
}
struct WordList {
word_list: &'static [&'static str],
word_map: HashMap<&'static str, usize>,
trimmed_word_map: HashMap<String, usize>,
unique_prefix_length: usize,
}
impl WordList {
fn new(word_list: &'static [&'static str], prefix_length: usize) -> WordList {
let mut lang = WordList {
word_list,
word_map: HashMap::new(),
trimmed_word_map: HashMap::new(),
unique_prefix_length: prefix_length,
};
for (i, word) in lang.word_list.iter().enumerate() {
lang.word_map.insert(word, i);
lang.trimmed_word_map.insert(trim(word, lang.unique_prefix_length).deref().clone(), i);
}
lang
}
}
static LANGUAGES_CELL: OnceLock<HashMap<Language, WordList>> = OnceLock::new();
#[allow(non_snake_case)]
fn LANGUAGES() -> &'static HashMap<Language, WordList> {
LANGUAGES_CELL.get_or_init(|| {
HashMap::from([
(Language::Chinese, WordList::new(include!("./words/zh.rs"), 1)),
(Language::English, WordList::new(include!("./words/en.rs"), 3)),
(Language::Dutch, WordList::new(include!("./words/nl.rs"), 4)),
(Language::French, WordList::new(include!("./words/fr.rs"), 4)),
(Language::Spanish, WordList::new(include!("./words/es.rs"), 4)),
(Language::German, WordList::new(include!("./words/de.rs"), 4)),
(Language::Italian, WordList::new(include!("./words/it.rs"), 4)),
(Language::Portuguese, WordList::new(include!("./words/pt.rs"), 4)),
(Language::Japanese, WordList::new(include!("./words/ja.rs"), 3)),
(Language::Russian, WordList::new(include!("./words/ru.rs"), 4)),
(Language::Esperanto, WordList::new(include!("./words/eo.rs"), 4)),
(Language::Lojban, WordList::new(include!("./words/jbo.rs"), 4)),
(Language::DeprecatedEnglish, WordList::new(include!("./words/ang.rs"), 4)),
])
})
}
fn checksum_index(words: &[Zeroizing<String>], lang: &WordList) -> usize {
let mut trimmed_words = Zeroizing::new(String::new());
for w in words {
*trimmed_words += &trim(w, lang.unique_prefix_length);
}
const fn crc32_table() -> [u32; 256] {
let poly = 0xedb88320u32;
let mut res = [0; 256];
let mut i = 0;
while i < 256 {
let mut entry = i;
let mut b = 0;
while b < 8 {
let trigger = entry & 1;
entry >>= 1;
if trigger == 1 {
entry ^= poly;
}
b += 1;
}
res[i as usize] = entry;
i += 1;
}
res
}
const CRC32_TABLE: [u32; 256] = crc32_table();
let trimmed_words = trimmed_words.as_bytes();
let mut checksum = u32::MAX;
for i in 0 .. trimmed_words.len() {
checksum = CRC32_TABLE[usize::from(u8::try_from(checksum % 256).unwrap() ^ trimmed_words[i])] ^
(checksum >> 8);
}
usize::try_from(!checksum).unwrap() % words.len()
}
// Convert a private key to a seed
#[allow(clippy::needless_pass_by_value)]
fn key_to_seed(lang: Language, key: Zeroizing<Scalar>) -> Seed {
let bytes = Zeroizing::new(key.to_bytes());
// get the language words
let words = &LANGUAGES()[&lang].word_list;
let list_len = u64::try_from(words.len()).unwrap();
// To store the found words & add the checksum word later.
let mut seed = Vec::with_capacity(25);
// convert to words
// 4 bytes -> 3 words. 8 digits base 16 -> 3 digits base 1626
let mut segment = [0; 4];
let mut indices = [0; 4];
for i in 0 .. 8 {
// convert first 4 byte to u32 & get the word indices
let start = i * 4;
// convert 4 byte to u32
segment.copy_from_slice(&bytes[start .. (start + 4)]);
// Actually convert to a u64 so we can add without overflowing
indices[0] = u64::from(u32::from_le_bytes(segment));
indices[1] = indices[0];
indices[0] /= list_len;
indices[2] = indices[0] + indices[1];
indices[0] /= list_len;
indices[3] = indices[0] + indices[2];
// append words to seed
for i in indices.iter().skip(1) {
let word = usize::try_from(i % list_len).unwrap();
seed.push(Zeroizing::new(words[word].to_string()));
}
}
segment.zeroize();
indices.zeroize();
// create a checksum word for all languages except old english
if lang != Language::DeprecatedEnglish {
let checksum = seed[checksum_index(&seed, &LANGUAGES()[&lang])].clone();
seed.push(checksum);
}
let mut res = Zeroizing::new(String::new());
for (i, word) in seed.iter().enumerate() {
if i != 0 {
*res += " ";
}
*res += word;
}
Seed(lang, res)
}
// Convert a seed to bytes
fn seed_to_bytes(lang: Language, words: &str) -> Result<Zeroizing<[u8; 32]>, SeedError> {
// get seed words
let words = words.split_whitespace().map(|w| Zeroizing::new(w.to_string())).collect::<Vec<_>>();
if (words.len() != CLASSIC_SEED_LENGTH) && (words.len() != CLASSIC_SEED_LENGTH_WITH_CHECKSUM) {
panic!("invalid seed passed to seed_to_bytes");
}
let has_checksum = words.len() == CLASSIC_SEED_LENGTH_WITH_CHECKSUM;
if has_checksum && lang == Language::DeprecatedEnglish {
Err(SeedError::DeprecatedEnglishWithChecksum)?;
}
// Validate words are in the language word list
let lang_word_list: &WordList = &LANGUAGES()[&lang];
let matched_indices = (|| {
let has_checksum = words.len() == CLASSIC_SEED_LENGTH_WITH_CHECKSUM;
let mut matched_indices = Zeroizing::new(vec![]);
// Iterate through all the words and see if they're all present
for word in &words {
let trimmed = trim(word, lang_word_list.unique_prefix_length);
let word = if has_checksum { &trimmed } else { word };
if let Some(index) = if has_checksum {
lang_word_list.trimmed_word_map.get(word.deref())
} else {
lang_word_list.word_map.get(&word.as_str())
} {
matched_indices.push(*index);
} else {
Err(SeedError::InvalidSeed)?;
}
}
if has_checksum {
// exclude the last word when calculating a checksum.
let last_word = words.last().unwrap().clone();
let checksum = words[checksum_index(&words[.. words.len() - 1], lang_word_list)].clone();
// check the trimmed checksum and trimmed last word line up
if trim(&checksum, lang_word_list.unique_prefix_length) !=
trim(&last_word, lang_word_list.unique_prefix_length)
{
Err(SeedError::InvalidChecksum)?;
}
}
Ok(matched_indices)
})()?;
// convert to bytes
let mut res = Zeroizing::new([0; 32]);
let mut indices = Zeroizing::new([0; 4]);
for i in 0 .. 8 {
// read 3 indices at a time
let i3 = i * 3;
indices[1] = matched_indices[i3];
indices[2] = matched_indices[i3 + 1];
indices[3] = matched_indices[i3 + 2];
let inner = |i| {
let mut base = (lang_word_list.word_list.len() - indices[i] + indices[i + 1]) %
lang_word_list.word_list.len();
// Shift the index over
for _ in 0 .. i {
base *= lang_word_list.word_list.len();
}
base
};
// set the last index
indices[0] = indices[1] + inner(1) + inner(2);
if (indices[0] % lang_word_list.word_list.len()) != indices[1] {
Err(SeedError::InvalidSeed)?;
}
let pos = i * 4;
let mut bytes = u32::try_from(indices[0]).unwrap().to_le_bytes();
res[pos .. (pos + 4)].copy_from_slice(&bytes);
bytes.zeroize();
}
Ok(res)
}
/// A Monero seed.
#[derive(Clone, PartialEq, Eq, Zeroize)]
pub struct Seed(Language, Zeroizing<String>);
impl Seed {
/// Create a new seed.
pub fn new<R: RngCore + CryptoRng>(rng: &mut R, lang: Language) -> Seed {
let mut scalar_bytes = Zeroizing::new([0; 64]);
rng.fill_bytes(scalar_bytes.as_mut());
key_to_seed(lang, Zeroizing::new(Scalar::from_bytes_mod_order_wide(scalar_bytes.deref())))
}
/// Parse a seed from a string.
#[allow(clippy::needless_pass_by_value)]
pub fn from_string(lang: Language, words: Zeroizing<String>) -> Result<Seed, SeedError> {
let entropy = seed_to_bytes(lang, &words)?;
// Make sure this is a valid scalar
let scalar = Scalar::from_canonical_bytes(*entropy);
if scalar.is_none().into() {
Err(SeedError::InvalidSeed)?;
}
let mut scalar = scalar.unwrap();
scalar.zeroize();
// Call from_entropy so a trimmed seed becomes a full seed
Ok(Self::from_entropy(lang, entropy).unwrap())
}
/// Create a seed from entropy.
#[allow(clippy::needless_pass_by_value)]
pub fn from_entropy(lang: Language, entropy: Zeroizing<[u8; 32]>) -> Option<Seed> {
Option::from(Scalar::from_canonical_bytes(*entropy))
.map(|scalar| key_to_seed(lang, Zeroizing::new(scalar)))
}
/// Convert a seed to a string.
pub fn to_string(&self) -> Zeroizing<String> {
self.1.clone()
}
/// Return the entropy underlying this seed.
pub fn entropy(&self) -> Zeroizing<[u8; 32]> {
seed_to_bytes(self.0, &self.1).unwrap()
}
}

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff