1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
//! Homoglyph detection: finds secrets obfuscated with lookalike Unicode characters.
//!
//! Attackers may replace 'a' with Cyrillic 'а' to bypass simple regexes.
//! This module provides a way to match patterns against homoglyph-expanded forms.
use std::collections::{HashMap, HashSet};
use std::sync::{LazyLock, OnceLock};
/// Returns a map of ASCII characters to their common Unicode homoglyphs.
fn homoglyph_map() -> &'static HashMap<char, Vec<char>> {
static MAP: OnceLock<HashMap<char, Vec<char>>> = OnceLock::new();
MAP.get_or_init(|| {
let mut m = HashMap::new();
m.insert('a', vec!['а', 'α', 'ａ']);
m.insert('b', vec!['Ь', 'β', 'ｂ']);
m.insert('c', vec!['с', 'ｃ']);
m.insert('e', vec!['е', 'ε', 'ｅ']);
m.insert('g', vec!['ɡ', 'ｇ']); // U+0261
m.insert('h', vec!['н', 'һ', 'ｈ']); // U+04BB for h
m.insert('i', vec!['і', 'ι', 'ｉ']);
m.insert('j', vec!['ј', 'ｊ']);
m.insert('k', vec!['к', 'κ', 'ｋ']);
m.insert('m', vec!['м', 'ｍ']);
m.insert('n', vec!['п', 'ν', 'ｎ']);
m.insert('o', vec!['о', 'ο', 'ｏ']);
m.insert('p', vec!['р', 'ρ', 'ｐ']);
m.insert('s', vec!['ѕ', 'ｓ']);
m.insert('t', vec!['т', 'τ', 'ｔ']);
m.insert('u', vec!['υ', 'ｕ']);
// 'l' confuses with the I/1/| cluster: Cyrillic/Greek dotless i and
// fullwidth l. The Greek/Cyrillic o-lookalikes (Ο/ο/о) are an 'o' cluster,
// not 'l', and only add a false-positive/automaton-bloat surface here.
m.insert('l', vec!['і', 'І', 'ι', 'Ι', 'ｌ']);
m.insert('x', vec!['х', 'χ', 'ｘ']);
m.insert('y', vec!['у', 'ｙ']);
m.insert('L', vec!['Ｌ']);
m.insert('A', vec!['А', 'Α', 'Ａ']);
m.insert('B', vec!['В', 'Β', 'Ｂ']);
m.insert('E', vec!['Е', 'Ε', 'Ｅ']);
m.insert('H', vec!['Н', 'Η', 'Ｈ']);
m.insert('I', vec!['І', 'Ι', 'Ｉ']);
m.insert('J', vec!['Ј', 'Ｊ']);
m.insert('K', vec!['К', 'Κ', 'Ｋ']);
m.insert('M', vec!['М', 'Ｍ']);
m.insert('N', vec!['Ν', 'Ｎ']);
m.insert('O', vec!['О', 'Ο', 'Ｏ']);
m.insert('P', vec!['Р', 'Ρ', 'Ｐ']);
m.insert('S', vec!['С', 'Ｓ']);
m.insert('T', vec!['Т', 'Τ', 'Ｔ']);
m.insert('X', vec!['Х', 'Χ', 'Ｘ']);
m.insert('Y', vec!['Υ', 'Ｙ']);
m
})
}
/// The `(ascii, confusable-glyphs)` entries of [`homoglyph_map`], sorted by the
/// ASCII key for deterministic iteration. Exposed (via the `testing` facade) so a
/// cross-map consistency gate can assert this AC/regex-expand map agrees with the
/// `unicode_hardening` normalize-path folds (`cyrillic_to_latin`/`greek_to_latin`)
/// on every shared codepoint (the two are separate scan paths that must not drift).
pub(crate) fn homoglyph_confusables() -> Vec<(char, Vec<char>)> {
let mut entries: Vec<(char, Vec<char>)> = homoglyph_map()
.iter()
.map(|(k, v)| (*k, v.clone()))
.collect();
entries.sort_by_key(|(k, _)| *k);
entries
}
/// The set of FIRST UTF-8 bytes of every confusable glyph in
/// [`homoglyph_map`], as a 256-entry table.
///
/// Derived from the map rather than hand-listed, so adding a glyph cannot
/// leave the table behind. Every confusable is non-ASCII, so an ASCII-only
/// text sets none of these.
fn confusable_lead_bytes() -> &'static [bool; 256] {
static LEADS: OnceLock<[bool; 256]> = OnceLock::new();
LEADS.get_or_init(|| {
let mut table = [false; 256];
let mut buffer = [0_u8; 4];
for glyphs in homoglyph_map().values() {
for glyph in glyphs {
let encoded = glyph.encode_utf8(&mut buffer).as_bytes();
table[usize::from(encoded[0])] = true;
}
}
table
})
}
fn exact_confusable_glyphs() -> &'static HashSet<char> {
static GLYPHS: LazyLock<HashSet<char>> =
LazyLock::new(|| homoglyph_map().values().flatten().copied().collect());
&GLYPHS
}
/// Whether `text` contains any confusable glyph.
///
/// The byte table is a cheap sound prefilter. A matching UTF-8 lead byte is not
/// sufficient by itself: unrelated characters such as the replacement glyph
/// `U+FFFD` share the fullwidth block's `0xEF` lead. Those false positives used
/// to compile and retain the complete Unicode residual matcher set for ordinary
/// invalid-UTF-8 input. Candidate texts therefore receive one exact character
/// membership pass against the same map that builds the matcher variants.
pub(crate) fn may_contain_confusable(text: &str) -> bool {
let leads = confusable_lead_bytes();
if !text.as_bytes().iter().any(|byte| leads[usize::from(*byte)]) {
return false;
}
let confusables = exact_confusable_glyphs();
text.chars()
.any(|character| confusables.contains(&character))
}
/// Expand a regex pattern to include homoglyphs.
/// e.g. "ghp_" -> "[gɡｇ][hнһｈ][pрρｐ]_"
pub(crate) fn expand_homoglyphs(pattern: &str) -> String {
let map = homoglyph_map();
// Every mapped ASCII char becomes a `[<ascii><glyphs>]` class (~8 bytes);
// reserve up front so expansion over all detector prefixes does not realloc
// as it grows. Byte-identical to building from an empty String.
let mut expanded = String::with_capacity(pattern.len() * 8);
// Simple implementation: replace ASCII chars with character classes
for ch in pattern.chars() {
if let Some(glyphs) = map.get(&ch) {
expanded.push('[');
expanded.push(ch);
for &g in glyphs {
expanded.push(g);
}
expanded.push(']');
} else {
push_regex_literal_char(&mut expanded, ch);
}
}
expanded
}
fn push_regex_literal_char(out: &mut String, ch: char) {
if matches!(
ch,
'\\' | '.'
| '+'
| '*'
| '?'
| '('
| ')'
| '|'
| '['
| ']'
| '{'
| '}'
| '^'
| '$'
| '#'
| '&'
| '-'
) {
out.push('\\');
}
out.push(ch);
}