| 1 |
<?php |
| 2 |
/** |
| 3 |
* Glob_Matcher — translate user-friendly glob patterns to PCRE for the |
| 4 |
* cache engine's exclusion checks. |
| 5 |
* |
| 6 |
* Supported syntax (shell-style glob, NOT full PCRE): |
| 7 |
* * → match any run of characters (greedy), including '/' |
| 8 |
* ? → match exactly one character |
| 9 |
* [abc] → character class |
| 10 |
* \* → literal asterisk (escape) |
| 11 |
* |
| 12 |
* Compiles each user pattern once, caches the compiled regex in a |
| 13 |
* static array for the request, then matches with a single preg_match. |
| 14 |
* |
| 15 |
* A bare substring (no glob metacharacters) keeps the legacy "contains" |
| 16 |
* semantics: `/cart` matches `/cart`, `/cart/items`, `/foo/cart/bar`. |
| 17 |
* Adding ANY of `* ? [` switches the pattern to anchored glob mode: |
| 18 |
* `/cart/*` matches `/cart/items` but NOT `/foo/cart/bar`. |
| 19 |
* |
| 20 |
* RAW REGEX: a pattern prefixed with `~` is treated as a raw PCRE |
| 21 |
* (unanchored) — `~utm_[a-z0-9_-]+` matches like a real regex. This lets |
| 22 |
* users paste LiteSpeed / WP Rocket exclusion lists (which are regex) |
| 23 |
* verbatim by adding the `~` marker. Invalid or over-long regex patterns |
| 24 |
* are rejected safely (never match, never fatal, never match-everything). |
| 25 |
* |
| 26 |
* NAME MODE (`matches_name()` / `any_match_name()`) matches an identifier — |
| 27 |
* a query parameter name — instead of a URL, and there "contains" is the |
| 28 |
* wrong contract: `ref` would swallow `preference`, `product_ref` and |
| 29 |
* `referrer`, collapsing genuinely different pages onto one cache key. In |
| 30 |
* name mode a bare pattern is an exact, case-insensitive match, and BOTH |
| 31 |
* the glob and `~regex` forms are anchored, so `~utm_[a-z0-9_-]+` matches |
| 32 |
* `utm_source` but not `my_utm_source`. URL matching keeps the contains |
| 33 |
* semantics above — a path exclusion of `/cart` is meant to catch |
| 34 |
* `/cart/items`. |
| 35 |
* |
| 36 |
* Cookie names deliberately do NOT use name mode: the shipped defaults are |
| 37 |
* prefixes of hash-suffixed real cookies (`comment_author` must match |
| 38 |
* `comment_author_<hash>`), so tightening them would serve shared cached |
| 39 |
* pages to commenters and logged-in-adjacent visitors. |
| 40 |
* |
| 41 |
* @package XSpeed |
| 42 |
*/ |
| 43 |
|
| 44 |
declare(strict_types=1); |
| 45 |
|
| 46 |
namespace XSpeed; |
| 47 |
|
| 48 |
defined( 'ABSPATH' ) || exit; |
| 49 |
|
| 50 |
final class Glob_Matcher { |
| 51 |
|
| 52 |
/** |
| 53 |
* @var array<string,string> pattern => compiled regex |
| 54 |
*/ |
| 55 |
private static $compiled = array(); |
| 56 |
|
| 57 |
/** |
| 58 |
* Does any of `$patterns` match `$subject`? |
| 59 |
* |
| 60 |
* @param string[] $patterns |
| 61 |
*/ |
| 62 |
public static function any_match( array $patterns, string $subject ): bool { |
| 63 |
foreach ( $patterns as $p ) { |
| 64 |
$p = (string) $p; |
| 65 |
if ( '' === $p ) { |
| 66 |
continue; |
| 67 |
} |
| 68 |
if ( self::matches( $p, $subject ) ) { |
| 69 |
return true; |
| 70 |
} |
| 71 |
} |
| 72 |
return false; |
| 73 |
} |
| 74 |
|
| 75 |
public static function matches( string $pattern, string $subject ): bool { |
| 76 |
$regex = self::compile( $pattern ); |
| 77 |
// Anchored glob (regex returned starts with '#^') vs substring |
| 78 |
// (regex returned starts with '#'). Both use preg_match the |
| 79 |
// same way; the anchoring is baked into the pattern. |
| 80 |
return 1 === preg_match( $regex, $subject ); |
| 81 |
} |
| 82 |
|
| 83 |
/** |
| 84 |
* Does any of `$patterns` match the identifier `$name`? |
| 85 |
* |
| 86 |
* Name mode: whole-string matching for every pattern form. See the class |
| 87 |
* docblock for why an identifier must not use contains semantics. |
| 88 |
* |
| 89 |
* @param string[] $patterns |
| 90 |
*/ |
| 91 |
public static function any_match_name( array $patterns, string $name ): bool { |
| 92 |
foreach ( $patterns as $p ) { |
| 93 |
$p = (string) $p; |
| 94 |
if ( '' === $p ) { |
| 95 |
continue; |
| 96 |
} |
| 97 |
if ( self::matches_name( $p, $name ) ) { |
| 98 |
return true; |
| 99 |
} |
| 100 |
} |
| 101 |
return false; |
| 102 |
} |
| 103 |
|
| 104 |
/** |
| 105 |
* Match one pattern against an identifier, whole-string and |
| 106 |
* case-insensitively. |
| 107 |
*/ |
| 108 |
public static function matches_name( string $pattern, string $name ): bool { |
| 109 |
return 1 === preg_match( self::compile_name( $pattern ), $name ); |
| 110 |
} |
| 111 |
|
| 112 |
/** |
| 113 |
* Compile a pattern for name mode. Cached separately from URL mode — |
| 114 |
* the same pattern compiles to a different regex in each — under a key |
| 115 |
* no user pattern can produce, since a NUL byte cannot survive the |
| 116 |
* settings sanitizer. |
| 117 |
*/ |
| 118 |
public static function compile_name( string $pattern ): string { |
| 119 |
$cache_key = "\0name:" . $pattern; |
| 120 |
if ( isset( self::$compiled[ $cache_key ] ) ) { |
| 121 |
return self::$compiled[ $cache_key ]; |
| 122 |
} |
| 123 |
|
| 124 |
if ( '' !== $pattern && '~' === $pattern[0] ) { |
| 125 |
// Anchored, unlike URL mode: an unanchored `~utm_[a-z0-9_-]+` |
| 126 |
// still contains-matches `my_utm_source`, which is the very |
| 127 |
// over-match name mode exists to stop. |
| 128 |
$regex = self::compile_regex( substr( $pattern, 1 ), true ); |
| 129 |
} elseif ( preg_match( '/(?<!\\\\)[*?\[]/', $pattern ) ) { |
| 130 |
$regex = '#^' . self::glob_to_regex( $pattern ) . '$#i'; |
| 131 |
} else { |
| 132 |
// Bare pattern → exact name. Resolve `\X` → `X` first so an |
| 133 |
// escaped metacharacter matches itself. |
| 134 |
$resolved = preg_replace( '/\\\\(.)/', '$1', $pattern ); |
| 135 |
$regex = '#^' . preg_quote( (string) $resolved, '#' ) . '$#i'; |
| 136 |
} |
| 137 |
|
| 138 |
self::$compiled[ $cache_key ] = $regex; |
| 139 |
return $regex; |
| 140 |
} |
| 141 |
|
| 142 |
/** |
| 143 |
* Compile a user pattern to a PCRE delimited with `#`. Cached |
| 144 |
* for the request lifetime. |
| 145 |
*/ |
| 146 |
public static function compile( string $pattern ): string { |
| 147 |
if ( isset( self::$compiled[ $pattern ] ) ) { |
| 148 |
return self::$compiled[ $pattern ]; |
| 149 |
} |
| 150 |
// Raw-regex mode: a leading `~` marks the rest as a PCRE pattern. |
| 151 |
// We validate it once and store either the usable regex or a |
| 152 |
// never-matching sentinel, so a malformed user pattern degrades to |
| 153 |
// "matches nothing" instead of fataling or matching everything. |
| 154 |
if ( '' !== $pattern && '~' === $pattern[0] ) { |
| 155 |
$regex = self::compile_regex( substr( $pattern, 1 ) ); |
| 156 |
self::$compiled[ $pattern ] = $regex; |
| 157 |
return $regex; |
| 158 |
} |
| 159 |
// Decide mode based on UNESCAPED glob metacharacters only. |
| 160 |
// `\*` alone keeps the pattern in substring mode (with escapes |
| 161 |
// resolved); `/cart/*` flips to anchored glob mode. |
| 162 |
$has_unescaped_glob = (bool) preg_match( '/(?<!\\\\)[*?\[]/', $pattern ); |
| 163 |
if ( $has_unescaped_glob ) { |
| 164 |
$regex = '#^' . self::glob_to_regex( $pattern ) . '$#'; |
| 165 |
} else { |
| 166 |
// Substring "contains" mode. Resolve `\X` → `X` first so |
| 167 |
// `\*` matches a literal asterisk anywhere in the subject. |
| 168 |
$resolved = preg_replace( '/\\\\(.)/', '$1', $pattern ); |
| 169 |
$regex = '#' . preg_quote( (string) $resolved, '#' ) . '#'; |
| 170 |
} |
| 171 |
self::$compiled[ $pattern ] = $regex; |
| 172 |
return $regex; |
| 173 |
} |
| 174 |
|
| 175 |
/** |
| 176 |
* A delimited PCRE that can never match any input — used as the safe |
| 177 |
* fallback for invalid / over-long user regex patterns. `(?!)` is the |
| 178 |
* empty negative lookahead: it fails at every position. |
| 179 |
*/ |
| 180 |
private const NEVER = '#(?!)#'; |
| 181 |
|
| 182 |
/** |
| 183 |
* Validate + delimit a user-supplied raw regex (the part after `~`). |
| 184 |
* Returns a `#…#` delimited PCRE (unanchored, so it matches like a |
| 185 |
* "contains" regex — anchored and case-insensitive in name mode), or the |
| 186 |
* NEVER sentinel when the pattern is empty, too long, or not a valid |
| 187 |
* PCRE. We never let a bad pattern through: a regex that errors at match |
| 188 |
* time would otherwise emit warnings on every cached request. |
| 189 |
* |
| 190 |
* @param string $body The pattern after the `~` marker. |
| 191 |
* @param bool $anchored Wrap in `^(?:…)$` and match case-insensitively. |
| 192 |
*/ |
| 193 |
private static function compile_regex( string $body, bool $anchored = false ): string { |
| 194 |
// Cap length to keep compile + match cheap and bound backtracking |
| 195 |
// exposure from pathological user input. |
| 196 |
if ( '' === $body || strlen( $body ) > 200 ) { |
| 197 |
return self::NEVER; |
| 198 |
} |
| 199 |
$escaped = str_replace( '#', '\\#', $body ); |
| 200 |
// Group before anchoring so a top-level alternation (`a|b`) anchors |
| 201 |
// as a whole rather than binding `^` to the first branch only. |
| 202 |
$regex = $anchored ? '#^(?:' . $escaped . ')$#i' : '#' . $escaped . '#'; |
| 203 |
// Validate by compiling against an empty subject. preg_match returns |
| 204 |
// false on a malformed pattern; suppress the warning it emits. |
| 205 |
// phpcs:ignore WordPress.PHP.NoSilencedErrors.Discouraged -- intentional: invalid user regex must degrade to never-match, not warn. |
| 206 |
if ( false === @preg_match( $regex, '' ) ) { |
| 207 |
return self::NEVER; |
| 208 |
} |
| 209 |
return $regex; |
| 210 |
} |
| 211 |
|
| 212 |
/** |
| 213 |
* Translate glob syntax → regex body (no delimiters, no anchors). |
| 214 |
* Mirrors fnmatch's FNM_PATHNAME-disabled semantics: `*` matches |
| 215 |
* across `/` so `/cart/*` correctly catches `/cart/items/sub`. |
| 216 |
*/ |
| 217 |
private static function glob_to_regex( string $glob ): string { |
| 218 |
$out = ''; |
| 219 |
$in_class = false; |
| 220 |
$len = strlen( $glob ); |
| 221 |
$escape = false; |
| 222 |
|
| 223 |
for ( $i = 0; $i < $len; $i++ ) { |
| 224 |
$ch = $glob[ $i ]; |
| 225 |
|
| 226 |
if ( $escape ) { |
| 227 |
$out .= preg_quote( $ch, '#' ); |
| 228 |
$escape = false; |
| 229 |
continue; |
| 230 |
} |
| 231 |
|
| 232 |
if ( '\\' === $ch ) { |
| 233 |
$escape = true; |
| 234 |
continue; |
| 235 |
} |
| 236 |
|
| 237 |
if ( $in_class ) { |
| 238 |
if ( ']' === $ch ) { |
| 239 |
$out .= ']'; |
| 240 |
$in_class = false; |
| 241 |
} else { |
| 242 |
// Inside a character class, dash + letters are passed |
| 243 |
// through; we still preg_quote dangerous chars. |
| 244 |
$out .= preg_quote( $ch, '#' ); |
| 245 |
} |
| 246 |
continue; |
| 247 |
} |
| 248 |
|
| 249 |
switch ( $ch ) { |
| 250 |
case '*': |
| 251 |
$out .= '.*'; |
| 252 |
break; |
| 253 |
case '?': |
| 254 |
$out .= '.'; |
| 255 |
break; |
| 256 |
case '[': |
| 257 |
$out .= '['; |
| 258 |
$in_class = true; |
| 259 |
break; |
| 260 |
default: |
| 261 |
$out .= preg_quote( $ch, '#' ); |
| 262 |
} |
| 263 |
} |
| 264 |
|
| 265 |
return $out; |
| 266 |
} |
| 267 |
|
| 268 |
/** |
| 269 |
* Test-only: clear the compile cache. Production code never needs |
| 270 |
* this (PHP request lifetime handles it). |
| 271 |
*/ |
| 272 |
public static function reset_cache(): void { |
| 273 |
self::$compiled = array(); |
| 274 |
} |
| 275 |
} |
| 276 |
|
