mediawiki-extensions-AbuseF.../AbuseFilterTokenizer.php
Ori Livneh 907de9a26c Increase AbuseFilterTokenizer cache TTL from 10m to 1d
AbuseFilterTokenizer is referentially transparent -- the mapping of input to
outputs does not vary on nonlocal state. So the cache TTL can be much longer.

Change-Id: I9e6ec4347dbb940c3d73538d550a0f045706264c
2015-10-24 10:30:24 -07:00

252 lines
6.4 KiB
PHP

<?php
/**
* Tokenizer for AbuseFilter rules.
*/
class AbuseFilterTokenizer {
/** @var int Tokenizer cache version. Increment this when changing the syntax. **/
const CACHE_VERSION = 1;
const COMMENT_START_RE = '/\s*\/\*/A';
const ID_SYMBOL_RE = '/[0-9A-Za-z_]+/A';
const OPERATOR_RE = '/(\!\=\=|\!\=|\!|\*\*|\*|\/|\+|\-|%|&|\||\^|\:\=|\?|\:|\<\=|\<|\>\=|\>|\=\=\=|\=\=|\=)/A';
const RADIX_RE = '/([0-9A-Fa-f]+(?:\.\d*)?|\.\d+)([bxo])?/Au';
const WHITESPACE = "\011\012\013\014\015\040";
// Order is important. The punctuation-matching regex requires that
// ** comes before *, etc. They are sorted to make it easy to spot
// such errors.
static $operators = array(
'!==', '!=', '!', // Inequality
'**', '*', // Multiplication/exponentiation
'/', '+', '-', '%', // Other arithmetic
'&', '|', '^', // Logic
':=', // Setting
'?', ':', // Ternery
'<=', '<', // Less than
'>=', '>', // Greater than
'===', '==', '=', // Equality
);
static $punctuation = array(
',' => AFPToken::TComma,
'(' => AFPToken::TBrace,
')' => AFPToken::TBrace,
'[' => AFPToken::TSquareBracket,
']' => AFPToken::TSquareBracket,
';' => AFPToken::TStatementSeparator,
);
static $bases = array(
'b' => 2,
'x' => 16,
'o' => 8
);
static $baseCharsRe = array(
2 => '/^[01]+$/',
8 => '/^[0-8]+$/',
16 => '/^[0-9A-Fa-f]+$/',
10 => '/^[0-9.]+$/',
);
static $keywords = array(
'in', 'like', 'true', 'false', 'null', 'contains', 'matches',
'rlike', 'irlike', 'regex', 'if', 'then', 'else', 'end',
);
/**
* @param string $code
* @return array
* @throws AFPException
* @throws AFPUserVisibleException
*/
static function tokenize( $code ) {
static $tokenizerCache = null;
if ( !$tokenizerCache ) {
$tokenizerCache = ObjectCache::newAccelerator( array(), 'hash' );
}
static $stats = null;
if ( !$stats ) {
$stats = RequestContext::getMain()->getStats();
}
$cacheKey = wfGlobalCacheKey( __CLASS__, self::CACHE_VERSION, crc32( $code ) );
$tokens = $tokenizerCache->get( $cacheKey );
if ( $tokens ) {
$stats->increment( 'AbuseFilter.tokenizerCache.hit' );
return $tokens;
}
$stats->increment( 'AbuseFilter.tokenizerCache.miss' );
$tokens = array();
$curPos = 0;
do {
$prevPos = $curPos;
$token = self::nextToken( $code, $curPos );
$tokens[ $token->pos ] = array( $token, $curPos );
} while ( $curPos !== $prevPos );
$tokenizerCache->set( $cacheKey, $tokens, 60 * 60 * 24 );
return $tokens;
}
/**
* @param string $code
* @param integer &$offset
* @return AFPToken
* @throws AFPException
* @throws AFPUserVisibleException
*/
protected static function nextToken( $code, &$offset ) {
$matches = array();
$start = $offset;
// Read past comments
while ( preg_match( self::COMMENT_START_RE, $code, $matches, 0, $offset ) ) {
$offset = strpos( $code, '*/', $offset ) + 2;
}
// Spaces
$offset += strspn( $code, self::WHITESPACE, $offset );
if ( $offset >= strlen( $code ) ) {
return new AFPToken( AFPToken::TNone, '', $start );
}
$chr = $code[$offset];
// Punctuation
if ( isset( self::$punctuation[$chr] ) ) {
$offset++;
return new AFPToken( self::$punctuation[$chr], $chr, $start );
}
// String literal
if ( $chr === '"' || $chr === "'" ) {
return self::readStringLiteral( $code, $offset, $start );
}
$matches = array();
// Operators
if ( preg_match( self::OPERATOR_RE, $code, $matches, 0, $offset ) ) {
$token = $matches[0];
$offset += strlen( $token );
return new AFPToken( AFPToken::TOp, $token, $start );
}
// Numbers
if ( preg_match( self::RADIX_RE, $code, $matches, 0, $offset ) ) {
$token = $matches[0];
$input = $matches[1];
$baseChar = isset( $matches[2] ) ? $matches[2] : null;
// Sometimes the base char gets mixed in with the rest of it because
// the regex targets hex, too.
// This mostly happens with binary
if ( !$baseChar && !empty( self::$bases[ substr( $input, - 1 ) ] ) ) {
$baseChar = substr( $input, - 1, 1 );
$input = substr( $input, 0, - 1 );
}
$base = $baseChar ? self::$bases[$baseChar] : 10;
// Check against the appropriate character class for input validation
if ( preg_match( self::$baseCharsRe[$base], $input ) ) {
$num = $base !== 10 ? base_convert( $input, $base, 10 ) : $input;
$offset += strlen( $token );
return ( strpos( $input, '.' ) !== false )
? new AFPToken( AFPToken::TFloat, floatval( $num ), $start )
: new AFPToken( AFPToken::TInt, intval( $num ), $start );
}
}
// IDs / Keywords
if ( preg_match( self::ID_SYMBOL_RE, $code, $matches, 0, $offset ) ) {
$token = $matches[0];
$offset += strlen( $token );
$type = in_array( $token, self::$keywords )
? AFPToken::TKeyword
: AFPToken::TID;
return new AFPToken( $type, $token, $start );
}
throw new AFPUserVisibleException(
'unrecognisedtoken', $start, array( substr( $code, $start ) ) );
}
/**
* @param string $code
* @param int &$offset
* @param int $start
* @return AFPToken
* @throws AFPException
* @throws AFPUserVisibleException
*/
protected static function readStringLiteral( $code, &$offset, $start ) {
$type = $code[$offset];
$offset++;
$length = strlen( $code );
$token = '';
while ( $offset < $length ) {
if ( $code[$offset] === $type ) {
$offset++;
return new AFPToken( AFPToken::TString, $token, $start );
}
// Performance: Use a PHP function (implemented in C)
// to scan ahead.
$addLength = strcspn( $code, $type . "\\", $offset );
if ( $addLength ) {
$token .= substr( $code, $offset, $addLength );
$offset += $addLength;
} elseif ( $code[$offset] == '\\' ) {
switch( $code[$offset + 1] ) {
case '\\':
$token .= '\\';
break;
case $type:
$token .= $type;
break;
case 'n';
$token .= "\n";
break;
case 'r':
$token .= "\r";
break;
case 't':
$token .= "\t";
break;
case 'x':
$chr = substr( $code, $offset + 2, 2 );
if ( preg_match( '/^[0-9A-Fa-f]{2}$/', $chr ) ) {
$chr = base_convert( $chr, 16, 10 );
$token .= chr( $chr );
$offset += 2; # \xXX -- 2 done later
} else {
$token .= 'x';
}
break;
default:
$token .= "\\" . $code[$offset + 1];
}
$offset += 2;
} else {
$token .= $code[$offset];
$offset++;
}
}
throw new AFPUserVisibleException( 'unclosedstring', $offset, array() );
}
}