Whitelist HTML sanitizing with PHP

The following is a single class written to perform comprehensive HTML input filtering with minimal dependencies (basically only Tidy) and should work in PHP 5.3+. This will be included in my forum script as the default filter.

This version captures URL encoded XSS attempts with deep attribute inspection (to a decoding depth of 6 by default) as well as scrubbing all non-whitelisted attributes, tags and conversion of surviving attribute data into HTML entities.

In addition, it will attempt to capture directory traversal attempts ( ../ or \\ or /~/ etc… ) which may give access to restricted areas of a site. Your web server should deny access to these URLs by default, however that won’t stop someone from posting links pointing elsewhere. This will reduce your liability should such a link be included in your site content by a user.

You can post sourcecode within <code> tags and it will be encoded by default.

<?php

/**
 * HTML parsing, filtering and sanitization
 * This class depends on Tidy which is included in the core since PHP 5.3
 *
 * @author Eksith Rodrigo <reksith at gmail.com>
 * @license http://opensource.org/licenses/ISC ISC License
 * @version 0.2
 */

class Html {
	
	/**
	 * @var array HTML filtering options
	 */
	public static $options = array( 
		'rx_url'	=> // URLs over 255 chars can cause problems
			'~^(http|ftp)(s)?\:\/\/((([a-z|0-9|\-]{1,25})(\.)?){2,7})($|/.*$){4,255}$~i',
		
		'rx_js'		=> // Questionable attributes
			'/((java)?script|eval|document)/ism',
		
		'rx_xss'	=> // XSS (<style> can also be a vector. Stupid IE 6!)
			'/(<(s(?:cript|tyle)).*?)/ism',
		
		'rx_xss2'	=> // More potential XSS
			'/(document\.|window\.|eval\(|\(\))/ism',
		
		'rx_esc'	=> // Directory traversal/escaping/injection
			'/(\\~\/|\.\.|\\\\|\-\-)/sm'	,
		
		'scrub_depth'	=> 6, // URL Decoding depth (fails on exceeding this)
		
		'nofollow'	=> true // Set rel='nofollow' on all links

	);
	
	/**
	 * @var array List of HTML Tidy output settings
	 * @link http://tidy.sourceforge.net/docs/quickref.html
	 */
	private static $tidy = array(
		// Preserve whitespace inside tags
		'add-xml-space'			=> true,
		
		// Remove proprietary markup (E.G. og:tags)
		'bare'				=> true,
		
		// More proprietary markup
		'drop-proprietary-attributes'	=> true,
		
		// Remove blank (E.G. <p></p>) paragraphs
		'drop-empty-paras'		=> true,
		
		// Wraps bare text in <p> tags
		'enclose-text'			=> true,
		
		// Removes illegal/invalid characters in URIs
		'fix-uri'			=> true,
		
		// Removes <!-- Comments -->
		'hide-comments'			=> true,
		
		// Removing indentation saves storage space
		'indent'			=> false,
		
		// Combine individual formatting styles
		'join-styles'			=> true,
		
		// Converts <i> to <em> & <b> to <strong>
		'logical-emphasis'		=> true,
		
		// Byte Order Mark isn't really needed
		'output-bom'			=> false,
		
		// Ensure UTF-8 characters are preserved
		'output-encoding'		=> 'utf8',
		
		// W3C standards compliant markup
		'output-xhtml'			=> true,
		
		// Had some unexpected behavior with this
		//'markup'			=> true,

		// Merge multiple <span> tags into one		
		'merge-spans'			=> true,
		
		// Only outputs <body> (<head> etc... not needed)
		'show-body-only'		=> true,
		
		// Removing empty lines saves storage
		'vertical-space'		=> false,
		
		// Wrapping tags not needed (saves bandwidth)
		'wrap'				=> 0
	);
	
	
	/**
	 * @var array Whitelist of tags. Trim or expand these as necessary
	 * @example 'tag' => array( of, allowed, attributes )
	 */
	private static $whitelist = array(
		'p'		=> array( 'style', 'class', 'align' ),
		'div'		=> array( 'style', 'class', 'align' ),
		'span'		=> array( 'style', 'class' ),
		'br'		=> array( 'style', 'class' ),
		'hr'		=> array( 'style', 'class' ),
		
		'h1'		=> array( 'style', 'class' ),
		'h2'		=> array( 'style', 'class' ),
		'h3'		=> array( 'style', 'class' ),
		'h4'		=> array( 'style', 'class' ),
		'h5'		=> array( 'style', 'class' ),
		'h6'		=> array( 'style', 'class' ),
		
		'strong'	=> array( 'style', 'class' ),
		'em'		=> array( 'style', 'class' ),
		'u'		=> array( 'style', 'class' ),
		'strike'	=> array( 'style', 'class' ),
		'del'		=> array( 'style', 'class' ),
		'ol'		=> array( 'style', 'class' ),
		'ul'		=> array( 'style', 'class' ),
		'li'		=> array( 'style', 'class' ),
		'code'		=> array( 'style', 'class' ),
		'pre'		=> array( 'style', 'class' ),
		
		'sup'		=> array( 'style', 'class' ),
		'sub'		=> array( 'style', 'class' ),
		
		// Took out 'rel' and 'title', because we're using those below
		'a'		=> array( 'style', 'class', 'href' ),
		
		'img'		=> array( 'style', 'class', 'src', 'height', 
					  'width', 'alt', 'longdesc', 'title', 
					  'hspace', 'vspace' ),
		
		'table'		=> array( 'style', 'class', 'border-collapse', 
					  'cellspacing', 'cellpadding' ),
					
		'thead'		=> array( 'style', 'class' ),
		'tbody'		=> array( 'style', 'class' ),
		'tfoot'		=> array( 'style', 'class' ),
		'tr'		=> array( 'style', 'class' ),
		'td'		=> array( 'style', 'class', 
					'colspan', 'rowspan' ),
		'th'		=> array( 'style', 'class', 'scope', 'colspan', 
					  'rowspan' ),
		
		'q'		=> array( 'style', 'class', 'cite' ),
		'cite'		=> array( 'style', 'class' ),
		'abbr'		=> array( 'style', 'class' ),
		'blockquote'	=> array( 'style', 'class' ),
		
		// Stripped out
		'body'		=> array()
	);
	
	
	
	/**#@+
	 * HTML Filtering
	 */
	
	
	/**
	 * Convert content between code blocks into code tags
	 * 
	 * @param $val string Value to encode to entities
	 */
	protected function escapeCode( $val ) {
		
		if ( is_array( $val ) ) {
			$out = self::entities( $val[1] );
			return '<code>' . $out . '</code>';
		}
		
	}
	
	
	/**
	 * Convert an unformatted text block to paragraphs
	 * 
	 * @link http://stackoverflow.com/a/2959926
	 * @param $val string Filter variable
	 */
	protected function makeParagraphs( $val ) {
		
		/**
		 * Convert newlines to linebreaks first
		 * This is why PHP both sucks and is awesome at the same time
		 */
		$out = nl2br( $val );
		
		/**
		 * Turn consecutive <br>s to paragraph breaks and wrap the 
		 * whole thing in a paragraph
		 */
		$out = '<p>' . preg_replace('#(?:<br\s*/?>\s*?){2,}#', 
			'<p></p><p>', $out ) . '</p>';
		
		/**
		 * Remove <br> abnormalities
		 */
		$out = preg_replace( '#<p>(\s*<br\s*/?>)+#', '</p><p>', $out );
		$out = preg_replace( '#<br\s*/?>(\s*</p>)+#', '<p></p>', $out );
		
		return $out;
	}
	
	
	/**
	 * Filters HTML content through whitelist of tags and attributes
	 * 
	 * @param $val string Value filter
	 */
	public function filter( $val ) {
		
		if ( !isset( $val ) || empty( $val ) ) {
			return '';
		}
		
		/**
		 * Escape the content of any code blocks before we parse HTML or 
		 * they will get stripped
		 */
		$out	= preg_replace_callback( "/\<code\>(.*)\<\/code\>/imu", 
				array( $this, 'escapeCode' ) , $val
			);
		
		/**
		 * Convert to paragraphs and begin
		 */
		$out	= $this->makeParagraphs( $out );
		$dom	= new DOMDocument();
		
		/**
		 * Hide parse warnings since we'll be cleaning the output anyway
		 */
		$err	= libxml_use_internal_errors( true );
		
		$dom->loadHTML( $out );
		$dom->encoding = 'utf-8';
		
		$body	= $dom->getElementsByTagName( 'body' )->item( 0 );
		$this->cleanNodes( $body, $badTags );
		
		/**
		 * Iterate through bad tags found above and convert them to 
		 * harmless text
		 */
		foreach ( $badTags as $node ) {
			if( $node->nodeName != "#text" ) {
				$ctext = $dom->createTextNode( 
						$dom->saveHTML( $node )
					);
				$node->parentNode->replaceChild( 
					$ctext, $node 
				);
			}
		}
		
		
		/**
		 * Filter the junk and return only the contents of the body tag
		 */
		$out = tidy_repair_string( 
				$dom->saveHTML( $body ), 
				self::$tidy
			);
		
		
		/**
		 * Reset errors
		 */
		libxml_clear_errors();
		libxml_use_internal_errors( $err );
		
		return $out;
	}
	
	
	protected function cleanAttributeNode( 
		&$node, 
		&$attr, 
		&$goodAttributes, 
		&$href 
	) {
		/**
		 * Why the devil is an attribute name called "nodeName"?!
		 */
		$name = $attr->nodeName;
		
		/**
		 * And an attribute value is still "nodeValue"?? Damn you PHP!
		 */
		$val = $attr->nodeValue;
		
		/**
		 * Default action is to remove the attribute completely
		 * It's reinstated only if it's allowed and only after 
		 * it's filtered
		 */
		$node->removeAttributeNode( $attr );
		
		if ( in_array( $name, $goodAttributes ) ) {
			
			switch ( $name ) {
				
				/**
				 * Validate URL attribute types
				 */
				case 'url':
				case 'src':
				case 'href':
				case 'longdesc':
					if ( self::urlFilter( $val ) ) {
						$href = $val;
					} else {
						$val = '';
					}
					break;
				
				/**
				 * Everything else gets default scrubbing
				 */
				default:
					if ( self::decodeScrub( $val ) ) {
						$val = self::entities( $val );
					} else {
						$val = '';
					}
			}
			
			if ( '' !== $val ) {
				$node->setAttribute( $name, $val );
			}
		}
	}
	
	
	/**
	 * Modify links to display their domains and add 'nofollow'.
	 * Also puts the linked domain in the title as well as the file name
	 */
	protected static function linkAttributes( &$node, $href ) {
		try {
			if ( !self::$options['nofollow'] ) {
				return;
			}
			
			$parsed	= parse_url( $href );
			$title	= $parsed['host'] . ' ';
			
			$f	= pathinfo( $parsed['path'] );
			$title	.= ' ( /' . $f['basename'] . ' ) ';
				
			$node->setAttribute( 
				'title', $title
			);
			
			if ( self::$options['nofollow'] ) {
				$node->setAttribute(
					'rel', 'nofollow'
				);
			}
			
		} catch ( Exception $e ) { }
	}
	
	
	/**
	 * Iterate through each tag and add non-whitelisted tags to the 
	 * bad list. Also filter the attributes and remove non-whitelisted ones.
	 * 
	 * @param htmlNode $node Current HTML node
	 * @param array $badTags Cumulative list of tags for deletion
	 */
	protected function cleanNodes( $node, &$badTags = array() ) {
		
		if ( array_key_exists( $node->nodeName, self::$whitelist ) ) {
			
			if ( $node->hasAttributes() ) {
				
				/**
				 * Prepare for href attribute which gets special 
				 * treatment
				 */
				$href = '';
				
				/**
				 * Filter through attribute whitelist for this 
				 * tag
				 */
				$goodAttributes = 
					self::$whitelist[$node->nodeName];
				
				
				/**
				 * Check out each attribute in this tag
				 */
				foreach ( 
					iterator_to_array( $node->attributes ) 
					as $attr ) {
					$this->cleanAttributeNode( 
						$node, $attr, $goodAttributes, 
						$href
					);
				}
				
				/**
				 * This is a link. Treat it accordingly
				 */
				if ( 'a' === $node->nodeName && '' !== $href ) {
					self::linkAttributes( $node, $href );
				}
				
			} // End if( $node->hasAttributes() )
			
			/**
			 * If we have childnodes, recursively call cleanNodes 
			 * on those as well
			 */
			if ( $node->childNodes ) {
				foreach ( $node->childNodes as $child ) {
					$this->cleanNodes( $child, $badTags );
				}
			}
			
		} else {
			
			/**
			 * Not in whitelist so no need to check its child nodes. 
			 * Simply add to array of nodes pending deletion.
			 */
			$badTags[] = $node;
			
		} // End if array_key_exists( $node->nodeName, self::$whitelist )
		
	}
	
	/**#@-*/
	
	
	/**
	 * Returns true if the URL passed value is harmless.
	 * This regex takes into account Unicode domain names however, it 
	 * doesn't check for TLD (.com, .net, .mobi, .museum etc...) as that 
	 * list is too long.
	 * The purpose is to ensure your visitors are not harmed by invalid 
	 * markup, not that they get a functional domain name.
	 * 
	 * @param string $v Raw URL to validate
	 * @returns boolean
	 */
	public static function urlFilter( $v ) {
		
		$v = strtolower( $v );
		$out = false;
		
		if ( filter_var( $v, 
			FILTER_VALIDATE_URL, FILTER_FLAG_SCHEME_REQUIRED ) ) {
			
			/**
			 * PHP's native filter isn't restrictive enough.
			 */
			if ( preg_match( self::$options['rx_url'], $v ) ) {
				$out = true;
			} else {
				$out = false;
			}
			
			if ( $out ) {
				$out = self::decodeScrub( $v );
			}
		} else {
			$out = false;
		}
		
		return $out;
	}
	
	
	/**
	 * Regular expressions don't work well when used for validating HTML.
	 * It really shines when evaluating text so that's what we're doing here
	 * 
	 * @param string $v string Attribute name
	 * @param int $depth Number of times to URL decode
	 * @returns boolean True if nothing unsavory was found.
	 */
	public static function decodeScrub( $v ) {
		if ( empty( $v ) ) {
			return true;
		}
		
		$depth		= self::$options['scrub_depth'];
		$i		= 1;
		$success	= false;
		$old		= '';
		
		
		while( $i <= $depth && !empty( $v ) ) {
			// Check for any JS and other shenanigans
			if (
				preg_match( self::$options['rx_xss'], $v ) || 
				preg_match( self::$options['rx_xss2'], $v ) || 
				preg_match( self::$options['rx_esc'], $v )
			) {
				$success = false;
				break;
			} else {
				$old	= $v;
				$v	= self::utfdecode( $v );
				
				/**
				 * We found the the lowest decode level.
				 * No need to continue decoding.
				 */
				if ( $old === $v ) {
					$success = true;
					break;
				}
			}
			
			$i++;
		}
		
		
		/**
		 * If after decoding a number times, we still couldn't get to 
		 * the original string, then there's something still wrong
		 */
		if ( $old !== $v && $i === $depth ) {
			return false;
		}
		
		return $success;
	}
	
	
	/**
	 * UTF-8 compatible URL decoding
	 * 
	 * @link http://www.php.net/manual/en/function.urldecode.php#79595
	 * @returns string
	 */
	public static function utfdecode( $v ) {
		$v = urldecode( $v );
		$v = preg_replace( '/%u([0-9a-f]{3,4})/i', '&#x\\1;', $v );
		return html_entity_decode( $v, null, 'UTF-8' );
	}
	
	
	/**
	 * HTML safe character entitites in UTF-8
	 * 
	 * @returns string
	 */
	public static function entities( $v ) {
		return htmlentities( 
			iconv( 'UTF-8', 'UTF-8', $v ), 
			ENT_NOQUOTES | ENT_SUBSTITUTE, 
			'UTF-8'
		);
	}	
}

Usage is pretty simple:

$data = $_POST['body'];
$html = new Html();
$data = $html->filter( $data );
Advertisement