Merge branch 'w15_MDL-25826_m21_htmlpurifier' of git://github.com/sko…

…dak/moodle

lib/htmlpurifier/HTMLPurifier.php

@@ -19,7 +19,7 @@
  */
 
 /*
-    HTML Purifier 4.2.0 - Standards Compliant HTML Filtering
+    HTML Purifier 4.3.0 - Standards Compliant HTML Filtering
     Copyright (C) 2006-2008 Edward Z. Yang
 
     This library is free software; you can redistribute it and/or
@@ -55,10 +55,10 @@ class HTMLPurifier
 {
 
     /** Version of HTML Purifier */
-    public $version = '4.2.0';
+    public $version = '4.3.0';
 
     /** Constant with version of HTML Purifier */
-    const VERSION = '4.2.0';
+    const VERSION = '4.3.0';
 
     /** Global configuration object */
     public $config;

lib/htmlpurifier/HTMLPurifier.safe-includes.php

@@ -119,6 +119,7 @@
 require_once $__dir . '/HTMLPurifier/AttrTransform/Length.php';
 require_once $__dir . '/HTMLPurifier/AttrTransform/Name.php';
 require_once $__dir . '/HTMLPurifier/AttrTransform/NameSync.php';
+require_once $__dir . '/HTMLPurifier/AttrTransform/Nofollow.php';
 require_once $__dir . '/HTMLPurifier/AttrTransform/SafeEmbed.php';
 require_once $__dir . '/HTMLPurifier/AttrTransform/SafeObject.php';
 require_once $__dir . '/HTMLPurifier/AttrTransform/SafeParam.php';
@@ -145,6 +146,7 @@
 require_once $__dir . '/HTMLPurifier/HTMLModule/Legacy.php';
 require_once $__dir . '/HTMLPurifier/HTMLModule/List.php';
 require_once $__dir . '/HTMLPurifier/HTMLModule/Name.php';
+require_once $__dir . '/HTMLPurifier/HTMLModule/Nofollow.php';
 require_once $__dir . '/HTMLPurifier/HTMLModule/NonXMLCommonAttributes.php';
 require_once $__dir . '/HTMLPurifier/HTMLModule/Object.php';
 require_once $__dir . '/HTMLPurifier/HTMLModule/Presentation.php';

lib/htmlpurifier/HTMLPurifier/AttrDef/CSS/FontFamily.php

@@ -2,11 +2,43 @@
 
 /**
  * Validates a font family list according to CSS spec
- * @todo whitelisting allowed fonts would be nice
  */
 class HTMLPurifier_AttrDef_CSS_FontFamily extends HTMLPurifier_AttrDef
 {
 
+    protected $mask = null;
+
+    public function __construct() {
+        $this->mask = '- ';
+        for ($c = 'a'; $c <= 'z'; $c++) $this->mask .= $c;
+        for ($c = 'A'; $c <= 'Z'; $c++) $this->mask .= $c;
+        for ($c = '0'; $c <= '9'; $c++) $this->mask .= $c; // cast-y, but should be fine
+        // special bytes used by UTF-8
+        for ($i = 0x80; $i <= 0xFF; $i++) {
+            // We don't bother excluding invalid bytes in this range,
+            // because the our restriction of well-formed UTF-8 will
+            // prevent these from ever occurring.
+            $this->mask .= chr($i);
+        }
+
+        /*
+            PHP's internal strcspn implementation is
+            O(length of string * length of mask), making it inefficient
+            for large masks.  However, it's still faster than
+            preg_match 8)
+          for (p = s1;;) {
+            spanp = s2;
+            do {
+              if (*spanp == c || p == s1_end) {
+                return p - s1;
+              }
+            } while (spanp++ < (s2_end - 1));
+            c = *++p;
+          }
+         */
+        // possible optimization: invert the mask.
+    }
+
     public function validate($string, $config, $context) {
         static $generic_names = array(
             'serif' => true,
@@ -15,6 +47,7 @@ public function validate($string, $config, $context) {
             'fantasy' => true,
             'cursive' => true
         );
+        $allowed_fonts = $config->get('CSS.AllowedFonts');
 
         // assume that no font names contain commas in them
         $fonts = explode(',', $string);
@@ -24,7 +57,9 @@ public function validate($string, $config, $context) {
             if ($font === '') continue;
             // match a generic name
             if (isset($generic_names[$font])) {
-                $final .= $font . ', ';
+                if ($allowed_fonts === null || isset($allowed_fonts[$font])) {
+                    $final .= $font . ', ';
+                }
                 continue;
             }
             // match a quoted name
@@ -40,6 +75,10 @@ public function validate($string, $config, $context) {
 
             // $font is a pure representation of the font name
 
+            if ($allowed_fonts !== null && !isset($allowed_fonts[$font])) {
+                continue;
+            }
+
             if (ctype_alnum($font) && $font !== '') {
                 // very simple font, allow it in unharmed
                 $final .= $font . ', ';
@@ -50,17 +89,103 @@ public function validate($string, $config, $context) {
             // shouldn't show up regardless
             $font = str_replace(array("\n", "\t", "\r", "\x0C"), ' ', $font);
 
-            // These ugly transforms don't pose a security
-            // risk (as \\ and \" might).  We could try to be clever and
-            // use single-quote wrapping when there is a double quote
-            // present, but I have choosen not to implement that.
-            // (warning: this code relies on the selection of quotation
-            // mark below)
-            $font = str_replace('\\', '\\5C ', $font);
-            $font = str_replace('"',  '\\22 ', $font);
-
-            // complicated font, requires quoting
-            $final .= "\"$font\", "; // note that this will later get turned into &quot;
+            // Here, there are various classes of characters which need
+            // to be treated differently:
+            //  - Alphanumeric characters are essentially safe.  We
+            //    handled these above.
+            //  - Spaces require quoting, though most parsers will do
+            //    the right thing if there aren't any characters that
+            //    can be misinterpreted
+            //  - Dashes rarely occur, but they fairly unproblematic
+            //    for parsing/rendering purposes.
+            //  The above characters cover the majority of Western font
+            //  names.
+            //  - Arbitrary Unicode characters not in ASCII.  Because
+            //    most parsers give little thought to Unicode, treatment
+            //    of these codepoints is basically uniform, even for
+            //    punctuation-like codepoints.  These characters can
+            //    show up in non-Western pages and are supported by most
+            //    major browsers, for example: "ＭＳ 明朝" is a
+            //    legitimate font-name
+            //    <http://ja.wikipedia.org/wiki/MS_明朝>.  See
+            //    the CSS3 spec for more examples:
+            //    <http://www.w3.org/TR/2011/WD-css3-fonts-20110324/localizedfamilynames.png>
+            //    You can see live samples of these on the Internet:
+            //    <http://www.google.co.jp/search?q=font-family+ＭＳ+明朝|ゴシック>
+            //    However, most of these fonts have ASCII equivalents:
+            //    for example, 'MS Mincho', and it's considered
+            //    professional to use ASCII font names instead of
+            //    Unicode font names.  Thanks Takeshi Terada for
+            //    providing this information.
+            //  The following characters, to my knowledge, have not been
+            //  used to name font names.
+            //  - Single quote.  While theoretically you might find a
+            //    font name that has a single quote in its name (serving
+            //    as an apostrophe, e.g. Dave's Scribble), I haven't
+            //    been able to find any actual examples of this.
+            //    Internet Explorer's cssText translation (which I
+            //    believe is invoked by innerHTML) normalizes any
+            //    quoting to single quotes, and fails to escape single
+            //    quotes.  (Note that this is not IE's behavior for all
+            //    CSS properties, just some sort of special casing for
+            //    font-family).  So a single quote *cannot* be used
+            //    safely in the font-family context if there will be an
+            //    innerHTML/cssText translation.  Note that Firefox 3.x
+            //    does this too.
+            //  - Double quote.  In IE, these get normalized to
+            //    single-quotes, no matter what the encoding.  (Fun
+            //    fact, in IE8, the 'content' CSS property gained
+            //    support, where they special cased to preserve encoded
+            //    double quotes, but still translate unadorned double
+            //    quotes into single quotes.)  So, because their
+            //    fixpoint behavior is identical to single quotes, they
+            //    cannot be allowed either.  Firefox 3.x displays
+            //    single-quote style behavior.
+            //  - Backslashes are reduced by one (so \\ -> \) every
+            //    iteration, so they cannot be used safely.  This shows
+            //    up in IE7, IE8 and FF3
+            //  - Semicolons, commas and backticks are handled properly.
+            //  - The rest of the ASCII punctuation is handled properly.
+            // We haven't checked what browsers do to unadorned
+            // versions, but this is not important as long as the
+            // browser doesn't /remove/ surrounding quotes (as IE does
+            // for HTML).
+            //
+            // With these results in hand, we conclude that there are
+            // various levels of safety:
+            //  - Paranoid: alphanumeric, spaces and dashes(?)
+            //  - International: Paranoid + non-ASCII Unicode
+            //  - Edgy: Everything except quotes, backslashes
+            //  - NoJS: Standards compliance, e.g. sod IE. Note that
+            //    with some judicious character escaping (since certain
+            //    types of escaping doesn't work) this is theoretically
+            //    OK as long as innerHTML/cssText is not called.
+            // We believe that international is a reasonable default
+            // (that we will implement now), and once we do more
+            // extensive research, we may feel comfortable with dropping
+            // it down to edgy.
+
+            // Edgy: alphanumeric, spaces, dashes and Unicode.  Use of
+            // str(c)spn assumes that the string was already well formed
+            // Unicode (which of course it is).
+            if (strspn($font, $this->mask) !== strlen($font)) {
+                continue;
+            }
+
+            // Historical:
+            // In the absence of innerHTML/cssText, these ugly
+            // transforms don't pose a security risk (as \\ and \"
+            // might--these escapes are not supported by most browsers).
+            // We could try to be clever and use single-quote wrapping
+            // when there is a double quote present, but I have choosen
+            // not to implement that.  (NOTE: you can reduce the amount
+            // of escapes by one depending on what quoting style you use)
+            // $font = str_replace('\\', '\\5C ', $font);
+            // $font = str_replace('"',  '\\22 ', $font);
+            // $font = str_replace("'",  '\\27 ', $font);
+
+            // font possibly with spaces, requires quoting
+            $final .= "'$font', ";
         }
         $final = rtrim($final, ', ');
         if ($final === '') return false;

lib/htmlpurifier/HTMLPurifier/AttrDef/CSS/URI.php

@@ -43,6 +43,15 @@ public function validate($uri_string, $config, $context) {
         // extra sanity check; should have been done by URI
         $result = str_replace(array('"', "\\", "\n", "\x0c", "\r"), "", $result);
 
+        // suspicious characters are ()'; we're going to percent encode
+        // them for safety.
+        $result = str_replace(array('(', ')', "'"), array('%28', '%29', '%27'), $result);
+
+        // there's an extra bug where ampersands lose their escaping on
+        // an innerHTML cycle, so a very unlucky query parameter could
+        // then change the meaning of the URL.  Unfortunately, there's
+        // not much we can do about that...
+
         return "url(\"$result\")";
 
     }

lib/htmlpurifier/HTMLPurifier/AttrDef/URI/Host.php

@@ -23,6 +23,12 @@ public function __construct() {
 
     public function validate($string, $config, $context) {
         $length = strlen($string);
+        // empty hostname is OK; it's usually semantically equivalent:
+        // the default host as defined by a URI scheme is used:
+        //
+        //      If the URI scheme defines a default for host, then that
+        //      default applies when the host subcomponent is undefined
+        //      or when the registered name is empty (zero length).
         if ($string === '') return '';
         if ($length > 1 && $string[0] === '[' && $string[$length-1] === ']') {
             //IPv6

lib/htmlpurifier/HTMLPurifier/AttrTransform/Nofollow.php

@@ -0,0 +1,41 @@
+<?php
+
+// must be called POST validation
+
+/**
+ * Adds rel="nofollow" to all outbound links.  This transform is
+ * only attached if Attr.Nofollow is TRUE.
+ */
+class HTMLPurifier_AttrTransform_Nofollow extends HTMLPurifier_AttrTransform
+{
+    private $parser;
+
+    public function __construct() {
+        $this->parser = new HTMLPurifier_URIParser();
+    }
+
+    public function transform($attr, $config, $context) {
+
+        if (!isset($attr['href'])) {
+            return $attr;
+        }
+
+        // XXX Kind of inefficient
+        $url = $this->parser->parse($attr['href']);
+        $scheme = $url->getSchemeObj($config, $context);
+
+        if (!is_null($url->host) && $scheme !== false && $scheme->browsable) {
+            if (isset($attr['rel'])) {
+                $attr['rel'] .= ' nofollow';
+            } else {
+                $attr['rel'] = 'nofollow';
+            }
+        }
+
+        return $attr;
+
+    }
+
+}
+
+// vim: et sw=4 sts=4

lib/htmlpurifier/HTMLPurifier/AttrTransform/SafeParam.php

@@ -19,6 +19,7 @@ class HTMLPurifier_AttrTransform_SafeParam extends HTMLPurifier_AttrTransform
 
     public function __construct() {
         $this->uri = new HTMLPurifier_AttrDef_URI(true); // embedded
+        $this->wmode = new HTMLPurifier_AttrDef_Enum(array('window', 'opaque', 'transparent'));
     }
 
     public function transform($attr, $config, $context) {
@@ -41,7 +42,7 @@ public function transform($attr, $config, $context) {
                 }
                 break;
             case 'wmode':
-                $attr['value'] = 'window';
+                $attr['value'] = $this->wmode->validate($attr['value'], $config, $context);
                 break;
             case 'movie':
             case 'src':

lib/htmlpurifier/HTMLPurifier/Bootstrap.php

@@ -37,7 +37,12 @@ class HTMLPurifier_Bootstrap
     public static function autoload($class) {
         $file = HTMLPurifier_Bootstrap::getPath($class);
         if (!$file) return false;
-        require HTMLPURIFIER_PREFIX . '/' . $file;
+        // Technically speaking, it should be ok and more efficient to
+        // just do 'require', but Antonio Parraga reports that with
+        // Zend extensions such as Zend debugger and APC, this invariant
+        // may be broken.  Since we have efficient alternatives, pay
+        // the cost here and avoid the bug.
+        require_once HTMLPURIFIER_PREFIX . '/' . $file;
         return true;
     }
 
@@ -65,10 +70,11 @@ public static function registerAutoload() {
         if ( ($funcs = spl_autoload_functions()) === false ) {
             spl_autoload_register($autoload);
         } elseif (function_exists('spl_autoload_unregister')) {
+            $buggy  = version_compare(PHP_VERSION, '5.2.11', '<');
             $compat = version_compare(PHP_VERSION, '5.1.2', '<=') &&
                       version_compare(PHP_VERSION, '5.1.0', '>=');
             foreach ($funcs as $func) {
-                if (is_array($func)) {
+                if ($buggy && is_array($func)) {
                     // :TRICKY: There are some compatibility issues and some
                     // places where we need to error out
                     $reflector = new ReflectionMethod($func[0], $func[1]);

lib/htmlpurifier/HTMLPurifier/CSSDefinition.php

@@ -219,6 +219,10 @@ protected function doSetup($config) {
             $this->doSetupTricky($config);
         }
 
+        if ($config->get('CSS.Trusted')) {
+            $this->doSetupTrusted($config);
+        }
+
         $allow_important = $config->get('CSS.AllowImportant');
         // wrap all attr-defs with decorator that handles !important
         foreach ($this->info as $k => $v) {
@@ -260,6 +264,23 @@ protected function doSetupTricky($config) {
         $this->info['overflow'] = new HTMLPurifier_AttrDef_Enum(array('visible', 'hidden', 'auto', 'scroll'));
     }
 
+    protected function doSetupTrusted($config) {
+        $this->info['position'] = new HTMLPurifier_AttrDef_Enum(array(
+            'static', 'relative', 'absolute', 'fixed'
+        ));
+        $this->info['top'] =
+        $this->info['left'] =
+        $this->info['right'] =
+        $this->info['bottom'] = new HTMLPurifier_AttrDef_CSS_Composite(array(
+            new HTMLPurifier_AttrDef_CSS_Length(),
+            new HTMLPurifier_AttrDef_CSS_Percentage(),
+            new HTMLPurifier_AttrDef_Enum(array('auto')),
+        ));
+        $this->info['z-index'] = new HTMLPurifier_AttrDef_CSS_Composite(array(
+            new HTMLPurifier_AttrDef_Integer(),
+            new HTMLPurifier_AttrDef_Enum(array('auto')),
+        ));
+    }
 
     /**
      * Performs extra config-based processing. Based off of