update HTMLPurifier; enable embedded flash video in articles

lib/htmlpurifier/library/HTMLPurifier/AttrDef/CSS/BackgroundPosition.php

@@ -59,7 +59,8 @@ class HTMLPurifier_AttrDef_CSS_BackgroundPosition extends HTMLPurifier_AttrDef
         $keywords = array();
         $keywords['h'] = false; // left, right
         $keywords['v'] = false; // top, bottom
-        $keywords['c'] = false; // center
+        $keywords['ch'] = false; // center (first word)
+        $keywords['cv'] = false; // center (second word)
         $measures = array();
 
         $i = 0;
@@ -79,6 +80,13 @@ class HTMLPurifier_AttrDef_CSS_BackgroundPosition extends HTMLPurifier_AttrDef
             $lbit = ctype_lower($bit) ? $bit : strtolower($bit);
             if (isset($lookup[$lbit])) {
                 $status = $lookup[$lbit];
+                if ($status == 'c') {
+                    if ($i == 0) {
+                        $status = 'ch';
+                    } else {
+                        $status = 'cv';
+                    }
+                }
                 $keywords[$status] = $lbit;
                 $i++;
             }
@@ -101,20 +109,19 @@ class HTMLPurifier_AttrDef_CSS_BackgroundPosition extends HTMLPurifier_AttrDef
 
         if (!$i) return false; // no valid values were caught
 
-
         $ret = array();
 
         // first keyword
         if     ($keywords['h'])     $ret[] = $keywords['h'];
-        elseif (count($measures))   $ret[] = array_shift($measures);
-        elseif ($keywords['c']) {
-            $ret[] = $keywords['c'];
-            $keywords['c'] = false; // prevent re-use: center = center center
+        elseif ($keywords['ch']) {
+            $ret[] = $keywords['ch'];
+            $keywords['cv'] = false; // prevent re-use: center = center center
         }
+        elseif (count($measures))   $ret[] = array_shift($measures);
 
         if     ($keywords['v'])     $ret[] = $keywords['v'];
+        elseif ($keywords['cv'])    $ret[] = $keywords['cv'];
         elseif (count($measures))   $ret[] = array_shift($measures);
-        elseif ($keywords['c'])     $ret[] = $keywords['c'];
 
         if (empty($ret)) return false;
         return implode(' ', $ret);

lib/htmlpurifier/library/HTMLPurifier/AttrDef/CSS/Color.php

@@ -9,7 +9,7 @@ class HTMLPurifier_AttrDef_CSS_Color extends HTMLPurifier_AttrDef
     public function validate($color, $config, $context) {
 
         static $colors = null;
-        if ($colors === null) $colors = $config->get('Core', 'ColorKeywords');
+        if ($colors === null) $colors = $config->get('Core.ColorKeywords');
 
         $color = trim($color);
         if ($color === '') return false;

lib/htmlpurifier/library/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 @@ class HTMLPurifier_AttrDef_CSS_FontFamily extends HTMLPurifier_AttrDef
             'fantasy' => true,
             'cursive' => true
         );
+        $allowed_fonts = $config->get('CSS.AllowedFonts');
 
         // assume that no font names contain commas in them
         $fonts = explode(',', $string);
@@ -24,7 +57,9 @@ class HTMLPurifier_AttrDef_CSS_FontFamily extends HTMLPurifier_AttrDef
             if ($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
@@ -34,50 +69,122 @@ class HTMLPurifier_AttrDef_CSS_FontFamily extends HTMLPurifier_AttrDef
                 $quote = $font[0];
                 if ($font[$length - 1] !== $quote) continue;
                 $font = substr($font, 1, $length - 2);
-
-                $new_font = '';
-                for ($i = 0, $c = strlen($font); $i < $c; $i++) {
-                    if ($font[$i] === '\\') {
-                        $i++;
-                        if ($i >= $c) {
-                            $new_font .= '\\';
-                            break;
-                        }
-                        if (ctype_xdigit($font[$i])) {
-                            $code = $font[$i];
-                            for ($a = 1, $i++; $i < $c && $a < 6; $i++, $a++) {
-                                if (!ctype_xdigit($font[$i])) break;
-                                $code .= $font[$i];
-                            }
-                            // We have to be extremely careful when adding
-                            // new characters, to make sure we're not breaking
-                            // the encoding.
-                            $char = HTMLPurifier_Encoder::unichr(hexdec($code));
-                            if (HTMLPurifier_Encoder::cleanUTF8($char) === '') continue;
-                            $new_font .= $char;
-                            if ($i < $c && trim($font[$i]) !== '') $i--;
-                            continue;
-                        }
-                        if ($font[$i] === "\n") continue;
-                    }
-                    $new_font .= $font[$i];
-                }
-
-                $font = $new_font;
             }
+
+            $font = $this->expandCSSEscape($font);
+
             // $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 . ', ';
                 continue;
             }
 
-            // complicated font, requires quoting
+            // bugger out on whitespace.  form feed (0C) really
+            // shouldn't show up regardless
+            $font = str_replace(array("\n", "\t", "\r", "\x0C"), ' ', $font);
 
-            // armor single quotes and new lines
-            $font = str_replace("\\", "\\\\", $font);
-            $font = str_replace("'", "\\'", $font);
+            // 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, ', ');

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

@@ -34,20 +34,25 @@ class HTMLPurifier_AttrDef_CSS_URI extends HTMLPurifier_AttrDef_URI
             $uri = substr($uri, 1, $new_length - 1);
         }
 
-        $keys   = array(  '(',   ')',   ',',   ' ',   '"',   "'");
-        $values = array('\\(', '\\)', '\\,', '\\ ', '\\"', "\\'");
-        $uri = str_replace($values, $keys, $uri);
+        $uri = $this->expandCSSEscape($uri);
 
         $result = parent::validate($uri, $config, $context);
 
         if ($result === false) return false;
 
-        // escape necessary characters according to CSS spec
-        // except for the comma, none of these should appear in the
-        // URI at all
-        $result = str_replace($keys, $values, $result);
+        // extra sanity check; should have been done by URI
+        $result = str_replace(array('"', "\\", "\n", "\x0c", "\r"), "", $result);
 
-        return "url($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\")";
 
     }