From f4f0f80d2118437e5047ba266f92d7acb3c38fb7 Mon Sep 17 00:00:00 2001 From: Andrew Dolgov Date: Mon, 11 Apr 2011 16:41:01 +0400 Subject: update HTMLPurifier; enable embedded flash video in articles --- .../HTMLPurifier/AttrDef/CSS/FontFamily.php | 177 +++++++++++++++++---- 1 file changed, 142 insertions(+), 35 deletions(-) mode change 100755 => 100644 lib/htmlpurifier/library/HTMLPurifier/AttrDef/CSS/FontFamily.php (limited to 'lib/htmlpurifier/library/HTMLPurifier/AttrDef/CSS/FontFamily.php') diff --git a/lib/htmlpurifier/library/HTMLPurifier/AttrDef/CSS/FontFamily.php b/lib/htmlpurifier/library/HTMLPurifier/AttrDef/CSS/FontFamily.php old mode 100755 new mode 100644 index 705ac893d..0d9a4e12c --- a/lib/htmlpurifier/library/HTMLPurifier/AttrDef/CSS/FontFamily.php +++ b/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 = $this->expandCSSEscape($font); - $font = $new_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); + + // 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 + // . See + // the CSS3 spec for more examples: + // + // You can see live samples of these on the Internet: + // + // 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); - // armor single quotes and new lines - $font = str_replace("\\", "\\\\", $font); - $font = str_replace("'", "\\'", $font); + // font possibly with spaces, requires quoting $final .= "'$font', "; } $final = rtrim($final, ', '); -- cgit v1.2.3