Escape story fields and clean story HTML with HTML Purifier.

Add HTML Purifier 4.19.1. Stories are purified when they are saved and
when they are shown.
This commit is contained in:
tim
2026-09-25 21:34:10 +02:00
parent e881ddb458
commit 00e1829f8b
385 changed files with 33174 additions and 12 deletions
@@ -0,0 +1,136 @@
<?php
/**
* Implements data: URI for base64 encoded images supported by GD.
*/
class HTMLPurifier_URIScheme_data extends HTMLPurifier_URIScheme
{
/**
* @type bool
*/
public $browsable = true;
/**
* @type array
*/
public $allowed_types = array(
// you better write validation code for other types if you
// decide to allow them
'image/jpeg' => true,
'image/gif' => true,
'image/png' => true,
);
// this is actually irrelevant since we only write out the path
// component
/**
* @type bool
*/
public $may_omit_host = true;
/**
* @param HTMLPurifier_URI $uri
* @param HTMLPurifier_Config $config
* @param HTMLPurifier_Context $context
* @return bool
*/
public function doValidate(&$uri, $config, $context)
{
$result = explode(',', $uri->path, 2);
$is_base64 = false;
$charset = null;
$content_type = null;
if (count($result) == 2) {
list($metadata, $data) = $result;
// do some legwork on the metadata
$metas = explode(';', $metadata);
while (!empty($metas)) {
$cur = array_shift($metas);
if ($cur == 'base64') {
$is_base64 = true;
break;
}
if (substr($cur, 0, 8) == 'charset=') {
// doesn't match if there are arbitrary spaces, but
// whatever dude
if ($charset !== null) {
continue;
} // garbage
$charset = substr($cur, 8); // not used
} else {
if ($content_type !== null) {
continue;
} // garbage
$content_type = $cur;
}
}
} else {
$data = $result[0];
}
if ($content_type !== null && empty($this->allowed_types[$content_type])) {
return false;
}
if ($charset !== null) {
// error; we don't allow plaintext stuff
$charset = null;
}
$data = rawurldecode($data);
if ($is_base64) {
$raw_data = base64_decode($data);
} else {
$raw_data = $data;
}
if ( strlen($raw_data) < 12 ) {
// error; exif_imagetype throws exception with small files,
// and this likely indicates a corrupt URI/failed parse anyway
return false;
}
// XXX probably want to refactor this into a general mechanism
// for filtering arbitrary content types
if (function_exists('sys_get_temp_dir')) {
$file = tempnam(sys_get_temp_dir(), "");
} else {
$file = tempnam("/tmp", "");
}
file_put_contents($file, $raw_data);
if (function_exists('exif_imagetype')) {
$image_code = exif_imagetype($file);
unlink($file);
} elseif (function_exists('getimagesize')) {
set_error_handler(array($this, 'muteErrorHandler'));
$info = getimagesize($file);
restore_error_handler();
unlink($file);
if ($info == false) {
return false;
}
$image_code = $info[2];
} else {
throw new Exception("could not find exif_imagetype or getimagesize functions");
}
$real_content_type = image_type_to_mime_type($image_code);
if ($real_content_type != $content_type) {
// we're nice guys; if the content type is something else we
// support, change it over
if (empty($this->allowed_types[$real_content_type])) {
return false;
}
$content_type = $real_content_type;
}
// ok, it's kosher, rewrite what we need
$uri->userinfo = null;
$uri->host = null;
$uri->port = null;
$uri->fragment = null;
$uri->query = null;
$uri->path = "$content_type;base64," . base64_encode($raw_data);
return true;
}
/**
* @param int $errno
* @param string $errstr
*/
public function muteErrorHandler($errno, $errstr)
{
}
}