[iacobdaniel]

Apps that parse the XML request body with minidom, sax, pulldom, xmltodict or defusedxml (which Django REST framework's XML parser uses) were not protected: the values never reached the injection checks, so an attack in the XML body went through. Instead of hooking each of these libraries one by one, this hooks the single place they all go through to read XML, the creation of Python's built-in expat parser, so one hook covers all of them and any other library built on expat. ElementTree and lxml don't use that parser, so their existing hooks are extended from fromstring to XML, parse and iterparse. Only XML that is the request body is collected, and values are collected as they are parsed, so a query run inside a sax callback or an iterparse loop is checked too.

[aikido-pr-checks]

🟠 High - Split lxml text is not scanned as the value the app consumes

An attacker sends an XML body whose lxml text value is split around a child or comment, while the application builds a query from the element's composed itertext() value. The new walker stores only element.text and element.tail as separate fragments and never stores the composed text, so the SQL-injection sink scans different values from the one used by the application and a payload can reach the sink without detection.

Suggested change

	texts = (element.text, element.tail)
	texts = (element.text, element.tail, "".join(element.itertext()))

_{More info - Reply on this comment to give feedback or ignore the issue.}

[iacobdaniel]

Not a regression. It will be handled in a separate PR.