Fix XML external entity processing in Python and Flask
======================================================

Parse untrusted XML with defusedxml, not lxml or the stdlib. Pass
forbid_dtd=True so any DOCTYPE is an error before an external entity is
dereferenced (CWE-611) or entities expand (CWE-776); forbid_entities and
forbid_external stay on. Do not try to strip '<!DOCTYPE' from the bytes: a
UTF-16 copy of the same document has no such bytes and is parsed and resolved
anyway.

Page: https://softwaresecurity.com/fix/xxe/python-flask/

What is in this archive
-----------------------
xxe/python-flask/   the project: three implementations of the same code and one test suite
  impl/vulnerable.py   (vulnerable)
  impl/fixed.py   (fixed)
  impl/wrong.py   (a plausible fix that still fails)
run-all.mjs   runs the tests in all three modes and checks the results (Node.js only)
LICENSE-CODE.txt

Run it
------
  Tested with CPython 3.14.8 (libexpat 2.8.5). run-all.mjs needs Node.js 22 or
  later. Supported versions: https://softwaresecurity.com/fix/versions/

  From the xxe/python-flask folder, install the dependencies once (this needs
  network access):

    python3 -m venv .venv
    .venv/bin/python -m pip install -r requirements.txt

  Then run the tests. IMPL chooses the implementation the tests load:

    # every test passes
    IMPL=fixed .venv/bin/python -m pytest
    # an exploit test fails
    IMPL=vulnerable .venv/bin/python -m pytest
    # a test fails
    IMPL=wrong .venv/bin/python -m pytest

  The JUnit report is written to reports/junit.xml. To run all three modes and
  check the results in one step, run the harness from this top folder (it needs
  only Node.js; --install runs the install step first):

    node run-all.mjs --install xxe/python-flask
    node run-all.mjs xxe/python-flask

Verified
--------
  11 October 2026, with python 3.14.8, libexpat 2.8.5, flask 3.1.3, lxml 6.1.3,
  defusedxml 0.7.1, pytest 9.1.1. Versions and results hold for that date: run
  the tests with your own versions before relying on the fix.

Licence
-------
  The code is MIT licensed (LICENSE-CODE.txt). The page text is CC BY 4.0.
