Bug 2218238

+++ This bug was initially created as a clone of Bug #2217906 +++

Hello,
In RHEL 9.3 and 8.9, we're planning to fix the long-standing CVE-2007-4559: Python's `tarfile` module makes it too easy to extract tarballs in an unsafe way.
Unfortunately, for the CVE to be considered fixed, this needs a behavior change. (If you don't think this is the case, let's bring it up with the security team.)
Upstream, Python will emit deprecation warnings for 2 releases, but in RHEL we change the behavior now, emit warnings, and provide ways for customers to restore earlier behavior.
To avoid the warning, software shipped by Red Hat will need a change.

For more details see upstream PEP 706: https://peps.python.org/pep-0706
and the Red Hat knowledge base draft: https://access.redhat.com/articles/7004769

---

In /usr/lib/python3.9/site-packages/sos/cleaner/archives/__init__.py, sos calls `archive.extractall(path). The call will emit a warning by default, and should be changed to something like:

if hasattr(tarfile, 'data_filter'):
    # Python with CVE-2007-4559 mitigation (PEP 706)
    archive.extractall(path, filter='data')
else:
    # Fallback to a possibly dangerous extraction (before PEP 706)
    archive.extractall(path)

The 'data' filter above attempts a "safe" extraction, intended for pure data archives. For example:
- prevents extracting outside the target directory, and to absolute paths (by raising an exception)
- prevents symlinks pointing outside the target directory, and to absolute paths
- adjusts permissions (for the owner, only the executable bit is honored)
See PEP 706 for details: https://peps.python.org/pep-0706/#filters

If you trust that the archive is not malicious, use `filter='fully_trusted'` instead. That will preserve the existing behavior.

---

Let me know if you have any questions!