Apache Tika 1.19 included an incomplete fix for CVE-2018-11761 which added an entity expansion limit for XML parsing. However, Tika reuses SAXParsers and calls reset() after each parse, which, for Xerces2 parsers, as per the documentation, removes the user-specified SecurityManager and thus removes entity expansion limits after the first parse. Apache Tika 1.19 is therefore still vulnerable to entity expansions which can lead to a denial of service attack.

External Reference:

https://lists.apache.org/thread.html/88de8350cda9b184888ec294c813c5bd8a2081de8fd3666f8904bc05@%3Cdev.tika.apache.org%3E

Upstream Issue:

https://issues.apache.org/jira/projects/TIKA/issues/TIKA-2727

Upstream Patch:

https://github.com/apache/tika/commit/86d4ba1e

In Apache Tika 1.19 to 1.21, a carefully crafted 2003ml or 2006ml file could consume all available SAXParsers in the pool and lead to very long hangs. Apache Tika users should upgrade to 1.22 or later.