Let’s do a quick refresher on the lifecycle of OpenVox facts.
When the agent starts a run, it generates facts and sends them to
the server along with the catalog request. The server combines
them with trusted information from the agent’s SSL certificate and
uses them during compilation. It also caches these facts in
OpenVoxDB where they can be used for other purposes; dashboards,
OpenBolt target lists, puppetdb_query() calls, etc.
Outside of regular catalog compilation, agents can update their
own cached facts using the /puppet/v3/facts endpoint. The flaw
fixed in this release is that OpenVoxDB was not validating that
the factset being stored with this endpoint actually represented
the agent submitting the facts!
Now, to be clear, this doesn’t directly affect classification or
Hiera lookups. Those will always use the fresh factset submitted
by the agent. What is affected is any classification using or
gated behind puppetdb_query() calls or any OpenBolt plans that use
a puppetdb_query() or OpenBolt tasks that use PQL to select a list
of targets. This would also affect any custom administration
tooling that used facts from OpenVoxDB.
In particular, this could lead to administrative actions intended
for one node actually executing on another node, which a skilled
attacker could leverage via social engineering to achieve
privilege escalation.