It's too easy to accidentally use all available disk space on the controller, causing trouble (re-saving XML files may fail).

If we can hook into ArtifactArchiver, specifically the controller side of the Tar*Stream, we might be able to abort it if the Jenkins controller runs out of disk space.

This would protect not just form weird cyclic symlink problems, but also overly big artifacts. Something like 1-10 GB of disk space (perhaps reusing the node monitors) could be the threshold to just reject further archiving to the controller.