Before this change we ran the tests and the mount in the same process.
This could cause deadlocks and often did, and made the mount tests
very unreliable.
This fixes the problem by running the mount in a seperate process and
commanding it via a pipe over stdin/stdout.