It`s that in the *real world*, getting the immensely-complicated
policy
machinery correct is next-to-impossible. And by correct, I mean
``provides security, and never causes unwanted failures of
applications``.
For the web servers I'm running it was a simple matter of reading the
manual and relabelling the relevant content to indicate if it was web
accessible.
A properly-configured Linux server, even without SELinux, but with
other
security features like firewalling turned on, is likely secure-enough
in many environments.
In some perhaps. The big cases it helps are desktop (mostly protecting
against browser stuff) - where it usually just works, and web serving,
where it's most definitely valuable but does mean reading the docs.
Mind you people used to say weak passwords were ok, unencrpyted sessions
were ok, putting . in your path was ok, file permissions were a nuisance
so login as root.
The threat model has changed and continues to evolve.