If I get the state of things right, we can either:

  - Use "mount points" and "file ids on these mount points" (as we do
    now), but that will have problems working with unionfs and -- though
    I'm not sure about that, because AFAIK Veriexec worked that way --
    NFS.

    How is that solution dealing with unionfs in the long term? I know
    we have (had?) some plans on making it easier; we've had this
    discussion before IIRC.

  - Use "file handles" to uniquely identify files, in which case we
    lose in the short-term with unionfs and NFS, but have better chance
    of supporting them in the future hypothetically assuming someone
    will write code to make it work. (not me)

For the reasons *I* wanted to add metahook(9) (or whatever you'd like
to name it) it doesn't matter which of the above is used. Can we please
reach a concensus and not just drop the ball on this issue because no
solution is perfect in the short term?

I'll just commit the code I have with dev_t -> struct mount * if no
concensus is reached within reasonable amount of time and there's no
real argument against it.

-e.

-- 
Elad Efrat