[LWN.net]

Posted May 14, 2009 3:30 UTC (Thu) by jamesmrh (guest, #31622)
Parent article: Seccomp and sandboxing

It's like TCP or Unix, which people keep reinventing poorly.

They start out with an idea which superficially seems simple and efficient, yet once all of the hard-learned lessons of the past are applied with all of their subtleties and nuances, the end result is just some variation on an existing scheme, but without the benefit of having been closely scrutinized and shaken-out over time.

That's what I'm sensing in this case, although I'm more than happy to be proven wrong.

Posted May 17, 2009 14:05 UTC (Sun) by davecb (subscriber, #1574) [Link]

jamesmrh wrote: It's like TCP or Unix, which people keep reinventing poorly.

A useful area to look at for previous successful solutions is MAC, or Mandatory Access Control, which is a necessary and sufficient component of a secure system, from work done back in 1985.

Besides being part of SE Linux, it's also one of the building blocks of the Solaris version of kernel virtual machines, "zones", so it's not just well-understood, it's well-tested.

For the original wheel, see the Department of Defense Trusted Computer System Evaluation Criteria. Accept no substitutes: the "common criteria" are watered-down political compromises with no technical content (;-))

--dave

hammer