Re: [PEN-TEST] Crusoe chip.

[Bennett Todd <bet () RAHUL NET>]

As Craig said, the good folks on Bugtraq have demonstrated that
preventing execution in the stack doesn't actually add important
protection, it just changes the way you have to mount your attack.

Furthermore, it would break various techniques that various language
implementations use, that legitimately require executing in the
stack. Some compilers like to generate code that installs trampoline
instructions into the stack (I believe this is mostly to help ease
interfacing between wildly different calling conventions); some
compile-n-go implementations might want to execute out of stack
storage.

If there were a real and important security benefit to a non-exec
stack, then the potential compatibility problems could be lived
with, as each could be fixed if the implementor chose. But they
point up a potential cost, and as the only benefit to a non-exec
stack is effectively security through obscurity --- if the attacker
knows you're doing it they can dodge --- it just doesn't seem worth
implementing. Of course the benefit would be greatest if you did a
private, one-off implementation. But implementation costs, and costs
of dealing with any resulting compatibility problems, are the
highest --- because they're not shared --- for such one-offs.

-Bennett

Attachment: _bin
Description: