[fw-wiz] The Mathematics of Relative Security
From: Chris Pugrud (chris_at_pugrud.net)
To: firstname.lastname@example.org Date: Mon, 20 Sep 2004 10:06:17 -0700 (PDT)
In attempting to evaluate the relative security and exposure of interconnected
subsets of computers there is a distinct shortage of language and tools to
algorithmically evaluate the risks between those groups.
Set theory and discrete mathematics give us a good foundation to evaluate the
risk exposure between groups, but those tools only work with absolutes. A
point is either a member of a set, or it isn't. If two networks are airgapped,
they are logically and provably separate. If two groups are joined with a
"firewall" policy of "permit ip any any" they are logically and obviously
joined, with the grouping inheriting the policy and exposure of the weakest
member(s). There is nothing to take into account one-way transactions, as TCP
permits, other than to wave a wand over distinctive sets. If two sets can
initiate communication into the intersection, but the intersection can not
initiate communication out, then the sets can, observably, be show to be
disjoint, outside of the intersection.
All of this noticably falls apart in the real world, where we have to allow
selected ports to cross boundaries with minimal controls. How many ports
necessitate a union, should ports be allowed higher weight than others? Can it
be convincingly argued that port 135 carries a measurably higher risk than port
123, or application IIS over the innumerable incarnations of application Apache
(with which extensions, options, and controls included?) on port 80?
I know I'm not the first person to evaluate these issues, or to initiate this
conversation in this group. I think that this is fundamentably possible at a
higher level, only looking at connections and direction, and provably
unsolvable at the lowest levels of ports and protocols (reducability to the
halting problem). I'm searching for the people here who have already done some
of the heavy lifting and can at least point me in the right direction to enable
some more quantifiable analysis of highly complex security environments.
Let me know,
firewall-wizards mailing list