Why Capabilities and Persistence are Essential Jonathan Shapiro (shapj@us.ibm.com)
Tue, 9 Nov 1999 14:18:34 -0500

In this note, I want to define some terms, explore the conditions under which capabilities and persistence become essential, and identify some general requirements. By persistence, I mean *process* persistence.

For this thread, I am not interested here in a debate about relative merits with regard to security; only in a debate about when the *mechanism* of capabilities (or some close equivalent) is required for functional reasons, and when it must be extended to the store. If we can come to agreement that capabilities are required independent of the merit of the security arguments, then our discussion has been reduced to "... now that we have capabilities, how far can we push them and what else (if anything) do we need?", which is the subject of the next two threads.

First, I want to define what protection mechanisms and security policies are (albeit somewhat informally), and what requirements must be satisfied by a protection environment.

Protection Mechanism:

A protection mechanism is a decision procedure. Given a process, an object, and an operation on that object, it answers "yes" or "no" as to whether that process is permitted to perform that operation on that object.

Security Policy:

A security policy is a statement about what decision procedure outcomes should be possible in some dynamic execution of the system (or rather, about which decision outcomes must be false in ALL executions of the system, given an initial condition).

For our purposes, we can ignore security policies that are logically nonsensical in considering the merits of a protection mechanism. We cannot ignore security policies that are economically differentiated. It is a reasonable policy objective to alter the cost or required specificity of an attack. In particular, many attacks are facilitated by prior knowledge, so understanding the potential spread of knowledge (even if we cannot prevent it) is desirable.

The security community distinguishes two types of security policy: DISCRETIONARY and MANDATORY. Discretionary policies are those applied volitionally by some user. Mandatory policies are those applied whether the user wants them or not. Note that discretionary policies can usually be overcome by viruses, as there is no way to distinguish between intent expressed by the user via a well-behaved program and intent expressed by a virus suborning some program that holds equivalent authority.

Confinement is an interesting middle ground policy. It is discretionary in the sense that the user may elect to run a program in an unconfined fashion. It is mandatory in the sense that once started, the program cannot violate the confinement boundary. The existence of such a mugwump policy suggests that the whole terminology framing of security discussions needs a re-examination.

In general, real users are imperfectly consistent in applying voluntary policies, it is therefore sometimes necessary to engineer systems that apply them without allowing the user volition where extremely sensitive content is concerned.

Trusted Program:

A trusted program is one whose operation has been assured (usually by hand) and is known not to misbehave. That is, it is a program that we know obeys the intent of its user. Such assurance requires both inspection of the program and constraints on the integrity of the environment in which it executes.

In general, assuring programs is exceedingly expensive. In secure systems, on therefore seeks to minimize the number of trusted programs, and assumes that all other programs are untrusted. Security policies must be enforced based on the assumption that the majority of programs are untrusted.

A trusted program is a program that can be run safely without external controls.

Capabilities are a particular family of decision rules. given a capability {o, sig}, where /o/ is some object and sig is a set of operations on things of type /o/, the decision procedure is defined in one of the following two ways (depending on the specific system)

           [implicit use design:]
	   allow(p, o, op) :=
                (exists SIG s.t. OP in SIG) and
                ({O, SIG} in p)

           [explicit use design:]
	   allow(p, c, o, op) :=
	        (exists C in caps(P)) and
                (OP in sig(C)) and

                (O == object(C))

In contrast, ACLs use auxiliary tagging information. The decision procedure for a general ACL system is

	   allow(p, o, op) :=
                exists U in users(P), exists A in acls(O) s.t.
                   (op in sig(A)) AND
                   (U in users(A))

Because of the additional levels of indirection in the ACL decision procedure (users(A) and users(P)), the two models are not formally equivalent as they are customarily implemented.


Capabilities are used as an efficiency in POSIX systems. File descriptors are (approximately) capabilities. The question is therefore not "do we need capabilities", but rather "when do we need to store capabilities?"

Stored capabilities are not needed unless process instances are persistent. Providing them in the absence of process persistence may add value, but it is not required to have a useful computing system. Once processes are saved, however, we require some representation capturing the decision procedure state that can be tied to particular process instances rather than to process binaries. The most efficient such representation known is capabilities.

That is: there is a major requirements difference in power between a system that can say "the password *program* runs with extra authorities" and a system that can say "this program instance runs with different authorities than that program instance, even though both run on behalf of the same user." Once we wish to be able to say that we need capabilities to be stored, and we open the pandoras box of issues about whether we will permit their transfer. [It may or may not still be desirable to have other protection models -- I haven't gotten to that.]

Another way to say that is this: if all we want is setuid programs, we can associate attributes with the file image from which the program runs, either in the form of a setuid bit or in the form of a side table of exceptional authorities (as in VMS, or using cryptography as in Java).

It is fairly easy to convince people based on arguments about functionality that per-process-instance authorities are desirable. For a simple example, look to wallet programs. You clearly want your Schwab trading program to use a different trading wallet than your Quicken program. Thus, the requirement for persistent capabilities or some closely comparable representation can be based on arguments of function rather than security.

Okay, a few requirements and I'll get on to the challenge problems where we can enjoy a controversial discussion:

Requirement: When things go wrong, we need to know where to look. In particular, we desire a system where inadvertant transmission of authority (either through accident or trojan malice) is rapidly detected and either the offending user is educated about how to avoid the mistake in the future or the compromised program is expunged.

Requirement: The above requirement demands that some means of traceability exist.

Requirement: the fundamental requirement of a protection mechanism is that it be able to support those security policies that are enforceable and required by the human being(s) holding administrative authority over the system.

Jonathan S. Shapiro, Ph. D.
Research Staff Member
IBM T.J. Watson Research Center
Email: shapj@us.ibm.com
Phone: +1 914 784 7085 (Tieline: 863)
Fax: +1 914 784 7595