Lecture Notes for CS 140
Spring 2014
John Ousterhout

  • Readings for this topic from Operating Systems: Principles and Practice: none.
  • Protection: mechanisms that prevent accidental or intentional misuse of a system.
    • Accidents: generally easier to solve (make them unlikely)
    • Malicious abuse: much more difficult to eliminate (can't leave any loopholes, can't use probabilities).
  • Three aspects to a protection mechanism:
    • Authentication: identify a responsible party (principal) behind each action.
    • Authorization: determine which principals are allowed to perform which actions.
    • Access enforcement: combine authentication and authorization to control access.
    A tiny flaw in any of these areas can compromise the entire protection mechanism.


  • Typically done with passwords:
    • A secret piece of information used to establish identity of a user.
    • Passwords should be relatively long and obscure (only useful if hard to guess).
    • The password database is a vulnerability and must be carefully protected; for example, don't store passwords in a directly-readable form (use one-way transformations).
  • Alternate form of authentication: badge or key.
    • Does not have to be kept secret.
    • Can be stolen, but owner will know if it is.
    • Should not be forgable or copyable.
  • Paradox: key must be cheap to make, hard to duplicate.
  • Once authentication is complete, the identity of the principal must be protected from tampering, since other parts of the system will rely on it.
  • After you log in, your user id is associated with every process executed under that login: each process inherits the user id from its parent.


  • Goal: determine which principals can perform which operations on which objects.
  • Logically, authorization information represented as an access matrix:
    • One row per principal.
    • One column per object.
    • Each entry indicates what that principle can do to that object.
  • In practice a full access matrix would be too bulky, so it gets stored in one of two compressed ways: access control lists or capabilities.
  • Access Control Lists (ACLs): organize by columns.
    • With each object, store information about which users are allowed to perform which operations.
    • Most general form: list of <user, privilege> pairs.
    • For simplicity, users can be organized into groups, with a single ACL entry for an entire group.
    • ACLs can be very general (Windows) or simplified (Unix).
    • UNIX: 9 bits per file:
      • owner, group, anyone
      • read, write, execute permissions for each of the above
      • In addition, user "root" has all permissions for everything
    • ACLs are simple and are used in almost all file systems.
  • Capabilities: organize by rows.
    • With each user, indicate which objects may be accessed, and in what ways.
    • Store a list of <object, privilege> pairs with each user. This is called a capability list.
    • Typically, capabilities also act as names for objects: can't even name objects not referred to in your capability list.
  • Systems based on ACLs encourage visibility of objects: shared public namespace.
  • Capability systems discourage visibility; namespaces are private by default.
  • Capabilities have been used in experimental systems attempting to be secure. However, they have proven to be clumsy to use (painful to share things), so they have mostly fallen out of favor for managing objects such as files.

Access Enforcement

  • Some part of the system must be responsible for enforcing access controls and protecting authentication and authorization info.
  • This portion of the system has total power, so it should be as small and simple as possible. Example: the portion of the system that sets up page tables.
  • One possible approach: Security kernel
    • An inner layer of the operating system that enforces security; only this layer has total power.
    • Most operating systems have no security kernel: the entire OS has unlimited power.

Miscellaneous Issues

  • Rights amplification
    • A mechanism that causes a callee to acquire more privileges (or different privileges) than its caller.
    • Simple example: kernel call
    • Another example: Unix set user id (setuid):
      • Each file has one extra protection bit "s" (for setuid).
      • Normally, each process runs with the same user id as the process that created it.
      • If an executable is invoked with setuid set, the effective user id for that process changes to the owner of the executable file.
      • Typical use: setuid to root to perform protected operations in a safe and controlled fashion.
  • It is extremely difficult to make all of these mechanisms work with no loopholes that can be exploited by evil-doers. Take CS 155 to learn more.