Back to Dmytro Taranovsky's home page.
Dmytro Taranovsky
November 7, 2001
Last Modified:  July 29, 2002

Operating System Architecture

By gaining control of computers, computer viruses, worms, trojans, software bugs, and bad people can create extraordinary damage by shutting down infrastructure, using online banking to steal money, or using robots to attack people. Our civilization is increasingly depended on computers for survival. Therefore, a fully secure operating system is essential for the society. Below is the architecture by which maximum security against viruses and other threats can be achieved.

Operating system and the rest of the software should be modular. Each small module performs a function and may take parameters. For example, a web browser uses an HTML renderer to display a webpage. In turn, the HTML renderer uses a jpg-renderer to display jpg images. An e-mail program might use the same HTML renderer. The rich collection of functions (with application program interfaces) makes writing programs much easier, greatly reduces the need to write the same functionality many times, and allows the same program to work on different computers. Modularity is also essential for stability and internal security of a complex program--failure of some modules to work properly should affect only the functions the modules perform and (usually) should not cause the program to crash or become unnecessarily insecure. Moreover, proper multi-tasking (see below) ensures that a delay in a non-essential module does not excessively delay the program.

While the code for a function is loaded once into the memory, multiple executions can occur simultaneously. Functions remain in memory until the memory (RAM) is about to run out; when that happens, functions least likely to be called are unloaded or put into the swap space.

Each process has a set of rights and priorities. The rights include the type of resources the process can access such as what devices (speakers, networks, etc.) and files the process can access and at what permission level. For example, a process may be allowed to modify only a certain region of the monitor. A process can be denied access to a file, given a read-only access, allowed to read and append, or allowed read-write access. Ideally, each process is given just enough rights to perform all of the functions it should perform.

The priorities are specified for limited resources such as the amount of memory, processing time, bandwidth, and disk space. The priorities are determined based on the importance of each process. The process managers allocate scarce resources based on the priorities. Real-time multitasking ensures that each process gets a fair share and that essential processes are performed. (As a process is granted more and more resources, its priority decreases; as a process is denied resources, its priority increases; thus, a malfunctioning process will not be able to occupy too much of vital resources.)

For example, a device driver can communicate with the hardware device and with the device manager. Its priorities depend on the importance of the hardware. A device driver is not allowed to modify files or otherwise damage the computer.

A process can use any function (and execute any code) that it can read, and can create new processes. The parent process decides (and can later change) what rights and priorities to give to its child. (A function used by a process is either a part of that process or its child.)  However, the child process cannot have more rights than the parent process. Moreover, the total priorities for the parent and all of its derivative processes cannot exceed the original priority of the parent. This way processes will be unable to circumvent their rights and priorities.

To prevent security vulnerabilities, the operating system uses microkernel architecture. Each resource (such as a file system) has a device manager. The resource can only be accessed through the device manager. The device manager uses the process rights and priorities to determine whether to grant access. To prevent security flaws, the device manager does not do much else and uses the device driver to actually access the device. For efficiency, CPU acts as a basic memory manager preventing unauthorized access.

Programs can be run as a team of mutually trusted cooperating processes. Such processes can send messages to each other. The messaging device manager(s) control the form and the existence of messages. (Otherwise, malicious programs could use messaging to gain control of processes with many rights.) The limited messaging right is given to cooperating processes (only to communication among themselves), basic resource/device managers, log-in managers, and other processes as necessary.

A process may ask the user a special permission for a resource. To do so it sends a message (permission request and reason) to an appropriate log-in manager.  A log-in manager is a program that has the right to change certain rights of processes.  The log-in manager decides whether to grant permission and may prompt the user with the request and the text of the reason for the request. If the user agrees (and both the user and the manager have the permission), the request is granted. For security, messaging device manager limits access to log-in managers.  The log-in manager can later terminate the special rights it had granted.  A log-in manager may request authentification (username/password, digital signature, and/or satisfactory results from a special authentification program).  For greatest compatibility with legacy programs, a process's rights may specify for automatic transfer of a denied attempt to access a resource to a log-in manager for a possible authorization of access.  To prevent security breaches, in an ordinary configuration, only one log-in manager has broad power, and that manager may only give that power to a user who communicates only through approved channels (is at the computer).

For example, an editor does not need read-write access to all files: It simply relies on the user to grant a permission to save a file. A harmful macro in Microsoft Office will not be able to damage files except the few files for which the user explicitly gave the access. (Since the damage is quickly detected, Microsoft Office can be terminated and possibly reinstalled before the macro damages too many files.) Of course, Microsoft Office should have internal process and document rights management system to prevent a malicious document from making damage.

Digital signature module is used to make and verify high security digital signatures. Encryption module uses public key cryptography to prevent unauthorized access to secret data. These modules allow secure (and confidential) interaction between computers even if the network itself (including network device driver) is not. Certain functions (such as processing time of a remote computer) are not free. Financial module allows controlled money transfer through digitally signed authorization to take the stated amount of money from the bank account. Digital signature prevents fraud. Of course, the access to digital signatures and finances is restricted. Different users/processes may have different signatures.  

Each process has a unique ID. The child process inherits the ID of the parent in addition to its own ID. This way a process along with all its derivative processes can be terminated (or abridged on rights/priorities) if so desired.

A user is just like a process, with certain rights and with the ability to create processes. Since each user has limited rights, the computer/network is safe from malicious users. Just like many processes can be active at the same time, multiple users can use the computer/network at the same time.

By default, users and processes own the processes they create. The process manager allows the user or process to control the rights and priorities of the processes it owns. Thus, users and processes can terminate the processes they own (including themselves).

Certain processes present a special identification to the user such as a certain message in a particular area of the screen. The right to present the identification is restricted to the authorized software and enforced by appropriate device managers.

The network manager provides basic functionality of sending inter-process messages across the network (such as a multiprocessor system, local area network, or the internet), for accessing functions over the network, and for transferring and creating processes from one computer on the network to another. This way the network manager makes the network transparent, with network services accessed just like ordinary programs. Because each process has limited rights and priorities, which are enforced, network transparency does not cause security vulnerabilities.

The architecture above when properly implemented resolves all security issues except which (authenticated) users/processes to trust with which resources and files.  Users and programs will not be able to what they are not authorized to, but at the same time they will be able to run any programs they want and programs will not be restricted in their internal structure.  The architecture allows the operating system to be very extensible and scalable. Moreover, the operating system is stable since a failure of almost any process will not cause the computer to crash or to excessively delay an essential process. Stability is essential in many cases: It is essential that computer-controlled automobiles do not crash every year.