14

Cybersecurity and Confidential Computing Architectures

This chapter introduces computing architectures suitable for applications that require an exceptional assurance of security. Critical application areas like national security systems and financial transaction processing demand this high level of protection. These systems must be resilient against a broad range of cybersecurity threats, including malicious code, covert channel attacks, and attacks enabled by physical access to the computing hardware. Topics addressed in this chapter include cybersecurity threats, encryption, digital signatures, and secure hardware and software design.

After completing this chapter, you will be able to identify the categories of cybersecurity threats a system will face and understand the security features of modern computer hardware. You will understand how to avoid security gaps in system architectures and how a secure computer architecture can help enforce security in software applications.

The following topics will be presented in this chapter:

• Cybersecurity threats
• Features of secure hardware
• Confidential computing
• Designing for security at the architectural level
• Ensuring security in system and application software

Technical requirements

The files for this chapter, including solutions to the exercises, are available at https://github.com/PacktPublishing/Modern-Computer-Architecture-and-Organization-Second-Edition.

Cybersecurity threats

The first step in ensuring the cybersecurity of a computer system is to understand the threats it is likely to encounter and must defend against. We can place these threats into a few broad categories and identify the key features of each category. With this information, we can design a computer system architecture with attributes that will provide an appropriate level of protection against those threats.

Unfortunately, developing a completely secure computer system is not a straightforward process. This is because new vulnerabilities are identified regularly in existing operating systems, software libraries, user applications, and web applications. It is not unusual for flaws to be identified in widely used cryptographic components such as encryption algorithms and authentication protocols. As new software products are developed, they often contain entirely new vulnerabilities that will be discovered sooner or later. Most importantly, experience shows the greatest security weaknesses related to the computer systems we use are created by the humans who operate them.

Given this reality, we can only strive to reduce the risk of a damaging cyberattack to an acceptably low level, rather than eliminate it entirely. In the process of hardening a system against attack, we must keep the cost of protective measures to an acceptable level, and the system cannot be overly cumbersome for users to operate. The protections must also avoid reducing the reliability of the system in terms of performing its intended function.

Cybersecurity threat categories

As a first step to understanding the landscape of cybersecurity threats, we will examine the categories of people and organizations who plan and carry out cyberattacks. Based on historical patterns of cyberattacks against individuals, governments, businesses, and other groups, these are the primary threat categories:

• National governments: Many countries, including some that may be considered “hostile” and others one might normally think are “friendly,” have been credibly identified as conducting aggressive cyber actions against other governments as well as businesses and other organizations in different countries, and even against specific individuals. Military cyber organizations plan and carry out cyberattacks against enemy nations to achieve tactical and strategic goals.
• Terrorists: Terrorist organizations have expressed the desire to conduct cyberattacks with potentially devastating effects against governments and industrial systems with potential outcomes ranging from widespread power blackouts to the destruction of large-scale infrastructure, such as dams, refineries, and pipelines. While the potential for such attacks has been repeatedly demonstrated, actual occurrences of destructive attacks have, so far, been rare.
• Industrial spies: In many industries, particularly those involving high technology, secret information about software components and hardware systems can be a prime target for hackers. If the attackers can gain access to the computer network where the desired information resides, they may be able to copy it and use it for their own purposes, or simply sell it to the highest bidder.
• Criminal groups: Groups of criminal hackers carry out attacks such as penetrating computer systems containing private information like credit card data with the goal of selling the information to other criminals or using it directly to make illegal purchases. Ransomware is another category of cybercrime. In a ransomware attack, the hackers load software onto the victim’s computer that encrypts valuable data files. The attacker demands payment from the victim before providing a key that will purportedly decrypt the user’s data. At the lower end of the technological scale, mass-emailing scams continue to proliferate. In these efforts, attackers attempt to lure recipients into engaging the criminals in a conversation that can lead to expensive disappointment for the unwary.
• Hacktivists: Some cyberattackers claim to be driven by just motives in their attacks against perceived immoral enemies such as disfavored corporations or governments. These individuals are labeled hacktivists, a combination of the words hacker and activist. Hacktivists perform actions such as bombarding the website of their target with so much traffic that the site becomes unavailable for regular users. Hacktivists may also attempt to steal sensitive data belonging to their target and then publish the data to embarrass or shame the organization that was the source of the information.
• White-hat hackers: Traditionally, hackers might investigate a target and then plan and execute a cyberattack against it for the pure intellectual thrill. In other cases, cybersecurity researchers will investigate a target, possibly a website or a digital device such as a smartphone, and identify one or more vulnerabilities. The researcher then shares the information with the owner of the site or the manufacturer of the device with the goal of encouraging improved security. After allowing sufficient time for the issues to be resolved, the researcher then makes the information about the vulnerability public. This process is referred to as white-hat hacking, a reference to early American Western films in which the protagonists wore white hats and the villains wore black hats.
• Insiders (witting and unwitting): The most common precipitating factor for cyber intrusions into protected networks is an action taken by an authorized user of a computer system on that network. In many cases, the action is something as simple as clicking on a link in an email that references a topic of interest to the recipient. This can lead to the installation of malware that spreads quickly throughout the supposedly protected network. In other cases, perhaps due to an employee’s discontent with the working environment, a user intentionally takes steps to enable intrusion into the network by malicious actors. In either case, the network is compromised, and the attackers have a foothold that enables further exploitation of the supposedly secure network. Once attackers have achieved this level of access, they can take any action they wish, such as extracting confidential company information or installing ransomware on critical computer systems.

As of 2021, ransomware is the most widespread malware threat to individuals, businesses, and other organizations.

Cyberattack techniques

Regardless of a cyberattacker’s motivating factors, these actors employ a few common types of attacks against computer systems, whether the target is a web server, an industrial control system, or a PC owned by an individual. These are some of the attack types that are most likely to occur:

• Phishing: Phishing describes email-based attempts to convince the recipient to take some action that will further the attacker’s goals. The email may request the recipient to click a link, or possibly download and open a file, or sometimes just respond to the sender with an answer indicating interest in the topic of the original email. In most cases, the goal of the phishing effort is to convince the recipient to take some action that results in the installation of malware, giving the malware sender control over the recipient’s computer.
• Botnets: A botnet (short for robot network) is a collection of computers that have been infected by malware and are under the control of a single attacker. This individual is known as a bot herder. The computers within a botnet can be commanded to perform nefarious tasks such as sending out phishing emails in an attempt to recruit more members for the botnet or conducting denial-of-service (DoS) attacks by attempting to overload websites with thousands of simultaneous requests for service from the botnet member computers.
• Password attacks: In most computer systems, the passwords users enter to log in to the system are not stored in the same format they are typed in at the keyboard. Instead, the passwords are passed through a cryptographic hashing algorithm that converts the password to an unrecognizable string of binary data.

  The important features of a cryptographic hash algorithm are that entering the same password will always produce the same hashed value, and knowledge of the hashed value does not give the attacker any easy way to recover the corresponding password. Even though possession of a hashed password value does not allow direct discovery of the password, it is still possible for attackers to determine a user’s password given its hashed value. The standard approach for discovering a user’s password given a known hash is to simply try all possible passwords until a matching hash value is found. This technique is referred to as a brute-force password attack. The best defense against this type of attack is to use long passwords (15 characters or longer) with a mixture of upper- and lower-case characters along with some symbols, such as * and &. Passwords that are short (less than 10 characters) or can be found in a dictionary are trivially easy to crack, as well as unfortunately common passwords such as 123456, qwerty, and password1.

• Vulnerability exploitation: White hat security researchers and black hat criminal hackers put great effort into discovering cybersecurity vulnerabilities in existing operating systems, applications, websites, and embedded devices. White hat hackers pursue the goal of improved cybersecurity for everyone by first informing the developer of the vulnerable system or application and then, after allowing time for the problem to be fixed, publicizing the vulnerability for everyone to see. The white hats are not entirely driven by altruistic motives – discovering a critical vulnerability in a major software system can lead to publicity, reputation enhancement, and respect from peer researchers. Black hat hackers use any vulnerability they discover to commit criminal activity against computers that are susceptible to the attack. Often, it is only after systems have been attacked using a previously unknown vulnerability that the system developers and independent white hat researchers dissect the offending malware and discover the vulnerability in the targeted system.

Vulnerabilities are categorized by the length of time (in days) that system administrators and users have to prepare defenses against a newly discovered vulnerability before that vulnerability comes under attack. In the worst-case scenario for the owners and users of the victim computer system, attacks can begin with no warning at all. In this case, the vulnerability is referred to as a zero-day vulnerability, which means there was no warning in advance of the attack.

Types of malware

Software installed on a computer system without the owner’s permission and that attempts to perform actions the computer owner does not want is called malware.

Malware, an abbreviation of malicious software, is a category of code that performs undesired actions and often interferes with the normal operation of the computer. Some of the most common types of malware are:

• Spyware: Spyware is software that collects personal information about the computer user and other valuable data and transmits it to the attacker. This may include information such as usernames and passwords for online accounts and the addresses of websites visited. Spyware may use this information to display pop-up advertisements targeted to the perceived interests of the user. The tactics used by spyware are similar to the methods commonly used in advertisements by legitimate businesses, so classifying a piece of apparent spyware as malware may not be a straightforward process.
• Ransomware: As indicated earlier in this chapter, ransomware most commonly encrypts the user’s data files and then demands payment in return for providing a key to decrypt the data. In another form of ransomware, the attacker steals personal information from an individual or confidential company information and demands payment to prevent the public release of the information. Ransomware attacks have frequently targeted businesses and organizations such as hospitals. These criminals usually demand payment in the form of cryptocurrency, which they perceive to be untraceable. While in many cases payment of the ransom results in the successful decryption of the encrypted data or the withholding of the private information from public release, there are no guarantees that paying the ransom will result in a favorable outcome for the victim.
• Viruses: Like a biological virus, a software virus replicates by infecting new hosts with copies of itself. In addition to functioning as spyware, ransomware, or some other form of cyberattack, a virus contains code that attempts to gain access to other computer systems accessible over the network and install copies of itself onto them. A virus inserts itself into an existing computer program by making modifications to the program to incorporate the virus code. The modified program then becomes a virus host and begins attempting to infect other computers. A program that has been modified to incorporate the virus is said to be virus-infected. The requirement for replication via a host software application distinguishes a computer virus from a worm, which does not require a host application for this purpose.
• Worms: Similar to a virus, a computer worm is a standalone program that attempts to replicate by installing copies of itself onto other computers accessible over the network. A worm does not require an infected host application in the manner of a computer virus. Like a virus, a worm may contain code to act as spyware, ransomware, or some other type of attack.
• Man in the middle: In a man in the middle (MITM) attack, the attacker attempts to position software in a communication path between two communicating applications or computers. If successful, an MITM attack permits the attacker to capture sensitive information such as usernames and passwords and allows the attacker to modify information as it passes between the communicating nodes.
• Denial of service: In a denial of service (DoS) attack, a target system such as a web server is bombarded with an excessive level of spurious requests. This attack may achieve its goal by making the targeted system unusable by legitimate users for an extended period. DoS attacks are popular among hacktivists, who employ the technique to cause problems and draw negative attention to businesses and other organizations that are targets of their ire.
• SQL injection: Web applications often use Structured Query Language (SQL), a database programming language, to communicate between the user interface and databases containing information such as user credentials and information on products for sale through a website. If the developers of the web application are not sufficiently careful in employing secure coding practices, it may be possible for users of the application to craft inputs that are interpreted as executable code by the database SQL interpreter. If successful, this enables the attacker to extract and modify data in the database, and it sometimes leads to a more devastating attack against the network on which the database resides.
• Key logger: A key logger is a type of malware that collects the sequence of keys a computer user presses on the keyboard and forwards this information to the attacker. A key logger can record important private information such as usernames and passwords for bank accounts and credit card accounts. More generally, key logger-type malware can perform functions such as capturing screen shots from the victim’s computer and even recording using the video camera and microphone attached to an infected computer or smartphone. Key logger functionality is often a component within a larger piece of malware that contains spyware, ransomware, or virus capabilities.
• Infrastructure attacks: Many types of large-scale infrastructure such as power plants, electrical transmission systems, dams, refineries, and pipelines rely on computerized control systems for their operation. Traditionally, the specialized computers used for these purposes have possessed only rudimentary security features. There is great concern among cybersecurity professionals that sophisticated malware attacks against these systems could lead to drastic consequences such as widespread power outages or explosions at refineries.

Some types of malware combine several of these capabilities and perform actions such as conducting an extensive analysis of the newly infected system before it decides if it will proceed with an attack. This analysis may attempt to determine factors such as the country where the infected system is located, the company the system belongs to, and the particular applications installed on the computer.

Some malware arrives in a mostly-encrypted form and only performs decryption as individual segments of code prepare to execute. A goal of code encryption is to make it as difficult as possible for cybersecurity researchers to examine the code and understand what it does. Code encryption also may make it possible for malicious code to slip through the automated defenses provided by antivirus software.

Post-exploitation actions

Once an attacker has gained remote access to a targeted system, a variety of options become available to further exploit that system as well as other computers on the same network. One of the first steps an attacker normally takes after gaining access to a victim system is to install software that enables continued access even if the currently logged-in user logs out, or if the system reboots. Many types of attacks take this step automatically as part of their initial entry into the victim computer. In other cases, an attacker conducting remote probing of a target system will install this code once access has been achieved.

The attacker’s first entry into a victim system often takes place at the privilege level of the user that performed the action enabling the access, for example, clicking a link that downloaded and installed a piece of malware. Many types of malicious activity require only the limited privilege level of an ordinary user. Some examples are ransomware that encrypts the user’s data files or botnet malware that bombards a targeted web server with requests using the limited privileges available to an ordinary user.

Some types of attacks, such as those attempting to gain access to protected information like a database containing customer information, require administrator-level privileges. Once access into a computer system at the unprivileged user level has been achieved and software has been installed to enable ongoing communication between the attacker and the victim system, the next step may be to attempt privilege escalation.

Privilege escalation is the process an attacker uses to achieve a higher privilege level within the target system, thereby enabling access to system resources such as files and databases that are off-limits to regular users.

Privilege escalation can be achieved by a variety of means. Often, known flaws exist in operating systems, driver software, or applications that run at higher privilege levels. If the system is not updated regularly to install fixes for known vulnerabilities, attackers can exploit these flaws in a manner that enables them to raise their privilege to a higher level, granting them access to protected system resources.

Having established persistent connectivity into a victim system and gained administrative privileges, the attacker has full control of the computer. Any data present on the computer is available for extraction and any software the attacker desires can be installed. In hacker terminology, a computer that is under the full administrative control of a remote attacker is considered to be owned by the attacker. If the attacker achieved this level of control without any notice by legitimate users or by software such as antivirus tools, the attack is said to be a perfect own, shortened to pwn (pronounced “pone”) in hacker slang.

One constraint the attacker must usually keep in mind is a need to avoid making the fact the computer is infected with malware overly obvious to its legitimate users. Most users are not too concerned if the computer is a bit more sluggish than usual, but seeing a command window pop up onscreen followed by the display of commands typed by the attacker would probably alert anyone that the system has been breached. Sophisticated attackers strive to avoid making moves that expose their presence to the victims of their attacks.

This section provided an overview of some of the more common cybersecurity threat categories and techniques. The methods and malware types listed here are not exhaustive and new methods for attacking digital systems and devices are under development continuously. For the computer system architect, it is not sufficient to understand the types of cyberattacks that have been common in the past. It is vital to understand the full range of attacks that are even theoretically possible, even if some of those attacks have not been observed in past attacks. The next section will examine some key features that computing devices must implement to provide a high level of assurance the system and its interfaces to other networks, users, and devices will remain secure for the lifetime of the system.

Features of secure hardware

When initiating the design of a new computer or digital device, or when revisiting the design of an existing system, it is vital that the computer architect considers security as a top-level requirement. Even the most basic decisions in the development process, such as selecting a processor model, are likely to have profound effects on the security of the resulting design. As a first step in this process, it is necessary to understand what important types of data and other technology-related information must be protected from disclosure to unauthorized individuals.

Identify what needs to be protected

Some types of information present on computers and networks that generally need to be protected against unauthorized disclosure are:

• Personal information such as passwords, social security numbers, financial data, and medical history.
• Confidential information belonging to a business including customer lists, product design data, and strategic plans.
• Proprietary technology such as the digital circuit design within a smartphone.
• Government information such as national defense information and intelligence data collected by law enforcement.

In designing a particular system, the computer architect must remain constantly aware of the types of information that need to be protected. These concerns relate to the digital data contained within a system as well as hardware features that may expose vulnerabilities that grant attackers access into the device.

The system security analysis must include an assessment of the degree to which the system can be physically accessed by potential attackers. For items delivered to end users, such as personal computers and smartphones, users can obviously take any steps they desire with the hardware, including disassembling it and examining the components under microscopes.

For hardware intended to be operated in a controlled environment, such as a cloud server farm, protection against outright physical attacks may be less of a concern. However, given the potential for attacks by malicious insiders, some thought must be given to at least detecting the occurrence of unacceptable tampering, even if such tampering may be the result of actions by well-intentioned employees.

A particular category of intellectual property worthy of protection is the firmware and software code installed in devices delivered to users. If the manufacturer desires to maintain this code as a trade secret, the development process must include steps to secure the code and prevent its exposure by even the most skilled and capable attackers.

Anticipate all types of attacks

In designing the security features of a digital system, it is important to factor in not just the types of attacks that have been observed in the past.

The architect must also cast a wide net to identify attack categories that may have never been seen but are at least theoretically possible. Such thinking may lead to technologies that seem to be of little threat in the near term (such as, perhaps, quantum computing) but nevertheless can be effectively mitigated with an acceptable expenditure of resources.

For computer systems that do not require access to the internet or to other external networks, it is common to construct the computing environment in a manner that ensures isolation from connection to outside networks. This configuration is referred to as air-gapped, which means there is a significant physical distance between the computer hardware and any potential connection to external networks.

While constructing an air-gapped computing environment theoretically provides a substantial security enhancement, in real-world usage the benefit of this architecture has proven to be limited. For any computer system to provide substantial value, it is usually necessary to regularly transfer information into the system in the form of software updates and updates to data used on the computer. For an air-gapped system, these updates typically arrive on optical disks or on portable hard drives. Despite the best intentions and security efforts of the system operators, this data transfer process offers a substantial opportunity for malware to travel into the protected system and, once infected, enables the malware to attempt to transfer data out of the system and back to the attackers using the same data transfer mechanisms.

Sophisticated hackers are continually working to develop attack methods against air-gapped computer systems and other high-security computing environments. The designers of secure computing systems must consider all the potential avenues these attackers may use in their efforts to gain access and exploit the system. Creativity is vital to ensuring a secure design.

All possible forms of information access and data leakage are fair game for a dedicated attacker. Even if a computer system is intended to never connect to an externally accessible network, other potentially exotic attacks may be viable for a determined adversary. Cyberattacks that enable transfer of data between processes that are not intended to have permission to communicate are called covert channel attacks. Some surprising types of attacks that have demonstrated at least some degree of success are listed here:

• Row hammer: The modern DRAM devices discussed in earlier chapters have been shown to be vulnerable to a type of attack called row hammering. The “row” in row hammer refers to the rows of bit cells in a DRAM device. Due to the tiny size of each bit cell and its proximity to neighboring cells in the same row and in adjacent rows, it becomes possible to change the state (“flip the bits”) in cells in adjacent rows under certain conditions.

  To cause this effect, code performs repeated accesses to a targeted DRAM row at a high rate. To be successful, the code performing this attack must ensure its memory access requests result in cache misses, thus ensuring the DRAM internal circuitry is activated. Row hammer attacks have demonstrated the ability to achieve privilege escalation in computers based on the x86 architecture.

• Power consumption fluctuations: Even if attackers succeed at installing malware on an air-gapped computer system, extraction of data collected by the malware remains a challenge. If it is not feasible to exfiltrate information via the disks used to move data to and from the air-gapped system, an alternative method must be identified. Researchers have demonstrated that malware running on a target computer system can create sufficient variations in the building’s electrical power consumption that it may be possible to monitor those fluctuations on the power lines outside the building. By carefully coding digital data into the fluctuations, the attackers can collect data from the air-gapped computer.
• Thermal fluctuations: Modern computers have fans and sometimes liquid cooling systems to manage the system temperature and keep it within prescribed limits. When computers are located close together, the temperature of one computer may affect the temperature readings of a nearby computer. If malware is present in two closely spaced computers, one on an air-gapped network and one connected to an external network, it has been demonstrated that causing intentional temperature fluctuations in one computer by inducing processor loading can induce measurable temperature changes in the adjacent computer. Using this technique, malware on the air-gapped computer can transmit digital data to the network-connected computer. Data can be transferred in the opposite direction in the same manner. While extremely slow, in terms of bit transfer rate, this technique permits the transfer of critical data such as encryption keys that can be used to enable further attacks.
• Electromagnetic emissions: Any time electrical current flows through a conductor such as a USB cable or across a trace on a printed circuit board an electromagnetic wave is transmitted into the surrounding environment. If a threat actor can place a receiving antenna within range of this signal, it may be possible for the attacker to collect the information that was originally transmitted. Of course, in an electrically noisy environment, with perhaps many computers operating simultaneously, it may be exceptionally difficult to extract the signal radiated from any single system. However, if malware is present on the targeted computer, the malware may take actions to intentionally generate fluctuations in radiated emissions in a pattern that can be detected and decoded by a suitable reception system. Stranger things have happened.

This section has presented some examples of the more exotic types of attacks that a mission-critical computer system may experience. While by no means complete, this list of attacks provides some examples of the types of cyberattacks a system architect must consider when designing a secure system.

Features of secure system design

Given these examples of possible attacks, we can list some important features a secure computer system hardware design must implement to provide a high level of security assurance.

Secure key storage

Any cryptographic keys used by the system to secure data must be stored in a manner that prohibits their extraction by any conceivable means. This typically means the keys must be stored within a device such as a processor chip in a manner that prevents their extraction by any software technique. It also means that attempts to extract the keys by methods such as disassembling the integrated circuit or using sophisticated tools such as a scanning electron microscope must be unsuccessful.

Encryption of data at rest

Any data stored in the system must be protected when the system is powered off. This means that even if the hardware is disassembled and the contents of individual memory devices are extracted by an attacker, the data within them remains secure. The most common method of achieving this level of security is to encrypt the data using an encryption key that is available for use by the computer but fully protected against disclosure outside of its intended use. One way to store this key is in special registers within a processor that cannot be accessed even by malware running at elevated privilege. Many modern processors, even tiny embedded processors, have begun to provide carefully designed cryptographic capabilities for this purpose.

Encryption of data in transit

Any data transferred from a processor or communication interface can be accessed by an attacker with physical access to the communication path. Whether the data is transferring through a trace on a circuit board or over a global network, any information that must remain secure needs to be protected over the entire path from source to destination. Once again, the common approach for implementing this protection is the use of encryption. Securing data in transit between two endpoints is more challenging than encrypting and decrypting data in local storage because it is necessary to perform the encryption and decryption at two different locations.

Conceptually, the most straightforward method to achieve this is to give the encryption key to the systems at both ends of the communication path. However, it is challenging to transport a secret key to both systems if there is not already a secure communication path between them. The standard approach used today to set up secure communication paths between systems that don’t already have a shared secret key is to use public key encryption to transfer a secret key from one end of the communication path to the other. The secret key is then used to encrypt and decrypt the data at each end of the communication path. While it would be possible to just use the public key encryption process on all the data shared between the systems, it turns out that using public key encryption is far more computationally intensive than using a shared secret key.

Cryptographically secure key generation

When a fresh secret key is needed, perhaps when setting up a secure communication path between two computer systems, it is vital that the key is completely unpredictable to any outside attackers. Any newly generated secret key should appear entirely random and completely unrelated to any prior or subsequent keys created by the same system or by other systems. Traditionally, the easiest way to create a random-looking number on a computer was to use the pseudorandom number generation capabilities available in many programming language libraries. It turned out that the number sequences produced by many of these algorithms did not produce random-like outputs and these algorithmic flaws could be exploited in ways that made it far easier to crack encryption algorithms than an initial analysis would make it appear. Modern cryptographic random number generators use specialized hardware to generate truly random numbers, enabling the creation of cryptographic keys that are as secure as possible.

Secure boot procedure

In a secure system, all code that executes at an elevated privilege level must be verified as authentic before being permitted to run. This includes all code that executes as part of the boot process as well as the operating system kernel and drivers. The standard approach for achieving this goal is to attach a digital signature to each piece of code. The digital signature contains an encrypted hash that is computed over the entire block of executable code, whether it resides as a file on disk or is stored in flash memory within the processor. The key used to decrypt the digital signature hash must be stored securely within the processor hardware. With this level of protection, any attempt to replace the legitimate code with maliciously modified code will fail because the attacker lacks the key necessary to prepare a valid digital signature for the modified code.

Tamper-resistant hardware design

The features of secure computer architectures described in this section all rely to some degree on processor hardware that is able to securely store secret information like cryptographic keys. To ensure long-term protection of the sensitive information processed by the computer, the hardware architecture must remain secure against any conceivable type of attack. Some examples of methods that an attacker might use to try to recover sensitive information given physical access to a computing device are:

• Physical access into the device: By performing actions such as carefully grinding or chemically eroding the casing of the integrated circuit, the attacker can gain access to internal circuit components. Having achieved this access, it may be possible to probe circuit components electrically and extract information from them.
• Monitoring electromagnetic emissions: After gaining internal access to a device containing sensitive information, it may be possible to use a microscopic antenna to monitor activity within particular components or information traveling through interconnections.
• Microscopic examination: It may be possible to use exotic laboratory instrumentation such as a scanning voltage microscope to measure the voltage distribution across the surface of a circuit such as a memory array.

While the deployment of some of the attack techniques described in this section against ordinary computer systems may seem to be extremely unlikely, you may be surprised at how accessible these techniques can be. Although most of us do not own or have access to a scanning voltage microscope, there are companies that own these devices and are willing to perform scans for customers for a surprisingly reasonable price. This is one reason why it is important for architects of computer systems that will contain extremely sensitive or valuable information to take all potential attack types into account during system design.

In the next section, we will discuss confidential computing, where strong security measures are employed to ensure data is protected through its processing lifecycle.

Confidential computing

Confidential computing is a recent development that aims to use cryptography and hardware-level security features to ensure data is always protected. Data can be in one of three states: at rest, in transit, or in use. Data at rest is typically located in files on a storage device. Data in transit refers to data traveling over some type of communication medium. Data in use is being actively operated on by a processor and resides in the processor’s main memory.

Confidential computing aims to ensure a comprehensive level of protection for data in all three of these possible states. Traditional security mechanisms focus on one state at a time, such as encrypting data on disk or while transferring information to and from a website. These approaches neglect the necessity to provide the same level of protection to data in use.

Securing data in use requires support from processor hardware to isolate applications from each other and to ensure the protection of sensitive code and data. As one example of hardware supporting confidential computing, the Intel Secure Guard Extensions (SGX) enable robust application isolation and protection of data in use.

Intel claims that SGX protects application data even when the BIOS, operating system, and the application itself have been compromised and the attacker has full control of the platform.

The SGX technology creates isolated memory regions called enclaves. Each enclave contains non-addressable memory regions that hold an application’s code and data in encrypted memory pages.

Applications built with SGX technology consist of two parts: an untrusted part and a trusted part. The untrusted part of the application creates the trusted part, which contains the secure enclave. Enclaves are considered trusted because they cannot be modified after they have been built. If an enclave is modified, the change will be detected and it will not be permitted to execute. Code running within the trusted part of the application accesses data within the enclave as clear (unencrypted) text. Any code that is outside the enclave, such as the BIOS, operating system, or even the untrusted part of the same application, is denied access to the enclave memory region. Even code outside the enclave that is executing at the kernel privilege level is unable to gain access to the secure data within the enclave.

Memory pages within a secure enclave can be evicted to secondary storage using traditional page swapping algorithms. The encryption of code and data within the enclave pages protects the information they contain when secure pages are in the untrusted swap file.

The SGX technology supports the concept of software attestation. Using a software attestation procedure, code that is remotely accessing the functionality offered by a secure enclave can validate it is communicating with the specific enclave it intends to work with and not with an imposter. The attestation procedure relies on the exchange of cryptographic digital signatures to reliably verify the identity of the enclave.

Confidential computing is particularly applicable in the context of remote computing. Remote computing describes the use of computing resources that are owned and operated by an untrusted party that is not the application owner.

This description applies, for example, to a company that uses a cloud service provider to operate its enterprise computing capabilities. A remote application can work with sensitive data in the untrusted computing environment and maintain security throughout all phases of operation.

An example of remote computing is a commerce web server running in a commercial cloud environment. The web application collects and retains sensitive information such as credit card details for the site’s customers. Using SGX technology and other standard cryptographic techniques, it is possible to encrypt each customer’ sensitive data within the user’s computer and transport the data into a secure enclave for order processing. To update the user’s record within the application database, the sensitive information is encrypted within the secure enclave before passing it out for storage in the database. At every instant after the user types the sensitive information into their browser, the information is secured cryptographically and protected against malicious actors who may have achieved a presence within the untrusted computing environment.

Intel SGX technology was released in 2015 and is present in most modern Intel processors. To make use of this technology, a computer system must provide BIOS support that takes the steps required to enable it. There is broad support for SGX across motherboards and computers for sale today, but it is not automatically available in all systems. If you decide your application requires SGX, you will need to ensure the processor, motherboard, BIOS, and operating system of the computer you plan to use all provide support for this technology.

Unfortunately, as with many earlier attempts to improve computer security through hardware advances, security researchers have identified vulnerabilities in the SGX technology. In fact, researchers have demonstrated the ability to extract cryptographic keys and other valuable information from applications running within SGX secure enclaves.

The technique attackers use against SGX rely on the eager (or speculative) execution features of modern processors. As we discussed Chapter 8, Performance-Enhancing Techniques, eager execution is an optimization technique in which a processor begins executing code along both paths leading from a branch instruction until it becomes clear which path from the branch the code will take. Eager execution results in data being stored in cache memory that will ultimately be discarded. This cached data is the source of the vulnerability.

The specific attack technique against SGX is called load value injection (LVI). An LVI attack can not only read values from the supposedly secure enclave – it can also inject data values into the enclave. Intel has identified steps that software developers can take to mitigate LVI; however, the mitigations may impose a substantial performance impact on software execution. Intel has also been working to provide updates to processor microcode that will resolve this vulnerability.

Designing for security at the architectural level

The design process for a highly secure computer system must factor in a wide range of security requirements from the very start of the process. All aspects of system design, such as processor selection and printed circuit board characteristics, must be evaluated in the light of how those components may enhance the system’s overall security, or the ways they might degrade security.

In addition to ensuring security at the lowest level of integrated circuits and printed circuit boards, it is also important to apply security design principles in a layered manner. For example, after selecting suitably secure digital components and designing a security-focused circuit layout for a digital device, it may be necessary to design a tamper-resistant case to enclose the circuit board. This case might include embedded wires intended to detect if an attacker attempts to cut or drill holes through the case and thereby gain access to the internal components. This approach is often used in financially critical end-user devices such as point-of-sale credit card readers.

The best system security comes from applying a secure design methodology at all layers of system design, both in hardware and software. The following sections will review some design principles that will help ensure a secure system design.

Avoid security through obscurity

One tempting approach that has been used over the years in digital system development is to try to make it as unobvious as possible what the different components and connections within a digital device are and what they do. One step in this effort might involve grinding off the nomenclature from some of the integrated circuits within a device, thereby making it more difficult for someone analyzing the device to identify the components used to construct it.

Another way of obscuring the functionality available within a system is to run some of the wiring on a printed circuit board in an apparently nonsensical manner. The goal here is to make it as challenging as possible for someone to examine the device and reverse engineer its design.

The term reverse engineering has long been used to describe the analytical process of understanding how a device or software program was constructed without access to any of the documentation that went into the development process. While there are legitimate reasons for performing reverse engineering (such as when repairing a system for which no documentation exists), our focus is on more malicious applications of this process. Malicious reverse engineering occurs when someone attempts to gain access to protected information such as trade secrets or copyrighted software with nefarious intent.

Historically, the greatest flaw in the use of intentionally obscure designs in valuable digital systems has been the degree to which the system designers have underestimated the capabilities and resourcefulness of reverse engineers. It seems the designers who relied on the obscurity-insertion process assumed that because they themselves would never endure the tedium required to reverse engineer an intentionally baroque, overly complicated design, no one else would try to do it either.

There have been many cases in which reverse engineers have been able to perform tasks such as identifying the individual signals associated with a debugging interface hidden within a complex circuit board design. With knowledge of those connections, they were able to connect a hardware debugging system and extract all of the proprietary code and other information contained within the device.

It is common among experts in the development of a particular system to assume that only they could possibly understand the design and behavior of such a complicated device, and their understanding is only possible because they have access to system documentation that is unavailable to an outside attacker. This assumption is often invalid for two reasons:

• First, many reverse engineers are highly intelligent and detail-oriented, enabling them to methodically map out a system’s features and behavior in a surprisingly complete manner.
• Second, it may not be reasonable to assume the system documentation will remain secure and hidden from attackers over the long term. Cyberattacks and industrial espionage against high technology companies are widespread. There is a significant possibility that determined attackers will gain access to some or all of the system documentation at some point. This possibility must be considered when evaluating the use of obscurity injection rather than using proven security mechanisms to protect critical information.

After dispensing with the approach of using obscure design processes to hinder attackers attempting to compromise a system, we must turn to effective, proven secure design approaches that can be expected to perform reliably now and in the future.

Comprehensive secure design

Using the approaches discussed earlier in this chapter, a highly secure computer system design must be approached from its fundamental components through to the final details to ensure the highest level of security. Where feasible, it should be possible to mathematically prove the system is secure by design. While it will not be possible to prove security in all aspects of a system architecture (because with even the most basic software application it is usually infeasible to demonstrate security mathematically) it may be possible to do so with critical aspects of a system design, such as the user login process.

By identifying all access points through which a cyberattacker may attempt to exploit a system and demonstrating a comprehensive security implementation at each of those points, system designers can demonstrate a level of confidence in the security of the system against attacks. This analysis must incorporate all the potential exploitation methods described earlier in this chapter, including those methods that are perceived to be most unlikely, if a very high level of system security is required.

In addition to incorporating security concerns across all aspects of the system hardware and software design, it is important to ensure that users of the system operate with the minimum level of privilege they require to perform their job duties. This is the topic of the next section.

The principle of least privilege

While some users require privileged access to manage and maintain a secure computer system, many other users require only the minimal privilege of a general user to do their jobs. The principle of least privilege summarizes the idea that each user should have no higher privilege on the computer system than is required for their job responsibilities. For example, a user that needs to examine and update information in a corporate database should have the privileges required to perform those tasks, but should not have additional privileges such as those required for performing system administration tasks that are not part of the job description.

The principle of least privilege ensures that users have the authorizations and access rights they require to do their jobs, but nothing beyond that.

When a user is transferred to a different job function, or is terminated, it is particularly important for a security-conscious organization to update the employee’s access rights immediately upon the change in duties to ensure the employee (who may be disgruntled, if disciplined or terminated) no longer has access to information that is no longer part of their job description.

The effective use of the principle of least privilege requires more than a careful design process that grants minimal privileges to users and applications during system development. It is important for system administrators and operators to ensure on an ongoing basis that any newly introduced applications and users are given only the minimum privilege levels required to perform authorized functions.

Zero trust architecture

The traditional approach to secure computer system design described in the previous sections relies on a layered security model, with the goal that even if a particular security feature fails, the remaining layers will be sufficient to maintain overall security.

Given the wide range of successful attacks we see regularly publicized that feature total failure of security in systems that had been trusted to protect critical personal, business, and government data, this approach clearly has some limitations.

In the traditional security model, the outer perimeter of the network is considered the security boundary. Inside this protective layer, communication on the internal network is considered trusted and many types of access are available to computer systems and users.

In the zero trust architectural model, any communication that reaches a computer system containing protected information must be treated as potentially hostile. The request will only be processed after the source has been authenticated and the validity of the requested action has been confirmed.

The United States National Security Agency (NSA) has identified a set of guiding principles to apply when developing a zero trust computer architecture:

• Never trust, always verify: Every computer system, user, network device, or other source of data on the network must always be treated as untrusted. Every recipient of user input or other consumer of data must explicitly authenticate every piece of input data received. In addition, when granting privileges to users or applications the principle of least privilege dictates that only the minimum required privileges are granted.
• Assume breach: Assume at all times that hostile actors have penetrated the network and operate accordingly. This means the default response to any input or service request is to ignore or deny the request and only respond after rigorously validating the source and confirming the requested action is allowable. System administrators and automated tools must continuously monitor all aspects of the network configuration and identify when unauthorized access attempts or configuration changes occur and react promptly to stop any malicious activity and restore to an approved operating configuration. All relevant user, application, and network activity must be logged and inspected to rapidly detect any deviation from approved operations.
• Verify explicitly: Each access attempt to protected resources by a user or application requires a separate verification operation. The verification process must incorporate multiple attributes of the requesting user or application to reliably authenticate the source of the request. For example, the system can require the use of two-factor authentication, where a user inserts a cryptographically secure access card into a reading device and then enters a PIN. The card and the PIN represent two independent attributes that enable authentication as a trusted user.

To fully implement a zero trust architecture requires the implementation of a decision engine that assesses access decisions in the context of all available information about the requester and the destination of the request. In addition to authenticating the requester and verifying that the least privilege is applied to the request, the decision engine must address any additional information that may indicate an increased risk associated with the request. Only after validating all available information and determining that the risk associated with granting the request does not exceed a predefined threshold is the user’s request permitted to proceed.

The next section will examine some of the ways in which security weaknesses in software running on otherwise secure hardware can introduce vulnerabilities into a computer system.

Ensuring security in system and application software

When developing a secure system design at the hardware level, it is important to follow through with a secure and verified design approach at all levels of the software architecture. We will next examine some of the ways that software code can introduce vulnerabilities into otherwise secure computer systems.

Common software weaknesses

This section lists several categories of software weaknesses that have traditionally caused severe security issues in operating systems, applications, and web servers. These vulnerabilities sometimes occur because software developers make assumptions about user behavior that turn out to be invalid. At other times, developers are simply unaware that particular software patterns lead to insecure designs.

Some of the techniques listed below are more likely to occur in specific programming languages, but software developers should be aware that it is possible to create insecure code in any programming language.

This is a list of some of the most common software weaknesses that have been observed in recent years.

Buffer overflow

A buffer overflow occurs when an input data set is larger than the memory allocated for input and the excess data overwrites memory that may contain other important data. This has traditionally been an issue with software written in the C and C++ programming languages. In a typical scenario, the code prompts a user to enter a text string that is expected to be short, such as a username.

The developer may have allocated a buffer of perhaps 80 characters to receive the input, assuming no username would be even close to that long. If the code receiving the input accepts more than 80 characters of input and stores it into memory, the data at addresses beyond the 80-character buffer will be overwritten. In traditional C language implementations, the input buffer is likely to be located within the processor stack. By inserting a carefully constructed input string, an attacker may be able to insert code containing processor instructions of the attacker’s choosing and also overwrite the return address for the current function, which is stored on the stack. By overwriting the return address with the address of the malicious code, the attacker can execute any instructions they desire beginning when the input function returns. The initial goal of the attacker is usually an attempt to achieve remote access to a command prompt (also called a command shell) on the victim system. For this reason, the injected code that executes this type of exploit is called shellcode.

Cross-site scripting

Cross-site scripting is a vulnerability specific to web applications. In a cross-site scripting attack, the attacker finds a way of injecting malicious executable scripts into web pages that are then delivered from a reputable site to its users. These scripts execute in the context of the victim’s browser and can potentially perform actions such as forwarding the victim’s login credentials (such as an authorization cookie) to the attacker. This allows the attacker to impersonate the victim on the reputable website and gain access to the victim’s personal information. Several mechanisms are available for software developers to validate and sanitize input from website users that will effectively prevent cross-site scripting. Unfortunately, not all web developers take full advantage of these security features, so this vulnerability continues to exist on many websites.

SQL injection

As discussed earlier in this chapter, many web applications use a database to store user information such as usernames, hashed passwords, and site content like user posts and uploaded images. Many sites use SQL to manipulate information within the database and insert new data in response to user inputs. A potential vulnerability arises if a user intentionally provides input to the site that may be interpreted as SQL code, and the site fails to sanitize its input in a manner that prevents the execution of the user-provided code.

As an example, this server code retrieves the username typed into a box on a web page after a user clicks the Submit button:

txtUserName = getRequestString("UserName"); 

In the next processing step, a naïve implementation of server code might create a command in SQL syntax that will retrieve the data record associated with the provided username from the database:

txtSqlCmd = 'SELECT * FROM Users where UserName = "' + txtUserName + '"';

For example, if the user entered Alice as the username, the text string containing the resulting SQL command would be:

SELECT * FROM Users where UserName = "Alice"

The server then passes this command to the database command interpreter and, if a valid username was provided, the user’s data record will be returned.

The problem with this approach is the use of the text string entered by the user directly within the SQL command string. A malicious user can use SQL syntax to alter the behavior of the database access operation. For example, instead of entering a valid username the attacker might enter the following text string:

" or ""="

The SQL command that results when this string is provided as the username is:

SELECT * FROM Users where UserName = "" or ""=""

This command tells the database to return all records where the username is an empty string (which is probably none of them) or where an empty string is equal to an empty string. Since an empty string is always equal to an empty string, the condition will be true for all records and the database will return all the user records in the database. If the server code then presents the results of the SQL command to the user’s browser, the attacker will have succeeded in extracting the entire user database for the website.

While this might seem like an obscure method for attacking a website, you should understand that many websites, often owned by famous and richly valued companies, have fallen prey to devastating attacks very similar to the one described here.

Path traversal

A path traversal vulnerability occurs when a network application, typically a web server, unintentionally grants some degree of access to its directory structure to users of the application. Normally, the site operators intend for users to access subdirectories beneath a main application directory to retrieve data organized in those directories. The vulnerability may arise if the logic in the web server allows users to go up one or more directory levels using this technique.

In both the Windows and Linux operating systems, a directory path element consisting of two adjacent periods means to go up a directory level. As an example, the following URL shows how to attempt to retrieve the file containing encrypted passwords on a Linux system running a standard web server:

http://www.example.com/../../../../etc/shadow

If successful, visiting this URL in a browser enables the attacker to retrieve the hashed versions of all passwords on the system. The attacker can then use brute force password cracking techniques to attempt to recover the users’ passwords.

This section has listed just a few of the most common software vulnerabilities that have been exploited historically to cause severe harm to companies and individuals with private information stored on computer systems.

The Common Weakness Enumeration (CWE) database at https://cwe.mitre.org/index.html contains a list of software and hardware weaknesses derived from inputs provided by users around the world. In particular, the Top 25 Most Dangerous Software Weaknesses are listed at https://cwe.mitre.org/data/definitions/1337.html. This list provides a broad overview of the currently observed software weaknesses causing serious security issues.

Source code security scans

One way to quickly gain information about security weaknesses present in an existing body of software is to use an automated source code security scanning tool to assess the codebase and categorize the issues identified within it by severity.

Depending on the programming language or languages used to write your code, you may be able to find free tools to perform this type of scan. For any free code scanner you are considering using, be sure to examine reviews from users and ensure you are comfortable that it is a legitimately useful tool. For some programming languages, it may be necessary to purchase a security scanning tool, which may be quite expensive.

Automated security scanning tools can identify many categories of problems with source code, including the weaknesses listed above as well as others identified in the CWE database. These tools also identify other issues in the code such as the presence of deprecated features and the use of performance-compromising constructs.

Rather than attempt to list these tools here, I suggest you perform web searches for automated security scanning tools targeted at the languages you use for your critical applications.

Summary

This chapter introduced computing architectures suitable for uses that require an exceptional assurance of security. Critical application areas such as national security systems and financial transaction processing demand this high level of protection. These systems must be resilient against a broad range of cybersecurity threats including malicious code, covert channel attacks, and attacks enabled by physical access to the computing hardware. Topics addressed in this chapter included cybersecurity threats, encryption, digital signatures, and secure hardware and software design.

Having completed this chapter, you are able to identify many of the categories of cybersecurity threats a system will face and understand the security features of modern computer hardware. You understand some best practices for avoiding security gaps in system architectures and know how a secure computer architecture can help enforce security in software applications.

In the next chapter, we will introduce the concepts associated with blockchain, a public, cryptographically secured ledger recording a sequence of transactions. The chapter provides an overview of the bitcoin mining process and discusses the hardware architectures of bitcoin mining computer systems.

Exercises

1. Where supported, set up two-factor authentication for all your internet-accessible accounts containing data that you care about. This includes bank accounts, email accounts, social media, code repositories (if you are a software developer), medical services, and anything else you value. Ensure at all stages that you are using only information and software applications from trusted sources.
2. Create strong passwords for all your internet-accessible accounts containing information of value that cannot be protected by two-factor authentication. A strong password is long (15 characters or more) and includes uppercase, lowercase, numerical, and special characters (for example: ! " # $ % & ' ( ) * +). To keep track of these complicated passwords, install and use a reputable password safe application. Use care when selecting a password safe and consider its source.
3. Update the operating system and other applications and services (such as Java) on all computers and other devices under your control. This will ensure the security updates included in those updates start working to protect you soon after they become available. Set up a plan to continue regularly installing updates as they are released to ensure you are protected in the future.

Join our community Discord space

Join the book’s Discord workspace for a monthly Ask me Anything session with the author: https://discord.gg/7h8aNRhRuY