This article by Peter Waher, author of the book, Learning Internet of Things, we will focus on the security, interoperability, and what issues we need to address during the design of the overall architecture of Internet of Things (IoT) to avoid many of the unnecessary problems that might otherwise arise and minimize the risk of painting yourself into a corner. You will learn the following:

Risks with IoT
Modes of attacking a system and some counter measures
The importance of interoperability in IoT

(For more resources related to this topic, see here.)

Understanding the risks

There are many solutions and products marketed today under the label IoT that lack basic security architectures. It is very easy for a knowledgeable person to take control of devices for malicious purposes. Not only devices at home are at risk, but cars, trains, airports, stores, ships, logistics applications, building automation, utility metering applications, industrial automation applications, health services, and so on, are also at risk because of the lack of security measures in their underlying architecture. It has gone so far that many western countries have identified the lack of security measures in automation applications as a risk to national security, and rightly so. It is just a matter of time before somebody is literally killed as a result of an attack by a hacker on some vulnerable equipment connected to the Internet. And what are the economic consequences for a company that rolls out a product for use on the Internet that results into something that is vulnerable to well-known attacks?

How has it come to this? After all the trouble Internet companies and applications have experienced during the rollout of the first two generations of the Web, do we repeat the same mistakes with IoT?

Reinventing the wheel, but an inverted one

One reason for what we discussed in the previous section might be the dissonance between management and engineers. While management knows how to manage known risks, they don't know how to measure them in the field of IoT and computer communication. This makes them incapable of understanding the consequences of architectural decisions made by its engineers. The engineers in turn might not be interested in focusing on risks, but on functionality, which is the fun part.

Another reason might be that the generation of engineers who tackle IoT are not the same type of engineers who tackled application development on the Internet. Electronics engineers now resolve many problems already solved by computer science engineers decades earlier. Engineers working on machine-to-machine (M2M) communication paradigms, such as industrial automation, might have considered the problem solved when they discovered that machines could talk to each other over the Internet, that is, when the message-exchanging problem was solved. This is simply relabeling their previous M2M solutions as IoT solutions because the transport now occurs over the IP protocol. But, in the realm of the Internet, this is when the problems start. Transport is just one of the many problems that need to be solved.

The third reason is that when engineers actually re-use solutions and previous experience, they don't really fit well in many cases. The old communication patterns designed for web applications on the Internet are not applicable for IoT. So, even if the wheel in many cases is reinvented, it's not the same wheel. In previous paradigms, publishers are a relatively few number of centralized high-value entities that reside on the Internet. On the other hand, consumers are many but distributed low-value entities, safely situated behind firewalls and well protected by antivirus software and operating systems that automatically update themselves. But in IoT, it might be the other way around: publishers (sensors) are distributed, very low-value entities that reside behind firewalls, and consumers (server applications) might be high-value centralized entities, residing on the Internet. It can also be the case that both the consumer and publisher are distributed, low-value entities who reside behind the same or different firewalls. They are not protected by antivirus software, and they do not autoupdate themselves regularly as new threats are discovered and countermeasures added. These firewalls might be installed and then expected to work for 10 years with no modification or update being made. The architectural solutions and security patterns developed for web applications do not solve these cases well.

Knowing your neighbor

When you decide to move into a new neighborhood, it might be a good idea to know your neighbors first. It's the same when you move a M2M application to IoT. As soon as you connect the cable, you have billions of neighbors around the world, all with access to your device.

What kind of neighbors are they? Even though there are a lot of nice and ignorant neighbors on the Internet, you also have a lot of criminals, con artists, perverts, hackers, trolls, drug dealers, drug addicts, rapists, pedophiles, burglars, politicians, corrupt police, curious government agencies, murderers, demented people, agents from hostile countries, disgruntled ex-employees, adolescents with a strange sense of humor, and so on. Would you like such people to have access to your things or access to the things that belong to your children?

If the answer is no (as it should be), then you must take security into account from the start of any development project you do, aimed at IoT. Remember that the Internet is the foulest cesspit there is on this planet. When you move from the M2M way of thinking to IoT, you move from a nice and security gated community to the roughest neighborhood in the world. Would you go unprotected or unprepared into such an area? IoT is not the same as M2M communication in a secure and controlled network. For an application to work, it needs to work for some time, not just in the laboratory or just after installation, hoping that nobody finds out about the system. It is not sufficient to just get machines to talk with each other over the Internet.

Modes of attack

To write an exhaustive list of different modes of attack that you can expect would require a book by itself. Instead, just a brief introduction to some of the most common forms of attack is provided here. It is important to have these methods in mind when designing the communication architecture to use for IoT applications.

Denial of Service

A Denial of Service (DoS) or Distributed Denial of Service (DDoS) attack is normally used to make a service on the Internet crash or become unresponsive, and in some cases, behave in a way that it can be exploited. The attack consists in making repetitive requests to a server until its resources gets exhausted. In a distributed version, the requests are made by many clients at the same time, which obviously increases the load on the target. It is often used for blackmailing or political purposes.

However, as the attack gets more effective and difficult to defend against when the attack is distributed and the target centralized, the attack gets less effective if the solution itself is distributed. To guard against this form of attack, you need to build decentralized solutions where possible. In decentralized solutions, each target's worth is less, making it less interesting to attack.

Guessing the credentials

One way to get access to a system is to impersonate a client in the system by trying to guess the client's credentials. To make this type of attack less effective, make sure each client and each device has a long and unique, perhaps randomly generated, set of credentials. Never use preset user credentials that are the same for many clients or devices or factory default credentials that are easy to reset. Furthermore, set a limit to the number of authentication attempts per time unit permitted by the system; also, log an event whenever this limit is reached, from where to which credentials were used. This makes it possible for operators to detect systematic attempts to enter the system.

Getting access to stored credentials

One common way to illicitly enter a system is when user credentials are found somewhere else and reused. Often, people reuse credentials in different systems. There are various ways to avoid this risk from happening. One is to make sure that credentials are not reused in different devices or across different services and applications. Another is to randomize credentials, lessening the desire to reuse memorized credentials. A third way is to never store actual credentials centrally, even encrypted if possible, and instead store hashed values of these credentials. This is often possible since authentication methods use hash values of credentials in their computations. Furthermore, these hashes should be unique to the current installation. Even though some hashing functions are vulnerable in such a way that a new string can be found that generates the same hash value, the probability that this string is equal to the original credentials is miniscule. And if the hash is computed uniquely for each installation, the probability that this string can be reused somewhere else is even more remote.

Man in the middle

Another way to gain access to a system is to try and impersonate a server component in a system instead of a client. This is often referred to as a Man in the middle (MITM) attack. The reason for the middle part is that the attacker often does not know how to act in the server and simply forwards the messages between the real client and the server. In this process, the attacker gains access to confidential information within the messages, such as client credentials, even if the communication is encrypted. The attacker might even try to modify messages for their own purposes.

To avoid this type of attack, it's important for all clients (not just a few) to always validate the identity of the server it connects to. If it is a high-value entity, it is often identified using a certificate. This certificate can both be used to verify the domain of the server and encrypt the communication. Make sure this validation is performed correctly, and do not accept a connection that is invalid or where the certificate has been revoked, is self-signed, or has expired.

Another thing to remember is to never use an unsecure authentication method when the client authenticates itself with the server. If a server has been compromised, it might try to fool clients into using a less secure authentication method when they connect. By doing so, they can extract the client credentials and reuse them somewhere else. By using a secure authentication method, the server, even if compromised, will not be able to replay the authentication again or use it somewhere else. The communication is valid only once.

Sniffing network communication

If communication is not encrypted, everybody with access to the communication stream can read the messages using simple sniffing applications, such as Wireshark. If the communication is point-to-point, this means the communication can be heard by any application on the sending machine, the receiving machine, or any of the bridges or routers in between. If a simple hub is used instead of a switch somewhere, everybody on that network will also be able to eavesdrop. If the communication is performed using multicast messaging service, as can be done in UPnP and CoAP, anybody within the range of the Time to live (TTL) parameter (maximum number of router hops) can eavesdrop.

Remember to always use encryption if sensitive data is communicated. If data is private, encryption should still be used, even if the data might not be sensitive at first glance. A burglar can know if you're at home by simply monitoring temperature sensors, water flow meters, electricity meters, or light switches at your home. Small variations in temperature alert to the presence of human beings. Change in the consumption of electrical energy shows whether somebody is cooking food or watching television. The flow of water shows whether somebody is drinking water, flushing a toilet, or taking a shower. No flow of water or a relatively regular consumption of electrical energy tells the burglar that nobody is at home. Light switches can also be used to detect presence, even though there are applications today that simulate somebody being home by switching the lights on and off.

If you haven't done so already, make sure to download a sniffer to get a feel of what you can and cannot see by sniffing the network traffic. Wireshark can be downloaded from https://www.wireshark.org/download.html.

Port scanning and web crawling

Port scanning is a method where you systematically test a range of ports across a range of IP addresses to see which ports are open and serviced by applications. This method can be combined with different tests to see the applications that might be behind these ports. If HTTP servers are found, standard page names and web-crawling techniques can be used to try to figure out which web resources lie behind each HTTP server. CoAP is even simpler since devices often publish well-known resources. Using such simple brute-force methods, it is relatively easy to find (and later exploit) anything available on the Internet that is not secured.

To avoid any private resources being published unknowingly, make sure to close all the incoming ports in any firewalls you use. Don't use protocols that require incoming connections. Instead, use protocols that create the connections from inside the firewall. Any resources published on the Internet should be authenticated so that any automatic attempt to get access to them fails.

Always remember that information that might seem trivial to an individual might be very interesting if collected en masse. This information might be coveted not only by teenage pranksters but by public relations and marketing agencies, burglars, and government agencies (some would say this is a repetition).

Search features and wildcards

Don't make the mistake of thinking it's difficult to find the identities of devices published on the Internet. Often, it's the reverse. For devices that use multicast communication, such as those using UPnP and CoAP, anybody can listen in and see who sends the messages. For devices that use single-cast communication, such as those using HTTP or CoAP, port-scanning techniques can be used. For devices that are protected by firewalls and use message brokers to protect against incoming attacks, such as those that use XMPP and MQTT, search features or wildcards can be used to find the identities of devices managed by the broker, and in the case of MQTT, even what they communicate.

You should always assume that the identity of all devices can be found, and that there's an interest in exploiting the device. For this reason, it's very important that each device authenticates any requests made to it if possible. Some protocols help you more with this than others, while others make such authentication impossible.

XMPP only permits messages from accepted friends. The only thing the device needs to worry about is which friend requests to accept. This can be either configured by somebody else with access to the account or by using a provisioning server if the device cannot make such decisions by itself. The device does not need to worry about client authentication, as this is done by the brokers themselves, and the XMPP brokers always propagate the authenticated identities of everybody who send them messages.

MQTT, on the other hand, resides in the other side of the spectrum. Here, devices cannot make any decision about who sees the published data or who makes a request since identities are stripped away by the protocol. The only way to control who gets access to the data is by building a proprietary end-to-end encryption layer on top of the MQTT protocol, thereby limiting interoperability.

In between the two resides protocols such as HTTP and CoAP that support some level of local client authentication but lacks a good distributed identity and authentication mechanism. This is vital for IoT even though this problem can be partially solved in local intranets.

Breaking ciphers

Many believe that by using encryption, data is secure. This is not the case, as discussed previously, since the encryption is often only done between connected parties and not between end users of data (the so-called end-to-end encryption). At most, such encryption safeguards from eavesdropping to some extent. But even such encryption can be broken, partially or wholly, with some effort.

Ciphers can be broken using known vulnerabilities in code where attackers exploit program implementations rather than the underlying algorithm of the cipher. This has been the method used in the latest spectacular breaches in code based on the OpenSSL library. To protect yourselves from such attacks, you need to be able to update code in devices remotely, which is not always possible.

Other methods use irregularities in how the cipher works to figure out, partly or wholly, what is being communicated over the encrypted channel. This sometimes requires a considerable amount of effort. To safeguard against such attacks, it's important to realize that an attacker does not spend more effort into an attack than what is expected to be gained by the attack. By storing massive amounts of sensitive data centrally or controlling massive amounts of devices from one point, you increase the value of the target, increasing the interest of attacking it. On the other hand, by decentralizing storage and control logic, the interest in attacking a single target decreases since the value of each entity is comparatively lower. Decentralized architecture is an important tool to both mitigate the effects of attacks and decrease the interest in attacking a target. However, by increasing the number of participants, the number of actual attacks can increase, but the effort that can be invested behind each attack when there are many targets also decreases, making it easier to defend each one of the attacks using standard techniques.

Tools for achieving security

There are a number of tools that architects and developers can use to protect against malicious use of the system. An exhaustive discussion would fill a smaller library. Here, we will mention just a few techniques and how they not only affect security but also interoperability.

Virtual Private Networks

A method that is often used to protect unsecured solutions on the Internet is to protect them using Virtual Private Networks (VPNs). Often, traditional M2M solutions working well in local intranets need to expand across the Internet. One way to achieve this is to create such VPNs that allow the devices to believe they are in a local intranet, even though communication is transported across the Internet.

Even though transport is done over the Internet, it's difficult to see this as a true IoT application. It's rather a M2M solution using the Internet as the mode of transport. Because telephone operators use the Internet to transport long distance calls, it doesn't make it Voice over IP (VoIP). Using VPNs might protect the solution, but it completely eliminates the possibility to interoperate with others on the Internet, something that is seen as the biggest advantage of using the IoT technology.

X.509 certificates and encryption

We've mentioned the use of certificates to validate the identity of high-value entities on the Internet. Certificates allow you to validate not only the identity, but also to check whether the certificate has been revoked or any of the issuers of the certificate have had their certificates revoked, which might be the case if a certificate has been compromised. Certificates also provide a Public Key Infrastructure (PKI) architecture that handles encryption. Each certificate has a public and private part. The public part of the certificate can be freely distributed and is used to encrypt data, whereas only the holder of the private part of the certificate can decrypt the data.

Using certificates incurs a cost in the production or installation of a device or item. They also have a limited life span, so they need to be given either a long lifespan or updated remotely during the life span of the device. Certificates also require a scalable infrastructure for validating them. For these reasons, it's difficult to see that certificates will be used by other than high-value entities that are easy to administer in a network. It's difficult to see a cost-effective, yet secure and meaningful, implementation of validating certificates in low-value devices such as lamps, temperature sensors, and so on, even though it's theoretically possible to do so.

Authentication of identities

Authentication is the process of validating whether the identity provided is actually correct or not. Authenticating a server might be as simple as validating a domain certificate provided by the server, making sure it has not been revoked and that it corresponds to the domain name used to connect to the server. Authenticating a client might be more involved, as it has to authenticate the credentials provided by the client. Normally, this can be done in many different ways. It is vital for developers and architects to understand the available authentication methods and how they work to be able to assess the level of security used by the systems they develop.

Some protocols, such as HTTP and XMPP, use the standardized Simple Authentication and Security Layer (SASL) to publish an extensible set of authentication methods that the client can choose from. This is good since it allows for new authentication methods to be added. But it also provides a weakness: clients can be tricked into choosing an unsecure authentication mechanism, thus unwittingly revealing their user credentials to an impostor. Make sure clients do not use unsecured or obsolete methods, such as PLAIN, BASIC, MD5-CRAM, MD5-DIGEST, and so on, even if they are the only options available. Instead, use secure methods such as SCRAM-SHA-1 or SCRAM-SHA-1-PLUS, or if client certificates are used, EXTERNAL or no method at all. If you're using an unsecured method anyway, make sure to log it to the event log as a warning, making it possible to detect impostors or at least warn operators that unsecure methods are being used.

Other protocols do not use secure authentication at all. MQTT, for instance, sends user credentials in clear text (corresponding to PLAIN), making it a requirement to use encryption to hide user credentials from eavesdroppers or client-side certificates or pre-shared keys for authentication. Other protocols do not have a standardized way of performing authentication. In CoAP, for instance, such authentication is built on top of the protocol as security options. The lack of such options in the standard affects interoperability negatively.

Usernames and passwords

A common method to provide user credentials during authentication is by providing a simple username and password to the server. This is a very human concept. Some solutions use the concept of a pre-shared key (PSK) instead, as it is more applicable to machines, conceptually at least.

If you're using usernames and passwords, do not reuse them between devices, just because it is simple. One way to generate secure, difficult-to-guess usernames and passwords is to randomly create them. In this way, they correspond more to pre-shared keys.

One problem in using randomly created user credentials is how to administer them. Both the server and the client need to be aware of this information. The identity must also be distributed among the entities that are to communicate with the device. Here, the device creates its own random identity and creates the corresponding account in the XMPP server in a secure manner. There is no need for a common factory default setting. It then reports its identity to a thing registry or provisioning server where the owner can claim it and learn the newly created identity. This method never compromises the credentials and does not affect the cost of production negatively.

Furthermore, passwords should never be stored in clear text if it can be avoided. This is especially important on servers where many passwords are stored. Instead, hashes of the passwords should be stored. Most modern authentication algorithms support the use of password hashes. Storing hashes minimizes the risk of unwanted generation of original passwords for attempted reuse in other systems.

Using message brokers and provisioning servers

Using message brokers can greatly enhance security in an IoT application and lower the complexity of implementation when it comes to authentication, as long as message brokers provide authenticated identity information in messages it forwards.

In XMPP, all the federated XMPP servers authenticate clients connected to them as well as the federated servers themselves when they intercommunicate to transport messages between domains. This relieves clients from the burden of having to authenticate each entity in trying to communicate with it since they all have been securely authenticated. It's sufficient to manage security on an identity level. Even this step can be relieved further by the use of provisioning.

Unfortunately, not all protocols using message brokers provide this added security since they do not provide information about the sender of packets. MQTT is an example of such a protocol.

Centralization versus decentralization

Comparing centralized and decentralized architectures is like comparing the process of putting all the eggs in the same basket and distributing them in many much smaller baskets. The effect of a breach of security is much smaller in the decentralized case; fewer eggs get smashed when you trip over. Even though there are more baskets, which might increase the risk of an attack, the expected gain of an attack is much smaller. This limits the motivation of performing a costly attack, which in turn makes it simpler to protect it against. When designing IoT architecture, try to consider the following points:

Avoid storing data in a central position if possible. Only store the data centrally that is actually needed to bind things together.
Distribute logic, data, and workload. Perform work as far out in the network as possible. This makes the solution more scalable, and it utilizes existing resources better.
Use linked data to spread data across the Internet, and use standardized grid computation technologies to assemble distributed data (for example, SPARQL) to avoid the need to store and replicate data centrally.
Use a federated set of small local brokers instead of trying to get all the devices on the same broker. Not all brokered protocols support federation, for example, XMPP supports it but MQTT does not.
Let devices talk directly to each other instead of having a centralized proprietary API to store data or interpret communication between the two.
Contemplate the use of cheap small and energy-efficient microcomputers such as the Raspberry Pi in local installations as an alternative to centralized operation and management from a datacenter.

The need for interoperability

What has made the Internet great is not a series of isolated services, but the ability to coexist, interchange data, and interact with the users. This is important to keep in mind when developing for IoT. Avoid the mistakes made by many operators who failed during the first Internet bubble. You cannot take responsibility for everything in a service. The new Internet economy is based on the interaction and cooperation between services and its users.

Solves complexity

The same must be true with the new IoT. Those companies that believe they can control the entire value chain, from things to services, middleware, administration, operation, apps, and so on, will fail, as the companies in the first Internet bubble failed. Companies that built devices with proprietary protocols, middleware, and mobile phone applications, where you can control your things, will fail. Why? Imagine a future where you have a thousand different things in your apartment from a hundred manufacturers. Would you want to download a hundred smart phone apps to control them? Would you like five different applications just to control your lights at home, just because you have light bulbs from five different manufacturers? An alternative would be to have one app to rule them all. There might be a hundred different such apps available (or more), but you can choose which one to use based on your taste and user feedback. And you can change if you want to. But for this to be possible, things need to be interoperable, meaning they should communicate using a commonly understood language.

Reduces cost

Interoperability does not only affect simplicity of installation and management, but also the price of solutions. Consider a factory that uses thousands (or hundreds of thousands) of devices to control and automate all processes within. Would you like to be able to buy things cheaply or expensively? Companies that promote proprietary solutions, where you're forced to use their system to control your devices, can force their clients to pay a high price for future devices and maintenance, or the large investment made originally might be lost.

Will such a solution be able to survive against competitors who sell interoperable solutions where you can buy devices from multiple manufacturers? Interoperability provides competition, and competition drives down cost and increases functionality and quality. This might be a reason for a company to work against interoperability, as it threatens its current business model. But the alternative might be worse. A competitor, possibly a new one, might provide such a solution, and when that happens, the business model with proprietary solutions is dead anyway. The companies that are quickest in adapting a new paradigm are the ones who would most probably survive a paradigm shift, as the shift from M2M to IoT undoubtedly is.

Allows new kinds of services and reuse of devices

There are many things you cannot do unless you have an interoperable communication model from the start. Consider a future smart city. Here, new applications and services will be built that will reuse existing devices, which were installed perhaps as part of other systems and services. These applications will deliver new value to the inhabitants of the city without the need of installing new duplicate devices for each service being built. But such multiple use of devices is only possible if the devices communicate in an open and interoperable way. However, care has to be taken at the same time since installing devices in an open environment requires the communication infrastructure to be secure as well. To achieve the goal of building smart cities, it is vitally important to use technologies that allow you to have both a secure communication infrastructure and an interoperable one.

Combining security and interoperability

As we have seen, there are times where security is contradictory to interoperability. If security is meant to be taken as exclusivity, it opposes the idea of interoperability, which is by its very nature inclusive. Depending on the choice of communication infrastructure, you might have to use security measures that directly oppose the idea of an interoperable infrastructure, prohibiting third parties from accessing existing devices in a secure fashion.

It is important during the architecture design phase, before implementation, to thoroughly investigate what communication technologies are available, and what they provide and what they do not provide. You might think that this is a minor issue, thinking that you can easily build what is missing on top of the chosen infrastructure. This is not true. All such implementation is by its very nature proprietary, and therefore not interoperable. This might drastically limit your options in the future, which in turn might drastically reduce anyone else's willingness to use your solution.

The more a technology includes, in the form of global identity, authentication, authorization, different communication patterns, common language for interchange of sensor data, control operations and access privileges, provisioning, and so on, the more interoperable the solution becomes. If the technology at the same time provides a secure infrastructure, you have the possibility to create a solution that is both secure and interoperable without the need to build proprietary or exclusive solutions on top of it.

Summary

In this article, we presented the basic reasons why security and interoperability must be contemplated early on in the project and not added as late patchwork because it was shown to be necessary. Not only does such late addition limit interoperability and future use of the solution, it also creates solutions that can jeopardize not only yourself your company and your customers, but in the end, even national security. This article also presented some basic modes of attack and some basic defense systems to counter them.