Currently, I am working for a telecom company specialized in IoT devices. Essentially, we provide cellular communication options for IoT projects worldwide. I interact daily with customers deploying their devices in a variety of fields: vehicle tracking, energy monitoring and provisioning, security and industrial applications. While their applications are all different, I noticed that one thing never changes: the security threats to those devices. Regularly, every month, I have to deal with customers whose devices have been infected by malwares. While the damage to their operation is usually minimal, the data communication charges are not, leading to significant monetary consequences.
Given the prevalence of these situations, I thought I would dig deeper into the world of IoT security, to understand why these excess charges are happening. During this quest, I discovered the source of IoT malwares: Mirai, a botnet developed in 2016. Therefore, I used the opportunity of this paper to study Mirai, how it works and how we can defend from it.
In the fall of 2016, multiple high-profile websites (e.g. Netflix, GitHub, Reddit) found themselves rendered inaccessible by a very powerful DDoS attack, estimated, by some accounts, to have reached 1.2 Tbps (The Economist, 2016), an unprecedented size.
After that attack, the source code and instructions of a new IoT malware named Mirai was release as open-source on a hacking forum. Since then, multiple derivative malwares have been created that are now bringing into spotlight the multiple security issues of the IoT world.
Mirai is a malware engineered to perform DDoS attacks (Distributed Denial of Service). The aim of those attacks is to overwhelm the target server by flooding it with superfluous requests, in order to interrupt its services and prevent legitimate clients to acces it (C. Douligeris, 2004). To achieve this aim, Mirai discovers, infects and controls unprotected (or loosely protected) IoT devices connected to the internet, to use them as bots and coordinating them to carry DDoS operations (Margolis, et al., 2017).
Its architecture is shown in Diagram 1. Initially, the Command-and-Control (C&C) servers runs two socket listeners: one on port 23 for Telent connections and one on port 101 for programmatic API. The C&C server is written in Go.
Depending on the type of data sent to the Telnet socket (either a 4-byte integer or something else), an Admin Handler or Bot Handler are created. The Admin handler is an interactive prompt that allows users to manage bots and attacks (all data is stored in a MySQL DB); on the contrary, a Bot Handler handles the connection of a new bot. If the former successfully communicates with the latter, a new bot is created and its ID added to the above database.
The API socket, instead, has only one function: listening for orders of attacks. The syntax of these orders is as follows:
-n <attack type> <18.104.22.168[/8][,] …>
<attack length> [flag=val …]
The -n controls the number of bots to use, the IP can be provided either as a single value or as a range, while the attack length defines the duration of the attack in seconds (from 0 to 3600). The attack type, chosen from 10 different types, defines the “flavor” of the malicious operation.
To infect a new device, each bot randomly selects and IP and then checks it against a table of forbidden addresses; to note, the IPs of the US Postal Service and of the Hewlett-Packard Corp. are expressly protected from the attacks. If the IP is valid, the bot will then launch a dictionary attack using a list of 60 factory default login credentials that can be seen in Table 1.
After a new device is found, its IP and login credentials are sent to the loader which will contact the device, download and load a binary of the malware. This new bot will then start to infect new devices, creating a cycle that allows Mirai to grow its networks of bot in an exponential fashion.
It is exactly this ability to easily infect devices that makes Mirai a formidable threat. While the types of attacks it can perform are nothing new, its worming capabilities are exceptionally dangerous. In addition, the source code of Mirai has been released as open source since 2017 (Gamblin, 2017); consequently, many derivatives malware have been developed since then.
Mitigating the risks posed by Mirai is somewhat difficult in that it uses legitimate credentials to gain access to devices which are usually not scanned for malicious activity. Nevertheless, there are some important defensive actions that can significantly reduce the risk of infection:
- Change device credentials: after the deployment of the device, the user must change the access credentials for SSH or Telnet access. It is important to note that the credentials that can be changed via the device GUI are not always the credentials used for SSH or Telent connections.
- Closed unused ports: as described before, ports 22 and 23 are the ones exploited by Mirai to gain access. Hence, they should never be publicly accessible. Moreover, they should be blocked at router level to avoid any access to internal devices.
- Monitor ioctl: Mirai (and its derivatives) have the common pattern of sending an ioctl request to the watchdog timer. This is done to prevent the watchdog to restart unexpectedly. Given that the watchdog is a fundamental fail-safe system for Linux IoT devices (Weingel, 2007), its disabling should never occur. Hence, monitoring its activity is a very important step to identify the presence of Mirai.
- White-hat script: an automated white-hat penetration script can be used to identify vulnerable devices in the network. Once they are found, the script should attempt login in a manner similar to Mirai. This method could help in identifying exposed devices in very large network, where the risk of human error is greater.
Other security considerations for IoT
During my research, I also encountered many other challenges, other than malware, in securing an IoT project (Zhang, et al., 2014):
- Identifying objects in the network: until now, most IoT application used the DNS system to identify (and name) objects in the network. However, such system is still vulnerable to cache poisoning & man-in-the-middle attacks, which can inject fake DNS records in the target cache.
- Authentication & authorization: while many public-key (such as the ones studied during this course) provide a theoretically sound system from authentication & authorization, the absence of a global root Certificate Authority prevents those crypto-systems to be effective. In addition, it may prove impractical to provide a certificate to each IoT object given the sheer number of devices.
- Cryptosystems & security protocols: public-key cryptosystems are highly desirable because they generally provide advanced security features. However, they are very often inoperable on IoT systems, given the resource constraints that these objects have.
In addition, (Zarca, et al., 2018) also point out that Software Defined Networks (SDN) could help in devising new defensive approaches. According to the authors, SDN have a number of defensive benefits:
- Dynamic Flow Control: by decoupling the data from the network plane, there is now the possibility to enable dynamic access control functions depending on specified privileges and policies
- Traffic Isolation: which allows to flexibly isolate compromised sections of the network
- Network-wide visibility and monitoring: given that the SDN traffic is managed by a centralized controller, there is now the possibility of monitoring the traffic peak generated by compromised devices in the network.
The Economist, 2016. The internet of stings. [Online]
Available at: https://www.economist.com/science-and-technology/2016/10/08/the-internet-of-stings
[Accessed 3 February 2021].
C. Douligeris, A. M., 2004. DDoS attacks and defense mechanisms: classification and state-of-the-art.Computer Networks, 5 April, pp. 643-666.
Gamblin, J., 2017. GitHub Mirai-Source-Code. [Online]
Available at: https://github.com/jgamblin/Mirai-Source-Code
[Accessed 06 February 2021].
Weingel, C., 2007. The Linux Watchdog driver API. [Online]
Available at: https://www.kernel.org/doc/html/latest/watchdog/watchdog-api.html
[Accessed 6 February 2021].
Margolis, J. et al., 2017. An In-Depth Analysis of the Mirai Botnet. Altoona, PA,, International Conference on Software Security and Assurance (ICSSA).
Zhang, Z., Cho, M. C. Y., Wang, C. & Hsu, C., 2014. IoT Security: Ongoing Challenges and Research Opportunities. Matsue, Japan, IEEE 7th International Conference on Service-Oriented Computing and Applications.
Zarca, A. M., Bernabe, J. B., Farris, I. & Khettab, Y., 2018. Enhancing IoT security through network softwarizationand virtual security appliances. International Journal of Network Management, 28(5).