Embedded Systems Academy

In a blog entry from last year, we announced a “Two-year project for security of CANopen and other small-packet networks.” It is now time to give you an update on where we are with our SPsec (Securing Small-Packet Networks) project.

It comes as no surprise that adding security to small-packet networks like CAN, I2C, LIN, Modbus, and other fieldbuses is a challenge. The small-packet sizes offer only limited space for security information like an authentication tag and often, these networks are handled by microcontrollers with limited computational and memory resources. We are now aiming at protecting all communication in such a network when our initial goal was to protect only selected communication channels. The reason here is that for many industrial applications, recent acts and regulations like the European Cyber Resilience Act (CRA) will require security-by-design in the near future. For several use cases, they will also request that all data at rest and in motion is both authenticated and encrypted.

We defined the following SPsec key points and cryptographic primitives:

• Minimal hardware requirements of participating MCUs
• Cryptographic functions used
• Point-to-point security for configurations or communications with an limited amount of communication channels
• Time-based rolling key derivation for automated refreshing of keys
• Group security for multicast network technologies like CAN

For more detailed information see our white paper “Cybersecurity Primitives for Small-Packet Networks“.

Our first proof-of-concept implementation will be based on the PCAN-Router FD from PEAK-System. These devices have two CAN (or CAN FD) interfaces from which we use one for unprotected communication from a host system. The router implements a SPsec sub layer and uses the second interface for the secure communication. This allows for easy test and debugging, as there will be one CAN bus with the protected and one with the unprotected communication allowing a direct comparison.

Later the SPsec sub layer will be added to our Micro CANopen source code and integrated into various CANopen or CANopen FD devices for further testing.

Stay informed by following this blog or our linkedin page for up-to-date developments.

The current status of the EU Cyber Resilience Act (CRA) is that manufacturers of devices with digital elements or any software have until 2027 to comply with the outlined rules and regulations. These include compliance issues like overall risk assessment, documentation and incident reporting – which have a huge organizational impact. Technology details mentioned in the CRA are limited, so there is some interpretation as to what it all means for embedded systems and fieldbus communication. When it comes to specifics, the annex talks about how to treat data in transit:

Text excerpts from Annex I, 1. (3) (c) and (d) (emphasis ours):

• Products shall protect the confidentiality of stored, transmitted or otherwise processed data, personal or other, such as by encrypting relevant data at rest or in transit by state of the art mechanisms.
• Products shall protect the integrity of stored, transmitted or otherwise processed data, personal or other, commands, programs and configuration against any manipulation or modification not authorised by the user, as well as report on corruptions.

In other words:

• When communicating or storing (to non-volatile memory) relevant data then encrypt it.
• When communicating or storing (to non-volatile memory) any data then authenticate it.

There is not much room for exceptions here, discussing what is relevant might be challenging. If it is not relevant, then why communicate or store it in the first place?

There might be some relief in Annex I, 1. (1) which says:

• Products shall be designed, developed and produced in such a way that they ensure an appropriate level of cybersecurity based on the risks.

This can be a way out for some applications if in the risk assessment a security expert can show that there is no risk in not protecting certain data. Nevertheless, the security assessment must also reflect the following two statements:

1. Apply security-by-design to the entire lifecycle (min 5 years), from development process to production, deployment, and use/maintenance.
2. Products integrated in or connected to a larger electronic information system can serve as an attack vector for malicious actors.

These two statements have a huge impact on all microcontroller communication like plain UART, RS-485, CAN or other fieldbuses. The second statement boils down to not making a difference between important and lesser important communication. Even the lesser important communication may serve as an attack vector to the system.

The first statement requires layered security mechanisms given that protecting a single aspect won’t be sufficient. Taking CAN as an example, let us have a look at the known three stages of most common attacks on CAN systems:

1. Get access to the CAN communication.
   For example by accessing the CAN wires or hijacking a device (or interface to another network) already connected.
2. Monitor the CAN communication to learn from it.
   The attacker learns which CAN frames are used for what.
3. Inject or replay CAN communication to maliciously trigger behaviour.
   The attacker “takes over” certain aspects of the system.

By protecting any one of these three stages, the attacker’s success can be thwarted, and the system might seem “secure”. However, what if in the foreseeable future an attacker finds a way around that single protection? Security-by-design requires that we pay attention to all possible stages of an attack and do not focus on a single point of potential failure.

In any future risk assessment of a system using any form of communication (fieldbus or application specific communications), manufacturers will need to show which steps they took to protect all aspects:

1. Which steps were taken to minimize physical access?
   How easy is it to access the network wires? Can diagnostic ports be protected? Which interfaces to other networks are there? How are they protected?
2. Which steps were taken to keep communication confidential?
   To prevent attackers from learning anything about the system, encrypt all relevant communication.
3. Which steps were taken to ensure communication integrity?
   To prevent attackers from manipulating frames, authenticate all relevant communication.

In summary, to be EU Cyber Resilience Act compliant, a lot of the future fieldbus communication needs to be both authenticated and encrypted. To simplify the risk assessment and documentation, this should be done for all communication. Otherwise, manufacturers need to be prepared to have a security expert document every unprotected communication as to why this specific data set is irrelevant enough so that even if read or manipulated it won’t possibly constitute a cybersecurity risk.

Follow this blog and/or our LinkedIn page to learn about latest related developments including our upcoming security solutions for CAN, CAN FD, CANopen and CANopen FD.

Meltdown and Spectre are considered by many to be the biggest security flaws in the history of computing, both in terms of numbers of affected devices (billions) and time they have been laying dormant (20 years). Whenever security issues like these that affect PCs and mobile devices become public, we take a look at how they might affect Embedded Systems as well. An inconvenient truth in our industry is that software in Embedded Systems does not get updated, to put it mildly, as often as regular desktop PCs. Sometimes that means “never”. That is why even “ancient” attack vectors like the WannaCry and its descendants such as Petya and NotPetya ramsomware can still cause major damage in various systems, even months or years after the underlying security issues have been made public.

The core issue behind Meltdown and Spectre is that parts of a memory protection and isolation system are being compromised on a hardware level. Such isolation is meant to ensure that one task or program can not access the memory used by another task or program and potentially spy out sensitive information. The “good news” for most older chips and many embedded microcontroller devices first: They often don’t have a vulnerable memory isolation logic (involving out-of-order or speculative code execution) in the first place. It is actually worse: The memory in most lower-end embedded chips is wide open to all running tasks. While some microcontrollers do provide an MPU (Memory Protection Unit, see ARM Community for an example), it is often limited in terms of number of memory areas, sizes and number of levels/tasks supported. From our experience it is safe to say that a large number of embedded applications doesn’t make use of it at all. And when an MPU is used, then the primary goal is often to protect code against memory-crossing bugs to make it safer against failure, but not attacks. With these types of systems, once a hacker manages to execute some code on an embedded device, this code should be assumed to immediately have access to all resources of the chip, including the memory.

This looks like a devastating assessment from a security standpoint, however, injecting code into an embedded microcontroller is not easy. Many such systems do not use an operating system at all, have no command line or only a very limited user interface without the option to load and start a piece of code. Typically the only way to inject code is through a bootloader or a debug interface, if at all. It is up to the system designers, sometimes the factory programming and the program running on an embedded microcontroller to disable casual access to these functions.

We know that for many designers of embedded systems, the time they can spend on security issues is limited. If you are part of this group, you may use the publicity around Meltdown and Spectre to justify some extra time to review potentially vulnerabilities to attacks that are based on the same principle: to load or inject malicious code that spies out or manipulates data in your embedded system.

For such a review, first look for all options how code could be injected into your system or altered. Could an attacker make use of any of the provided bootloader mechanisms or the debug interface? If you can’t disable all of these because you need to be able to update “legitimate” code, then authentication is mandatory and encryption during transmission highly recommended. Preferably implement different layers of authentication, for example one to access the interface to update code and another one to protect the code itself. For an example see the secure secondary bootloader we implemented for NXP. Also, review if your microcontroller has a MPU or similar and how you can make best use of it not only to protect the system from buggy code but also from intentional attacks.

The Embedded Systems Academy now has the commercial version of CANcrypt available. It includes the hardcover, full-color version of the book and examples for both, the pairing and grouping modes. Demo implementations are provided for various NXP LPC processors (LPC23xx, LPC17xx and LPC11Cxx), the STM32F0xx as well as for the PEAK PCAN basic library supporting PEAK CAN interfaces.

The pairing demo shows how a 128bit key is securely transmitted from one device to another. The grouping demo shows secure communication between three devices. Here security is based on a session key that is continuously updated and gets saved when ungrouping. You can find more information including the software manual and the license information in our online store at www.esacademystore.eu/CANcrypt-Commercial/en.

A summary of the technical features used by CANcrypt

By Olaf Pfeiffer, Embedded Systems Academy GmbH, 26th of February 2016

At the Embedded World 2016 in Nuremberg, Embedded Systems Academy GmbH announced their book “Implementing Scalable CAN Security with CANcrypt”. The corresponding CANcrypt demo code will be published using an open license. At the Embedded World we have seen a lot of interest in the technical details. For those who do not want to wait until the publication of the book this article summarizes the key technical features of CANcrypt (also see our CANcrypt.eu web page for more information).

Core Functionality of CANcrypt

CANcrypt provides the following services:

• Pairing: dynamic generation of a random key that is only known by the paired devices; optionally, one device can enforce a preset key to the other.
  • generate and exchange keys
  • optional storing of keys in non-volatile memory for permanent pairing
  • support of a key hierarchy when multiple keys are stored
  • maintain dynamically changing key (pseudo one-time pad)
  • dynamic key updated using shared random bit
• Grouping: multiple devices share a common dynamic key
  • originally assigned through pairing
  • maintain dynamically changing key (pseudo one-time pad)
  • dynamic key cyclically updated by all grouped devices
• Safety communication: any secure communication uses a preamble message
  • messages received are only accepted and passed on to application if together with the preamble the authentication and decryption is verified successfully
  • preamble identifies message CAN ID, security features used, has a counter and a signature
  • secure messages must be received within 10ms after the preamble to be valid

CAN message IDs required:

• one CAN ID for each participating device
• used for preamble and control messages
• a CAN ID pair used for the random bit generation cycle

Cipher methods used

CANcrypt keys are symmetrical and dynamic, they are continuously updated. From the dynamic key and a message counter a pseudo one-time pad is generated that is used for the simple, customizable encryption.

If the secure pairing is only active for two nodes, a random bit generation cycle is used continuously in the background to introduce new bits to the dynamic key. If multiple nodes are paired, then the dynamic key update information is sent via an encrypted message.

The system pairing process is started using a CANcrypt configurator device. This can be done by a system builder or integrator once the CAN system is installed. It must happen in a secure environment. The keys generated at that time are stored locally in the devices connected – there is no need to keep any further copy of this key outside the system, minimizing the effort placed on key management. The keys cannot be duplicated. If a new device is added (or one exchanged), all keys need to be erased and newly generated.

As stored keys in each device make up a hierarchy, we can guarantee that erasing and regenerating keys can only happen when the configurator used is logged-in to the system based on a key high enough in the hierarchy to allow erasing and re-paring.

Operating principle for random bit generation

Bit generation cycle

Solely by monitoring CAN messages, one cannot identify the device that sent any individual message, because at that level, any device can transmit any message. As an example, let us allow two devices (named initiator and responder) to transmit messages with the CAN IDs 0010h and 0011h (and data length zero) within a “bit select time window”. Each node shall then randomly choose and send one of the two messages at a random time within the time window.

At the end of the bit select time window, a trace recording will show one of the following scenarios:

1. One or two messages of CAN ID 0010h
2. One each of CAN ID 0010h and 0011h
3. One or two messages of CAN ID 0011h

Let us have a closer look at case 2 – one each. If these are transmitted randomly within the bit response time window, then an observer has no way to identify which device sent which message. However, the devices themselves know it and use this information to derive a bit from it.

Unfortunately we cannot use case 1 and 3, so if those happen, both nodes need to recognize it and re-try, using another next bit select time window.

Note 1: If one device wants to enforce a specific bit to the other, it may generate a “flip bit” message at the end of the cycle to indicate to the other device that this bit needs to be flipped.

Note 2: A variation of this scheme is to not use a random delay, but instead ensure that both devices transmit their message immediately after the trigger message. Then both messages arbitrate the bus at the same time and in a trace recording we will always see 0010h followed by 0011h.

Potential attacks: As usual, a denial-of-service kind of attack is always possible. By injecting messages an attacker can break the cycle, the devices would not be able to exchange a key in the first place. If an attacker has full physical access (oscilloscope, transceiver), he can determine which node sent which message. However, there is still some effort required to recognize which bits were actually generated (as participating devices can change interpretation). Last but not least anything “random” is always an attack vector. The participating devices need a reasonably good random number generator.

Nuremberg, 22nd of February 2016: Embedded Systems Academy announces their new book “Implementing Scalable CAN Security with CANcrypt”. You can meet the authors at the Embedded World 2016 from February 23rd to 25th in hall 1, booth 620 – the booth of our partner PEAK-System.

The book covers authentication and encryption for CANopen and other Controller Area Network protocols and will be published in Q2/2016. The introduced CANcrypt system by ESAcademy adds multiple levels of security to CAN. CANcrypt supports the grouping of multiple devices and the encrypted and authenticated communication between them. The CANcrypt security layer sits between CAN driver and higher layers and is therefore independent of higher-layer protocols or applications used.

The required system resources are minimal compared to traditional cryptography methods and can be scaled to the application’s security requirements. A key hierarchy enables implementing of smart, simplified key management that supports manufacturers, system builders/integrators and owners.

Demo and example code will be published using the BSD license.
For more information see www.cancrypt.net