skip to main content

Title: Understanding the Security of ARM Debugging Features
Processors nowadays are consistently equipped with debugging features to facilitate the program analysis. Specifically, the ARM debugging architecture involves a series of CoreSight components and debug registers to aid the system debugging, and a group of debug authentication signals are designed to restrict the usage of these components and registers. Meantime, the security of the debugging features is under-examined since it normally requires physical access to use these features in the traditional debugging model. However, ARM introduces a new debugging model that requires no physical access since ARMv7, which exacerbates our concern on the security of the debugging features. In this paper, we perform a comprehensive security analysis of the ARM debugging features, and summarize the security and vulnerability implications. To understand the impact of the implications, we also investigate a series of ARM-based platforms in different product domains (i.e., development boards, IoT devices, cloud servers, and mobile devices). We consider the analysis and investigation expose a new attacking surface that universally exists in ARM-based platforms. To verify our concern, we further craft Nailgun attack, which obtains sensitive information (e.g., AES encryption key and fingerprint image) and achieves arbitrary payload execution in a high-privilege mode from a low-privilege mode via more » misusing the debugging features. This attack does not rely on software bugs, and our experiments show that almost all the platforms we investigated are vulnerable to the attack. The potential mitigations are discussed from different perspectives in the ARM ecosystem. « less
Authors:
;
Award ID(s):
1738929
Publication Date:
NSF-PAR ID:
10108062
Journal Name:
Proceedings - IEEE Symposium on Security and Privacy
ISSN:
1081-6011
Sponsoring Org:
National Science Foundation
More Like this
  1. Processors nowadays are consistently equipped with debugging features to facilitate the program analysis. Specifically, the ARM debugging architecture involves a series of CoreSight components and debug registers to aid the system debugging, and a group of debug authentication signals are designed to restrict the usage of these components and registers. Meantime, the security of the debugging features is under-examined since it normally requires physical access to use these features in the traditional debugging model. However, ARM introduces a new debugging model that requires no physical access since ARMv7, which exacerbates our concern on the security of the debugging features. Inmore »this paper, we perform a comprehensive security analysis of the ARM debugging features, and summarize the security and vulnerability implications. To understand the impact of the implications, we also investigate a series of ARM-based platforms in different product domains (i.e., development boards, IoT devices, cloud servers, and mobile devices). We consider the analysis and investigation expose a new attacking surface that universally exists in ARM-based platforms. To verify our concern, we further craft NAILGUN attack, which obtains sensitive information (e.g., AES encryption key and fingerprint image) and achieves arbitrary payload execution in a high-privilege mode from a low-privilege mode via misusing the debugging features. This attack does not rely on software bugs, and our experiments show that almost all the platforms we investigated are vulnerable to the attack. The potential mitigations are discussed from different perspectives in the ARM ecosystem.« less
  2. Home automation platforms provide a new level of convenience by enabling consumers to automate various aspects of physical objects in their homes. While the convenience is beneficial, security flaws in the platforms or integrated third-party products can have serious consequences for the integrity of a user's physical environment. In this paper we perform a systematic security evaluation of two popular smart home platforms, Google's Nest platform and Philips Hue, that implement home automation "routines" (i.e., trigger-action programs involving apps and devices) via manipulation of state variables in a centralized data store. Our semi-automated analysis examines, among other things, platform accessmore »control enforcement, the rigor of non-system enforcement procedures, and the potential for misuse of routines. This analysis results in ten key findings with serious security implications. For instance, we demonstrate the potential for the misuse of smart home routines in the Nest platform to perform a lateral privilege escalation, illustrate how Nest's product review system is ineffective at preventing multiple stages of this attack that it examines, and demonstrate how emerging platforms may fail to provide even bare-minimum security by allowing apps to arbitrarily add/remove other apps from the user's smart home. Our findings draw attention to the unique security challenges of platforms that execute routines via centralized data stores, and highlight the importance of enforcing security by design in emerging home automation platforms.« less
  3. Home automation platforms enable consumers to conveniently automate various physical aspects of their homes. However, the security flaws in the platforms or integrated third-party products can have serious security and safety implications for the user’s physical environment. This article describes our systematic security evaluation of two popular smart home platforms, Google’s Nest platform and Philips Hue, which implement home automation “routines” (i.e., trigger-action programs involving apps and devices) via manipulation of state variables in a centralized data store . Our semi-automated analysis examines, among other things, platform access control enforcement, the rigor of non-system enforcement procedures, and the potential formore »misuse of routines, and it leads to 11 key findings with serious security implications. We combine several of the vulnerabilities we find to demonstrate the first end-to-end instance of lateral privilege escalation in the smart home, wherein we remotely disable the Nest Security Camera via a compromised light switch app. Finally, we discuss potential defenses, and the impact of the continuous evolution of smart home platforms on the practicality of security analysis. Our findings draw attention to the unique security challenges of smart home platforms and highlight the importance of enforcing security by design.« less
  4. Obeid, I. ; Selesnik, I. ; Picone, J. (Ed.)
    The Neuronix high-performance computing cluster allows us to conduct extensive machine learning experiments on big data [1]. This heterogeneous cluster uses innovative scheduling technology, Slurm [2], that manages a network of CPUs and graphics processing units (GPUs). The GPU farm consists of a variety of processors ranging from low-end consumer grade devices such as the Nvidia GTX 970 to higher-end devices such as the GeForce RTX 2080. These GPUs are essential to our research since they allow extremely compute-intensive deep learning tasks to be executed on massive data resources such as the TUH EEG Corpus [2]. We use TensorFlow [3]more »as the core machine learning library for our deep learning systems, and routinely employ multiple GPUs to accelerate the training process. Reproducible results are essential to machine learning research. Reproducibility in this context means the ability to replicate an existing experiment – performance metrics such as error rates should be identical and floating-point calculations should match closely. Three examples of ways we typically expect an experiment to be replicable are: (1) The same job run on the same processor should produce the same results each time it is run. (2) A job run on a CPU and GPU should produce identical results. (3) A job should produce comparable results if the data is presented in a different order. System optimization requires an ability to directly compare error rates for algorithms evaluated under comparable operating conditions. However, it is a difficult task to exactly reproduce the results for large, complex deep learning systems that often require more than a trillion calculations per experiment [5]. This is a fairly well-known issue and one we will explore in this poster. Researchers must be able to replicate results on a specific data set to establish the integrity of an implementation. They can then use that implementation as a baseline for comparison purposes. A lack of reproducibility makes it very difficult to debug algorithms and validate changes to the system. Equally important, since many results in deep learning research are dependent on the order in which the system is exposed to the data, the specific processors used, and even the order in which those processors are accessed, it becomes a challenging problem to compare two algorithms since each system must be individually optimized for a specific data set or processor. This is extremely time-consuming for algorithm research in which a single run often taxes a computing environment to its limits. Well-known techniques such as cross-validation [5,6] can be used to mitigate these effects, but this is also computationally expensive. These issues are further compounded by the fact that most deep learning algorithms are susceptible to the way computational noise propagates through the system. GPUs are particularly notorious for this because, in a clustered environment, it becomes more difficult to control which processors are used at various points in time. Another equally frustrating issue is that upgrades to the deep learning package, such as the transition from TensorFlow v1.9 to v1.13, can also result in large fluctuations in error rates when re-running the same experiment. Since TensorFlow is constantly updating functions to support GPU use, maintaining an historical archive of experimental results that can be used to calibrate algorithm research is quite a challenge. This makes it very difficult to optimize the system or select the best configurations. The overall impact of all of these issues described above is significant as error rates can fluctuate by as much as 25% due to these types of computational issues. Cross-validation is one technique used to mitigate this, but that is expensive since you need to do multiple runs over the data, which further taxes a computing infrastructure already running at max capacity. GPUs are preferred when training a large network since these systems train at least two orders of magnitude faster than CPUs [7]. Large-scale experiments are simply not feasible without using GPUs. However, there is a tradeoff to gain this performance. Since all our GPUs use the NVIDIA CUDA® Deep Neural Network library (cuDNN) [8], a GPU-accelerated library of primitives for deep neural networks, it adds an element of randomness into the experiment. When a GPU is used to train a network in TensorFlow, it automatically searches for a cuDNN implementation. NVIDIA’s cuDNN implementation provides algorithms that increase the performance and help the model train quicker, but they are non-deterministic algorithms [9,10]. Since our networks have many complex layers, there is no easy way to avoid this randomness. Instead of comparing each epoch, we compare the average performance of the experiment because it gives us a hint of how our model is performing per experiment, and if the changes we make are efficient. In this poster, we will discuss a variety of issues related to reproducibility and introduce ways we mitigate these effects. For example, TensorFlow uses a random number generator (RNG) which is not seeded by default. TensorFlow determines the initialization point and how certain functions execute using the RNG. The solution for this is seeding all the necessary components before training the model. This forces TensorFlow to use the same initialization point and sets how certain layers work (e.g., dropout layers). However, seeding all the RNGs will not guarantee a controlled experiment. Other variables can affect the outcome of the experiment such as training using GPUs, allowing multi-threading on CPUs, using certain layers, etc. To mitigate our problems with reproducibility, we first make sure that the data is processed in the same order during training. Therefore, we save the data from the last experiment and to make sure the newer experiment follows the same order. If we allow the data to be shuffled, it can affect the performance due to how the model was exposed to the data. We also specify the float data type to be 32-bit since Python defaults to 64-bit. We try to avoid using 64-bit precision because the numbers produced by a GPU can vary significantly depending on the GPU architecture [11-13]. Controlling precision somewhat reduces differences due to computational noise even though technically it increases the amount of computational noise. We are currently developing more advanced techniques for preserving the efficiency of our training process while also maintaining the ability to reproduce models. In our poster presentation we will demonstrate these issues using some novel visualization tools, present several examples of the extent to which these issues influence research results on electroencephalography (EEG) and digital pathology experiments and introduce new ways to manage such computational issues.« less
  5. Post-quantum schemes are expected to replace existing public-key schemes within a decade in billions of devices. To facilitate the transition, the US National Institute for Standards and Technology (NIST) is running a standardization process. Multivariate signatures is one of the main categories in NIST's post-quantum cryptography competition. Among the four candidates in this category, the LUOV and Rainbow schemes are based on the Oil and Vinegar scheme, first introduced in 1997 which has withstood over two decades of cryptanalysis. Beyond mathematical security and efficiency, security against side-channel attacks is a major concern in the competition. The current sentiment is thatmore »post-quantum schemes may be more resistant to fault-injection attacks due to their large key sizes and the lack of algebraic structure. We show that this is not true. We introduce a novel hybrid attack, QuantumHammer, and demonstrate it on the constant-time implementation of LUOV currently in Round 2 of the NIST post-quantum competition. The QuantumHammer attack is a combination of two attacks, a bit-tracing attack enabled via Rowhammer fault injection and a divide and conquer attack that uses bit-tracing as an oracle. Using bit-tracing, an attacker with access to faulty signatures collected using Rowhammer attack, can recover secret key bits albeit slowly. We employ a divide and conquer attack which exploits the structure in the key generation part of LUOV and solves the system of equations for the secret key more efficiently with few key bits recovered via bit-tracing. We have demonstrated the first successful in-the-wild attack on LUOV recovering all 11K key bits with less than 4 hours of an active Rowhammer attack. The post-processing part is highly parallel and thus can be trivially sped up using modest resources. QuantumHammer does not make any unrealistic assumptions, only requires software co-location (no physical access), and therefore can be used to target shared cloud servers or in other sandboxed environments.« less