Under the monolithic architecture of the Linux kernel, all its components operate within the same address space. Notably, device drivers constitute over half of the kernel codebase yet are particularly prone to bugs. Therefore, exploring vulnerabilities in drivers is critical for ensuring kernel security. Extensive research has been done to fuzz kernel drivers through system calls and hardware interrupts. Through a comprehensive study of the Linux Kernel Device Model, we identified that the execution of device drivers is also influenced by runtime parameters, including device attributes and kernel module parameters. Our analysis reveals that large portions of the uncovered code are masked by these parameters, which are exposed to the userspace through a specialized virtual file system known as sysfs. Furthermore, adjacent devices interconnected within the same device tree also impact drivers' behavior.

The interpreter and the Just-In-Time (JIT) compiler are two core components of modern JavaScript engines, both of which take bytecodes as input. Most bugs in these components are closely related to specific bytecodes. Therefore, effective fuzzing should pay close attention to how bytecode is generated and exercised. However, previous work fails to consider this aspect and instead focuses primarily on the syntactic and semantic validity of test cases. This causes two major issues: 1) certain bytecodes are never exercised during fuzzing; 2) some bytecodes are exercised infrequently. In this paper, we propose BCFuzz, a bytecode-driven fuzzing approach designed to enhance the diversity of generated bytecode and increase testing opportunities for low-frequency bytecodes. Specifically, we introduce a parser-oriented probing technique to identify the necessary conditions for generating specific bytecodes and use this information to enhance the input generation process. To better test low-frequency bytecodes, we propose bytecode-aware seed preservation, scheduling, and mutation strategies. We evaluate BCFuzz on four mainstream JavaScript engines. In 72 hours of testing, BCFuzz discovers 1.73× and 1.67× more bugs than DIE and Fuzzilli, respectively. In total, BCFuzz uncovered 20 previously unknown bugs. Of these, 17 have already been fixed and one has been assigned a CVE. All the discovered bugs are related to bytecodes.

Deep neural networks (DNN) has been widely applied in big data-driven Internet of Things (IoT) for excellent learning ability, while the black-box nature of DNN leads to uncertainty of inference results. With higher interpretability, convolutional fuzzy neural network (CFNN) becomes an alternative choice for the model of IoT applications. IoT applications are often latency-sensitive. By jointly utilizing computing power of IoT devices and edge servers, end-edge collaborative CFNN inference improves the insufficiency of local computing resources and reduces the latency of computing-intensive CFNN inference. However, the calculation amount of fuzzy layers is hard to get directly, bringing difficulty to CFNN partition. In addition, the profit of service providers is often ignored in existing work on distributed inference. In this article, an end-edge collaborative inference framework of CFNNs for big data-driven IoT, named DisCFNN, is proposed. Specifically, a novel CFNN structure and a method of fuzzy layer calculation amount assessment are designed at first. Next, computing resource allocation and CFNN partition decisions are generated on each edge server based on deep reinforcement learning. Then, each IoT device sends the request of CFNN inference service to a certain edge server or infer the whole CFNN locally according to the task offloading strategy obtained through many-to-one matching game. Finally, the effectiveness of DisCFNN is evaluated through extensive experiments.

Deep Neural Network (DNN) Inference, as a key enabler of intelligent applications, is often computation-intensive and latency-sensitive. Combining the advantages of cloud computing (abundant computing resources) and edge computing (fast transmission), edge-cloud collaborative DNN inference is a powerful solution to these problems. However, in edge-cloud networks with heterogeneous resources, how to obtain reasonable decisions on server selection, model partition and resource allocation for efficient distributed DNN inference is a hard challenge. Furthermore, it is non-trivial to design suitable resource prices to maximize the social welfare. These challenges even escalate in dynamic edge-cloud networks where decisions should be generated as soon as each user arrives without future information. Therefore, we design a combinatorial auction for dynamic distributed DNN inference scheduling, named CADEC. CADEC first constructs a bid set for each user based on convex optimization theory for optimal solution searching. Next, prices of resources in the edge-cloud network are adjusted according to changes in supply-demand relationship, and whether to admit the request of each user is decided. Finally, the dynamic distributed inference scheduling decisions are generated through the primal-dual algorithm to maximize the social welfare. Theoretical analysis shows the good competitive ratio and polynomial time complexity of CADEC. Results of simulation experiments present that CADEC improves social welfare by up to 224% compared with state-of-the-art distributed DNN inference schemes.