Trusted execution environments (TEEs), like TrustZone, are pervasively employed to protect security sensitive programs and data from various attacks issued by untrusted rich execution environments (REEs) while they execute compact TEE operating systems which implement minimum security-critical operations but have poor device driver support. In this paper, we propose a twin driver approach where a pair of TEE and REE drivers is generated and cooperate to enable secure and efficient TEE driver support. To begin with, we propose a driver data flow analysis framework named driver analyzer (DrvAna) to automatically analyze the shared states between the TEE and REE driver where a novel data structure named value-type tree is investigated to facilitate field-sensitive data flow analysis upon the driver state. Furthermore, in order to maintain a minimal trusted computing base, we propose a Linux driver runtime (LDR) inside the TEE, a sandbox environment that confines the TEE driver based on the ARM domain access control features and mediates the driver's interaction with the TEE. We implement a DrvAna prototype based on LLVM as well as an LDR prototype on an NXP IMX6Q SABRE-SD evaluation board, adapt 6 existing Linux drivers into LDR, and evaluate their performance. The experimental results show that the LDR drivers can achieve comparable performance with their Linux counterparts with negligible overheads.

Network Covert Channels (NCC) enhance covertness by concealing the existence of information transmission. However, traditional NCCs remain vulnerable to traffic analysis. Once NCC is detected, adversaries can breach anonymity by uncovering users' network identities and even communication relationships. While certain indirect NCCs offer limited anonymity to protect the identity of at most one party and the relationship, this level proves insufficient. This paper proposes ANCC, an innovative Anonymous Network Covert Channel that is the first to achieve comprehensive anonymity for the sender, the receiver and the communication relationship. By leveraging the Tor network's Hidden Service Directories (HSDirs) as intermediate nodes, Tor-based ANCC modulates covert information through the publication and retrieval statuses of hidden services distributed on multiple HSDirs. This mechanism allows ANCC traffic to blend seamlessly into legitimate Tor traffic, ensuring both robust covertness and high-level anonymity. Theoretical analysis demonstrates that even against a powerful adversary compromising fifty intermediate nodes, the detection probability remains below 0.25%, with the risk of identity or relationship exposure staying negligible (under 0.0021% and 0.00002% respectively). Additionally, the multiple HSDirs supporting parallel transmission enhance the channel capacity and error correction encoding strengthens the robustness. Extensive evaluation within the real-world Tor network demonstrates a transmission accuracy exceeding 99.6% and a channel capacity of around 3 Kbps, proving its effectiveness for practical applications.

Anonymous networks facilitate privacy-preserving information transmission by ensuring data confidentiality and communication anonymity. While encryption effectively secures content, anonymity remains vulnerable to traffic analysis, which can expose users' network identities and communication relationships. Consequently, anonymity alone is insufficient, necessitating integrated technologies to further strengthen communication security. This paper introduces covert communication to anonymous networks by proposing a novel Covert Information Transmission Framework over Anonymous Networks, designed to realize both covert and anonymous data transfer. To the best of our knowledge, this is the first work to propose covert communication schemes specifically within anonymous networks. In our framework, we establish covert channels in the Tor network and transmit covert information based on Tor's hidden service protocol. Utilizing Tor's anonymous circuits, we strategically select suitable protocol data fields as information carriers and propose three covert communication schemes with distinct characteristics. These schemes enable secure data exchanges between clients and hidden services, catering to diverse communication requirements. Finally, we implement the proposed framework in a real-world Tor network and conduct extensive experiments to evaluate its feasibility and performance.

We present DarkEE, a framework that extracts document-level event intelligence from Dark Web pages to support the Cyber Threat Intelligence (CTI) pipeline. It is evaluated on DarkEvents, a manually annotated dataset distilled from 15.4 million Tor pages. Containing 164 full-page documents of up to 20,000 words, it preserves the noise and length of real-world data. Our schema defines 11 dark-web-specific event categories (e.g., Hacking, Counterfeits) along with 16 semantic argument roles (e.g., Payment Method, Communication Channel) tailored to the domain. To handle these challenging inputs, DarkEE employs a clues-enhanced two-stage framework utilizing Large Language Models. It incorporates eventrelevant textual clues as in-context demonstrations to guide reasoning chains, enabling robust extraction from uncurated text. We have conducted experiments on the overall Document-level Event Extraction (DEE) task and its two subtasks corresponding to our two-stage pipeline: Documentlevel Event Detection (DED) and Document-level Event Argument Extraction (DEAE). The results demonstrate that our framework achieves significant improvements in Precision, Recall, and F1score across all three tasks compared to strong baselines. We contribute our DarkEvents to the community to promote its advancement through further research: https://anonymous.4open.science/ r/dw darkevents-7E6F.

Recently, there has been an emerging radio frequency fingerprint identification (RFFI) technology that enhances fingerprint distinguishability by deliberately injecting an RF fingerprint into the device's Wi-Fi baseband signal. The current RF fingerprint injection methods are impractical, degrading the communication quality between Wi-Fi devices while offering limited improvements in distinguishability among a set of devices. To address these issues, we propose injecting I/Q imbalance into a short training field (STF) instead of the entire baseband signal. Our findings indicate that this method can effectively preserve the quality of the original wireless communication. Besides, a temperature-independent RF feature differential carrier frequency offset (DCFO) is proposed as an extended feature for the enhancement of fingerprint distinguishability. Building upon these, we introduce a fingerprinting scheme called PR-RFFI that generates distinguishable fingerprints for a set of devices by injecting appropriate I/Q imbalance and DCFO into the STF. Leveraging the short-term invariance of the channel, we design a practical I/Q imbalance extraction method based on the communication-quality preserving injection. Moreover, we design an optimal assignment method for I/Q imbalance and DCFO to maximize the distinguishability of RF fingerprints for all devices. Finally, we implement the PR-RFFI solution and conduct experiments in real-world and simulation scenarios. The experimental results demonstrate that PR-RFFI consistently maintains good communication quality, and achieves over 98% precision, recall, and F1-score.

In recent years, the Internet of Things (IoT) has rapidly advanced, with applications ranging from smart homes to industrial manufacturing, often involving densely deployed nodes such as temperature and humidity sensors. Since these nodes have limited computation and energy, the use of stuffed Wi-Fi management frames for data transmission has emerged as a promising way to avoid the association overhead of the traditional transmission mode. However, this unassociated data transmission mode continues to encounter significant channel contention in dense deployments. To this end, we propose ODGMAC, an on-demand grouping-based MAC solution that dynamically groups transmission-awaiting nodes and allocates time slots on a per-group basis, thereby enabling intra-group contention to improve transmission efficiency and reduce node energy consumption. Firstly, we present a fuzzy control-based algorithm at the access point (AP) to dynamically identify nodes with transmission demands in the current beacon period. On this basis, we then propose a hierarchical group-based time slot allocation methodology. Specifically, the nodes are initially clustered according to their per-packet airtime requirements. Within each cluster, we evenly partition nodes into multiple groups and assign each group to a unique time slot for channel contention, where the optimal slot count is determined by a renewal-theory-based analytical model with a discrete search over candidate counts. Finally, we implement the ODGMAC testbed with one AP and 100 IoT nodes, and conduct real-world experiments in a dense environment. The experimental results show that our solution outperforms existing methods in terms of both data delivery rate and node power consumption. Specifically, under severe channel collision conditions, our solution achieves an average increase of 14.77% in data delivery rate and an average reduction of 8.21% in node power consumption, while maintaining excellent fairness. Moreover, extended simulations show that our solution scales to 1000 nodes and maintains excellent performance under node mobility.

Home Internet-of-Things (H-IoT) automation is increasingly threatened by single-rule anomalies (SRA) and cross-rule interference (CRI). These threats can produce outcomes diverging from or even contradicting user expectations, compromising H-IoT system security and potentially endangering user safety and property. Existing detection techniques for these threats are fundamentally limited as they overlook undocumented command side effects, creating a critical blind spot in security analysis. This paper introduces Cet-Miner, a method that obtains and models these side effects through black-box command testing. Cet-Miner models system behavior as a Bayesian-smoothed Markov Decision Process (MDP) to characterize the non-deterministic side effects. For side effects that are not immediately observable, it uncovers them by comparing semantic inferences with test results, and represents them via a latent state structure introduced into the MDP. Building on this method, we propose HA-Inspector, an end-to-end threat detection system that implements both a runtime SRA monitor and a CRI model checker. We evaluate our method and system on a real-world H-IoT testbed comprising 38 devices from 9 vendors. Results show Cet-Miner discovers 5 distinct types of side effects, and HA-Inspector outperforms state-of-the-art tools in detecting both SRA and CRI threats.

The anonymity of the Tor dark web has fostered numerous hidden online marketplaces. These markets typically present products using multimodal information, including images and text. Sellers often pair real images of illegal goods with simplified, vague, or inaccurate text descriptions. However, existing dark web market analyses generally rely on single-modal text classification or isolated Natural Language Processing (NLP) tasks, and thus fall short of delivering a comprehensive understanding. In this paper, we propose a multimodal framework, DarkIntelligence, for holistic market analysis. First, DarkIntelligence leverages a LoRA-fine-tuned vision-language model to process the multimodal product information, generating unified and comprehensive records of both product details and seller profiles. Next, it applies Density-Based Spatial Clustering of Applications with Noise (DBSCAN) to adaptively cluster products and construct a hierarchical taxonomy. A large language model then assigns human-interpretable labels to each cluster. The rich information extracted by DarkIntelligence enables accurate and explainable analysis of dark web markets from multiple perspectives, including product structure, inventory, pricing, and seller profiles. We evaluate DarkIntelligence on a dataset of 27,726 multimodal entries collected from five realistic and representative Tor markets. The framework achieves 94.30% accuracy in product field information extraction and effectively discovers more hidden categories.

Tor is a widely used network for anonymous communication, employing onion encryption and multi-hop routing to ensure anonymity for its users and service providers. Despite its robust design, Tor has been the target of numerous attacks, such as denial-of-service (DoS) attacks and deanonymization attacks. However, these attacks often rely on resource-intensive methods, such as bandwidth inflation or controlling large-scale nodes. They face limitations due to high costs, limited scalability, and countermeasures that Tor already has in place. In this paper, we identify a new vulnerability, termed the Descriptor Flood, in Tor's memory management mechanism and service publication protocol. By exploiting Descriptor Flood, attackers can flood Tor nodes with malicious descriptors of onion services, causing severe memory fragmentation, exhaustion, and eventual node crash. Unlike conventional attacks, our method leverages a fundamental design flaw, allowing cost-effective and scalable exploitation without requiring substantial resources. To demonstrate the practical impact of this vulnerability, we propose the Tordos Attack, a three-phase strategy that efficiently disables Tor nodes and executes DoS and deanonymization attacks against onion services via tearing down specific nodes in Tor. The attack addresses key challenges, such as measuring node memory capacity, inducing fragmentation, and disabling critical nodes to maximize disruption. Our extensive experimental results indicate that the attack can disable Tor nodes and onion services within 9.1 minutes and expose the onion service's real identity in 6.1 hours, potentially leading to the collapse of the entire Tor network.

Freenet is a well-known peer-to-peer network for anonymous file sharing. It preserves user anonymity by concealing the originating node, either an uploader or a downloader, among the relay nodes that form the multi-hop routing path. Prior work has examined protocol-level behavioral differences between originating and relay nodes, while leaving the underlying data transmission mechanism largely unexplored. In this paper, we identify a critical node role leakage during data transmission. Specifically, an uploader sends the entire block from its local storage in pieces, whereas a relay node forwards only the block pieces it has received. Although both uploaders and relay nodes encapsulate piece messages into fixed-size packets, a relay node may emit relay-specific under-filled packets when subsequent pieces are unavailable. Leveraging this role leakage, we develop a deanonymization attack that identifies uploaders by actively triggering relay-specific packets during the handling of block upload requests. Real-world experiments demonstrate that, by aggregating 10 block-upload observations, the attacker can reduce the false positive rate (FPR) of uploader identification to close to 0.

Freenet is a widely used anonymous communication system designed for file sharing. It preserves anonymity for both uploaders and downloaders via hop-by-hop routing and the enforcement of uniform protocols across nodes, preventing identification of the originating node along the routing path. Previous work has shown that the originating node can be deanonymized based on observable differences in interaction behaviors between nodes, while overlooking unobservable internal differences. In this paper, we identify a fundamental distinction between uploaders and relay nodes in their patterns of inserting application-layer messages into in-memory message queues. Although this difference is unobservable to a malicious node, we show that the FIFO (First-In, First-Out) property of the message queue allows the internal message insertion state to be mirrored in transmitted messages by triggering a beacon message. This insight enables a novel deanonymization attack against uploaders. We further address two challenges in conducting the attack: preventing the internal difference from being undermined and ensuring that it is effectively mirrored in transmitted messages. Real-world experiments demonstrate the feasibility and effectiveness of our attack, achieving a nearly 100% true positive rate with a maximum 4.17% false positive rate. Our work demonstrates that even unobservable internal differences can be potential threats to Freenet.

Tor is a widely used network for anonymous communication, designed to safeguard the privacy and anonymity of both users and service providers through multi-layered encryption and multi-hop routing. Despite its robust architecture, Tor remains susceptible to various attacks, including denial-of-service (DoS) and deanonymization. However, these attacks are constrained by high resource requirements, scalability limitations, and the defenses implemented within the Tor network. Furthermore, they are particularly ineffective in identifying and circumventing the Guard nodes that protect onion services. In this paper, we uncover a novel vulnerability in Tor’s circuit construction process and bandwidth scheduling, termed the Circuit Circle vulnerability. Exploiting this flaw, attackers can create circular circuits, leading to bandwidth contention and overloading a Tor node. To demonstrate the severity of this vulnerability, we propose Duplicate-Node Attack, a three-phase strategy that identifies Guard nodes, exhausts their bandwidth, and performs coarse geolocation inference of onion services. Unlike conventional methods, our Guard node identification technique bypasses existing Tor Vanguard defenses without requiring control over any relays. Our extensive experimental results confirm that Duplicate-Node Attack can reliably identify Guard nodes, exhaust their bandwidth with minimal cost, and infer the geolocation of onion services, with an average latency bias of 35.4 ms.

Remote Procedure Call (RPC) services have become a primary gateway for users to access public blockchains. While they offer significant convenience, RPC services also introduce critical privacy challenges that remain insufficiently examined. Existing deanonymization attacks either do not apply to blockchain RPC users or incur costs like transaction fees assuming an active network eavesdropper. In this paper, we propose a novel deanonymization attack that can link an IP address of a RPC user to this user's blockchain pseudonym. Our analysis reveals a temporal correlation between the timestamps of transaction confirmations recorded on the public ledger and those of TCP packets sent by the victim when querying transaction status. We assume a strong passive adversary with access to network infrastructure, capable of monitoring traffic at network border routers or Internet exchange points. By monitoring network traffic and analyzing public ledgers, the attacker can link the IP address of the TCP packet to the pseudonym of the transaction initiator by exploiting the temporal correlation. This deanonymization attack incurs zero transaction fee. We mathematically model and analyze the attack method, perform large-scale measurements of blockchain ledgers, and conduct real-world attacks to validate the attack. Our attack achieves a high success rate of over 95% against normal RPC users on various blockchain networks, including Ethereum, Bitcoin and Solana.

Graph Neural Networks (GNNs) have shown considerable promise in handling graph-structured data, yet their use is restricted in privacy-sensitive environments, especially in distributed settings. In this setting, current methods for preserving privacy in GNNs often rely on unrealistic assumptions or fail to construct effective models. In response, this paper introduces Distributed Private Aggregation (DPA), a pioneering GNN aggregation method which is built upon Secure Multi-Party Computation protocols, and is designed to ensure node-level differential privacy. We implement DPA-GNN, which to our knowledge, is the most effective privacy-preserving GNN model suitable for distributed contexts. Through extensive experiments on six real-world datasets, DPA-GNN has proven to consistently surpass existing privacy preserving GNNs, offering an optimal balance between privacy and utility.

The Tor network, while offering anonymity through traffic routing across volunteer-operated nodes, remains vulnerable to attacks that aim to deanonymize users by correlating traffic patterns between colluded Entry and Exit nodes in circuits. This paper presents a novel approach for detecting anomalous circuits in the Tor network, and for the first time provides a more comprehensive identification of potential malicious accomplice nodes in Tor by taking roles of nodes in anomalous circuits into consideration. Our method strategically utilizes modified Middle nodes to capture traffic data, followed by a novel circuit classification based on traffic patterns to pinpoint concerned circuits. Two kinds of anomalies are identified: routing anomalies and usage anomalies, that respectively represent the anomalies with explicit or implicit violation of Tor's circuit construction guidelines. This leads to a successful revealing of totally 1,960 anomalous nodes in Tor. Furthermore, we apply clustering analysis with considering corresponding anomalous circuits and other key characteristics to the detected anomalous nodes, revealing potential hidden organizations behind these nodes that can threaten the network's security. Our findings highlight the necessity for the Tor project to adopt targeted mitigation strategies to enhance overall network security and privacy.

Due to the diversity of architectures and peripherals of Internet of Things (IoT) systems, blackbox fuzzing stands out as a prime option for discovering vulnerabilities of IoT devices. Existing blackbox fuzzing tools often rely on companion apps to generate valid fuzzing packets. However, existing methods encounter the challenges of bypassing the cloud server side validation when it comes to fuzz devices that rely on cloud-based communication. Moreover, they tend to concentrate their efforts on Java components within Android companion apps, limiting their effectiveness in assessing non-Java components such as JavaScript-based mini-apps. In this paper, we introduce a novel blackbox fuzzing method, named RIOTFUZZER, designed to remotely uncover vulnerabilities of IoT devices with the assistance of companion apps, particularly those powered by All-in-one Apps with the JavaScript-based mini-apps feature enabled. Our approach utilizes document-based control command extraction, hybrid analysis for mutation point identification and side-channel-guided fuzzing to effectively address the challenges of fuzzing IoT devices remotely. We apply RIOTFUZZER to 27 IoT devices on prominent platforms and discovered 11 vulnerabilities. All of them have been acknowledged by the corresponding vendors. 8 have been confirmed by the vendors and have been assigned 4 CVE IDs. Our experiment results also demonstrate that side-channel-guided fuzzing can significantly enhance the efficiency of fuzzing packets sent to IoT devices, with an average increase of 76.62% and a maximum increase of 362.62%.

Trusted execution environments (TEEs), like Trust-Zone, are pervasively employed to protect security sensitive programs and data from various attacks.We target compact TEE operating systems like OP-TEE, which implement minimum TEE internal core APIs.Such a TEE OS often has poor device driver support and we want to alleviate such issue by reusing existing Linux drivers inside TEE OSes.An intuitive approach is to port all its dependency functions into the TEE OS so that the driver can directly execute inside the TEE.But this approach significantly enlarges the trusted computing base (TCB), making the TEE OS no longer compact.In this paper, we propose a TEE driver execution environment-Linux driver runtime (LDR).A Linux driver needs two types of functions, library functions and Linux kernel subsystem functions that a compact TEE OS does not have.The LDR reuses the existing TEE OS library functions whenever possible and redirects the kernel subsystem function calls to the Linux kernel in the normal world.LDR is realized as a sandbox environment, which confines the Linux driver inside the TEE through the ARM domain access control features to address associated security issues.The sandbox mediates the driver's TEE functions calls, sanitizing arguments and return values as well as enforcing forward control flow integrity.We implement and deploy an LDR prototype on an NXP IMX6Q SABRE-SD evaluation board, adapt 6 existing Linux drivers into LDR, and evaluate their performance.The experimental results show that the LDR drivers can achieve comparable performance with their Linux counterparts with negligible overheads.We are the first to reuse functions in both the TEE OS and normal world Linux kernel to run a TEE device driver and address related security issues.

Adaptive video streaming over wireless networks has experienced tremendous growth in past few years. In order to guarantee users’ quality of experience (QoE), adaptive bitrate (ABR) algorithms have been extensively studied. With the recent emergency of high-density Wi-Fi networks, these solutions no longer perform well. On the one hand, bitrate decisions respond slowly to high network fluctuations, and on the other hand, wireless channel resources are insufficient especially under the cases that multiple clients compete for the limited bandwidth. Hence, users’ personalized QoE metrics need to be taken into account to cope with the QoE degradation. To this end, we design an access point (AP) assisted wireless dynamic adaptive video streaming over Hypertext Transfer Protocol (HTTP) solution called Wi-DASH, which aims to improve users’ QoE while considering channel utilization. In Wi-DASH, the video server aggregates real-time network status with the assistance of AP, and estimates clients’ player status through statistical analysis of chunk request logs. On this basis, a deep reinforcement learning (DRL) based ABR algorithm is designed, where the DRL model can deal with the complicated global status information including network status, player status, and QoE preferences. Finally, we implement Wi-DASH system, and conduct experiments with 4K resolution videos. Experimental results reveal that the Wi-DASH can more fully utilize wireless channel resources, and significantly improve users’ QoE.

Existing local differential privacy (LDP) techniques enable untrustworthy aggregators to perform only very simple data mining tasks on distributed private data, including statistical estimation and frequent item mining. There is currently no general LDP method that discovers relations between items. The main challenge lies in the curse of dimensionality, as the quantity of values to be estimated in mining relations is the square of the quantity of values to be estimated in mining item-level knowledge, leading to a considerable decrease in the final estimation accuracy. We propose LDP-RM, the first relation mining method under LDP. It represents items and relations in a matrix and utilizes singular value decomposition and low rank approximation to reduce the number of values to estimate from O(k(2)) to O(r), where k is the number of all considered items, and r < k is a parameter determined by the aggregator, signifying the rank of the approximation. LDP-RM serves as a fundamental privacy-preserving method for enabling various complex data mining tasks.