arXiv subjects

Christof Fetzer

Publications and source records attributed to Christof Fetzer.

At least 19 recordsLinked to original sources

arXiv2026

EnclaveX: End-to-End Confidential AI with CPU/GPU TEEs

Large Language Models (LLMs) have rapidly proliferated, driving widespread adoption of AI applications. Most deployments rely on centralized infrastructures such as Microsoft Azure, Google Cloud, or AWS, requiring users to share sensitive data and training or fine-tuning code. This dependence raises significant security and privacy concerns, as cloud providers must be trusted to ensure confidentiality and integrity. Trusted Execution Environments (TEEs) e.g., Intel SGX/TDX, AMD SEV-SNP, and ARM CCA have been introduced to mitigate these risks. More recently, NVIDIA has developed GPU TEEs (e.g., H100/H200), yet comprehensive evaluations of end-to-end workflows that integrate CPU and GPU TEEs remain limited. Critical aspects, including performance overhead, remote attestation, and security guarantees for AI/LLM applications, have not been sufficiently studied. This paper addresses this gap by presenting an end-to-end workflow that combines CPU and GPU TEEs. We propose mechanisms to ensure confidentiality and integrity at both the VM level (via Intel TDX and AMD SEV-SNP) and the application level, highlighting vulnerabilities such as Kubernetes administrators' ability to access confidential VM contents. Finally, we evaluate the performance overhead of our system using industry benchmarks, focusing on configurations that integrate Intel TDX with NVIDIA H200 GPUs.

cs.CR↗

arXiv2025

A Comprehensive Study on the Impact of Vulnerable Dependencies on Open-Source Software

Open-source libraries are widely used by software developers to speed up the development of products, however, they can introduce security vulnerabilities, leading to incidents like Log4Shell. With the expanding usage of open-source libraries, it becomes even more imperative to comprehend and address these dependency vulnerabilities. The use of Software Composition Analysis (SCA) tools does greatly help here as they provide a deep insight on what dependencies are used in a project, enhancing the security and integrity in the software supply chain. In order to learn how wide spread vulnerabilities are and how quickly they are being fixed, we conducted a study on over 1k open-source software projects with about 50k releases comprising several languages such as Java, Python, Rust, Go, Ruby, PHP, and JavaScript. Our objective is to investigate the severity, persistence, and distribution of these vulnerabilities, as well as their correlation with project metrics such as team and contributors size, activity and release cycles. In order to perform such analysis, we crawled over 1k projects from github including their version history ranging from 2013 to 2023 using VODA, our SCA tool. Using our approach, we can provide information such as library versions, dependency depth, and known vulnerabilities, and how they evolved over the software development cycle. Being larger and more diverse than datasets used in earlier works and studies, ours provides better insights and generalizability of the gained results. The data collected answers several research questions about the dependency depth and the average time a vulnerability persists. Among other findings, we observed that for most programming languages, vulnerable dependencies are transitive, and a critical vulnerability persists in average for over a year before being fixed.

cs.SE↗

arXiv2025

GROOT: General-Purpose Automatic Parameter Tuning Across Layers, Domains, and Use Cases

Modern software systems are executed on a runtime stack with layers (virtualization, storage, trusted execution, etc.) each incurring an execution and/or monetary cost, which may be mitigated by finding suitable parameter configurations. While specialized parameter tuners exist, they are tied to a particular domain or use case, fixed in type and number of optimization goals, or focused on a specific layer or technology. These limitations pose significant adoption hurdles for specialized and innovative ventures (SIVs) that address a variety of domains and use cases, operate under strict cost-performance constraints requiring tradeoffs, and rely on self-hosted servers with custom technology stacks while having little data or expertise to set up and operate specialized tuners. In this paper, we present Groot - a general-purpose configuration tuner designed to a) be explicitly agnostic of a particular domain or use case, b) balance multiple potentially competing optimization goals, c) support different custom technology setups, and d) make minimal assumptions about parameter types, ranges, or suitable values. Our evaluation on both real-world use cases and benchmarks shows that Groot reliably improves performance and reduces resource consumption in scenarios representative for SIVs.

cs.PF↗

arXiv2025

TICAL: Trusted and Integrity-protected Compilation of AppLications

During the past few years, we have witnessed various efforts to provide confidentiality and integrity for applications running in untrusted environments such as public clouds. In most of these approaches, hardware extensions such as Intel SGX, TDX, AMD SEV, etc., are leveraged to provide encryption and integrity protection on process or VM level. Although all of these approaches increase the trust in the application at runtime, an often overlooked aspect is the integrity and confidentiality protection at build time, which is equally important as maliciously injected code during compilation can compromise the entire application and system. In this paper, we present Tical, a practical framework for trusted compilation that provides integrity protection and confidentiality in build pipelines from source code to the final executable. Our approach harnesses TEEs as runtime protection but enriches TEEs with file system shielding and an immutable audit log with version history to provide accountability. This way, we can ensure that the compiler chain can only access trusted files and intermediate output, such as object files produced by trusted processes. Our evaluation using micro- and macro-benchmarks shows that Tical can protect the confidentiality and integrity of whole CI/CD pipelines with an acceptable performance overhead.

cs.CR↗

arXiv2024

CRISP: Confidentiality, Rollback, and Integrity Storage Protection for Confidential Cloud-Native Computing

Trusted execution environments (TEEs) protect the integrity and confidentiality of running code and its associated data. Nevertheless, TEEs' integrity protection does not extend to the state saved on disk. Furthermore, modern cloud-native applications heavily rely on orchestration (e.g., through systems such as Kubernetes) and, thus, have their services frequently restarted. During restarts, attackers can revert the state of confidential services to a previous version that may aid their malicious intent. This paper presents CRISP, a rollback protection mechanism that uses an existing runtime for Intel SGX and transparently prevents rollback. Our approach can constrain the attack window to a fixed and short period or give developers the tools to avoid the vulnerability window altogether. Finally, experiments show that applying CRISP in a critical stateful cloud-native application may incur a resource increase but only a minor performance penalty.

cs.CR↗

arXiv2024

Trustworthy confidential virtual machines for the masses

Confidential computing alleviates the concerns of distrustful customers by removing the cloud provider from their trusted computing base and resolves their disincentive to migrate their workloads to the cloud. This is facilitated by new hardware extensions, like AMD's SEV Secure Nested Paging (SEV-SNP), which can run a whole virtual machine with confidentiality and integrity protection against a potentially malicious hypervisor owned by an untrusted cloud provider. However, the assurance of such protection to either the service providers deploying sensitive workloads or the end-users passing sensitive data to services requires sending proof to the interested parties. Service providers can retrieve such proof by performing remote attestation while end-users have typically no means to acquire this proof or validate its correctness and therefore have to rely on the trustworthiness of the service providers. In this paper, we present Revelio, an approach that features two main contributions: i) it allows confidential virtual machine (VM)-based workloads to be designed and deployed in a way that disallows any tampering even by the service providers and ii) it empowers users to easily validate their integrity. In particular, we focus on web-facing workloads, protect them leveraging SEV-SNP, and enable end-users to remotely attest them seamlessly each time a new web session is established. To highlight the benefits of Revelio, we discuss how a standalone stateful VM that hosts an open-source collaboration office suite can be secured and present a replicated protocol proxy that enables commodity users to securely access the Internet Computer, a decentralized blockchain infrastructure.

cs.CR↗

arXiv2023

A Last-Level Defense for Application Integrity and Confidentiality

Our objective is to protect the integrity and confidentiality of applications operating in untrusted environments. Trusted Execution Environments (TEEs) are not a panacea. Hardware TEEs fail to protect applications against Sybil, Fork and Rollback Attacks and, consequently, fail to preserve the consistency and integrity of applications. We introduce a novel system, LLD, that enforces the integrity and consistency of applications in a transparent and scalable fashion. Our solution augments TEEs with instantiation control and rollback protection. Instantiation control, enforced with TEE-supported leases, mitigates Sybil/Fork Attacks without incurring the high costs of solving crypto-puzzles. Our rollback detection mechanism does not need excessive replication, nor does it sacrifice durability. We show that implementing these functionalities in the LLD runtime automatically protects applications and services such as a popular DBMS.

cs.CR↗

arXiv2023

Triad: Trusted Timestamps in Untrusted Environments

We aim to provide trusted time measurement mechanisms to applications and cloud infrastructure deployed in environments that could harbor potential adversaries, including the hardware infrastructure provider. Despite Trusted Execution Environments (TEEs) providing multiple security functionalities, timestamps from the Operating System are not covered. Nevertheless, some services require time for validating permissions or ordering events. To address that need, we introduce Triad, a trusted timestamp dispatcher of time readings. The solution provides trusted timestamps enforced by mutually supportive enclave-based clock servers that create a continuous trusted timeline. We leverage enclave properties such as forced exits and CPU-based counters to mitigate attacks on the server's timestamp counters. Triad produces trusted, confidential, monotonically-increasing timestamps with bounded error and desirable, non-trivial properties. Our implementation relies on Intel SGX and SCONE, allowing transparent usage. We evaluate Triad's error and behavior in multiple dimensions.

cs.CR↗

arXiv2023

SinClave: Hardware-assisted Singletons for TEEs

For trusted execution environments (TEEs), remote attestation permits establishing trust in software executed on a remote host. It requires that the measurement of a remote TEE is both complete and fresh: We need to measure all aspects that might determine the behavior of an application, and this measurement has to be reasonably fresh. Performing measurements only at the start of a TEE simplifies the attestation but enables "reuse" attacks of enclaves. We demonstrate how to perform such reuse attacks for different TEE frameworks. We also show how to address this issue by enforcing freshness - through the concept of a singleton enclave - and completeness of the measurements. Completeness of measurements is not trivial since the secrets provisioned to an enclave and the content of the filesystem can both affect the behavior of the software, i.e., can be used to mount reuse attacks. We present mechanisms to include measurements of these two components in the remote attestation. Our evaluation based on real-world applications shows that our approach incurs only negligible overhead ranging from 1.03% to 13.2%.

cs.CR↗

arXiv2022

PCRAFT: Capacity Planning for Dependable Stateless Services

Fault-tolerance techniques depend on replication to enhance availability, albeit at the cost of increased infrastructure costs. This results in a fundamental trade-off: Fault-tolerant services must satisfy given availability and performance constraints while minimising the number of replicated resources. These constraints pose capacity planning challenges for the service operators to minimise replication costs without negatively impacting availability. To this end, we present PCRAFT, a system to enable capacity planning of dependable services. PCRAFT's capacity planning is based on a hybrid approach that combines empirical performance measurements with probabilistic modelling of availability based on fault injection. In particular, we integrate traditional service-level availability mechanisms (active route anywhere and passive failover) and deployment schemes (cloud and on-premises) to quantify the number of nodes needed to satisfy the given availability and performance constraints. Our evaluation based on real-world applications shows that cloud deployment requires fewer nodes than on-premises deployments. Additionally, when considering on-premises deployments, we show how passive failover requires fewer nodes than active route anywhere. Furthermore, our evaluation quantify the quality enhancement given by additional integrity mechanisms and how this affects the number of nodes needed.

cs.DC↗

arXiv2022

Synergia: Hardening High-Assurance Security Systems with Confidential and Trusted Computing

High-assurance security systems require strong isolation from the untrusted world to protect the security-sensitive or privacy-sensitive data they process. Existing regulations impose that such systems must execute in a trustworthy operating system (OS) to ensure they are not collocated with untrusted software that might negatively impact their availability or security. However, the existing techniques to attest to the OS integrity fall short due to the cuckoo attack. In this paper, we first show a novel defense mechanism against the cuckoo attack, and we formally prove it. Then, we implement it as part of an integrity monitoring and enforcement framework that attests to the trustworthiness of the OS from 3.7x to 8.5x faster than the existing integrity monitoring systems. We demonstrate its practicality by protecting the execution of a real-world eHealth application, performing micro and macro-benchmarks, and assessing the security risk.

cs.CR↗

arXiv2022

A Sorted Datalog Hammer for Supervisor Verification Conditions Modulo Simple Linear Arithmetic

In a previous paper, we have shown that clause sets belonging to the Horn Bernays-Sch\"onfinkel fragment over simple linear real arithmetic (HBS(SLR)) can be translated into HBS clause sets over a finite set of first-order constants. The translation preserves validity and satisfiability and it is still applicable if we extend our input with positive universally or existentially quantified verification conditions (conjectures). We call this translation a Datalog hammer. The combination of its implementation in SPASS-SPL with the Datalog reasoner VLog establishes an effective way of deciding verification conditions in the Horn fragment. We verify supervisor code for two examples: a lane change assistant in a car and an electronic control unit of a supercharged combustion engine. In this paper, we improve our Datalog hammer in several ways: we generalize it to mixed real-integer arithmetic and finite first-order sorts; we extend the class of acceptable inequalities beyond variable bounds and positively grounded inequalities; and we significantly reduce the size of the hammer output by a soft typing discipline. We call the result the sorted Datalog hammer. It not only allows us to handle more complex supervisor code and to model already considered supervisor code more concisely, but it also improves our performance on real world benchmark examples. Finally, we replace the before file-based interface between SPASS-SPL and VLog by a close coupling resulting in a single executable binary.

cs.LO↗

arXiv2021

SecFL: Confidential Federated Learning using TEEs

Federated Learning (FL) is an emerging machine learning paradigm that enables multiple clients to jointly train a model to take benefits from diverse datasets from the clients without sharing their local training datasets. FL helps reduce data privacy risks. Unfortunately, FL still exist several issues regarding privacy and security. First, it is possible to leak sensitive information from the shared training parameters. Second, malicious clients can collude with each other to steal data, models from regular clients or corrupt the global training model. To tackle these challenges, we propose SecFL - a confidential federated learning framework that leverages Trusted Execution Environments (TEEs). SecFL performs the global and local training inside TEE enclaves to ensure the confidentiality and integrity of the computations against powerful adversaries with privileged access. SecFL provides a transparent remote attestation mechanism, relying on the remote attestation provided by TEEs, to allow clients to attest the global training computation as well as the local training computation of each other. Thus, all malicious clients can be detected using the remote attestation mechanisms.

cs.CR↗

arXiv2021

Transient Execution of Non-Canonical Accesses

Recent years have brought microarchitectural security intothe spotlight, proving that modern CPUs are vulnerable toseveral classes of microarchitectural attacks. These attacksbypass the basic isolation primitives provided by the CPUs:process isolation, memory permissions, access checks, andso on. Nevertheless, most of the research was focused on In-tel CPUs, with only a few exceptions. As a result, few vulner-abilities have been found in other CPUs, leading to specula-tions about their immunity to certain types of microarchi-tectural attacks. In this paper, we provide a black-box anal-ysis of one of these under-explored areas. Namely, we inves-tigate the flaw of AMD CPUs which may lead to a transientexecution hijacking attack. Contrary to nominal immunity,we discover that AMD Zen family CPUs exhibit transient ex-ecution patterns similar for Meltdown/MDS. Our analysisof exploitation possibilities shows that AMDs design deci-sions indeed limit the exploitability scope comparing to In-tel CPUs, yet it may be possible to use them to amplify othermicroarchitectural attacks.

cs.CR↗

arXiv2021

A Datalog Hammer for Supervisor Verification Conditions Modulo Simple Linear Arithmetic

The Bernays-Sch\"onfinkel first-order logic fragment over simple linear real arithmetic constraints BS(SLR) is known to be decidable. We prove that BS(SLR) clause sets with both universally and existentially quantified verification conditions (conjectures) can be translated into BS(SLR) clause sets over a finite set of first-order constants. For the Horn case, we provide a Datalog hammer preserving validity and satisfiability. A toolchain from the BS(LRA) prover SPASS-SPL to the Datalog reasoner VLog establishes an effective way of deciding verification conditions in the Horn fragment. This is exemplified by the verification of supervisor code for a lane change assistant in a car and of an electronic control unit for a supercharged combustion engine.

cs.LO↗

arXiv2021

Revizor: Testing Black-box CPUs against Speculation Contracts

Speculative vulnerabilities such as Spectre and Meltdown expose speculative execution state that can be exploited to leak information across security domains via side-channels. Such vulnerabilities often stay undetected for a long time as we lack the tools for systematic testing of CPUs to find them. In this paper, we propose an approach to automatically detect microarchitectural information leakage in commercial black-box CPUs. We build on speculation contracts, which we employ to specify the permitted side effects of program execution on the CPU's microarchitectural state. We propose a Model-based Relational Testing (MRT) technique to empirically assess the CPU compliance with these specifications. We implement MRT in a testing framework called Revizor, and showcase its effectiveness on real Intel x86 CPUs. Revizor automatically detects violations of a rich set of contracts, or indicates their absence. A highlight of our findings is that Revizor managed to automatically surface Spectre, MDS, and LVI, as well as several previously unknown variants.

cs.CR↗

arXiv2021

WELES: Policy-driven Runtime Integrity Enforcement of Virtual Machines

Trust is of paramount concern for tenants to deploy their security-sensitive services in the cloud. The integrity of VMs in which these services are deployed needs to be ensured even in the presence of powerful adversaries with administrative access to the cloud. Traditional approaches for solving this challenge leverage trusted computing techniques, e.g., vTPM, or hardware CPU extensions, e.g., AMD SEV. But, they are vulnerable to powerful adversaries, or they provide only load time (not runtime) integrity measurements of VMs. We propose WELES, a protocol allowing tenants to establish and maintain trust in VM runtime integrity of software and its configuration. WELES is transparent to the VM configuration and setup. It performs an implicit attestation of VMs during a secure login and binds the VM integrity state with the secure connection. Our prototype's evaluation shows that WELES is practical and incurs low performance overhead.

cs.CR↗

arXiv2021

Perun: Secure Multi-Stakeholder Machine Learning Framework with GPU Support

Confidential multi-stakeholder machine learning (ML) allows multiple parties to perform collaborative data analytics while not revealing their intellectual property, such as ML source code, model, or datasets. State-of-the-art solutions based on homomorphic encryption incur a large performance overhead. Hardware-based solutions, such as trusted execution environments (TEEs), significantly improve the performance in inference computations but still suffer from low performance in training computations, e.g., deep neural networks model training, because of limited availability of protected memory and lack of GPU support. To address this problem, we designed and implemented Perun, a framework for confidential multi-stakeholder machine learning that allows users to make a trade-off between security and performance. Perun executes ML training on hardware accelerators (e.g., GPU) while providing security guarantees using trusted computing technologies, such as trusted platform module and integrity measurement architecture. Less compute-intensive workloads, such as inference, execute only inside TEE, thus at a lower trusted computing base. The evaluation shows that during the ML training on CIFAR-10 and real-world medical datasets, Perun achieved a 161x to 1560x speedup compared to a pure TEE-based approach.

cs.LG↗