Linux Security Summit 2021

Confidential Cloud Computing is a powerful security model where the cloud tenants are not required to trust the SW stack provided by Cloud Service Providers (CSPs). This includes the Virtual Machine Monitor (VMM) that has been an internal part of VM guest’s TCB for decades. In recent years CPU vendors are coming forward with the technologies that make possible to support this changed threat model (AMD SEV, Intel TDX, etc.), but a lot of work also needs to be done on the VM guest SW stack to truly make this setup secure. This talk would present our efforts and methodology for hardening the mainline Linux kernel that can be used as a secure VM guest kernel. We will talk about the challenges we have faced, successful and failed approaches, as well as share some initial results. We also hope to start a discussion with the Linux community on how we all can work together to develop and integrate these hardening measures into the general practices for all involved components of the Linux guest SW stack.

Landlock is the first Mandatory Access Control available to unprivileged processes on Linux. It is available since Linux 5.13, which enables all applications to sandbox themselves. Landlock development started 5 years ago, and multiple approaches were tried (e.g. extending seccomp, using eBPF) before picking the good one. This talk first explains the goal of Landlock and the related consequences. This will enable to explain the kernel implementation constraints, the choices that led to the current design, and the potential and limits of the current and future features. More information about Landlock can be found on the official website: https://landlock.io

This talk presents FineIBT, a compiler-based enhancement that enables fine-grained forward-edge Control-Flow Integrity (CFI) policies on top of Intel's Control-flow Enforcement Technology (CET). By combining the new hardware features with compiler instrumentation, FineIBT anchors indirect control transfers to sanity checks, enabling policies more restrictive than those supported solely by CET and increasing its effectiveness against control-flow hijacking attacks. An evaluation through custom benchmarks shown that FineIBT provides similar security guarantees with less performance costs when compared to Clang CFI, retaining its penalty between 1% and 7% while the latter added overheads between 5% and 53%. Beyond that, FineIBT also has other perks, such as benefiting from the CET's hardening against transient execution attacks and not depending on Link-Time Optimizations. This talk will explore the FineIBT implementation recently sent to the kernel-hardening mailing list, then discuss specific scenarios, such as how it could be used in the Linux kernel, possible improvements and expected challenges. Technical reference: https://www.openwall.com/lists/kernel-hardening/2021/02/11/1

In 2019, AMD introduced SEV-SNP (Secure Nested Paging), the latest generation of AMD VM isolation technology designed for confidential computing. Now that SEV-SNP hardware is commercially available, AMD is focusing on upstream enablement of the various new security capabilities provided by this technology, including memory integrity protection, new attestation models, interrupt security, and more. In this talk, we will provide a brief overview of these new capabilities and the status of upstream enablement work in the Linux kernel, QEMU, and related projects. We’ll also discuss planned future areas of development and how anyone interested can get involved.

The kernel, along with several other important projects, continue to use fully decentralized means of collaboration that is based on sending patches and code reviews via email. Existing end-to-end email attestation mechanisms, such as PGP/MIME or S/MIME, have important drawbacks that limit their usefulness when it comes to attesting structured content like patches. Patatt is a small library that adopts the DKIM standard to introduce end-to-end cryptographic signing of patches. When incorporated into maintainer tools like b4, it allows for full end-to-end attestation of code, as well as public keyring management via the git repository itself.

System security and safety have common goals, yet often follow divergent development paths. We are taking a look at the Linux kernel configuration features, many of which were originally designed for security, which can be used to enable safety critical applications. In this talk, we will give an overview of our recent work researching existing kernel features important to enable safety critical applications. The kernel configurations are mapped onto Common Weakness Enumerations, but more significantly we demonstrate how they are specifically relevant to support basic safety features such as kernel memory or avoiding race conditions. The work is in the context of ELISA (https://elisa.tech), striving to promote the acceptance of Linux in industries such as avionics, medical devices, and automotive, for which safety is an essential requirement. Our goal is to discuss our work with the Linux kernel developers engaged in the Linux Self-Protection Project and others interested in this area.

A new approach for providing a /dev/random implementation is publicly available with the LRNG implementation and sent to the Linux kernel community for review. This new implementation provides the following benefits: * Sole use of contemporary cryptographic algorithms for data processing * Significant performance gains in performance critical interrupt handler * Availability of test interfaces allowing all execution steps to be validated including extracting of raw noise for entropy assessments * Flexible configuration including runtime-replacement of cryptographic components for crypto-agility * Clean design of combining multiple entropy sources With its API and ABI compliant interfaces to the existing /dev/random implementation the LRNG can be used as a drop-in replacement. The presentation is intended to introduce the different aspects of the LRNG and explain how the LRNG integrates with the Linux kernel. The goal is to allow peer kernel developers to understand the LRNG. The presentation also provides suggestions on how the LRNG may be integrated into the mainline kernel.

Memory Tagging Extension (MTE) is an ARM v8.5 feature that enables hardware-assisted validation of the correctness of memory accesses. In a nutshell, MTE allows assigning tags to memory allocations, as well as to pointers that refer to those allocations. When a pointer is accessed, the CPU performs a validity check that ensures that the memory tag matches the pointer tag. As of now, MTE is integrated into the Linux kernel. It is available in both mainline and the Android common kernels. This talk focuses on the way MTE is used to assert the validity of kernel memory accesses. The talk describes the current state of the newly added Hardware Tag-Based KASAN mode and its planned improvements.