CADOS: Configurability-Aware Development of Operating Systems (DFG: LO 1719/3-1)

Todays operating systems (as well as other system software) offer a great deal of static configurability to tailor them with respect to a specific application or hardware platform. Linux 4.2, for instance, provides (via its Kconfig models and tools) more than fifteen thousand configurable features for this purpose. Technically, the implementation of all these features is spread over multiple levels of the software generation process, including the configuration system, build system, C preprocessor, compiler, linker, and more. This enormous variability has become unmanageable in practice; in the case of Linux it already has led to thousands of variability defects within the lifetime of Linux. With this term, we denote bugs and other quality issues related to the implementation of variable features. Variability defects manifest as configuration consistency and configuration coverage issues.

In the CADOS project, we investigate scalable methods and tools to grasp the variability on every layer within the configuration and implementation space, visualize and analyze it and, if possible, adjust it while maintaining a holistic view on variability.

Our findings have already led to more than 100 accepted patches in the Linux mainline kernel (see our EuroSys '11 and SPLC '12 papers) and an approach for the automatic tailoring of Linux server systems in order to reduce the exploitable code base (see our HotDep '12 paper). The issue of configuration coverage is addressed in our USENIX '14 paper. Our scalable AST-based hashing method to detect relevant changes in C-source files got a best paper award at USENIX '17.

People

Publications

CADOS is the follow-up project to VAMOS, which set some important foundations. Please look here for a list including the relevant VAMOS publications.

EuroSys Conference
Multiverse: Compiler-Assisted Management of Dynamic Variability in Low-Level System Software
Florian Rommel, Christian Dietrich, Michael Rodin, Daniel LohmannFourteenth EuroSys Conference 2019 (EuroSys '19)ACM Press2019.
PDF 10.1145/3302424.3303959 [BibTex]
ICSE Conference
The List is the Process: Reliable Pre-Integration Tracking of Commits on Mailing Lists
Ralf Ramsauer, Daniel Lohmann, Wolfgang MauererProceedings of the 41st International Conference on Software Engineering (ICSE '19)2019To appear.
PDF Raw Data [BibTex]
USENIX Conference Best Paper Award
cHash: Detection of Redundant Compilations via AST Hashing
Christian Dietrich, Valentin Rothberg, Ludwig Füracker, Andreas Ziegler, Daniel LohmannProceedings of the 2017 USENIX Annual Technical Conference USENIX Association2017Best Paper Award.
PDF Details Slides Raw Data [BibTex]
VAMOS Workshop
Analyzing the Impact of Feature Changes in Linux
Andreas Ziegler, Valentin Rothberg, Daniel LohmannProceedings of the Tenth International Workshop on Variability Modelling of Software-intensive Systems2016.
PDF 10.1145/2866614.2866618 [BibTex]
VAMOS Workshop
Feature Models in Linux - From Symbols to Semantic
Valentin Rothberg, Nicolas Dintzner, Andreas Ziegler, Daniel LohmannProceedings of the Tenth International Workshop on Variability Modelling of Software-intensive Systems2016.
PDF 10.1145/2866614.2866624 [BibTex]
OpenSym Conference
Observing Custom Software Modifications: A Quantitative Approach of Tracking the Evolution of Patch Stacks
Ralf Ramsauer, Daniel Lohmann, Wolfgang MauererProceedings of the 12th International Symposium on Open Collaboration (OpenSym '16)2016.
PDF 10.1145/2957792.2957810 [BibTex]
GPCE Conference
Towards Scalable Configuration Testing in Variable Software
Valentin Rothberg, Christian Dietrich, Andreas Ziegler, Daniel LohmannProceedings of the 2016 International Conference on Generative Programming: Concepts and Experiences2016.
PDF 10.1145/2993236.2993252 [BibTex]
GPCE Conference
Automatic Feature Selection in Large-Scale System-Software Product Lines
Andreas Ruprecht, Bernhard Heinloth, Daniel LohmannProceedings of the 13th International Conference on Generative Programming and Component Engineering (GPCE '14)ACM Press2014.
PDF 10.1145/2658761.2658767 [BibTex]
USENIX Conference
Static Analysis of Variability in System Software: The 90,000 #ifdefs Issue
Reinhard Tartler, Christian Dietrich, Julio Sincero, Wolfgang Schröder-Preikschat, Daniel LohmannProceedings of the 2014 USENIX Annual Technical Conference USENIX Association2014.
PDF Raw Data [BibTex]

CADOS in the Wild

CADOS research students attended the LPC '14 and gave talks at the Development Tools Microconference.

Valentin Rothberg talked about his new tool undertaker-checkpatch, which analyzes patch files. The vampyr tool to configurability-aware compile test (or determine the variability of) source files was presented by Stefan Hengelein.Abstracts and slides for both talks can be found at the talk-specific sites.

The Linux Weekly News site summarizes the highlights of the Development Tools Microconference and features both CADOS talks.

Tools

Various tools assist our analysis of the variability in the Linux kernel and other software projects. Our most important tool is called the undertaker, which searches and reports dead and undead conditional CPP blocks. Since its inception within VAMOS, it has matured to a toolbox of utilities for answering various research questions related to variability in general. For further details, please look into the undertaker trac.

vampyr

This tool is a wrapper for undertaker to do coverage analysis or variability-aware static analysis on source files. For the static analysis compilers such as gcc or clang are employed. Please refer to the USENIX '14 paper for detailed information.

undertaker-checkpatch

This tool reports changes to defects such as newly introduced or fixed defects. Defects can also be correlated to changes in Kconfig and the build system (Make, Kbuild) and vice versa. Additionally, undertaker-checkpatch ships the functionality to further analyze the causes of defects, displaying contradictory Kconfig items, a block's precondition or the defect causing formula. Since version 1.6, undertaker is able to minimize defect formulas (thanks to the PicoMUS-Tool which is part of PicoSAT). This functionality can be used in undertaker-checkpatch to further analyze and understand the cause of defects.

Flipper: Lightweight Kernel Tailoring

Configuring Linux is hard. With over 14,000 options to choose from, making an informed decision about every single one of them takes a very long time. While distributions for standard day-to-day use simply enable as many features (drivers, supported platforms, ...) as possible, this is not a practical solution for embedded systems, where memory is scarce and must not be wasted. To make it easier for an engineer to derive a small starting point to configure the system, we developed Flipper. Flipper provides a lean method to trace which functionality was exerted in the kernel. Using the Tailor tool from the undertaker package, a small, use-case specific configuration for Linux can be generated from the collected data. Flipper is part of undertaker release (v1.6) and can be downloaded here, for detailed usage instructions please read the README file provided in the tailor/flipper subdirectory.

Wundertaker

Wundertaker is a Web GUI visualizing #ifdefs and undertaker's defect reports in Linux code files. The source including installation instructions can be downloaded here. A detailed description of the tool and its implementation can also be found in the Bachelor's thesis of Patrick Plagwitz.

Multiverse

While static variability can be completely resolved at compile time, dynamic variation points come at a cost arising from extra tests and branches in the control flow. Multiverse is an approach to handle dynamic variability efficiently by means of binary patching. It provides an extension to the C programming language that enables the developer to express dynamic variability in performance-critical paths. With specially annotated config variables, multiverse can generate multiple versions of a function and dynamically binary patch the running system to use the version of the current configuration. The goal is to narrow the gap between dynamic and static variability by allowing the developer to easily employ run-time configurability at zero or low cost.

PaStA: Patch-Stack Analysis

PaStA quantitatively analyses the evolution of patch stacks by mining git repositories and produces data that can serve as input for statistical analysis. It compares different releases of stacks and groups similar patches (patches that lead to similar modifications) into equivalence classes. This allows us to compare those classes against the base project to measure integrability and influence of the patch stack on the base project. PaStA is furthermore able to map mails from mailboxes (e.g., dumps of mailing lists or public inboxes) to commit hashes of repositories. This makes it possible to trace the evolution of patches on mailing lists, like the LKML.

Theses

Open Theses Topics

Currently Running Work

Run-Time Binary Patching for Systems with Dynamically Loadable Modules

Typ: Bachelorarbeit
Status: reserviert
Supervisors: Florian Rommel, Daniel Lohmann
Bearbeiter: Dominik Töllner
Explore the possibilities of using multiverse on systems with loadable modules by extending its semantics and implementation.

Finished Theses

Bintail: Binary Level Tailoring of ELF Executables via Feature Selection and Variant Elimination

Typ: Masterarbeit
Status: abgeschlossen
Supervisors: Florian Rommel, Christian Dietrich, Daniel Lohmann
Bearbeiter: Felix Herrmann (abgegeben: 04. Nov 2018)

Multiverse: Compiler-Assisted Dynamic Variability Management in the Linux Kernel

Typ: Masterarbeit
Status: abgeschlossen
Supervisors: Christian Dietrich, Andreas Ziegler, Wolfgang Schröder-Preikschat, Daniel Lohmann
Bearbeiter: Florian Rommel (abgegeben: 02. Nov 2017)
Use the multiverse compiler plugin to optimize paths in the Linux kernel [PDF]