Compositional and Scheduler-Independent Information Flow Security

• Main Author: Sudbrock, Henning
• Format: Others
• Language: English; en
• Published: 2013
• Online Access: http://tuprints.ulb.tu-darmstadt.de/3884/1/2014-04-12-sudbrock-dissertation-final.pdf; Sudbrock, Henning <http://tuprints.ulb.tu-darmstadt.de/view/person/Sudbrock=3AHenning=3A=3A.html> : Compositional and Scheduler-Independent Information Flow Security. Technische Universität, Darmstadt [Ph.D. Thesis], (2013)

Software pervades our society deeper with every year. This trend makes software security more and more important. For instance, software systems running critical infrastructures like power plants must withstand criminal or even terrorist attacks, but also smartphone apps used by consumers in their daily routine are usually expected to operate securely. In particular, before entrusting a program with confidential information (such as, e.g., image or audio data recorded by a smartphone), one wants to be sure that the program is trustworthy and does not leak the secrets to untrusted sinks (such as, e.g., an untrusted server on the Internet). Information flow properties characterize such confidentiality requirements by restricting the flow of confidential information, and an information flow analysis permits to check that a program respects those restrictions. The problem of information flow in multi-threaded programs is particularly challenging, because information flows can originate in subtle ways from the interplay between threads. Moreover, the existence of such information flows depends on the scheduler, which might not even be known when analyzing a program. To obtain high assurance that no leak is overlooked in an information flow analysis, formally well-founded analyses provide a rigorous solution. Such analyses are proven sound with respect to formal information flow properties that specify precisely what restrictions on information flow mean. In this thesis, we develop two novel information flow properties for multi-threaded programs, FSI-security and SIFUM-security. These properties are scheduler-independent, i.e., they characterize secure information flow for different schedulers simultaneously. Moreover, they are compositional, i.e., they permit to break down the analysis of a multi-threaded program to single threads. For both properties we develop a security analysis based on a security type system that is proven sound with respect to the property. Compared to existing scheduler-independent information flow properties, FSI-security is less restrictive. In particular, FSI-security is the first scheduler-independent information flow property that permits programs with nondeterministic behavior and programs whose control flow depends on secrets. The security analysis based on SIFUM-security is the first provably sound flow-sensitive information flow analysis for multi-threaded programs in the form of a security type system. Flow-sensitivity results in increased analysis precision by taking the order of program statements into account. The key in our development of SIFUM-security and the corresponding flow-sensitive analysis for multi-threaded programs was to adopt assumption-guarantee style reasoning to information flow security. We integrate FSI-security and SIFUM-security into the novel property FSIFUM-security, and we integrate the security analyses for FSI- and SIFUM-security into a security analysis for FSIFUM-security. Thereby, FSIFUM-security and the corresponding analysis inherit the advantages of both FSI- and SIFUM-security. In addition to developing novel type-based information flow analyses we also explore information flow analysis for multi-threaded programs with program dependence graphs (PDGs) which is used successfully to analyze sequential programs. To this end, we develop a formal connection between PDG-based and type-based information flow analysis for sequential programs. We exploit the connection to transfer concepts from our type-based analysis for multi-threaded programs to PDGs, resulting in a provably sound PDG-based information flow analysis for multi-threaded programs. Beyond this, we also use the connection to transfer concepts from PDGs to type systems and to precisely compare the precision of a type-based and a PDG-based information flow analysis. Our results provide foundations for more precise and more widely applicable information flow analysis for multi-threaded programs, and we hope that they contribute to a more wide-spread certification of information flow security for concurrent programs.