TL;DR
This paper introduces LIO, a novel dynamic information flow control (IFC) system implemented as a Haskell library. LIO uses a "floating label" approach, where a computation's current label dynamically adjusts to reflect observed sensitive data, and a "clearance" to restrict access, offering a flexible yet secure way to manage information flow in complex software.
What Problem Does This Paper Address?
Managing information flow in complex software systems is crucial for security. Traditional static IFC methods can be too rigid for dynamic environments where new data types and users emerge at runtime. Dynamic IFC approaches, while more flexible, often struggle with safely handling exceptions and preventing covert channels. This paper addresses the challenge of building a flexible, dynamic IFC system that can be implemented as a library, offering robust security guarantees, including safe exception handling and discretionary access control, without requiring language runtime modifications.
Methods
LIO employs a floating-label approach, where a single, mutable "current label" is associated with the entire computation's context. This label dynamically rises to match the highest label of any data observed. This current label then restricts writes to data with lower labels. To maintain flexibility, LIO also introduces labeled values, allowing programmers to explicitly tag data with labels, and a mechanism to encapsulate computations that might otherwise raise the current label. A key innovation is the current clearance, which acts as an upper bound on the current label, enabling discretionary access control and helping to prevent covert channels. LIO also integrates exception handling, where exceptions are labeled with the current label at the point of their occurrence, allowing for safe recovery from IFC violations without compromising security.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- Floating Labels as a Core Abstraction: The idea of a single, dynamic "current label" for an entire computation simplifies IFC management while providing a powerful mechanism for dynamic security enforcement.
- Clearance for Discretionary Access Control: Introducing a "clearance" alongside the current label is a clever way to implement discretionary access control, allowing specific code regions to have a defined boundary for accessing sensitive information. This is crucial for preventing covert channels.
- Labeled Values and Encapsulation: The ability to explicitly label values and encapsulate sensitive computations is vital for maintaining fine-grained control and preventing unintended label creep, especially when dealing with complex data structures or external libraries.
- Safe Exception Handling for Dynamic IFC: Integrating exception handling directly into the IFC model, where exceptions carry security labels, is a significant step towards making dynamic IFC practical. It allows programs to gracefully recover from security violations without leaking information.
- Library-Based IFC: Implementing IFC as a library in a language like Haskell, rather than modifying the language itself, makes the approach more accessible and easier to integrate into existing codebases.
Key Results
The paper presents a formal semantics for LIO using a λ-calculus extended with IFC features. They prove key security properties:
- Non-interference: Confidential information does not leak to observable outputs.
- Confinement: Computations cannot modify data below their current label or access data above their clearance. They also demonstrate the practical application of LIO through a conference reviewing system (Chair), showcasing how it can manage integrity and confidentiality in a dynamic, multi-user environment.
Limitations and Future Work
The authors acknowledge that while their approach is dynamic and flexible, determining the appropriate label for toLabeled calls might require non-trivial static analysis in some cases. They also mention that their current formalization is limited to a single exception type. Future work could explore extending LIO to handle concurrency and more complex scenarios, as well as investigating more sophisticated static analysis techniques to aid in label management.
Summary updated