Daniel Hedin

Information-Flow Security for a Core of JavaScript

Tracking information flow in dynamic languages remains an important and intricate problem. This paper makes substantial headway toward understanding the main challenges and resolving them. We identify language constructs that constitute a core of Java Script: objects, higher-order functions, exceptions, and dynamic code evaluation. The core is powerful enough to naturally encode native constructs as arrays, as well as functionalities of Java Scripts API from the document object model (DOM) related to document tree manipulation and event processing. As the main contribution, we develop a dynamic type system that guarantees information-flow security for this language.

JSFlow: tracking information flow in JavaScript and its APIs

JavaScript drives the evolution of the web into a powerful application platform. Increasingly, web applications combine services from different providers. The script inclusion mechanism routinely turns barebone web pages into full-fledged services built up from third-party code. Such code provides a range of facilities from helper utilities (such as jQuery) to readily available services (such as Google Analytics and Tynt). Script inclusion poses a challenge of ensuring that the integrated third-party code respects security and privacy. This paper presents JSFlow, a security-enhanced JavaScript interpreter for fine-grained tracking of information flow. We show how to resolve practical challenges for enforcing information-flow policies for the full JavaScript language, as well as tracking information in the presence of libraries, as provided by browser APIs. The interpreter is itself written in JavaScript, which enables deployment as a browser extension. Our experiments with the extension provide in-depth understanding of information manipulation by third-party scripts such as Google Analytics. We find that different sites intended to provide similar services effectuate rather different security policies for the users sensitive information: some ensure it does not leave the browser, others share it with the originating server, while yet others freely propagate it to third parties.

Timing Aware Information Flow Security for a JavaCard-like Bytecode

Common protection mechanisms fail to provide end-to-end security; programs with legitimate access to secret information are not prevented from leaking this to the world. Information-flow aware analyses track the flow of information through the program to prevent such leakages, but often ignore information flows through covert channels even though they pose a serious threat. A typical covert channel is to use the timing of certain events to carry information. We present a timing-aware information-flow type system for a low-level language similar to a non-trivial subset of a sequential Java bytecode. The type system is parameterized over the time model of the instructions of the language and over the algorithm enforcing low-observational equivalence, used in the prevention of implicit and timing flows.

Cryptographically-masked flows

Cryptographic operations are essential for many security-critical systems. Reasoning about information flow in such systems is challenging because typical (noninterference-based) information-flow definitions allow no flow from secret to public data. Unfortunately, this implies that programs with encryption are ruled out because encrypted output depends on secret inputs: the plaintext and the key. However, it is desirable to allow flows arising from encryption with secret keys provided that the underlying cryptographic algorithm is strong enough. In this article we conservatively extend the noninterference definition to allow safe encryption, decryption, and key generation. To illustrate the usefulness of this approach, we propose (and implement) a type system that guarantees noninterference for a small imperative language with primitive cryptographic operations. The type system prevents dangerous program behavior (e.g., giving away a secret key or confusing keys and nonkeys), which we exemplify with secure implementations of cryptographic protocols. Because the model is based on a standard noninterference property, it allows us to develop some natural extensions. In particular, we consider public-key cryptography and integrity, which accommodate reasoning about primitives that are vulnerable to chosen-ciphertext attacks.

Cryptographically-Masked flows

Cryptographic operations are essential for many security-critical systems. Reasoning about information flow in such systems is challenging because typical (noninterference-based) information-flow definitions allow no flow from secret to public data. Unfortunately, this implies that programs with encryption are ruled out because encrypted output depends on secret inputs: the plaintext and the key. However, it is desirable to allow flows arising from encryption with secret keys provided that the underlying cryptographic algorithm is strong enough. In this paper we conservatively extend the noninterference definition to allow safe encryption, decryption, and key generation. To illustrate the usefulness of this approach, we propose (and implement) a type system that guarantees noninterference for a small imperative language with primitive cryptographic operations. The type system prevents dangerous program behavior (e.g., giving away a secret key or confusing keys and non-keys), which we exemplify with secure implementations of cryptographic protocols. Because the model is based on a standard noninterference property, it allows us to develop some natural extensions. In particular, we consider public-key cryptography and integrity, which accommodate reasoning about primitives that are vulnerable to chosen-ciphertext attacks.

Noninterference in the presence of non-opaque pointers

A common theoretical assumption in the study of information flow security in Java-like languages is that pointers are opaque - i.e., that the only properties that can be observed of pointers are the objects to which they point, and (at most) their equality. These assumptions often fail in practice. For example, various important operations in Javas standard API, such as hashcodes or serialization, might break pointer opacity. As a result, information-flow static analyses which assume pointer opacity risk being unsound in practice, since the pointer representation provides an unchecked implicit leak. We investigate information flow in the presence of non-opaque pointers for an imperative language with records, pointer instructions and exceptions, and develop an information flow aware type system which guarantees noninterference

Value-Sensitive Hybrid Information Flow Control for a JavaScript-Like Language

Secure integration of third-party code is one of the prime challenges for securing todays web. Recent empirical studies give evidence of pervasive reliance on and excessive trust in third-party JavaScript, with no adequate security mechanism to limit the trust or the extent of its abuse. Information flow control is a promising approach for controlling the behavior of third-party code and enforcing confidentiality and integrity policies. While much progress has been made on static and dynamic approaches to information flow control, only recently their combinations have received attention. Purely static analysis falls short of addressing dynamic language features such as dynamic objects and dynamic code evaluation, while purely dynamic analysis suffers from inability to predict side effects in non-performed executions. This paper develops a value-sensitive hybrid mechanism for tracking information flow in a JavaScript-like language. The mechanism consists of a dynamic monitor empowered to invoke a static component on the fly. This enables us to achieve a sound yet permissive enforcement. We establish formal soundness results with respect to the security policy of non-interference. In addition, we demonstrate permissiveness by proving that we subsume the precision of purely static analysis and by presenting a collection of common programming patterns that indicate that our mechanism has potential to provide more permissiveness than dynamic mechanisms in practice.

Boosting the Permissiveness of Dynamic Information-Flow Tracking by Testing

Tracking information flow in dynamic languages remains an open challenge. It might seem natural to address the challenge by runtime monitoring. However, there are well-known fundamental limits of dynamic flow-sensitive tracking of information flow, where paths not taken in a given execution contribute to information leaks. This paper shows how to overcome the permissiveness limit for dynamic analysis by a novel use of testing. We start with a program supervised by an information-flow monitor. The security of the execution is guaranteed by the monitor. Testing boosts the permissiveness of the monitor by discovering paths where the monitor raises security exceptions. Upon discovering a security error, the program is modified by injecting an annotation that prevents the same security exception on the next run of the program. The elegance of the approach is that it is sound no matter how much coverage is provided by the testing. Further, we show that when the mechanism has discovered the necessary annotations, then we have an accuracy guarantee: the results of monitoring a program are at least as accurate as flow-sensitive static analysis. We illustrate our approach for a simple imperative language with records and exceptions. Our experiments with the QuickCheck tool indicate that random testing accurately discovers annotations for a collection of scenarios with rich information flows.

Securing Interactive Programs

This paper studies the foundations of information-flow security for interactive programs. Previous research assumes that the environment is total, that is, it must always be ready to feed new inputs into programs. However, programs secure under this assumption can leak the presence of input. Such leaks can be magnified to whole-secret leaks in the concurrent setting. We propose a framework that generalizes previous research along two dimensions: first, the framework breaks away from the totality of the environment and, second, the framework features fine-grained security types for communication channels, where we distinguish between the security level of message presence and message content. We show that the generalized framework features appealing compositionality properties: parallel composition of secure program results in a secure thread pool. We also show that modeling environments as strategies leads to strong compositionality: various types of composition (with and without scoping) follow from our general compositionality result. Further, we propose a type system that supports enforcement of security via fine-grained security types.

Architectures for Inlining Security Monitors in Web Applications

Securing JavaScript in the browser is an open and challenging problem. Code from pervasive third-party JavaScript libraries exacerbates the problem because it is executed with the same privileges as the code that uses the libraries. An additional complication is that the different stakeholders have different interests in the security policies to be enforced in web applications. This paper focuses on securing JavaScript code by inlining security checks in the code before it is executed. We achieve great flexibility in the deployment options by considering security monitors implemented as security-enhanced JavaScript interpreters. We propose architectures for inlining security monitors for JavaScript: via browser extension, via web proxy, via suffix proxy (web service), and via integrator. Being parametric in the monitor itself, the architectures provide freedom in the choice of where the monitor is injected, allowing to serve the interests of the different stake holders: the users, code developers, code integrators, as well as the system and network administrators. We report on experiments that demonstrate successful deployment of a JavaScript information-flow monitor with the different architectures.