TL;DR
This paper introduces HTTPi, a new protocol designed to provide end-to-end web content integrity without the performance and caching limitations of HTTPS. By allowing content to be cached at intermediate servers while still guaranteeing its authenticity, HTTPi aims to make the web more secure and efficient, especially in the face of widespread man-in-the-middle attacks.
What Problem Does This Paper Address?
The modern web, especially with the prevalence of public Wi-Fi, is vulnerable to man-in-the-middle (MITM) attacks. While HTTPS offers a solution by providing confidentiality and integrity, its widespread adoption is hampered by performance overhead and its inability to leverage caching infrastructure like Content Delivery Networks (CDNs). This forces website operators to choose between security (HTTPS, with its costs) and performance/scalability (HTTP, with its vulnerabilities). Furthermore, many websites use a mix of HTTP and HTTPS content, creating "mixed-content" issues that compromise security. This paper addresses the need for a protocol that provides the essential security guarantees of integrity and authentication for web content, without the performance penalties of full encryption, thereby enabling efficient caching and improving overall web security.
Methods
The core of the proposed solution is the HTTPi protocol. HTTPi builds upon the existing HTTP protocol, meaning no changes are required for intermediate servers like CDNs or caching proxies. The key innovation lies in how it ensures end-to-end integrity:
- Content Signing: HTTPi uses a content-signature-based scheme. Instead of signing the entire HTTP response (which would prevent progressive loading), it signs segments of the content. The unit of verification is set to the typical TCP segment size (1400 bytes) to minimize waiting time for clients.
- Progressive Loading: To support modern web browsing where pages load progressively, HTTPi segments are signed in a way that allows browsers to verify and render content as it arrives. This is achieved by signing hashes of segments, with mechanisms to handle both static and dynamically generated content.
- Amortized Signing: The signing cost is amortized over multiple segments. For static content, a single signature can cover multiple segments. For dynamic content, signatures are computed per chunk, but the overhead is still managed.
- Integrity Header: A new "Integrity" header is introduced to signal the use of HTTPi.
- Mixed Content Handling: The paper proposes mechanisms for browsers to handle mixed content (HTTP, HTTPi, HTTPS) from the same domain, leveraging Content Security Policy (CSP) for granular control.
- Access Control: Inspired by information flow models (Bell LaPadula and Biba), the paper defines default communication policies between different protocol types to maintain security.
Datasets
The paper does not introduce new datasets for training. Instead, it relies on an analysis of real-world web content to evaluate cacheability. Specifically, they analyzed the top 1,000 Alexa sites to determine the prevalence of publicly cacheable web objects served over HTTP and HTTPS. For performance evaluation, they used a 836KB test page and a 4.8 KB test page (representing typical page sizes) to measure end-to-end response times and server throughput.
Key Insights and Ideas
- Integrity is Key, Confidentiality is Optional for Caching: The paper's central insight is that for many web security concerns, particularly preventing MITM attacks and ensuring the Same-Origin Policy (SOP) is meaningful, end-to-end integrity and authentication are sufficient. Confidentiality (encryption) is not strictly necessary for these goals and is the primary reason HTTPS cannot be effectively cached.
- HTTPi as a "Sweet Spot": HTTPi occupies a sweet spot by offering integrity without confidentiality, enabling it to be cached by intermediate servers like CDNs, thus regaining performance and scalability benefits lost with HTTPS.
- Practicality Over Theory: While the idea of end-to-end integrity without confidentiality isn't entirely new (e.g., SHTTP), the paper's significant contribution is its practical system design and implementation that addresses previously overlooked challenges like progressive loading and mixed content.
- Mixed Content is a Major Problem: The paper highlights how mixed-content pages, common today, undermine the security of HTTPS pages. HTTPi offers a way to secure these components without forcing a complete switch to HTTPS.
- Browser Vendors Should Adopt HTTPi: The authors strongly advocate for browser and web server vendors to adopt HTTPi, suggesting it's a more practical and scalable path to widespread web integrity than solely relying on HTTPS.
Key Results
- Negligible Performance Overhead: HTTPi incurs negligible performance overhead over HTTP. Measurements showed throughput degradation of less than 1% compared to HTTP, while HTTPS showed a 25-75% drop in throughput due to SSL handshake overhead.
- Significant Caching Benefits: A study of the top 1,000 Alexa sites revealed that approximately 73% of web objects are publicly cacheable. HTTPi can leverage this cacheability for content currently served over HTTP or even for cacheable HTTPS content.
- Practical Implementation: The authors successfully implemented a prototype of HTTPi for Internet Explorer and IIS, demonstrating its feasibility.
- Reduced Latency: For a 836KB test page, HTTPi added only 0.7-2.0 seconds of latency over HTTP and HTTPS, a cost deemed acceptable for the security benefits.
- Server-Side Efficiency: HTTPi offers a significant performance incentive for web servers, with minimal throughput degradation compared to HTTP, unlike HTTPS.
Limitations and Future Work
- JavaScript Dynamic Content: The current implementation may miss dynamically generated links within JavaScript, which could lead to missed HTTPi opportunities. The authors mention they are working on solving this.
- Client-Side Optimization: While the server-side overhead is minimal, client-side operations (like signature verification) still incur some cost. Further client-side optimizations are suggested.
- Browser/Server Adoption: The success of HTTPi hinges on adoption by browser and web server vendors, which is a significant hurdle.
- Future Work: The paper suggests further work on refining mixed-content policies, exploring more advanced access control mechanisms, and potentially integrating HTTPi with existing security features like HSTS. The authors also note that while their implementation is for IE, the protocol itself is browser-agnostic.
Summary updated