Measuring CDNs susceptible to Domain Fronting

Karthika Subramani, Roberto Perdisci, Pierros Skafidas

TL;DR

This paper introduces an automated system to test Content Delivery Networks (CDNs) for domain fronting vulnerabilities. Their findings reveal that a significant number of CDNs, including major players like Akamai and Fastly, still allow this technique, posing risks for both censorship circumvention and malicious activities.

What Problem Does This Paper Address?

Domain fronting is a clever technique that exploits how CDNs work to mask the true destination of network traffic. By making a request appear to be for a legitimate domain hosted on a CDN (the "front domain"), attackers or users can actually send traffic to a different, hidden destination (the "target domain") also served by the same CDN. This has been a powerful tool for bypassing internet censorship, allowing access to services like Signal and Telegram in restrictive countries. However, it's also a favorite for malware command-and-control (C2) infrastructure, making it hard for security systems to detect and block.

While some major CDN providers (like Google and Amazon) have taken steps to disable domain fronting, it's been unclear how widespread these mitigations are and whether less popular CDNs or even parts of major CDNs still allow it. This leaves a gap in understanding the current landscape of domain fronting susceptibility, making it difficult for security professionals, activists, and even regular users to assess risks and choose services wisely.

Methods

The core of this paper is a novel, automated system designed to systematically test CDNs for domain fronting vulnerabilities without requiring the researchers to register new domains or subscribe to CDN services. The system has three main components:

  1. CDN Domain Discovery: This component uses passive and active DNS traffic analysis to identify domain names whose content is served by a given CDN. It works by looking for CNAME records that point to CDN-specific subdomains.
  2. URL Discovery: Once domains served by a CDN are identified, this component uses a custom Chromium-based web crawler to find actual URLs (pointing to web objects like images, JS, or CSS files) under those domains. This is crucial because simply requesting the root of a domain might not work; specific, existing URLs are needed for testing.
  3. Domain Fronting Tester: This is the heart of the system. For a given CDN and a pair of domains (a "front domain" and a "target domain") known to be served by that CDN, it performs a series of HTTPS requests.
  4. First, it makes a standard request to the target URL with the target domain in the TLS SNI and Host header to get a baseline response.
  5. Then, it attempts the domain fronting test: it requests the same target URL but sets the TLS SNI to the front domain while keeping the Host header as the target domain.
  6. A third request is made with both SNI and Host header set to the front domain to ensure the target object isn't also directly available under the front domain.

A test is considered successful if the response in the fronting test matches the baseline response, and the third request either fails or returns a different response. The system also includes validation steps to rule out false positives, such as checking if the front and target domains are related (e.g., siblings or part of the same SSL certificate).

Datasets

The study leverages several data sources:

  • Passive DNS Traffic: 10 days of passive DNS data collected from two large academic networks (with IRB approval) and the ActiveDNS project. This was used for the CDN Domain Discovery component.
  • Tranco Ranking List: Used to determine the popularity of discovered domains.
  • VirusTotal: Used to identify malicious domains hosted on CDNs.

The paper does not introduce a new benchmark dataset but rather uses existing data sources to build its own datasets of CDN-served domains and URLs for testing.

Key Insights and Ideas

  • Domain Fronting is Still Prevalent: Despite awareness and some mitigation efforts, a surprisingly large number of CDNs still support domain fronting. This suggests that the "cat and mouse" game between attackers and defenders is ongoing.
  • Popular CDNs Aren't Immune: It's not just obscure CDNs; major players like Akamai and Fastly are still susceptible. This is significant because these CDNs host a vast number of popular websites, providing ample "front domain" material for malicious actors.
  • Popular Domains Aren't Exclusive to Popular CDNs: The research found that popular domains (high Tranco ranks) are hosted across a wide range of CDNs, not just the top-tier ones. This means even less popular CDNs can be leveraged for domain fronting using well-known front domains.
  • Inconsistent Mitigation: Some CDNs show partial susceptibility, meaning domain fronting works for some domains/servers but not others within the same CDN. This points to infrastructure inconsistencies or phased rollouts of mitigation strategies.
  • Malicious Domains on CDNs: A notable percentage of CDNs found to be prone to domain fronting also host malicious domains, increasing the risk of abuse.

Key Results

  • 22 out of 30 tested CDNs are susceptible to domain fronting. This represents a significant portion of the tested infrastructure.
  • Popular CDNs like Akamai and Fastly were found to be susceptible. For Akamai, domain fronting was successful for 52% of tested domains, and for Fastly, 100% of tested domains allowed it.
  • 16 out of the 22 susceptible CDNs allowed domain fronting for all tested domains.
  • Approximately 31% of the domain fronting-prone CDNs served content from one or more malicious domains flagged by at least two security vendors.
  • 99.64% of discovered FQDNs consistently mapped to a single CDN over a 10-day period, validating the stability of their domain discovery method.

Limitations and Future Work

  • CDN Infrastructure Enumeration is Hard: The authors acknowledge that fully enumerating all CDN infrastructure is challenging due to their distributed and dynamic nature. Their system relies on known CDN SLDs, and discovering all of them is an ongoing challenge.
  • Limited URL Discovery: While their crawler is effective, it might not discover all possible URLs served by a CDN, especially for less common or dynamically generated content. They suggest expanding this for future work.
  • Incomplete Testing for Some CDNs: For a few CDNs, the number of discoverable domains and URLs was small, making the tests potentially insufficient to confirm full mitigation.
  • Future Work: The authors plan to continue their disclosure process with other CDNs and suggest that future work could involve more comprehensive CDN infrastructure discovery and testing. They also highlight the need for continued efforts to prevent domain fronting abuse.

Summary updated

Abstract

Domain fronting is a network communication technique that involves leveraging (or abusing) content delivery networks (CDNs) to disguise the final destination of network packets by presenting them as if they were intended for a different domain than their actual endpoint. This technique can be used for both benign and malicious purposes, such as circumventing censorship or hiding malware-related communications from network security systems. Since domain fronting has been known for a few years, some popular CDN providers have implemented traffic filtering approaches to curb its use at their CDN infrastructure. However, it remains unclear to what extent domain fronting has been mitigated. To better understand whether domain fronting can still be effectively used, we propose a systematic approach to discover CDNs that are still prone to domain fronting. To this end, we leverage passive and active DNS traffic analysis to pinpoint domain names served by CDNs and build an automated tool that can be used to discover CDNs that allow domain fronting in their infrastructure. Our results reveal that domain fronting is feasible in 22 out of 30 CDNs that we tested, including some major CDN providers like Akamai and Fastly. This indicates that domain fronting remains widely available and can be easily abused for malicious purposes.
BibTeX
@inproceedings{subramani2024measuring,
  author = {Karthika Subramani and Roberto Perdisci and Pierros Skafidas},
  title = {{Measuring CDNs susceptible to Domain Fronting}},
  booktitle = {WWW},
  year = {2024},
  url = {https://openreview.net/forum?id=M8cThPoTC2},
}

Loading related content...

Figure