TL;DR
While major tech giants like Google and Amazon reportedly "killed" domain fronting years ago, this paper reveals that the technique is alive and well. Using a new automated measurement system, researchers found that 22 out of 30 tested CDNs—including industry leaders like Akamai and Fastly—remain vulnerable to domain fronting, providing a persistent loophole for both censorship circumvention and malware orchestration.
What Problem Does This Paper Address?
Domain fronting is a clever networking "magic trick." It exploits a discrepancy in how Content Delivery Networks (CDNs) handle requests: the TLS Server Name Indication (SNI) is unencrypted and visible to network monitors, while the HTTP Host header is encrypted inside the TLS tunnel. By putting a "safe" domain in the SNI and a "hidden" domain in the Host header, a user can trick a firewall into thinking they are visiting a benign site (like a government portal) while actually communicating with a blocked service or a malware command-and-control (C2) server.
Following high-profile use by apps like Signal and Telegram, major CDNs faced pressure to disable this. However, measuring the actual state of the ecosystem has been difficult. Previous studies were manual, expensive (requiring researchers to buy and host their own domains on every CDN), and didn't scale. This paper seeks to provide the first large-scale, automated "health check" of the CDN ecosystem regarding domain fronting.
Methods
The authors developed an automated pipeline to test CDNs without the need to register new domains or pay for subscriptions. Their system consists of three main stages:
- CDN Domain Discovery: The system analyzes passive and active DNS logs to find CNAME records. If a domain points to a known CDN-owned suffix (e.g.,
*.edgekey.netfor Akamai), it maps that domain to that specific CDN. - URL Discovery: Once they have a list of domains for a CDN, they use a Chromium-based crawler (Puppeteer) to find specific URLs for static assets (like images or .js files). These serve as the "targets" for the fronting test.
- Automated Testing: The system crafts a specialized HTTPS request. It sets the TLS SNI to a "Front Domain" and the internal HTTP Host header to a "Target Domain" (both hosted on the same CDN).
- Validation: To ensure the test is valid, they compare the SHA1 hash of the response from the "fronted" request against a legitimate request to the target. If they match, the CDN is confirmed to be prone to domain fronting.
Datasets
- Passive DNS Logs: 10 days of traffic from two large academic networks.
- ActiveDNS Project: Openly available DNS data used to broaden the search.
- Tranco List: Used to rank the popularity of the discovered domains.
- VirusTotal: Used to check if any of the domains already associated with these CDNs were previously flagged as malicious.
Key Insights and Ideas
- The "Long Tail" of CDNs: We often think of the web as living on just three or four CDNs. The researchers found that even highly popular sites (top 10k) are distributed across a wide variety of less-known CDNs, all of which represent potential "fronts."
- Inconsistent Mitigation: Mitigation isn't always "all or nothing." For example, Akamai showed a 52% success rate. This suggests that CDNs often deploy protections on newer infrastructure or for new customers, while older "legacy" edge servers remain vulnerable.
- Malicious "Pre-Fronting": The study found that ~31% of the vulnerable CDNs were already hosting domains flagged as malicious by security vendors, suggesting that attackers are already actively exploiting these specific infrastructures.
- The Cost of Defense: The paper notes that CDNs often hesitate to block domain fronting because it requires their TLS proxies and business logic proxies to communicate and "double-check" each other, which can be computationally expensive at scale.
Key Results
- Widespread Vulnerability: 73% (22 out of 30) of the tested CDNs were prone to domain fronting.
- Major Players Involved: Fastly was 100% vulnerable across all test cases, and Akamai was vulnerable in 52% of cases.
- Successful Mitigators: Cloudflare and Cloudfront (Amazon) were among the few that successfully blocked domain fronting across their entire tested infrastructure (0% success rate).
- Scale of Discovery: The researchers identified over 124,000 distinct domains served by 38 different CDNs, proving their automated discovery method is highly effective compared to manual efforts.
Limitations and Future Work
- Crawler Reach: The researchers couldn't find valid URLs for 8 of the 38 CDNs they discovered. They suggest that a more "aggressive" or non-targeted web crawl might be needed to find testable assets for these niche providers.
- DNS Visibility: The system's effectiveness is tied to the quality of the DNS data it can access. More diverse DNS datasets would likely reveal even more vulnerable CDNs.
- Manual Seeding: The initial list of CDN-owned subdomains (SLDs) still requires manual curation. Automating the discovery of these "infrastructure domains" is a potential area for future research.
- Responsible Disclosure: The authors have begun disclosing these vulnerabilities to the affected CDNs; the long-term impact will depend on how these providers choose to update their legacy infrastructure.
Summary updated