Google’s QUIC (Quick UDP Internet Connections) protocol, announced in 2013 as a way to make the web faster, waited seven years before being implemented in the ad giant’s Chrome browser. But it still arrived before privacy could get there.
A trio of researchers from China have found that QUIC is more vulnerable to web fingerprinting than HTTPS, a shortcoming that could make it easier for an adversary to infer which websites an individual is visiting by scrutinizing network traffic.
Boffins Pengwei Zhana and Liming Wang of the Chinese Academy of Sciences, and Yi Tang, of Guangzhou University, funded by China’s National Key R&D Program, describe their findings in a paper distributed via ArXiv.
The paper, “Website Fingerprinting on Early QUIC Traffic,” has been submitted to Elsevier Computer Networks.
TCP and UDP, the paper explains, are two networking protocols in the TCP/IP suite. TCP has some limitations in certain situations, like head of line blocking and retransmission ambiguity. Google developed QUIC to solve issues like these and the protocol is being worked on in parallel by the Internet Engineering Task Force (IETF) as a standard. About five per cent of websites currently support QUIC, according to the paper, and Chromium-based browsers will try QUIC first before falling back to HTTPS if QUIC is unavailable.
QUIC! IETF sets November deadline for last comments on TCP-killer spawned by Google and Cloudflare
“QUIC is a UDP-based cryptographic protocol with built-in TLS function and optimized multiplexing, flow control, and congestion control mechanism, which solves TCP transmission performance shortcomings,” the paper says. “QUIC, which is equal to HTTP+TLS+UDP, can achieve the same or better transmission efficiency as HTTPS (equal to HTTP+TLS+TCP) in most network conditions and provides security equivalent to TLS.”
But HTTPS, the researchers found, remains more resistant to web fingerprinting. Web fingerprinting in this context is not browser/device fingerprinting, which captures browser-based data points like browser type, device brand, screen size, and so on, and uses those to calculate a hash identifier. Rather, it refers to data extracted by sniffing network traffic between internet users and the websites they visit.
Web fingerprinting of this sort involves eavesdropping on people’s encrypted network traffic – the contents of which remain protected – and using the observed packets to build distinct patterns that correspond with specific websites. The attack measures characteristics like packet size, packet order, total transmission site, and other metrics. The researchers define some of these characteristics as Simple features – eight general packet size categories – and others as Transfer features – e.g. packet order.
This network data forms the basis for a model that can subsequently be used to predict which websites are being visited, the sort of thing, say, a censorious government might wish to do.
The boffins claim that the maximum attack accuracy on QUIC is about 57 per cent, which is 73 per cent higher than on HTTPS. By using “early traffic” – the initial packets being exchanged – they claim QUIC attack accuracy can reach about 95 per cent with only 40 packets and Simple features, compared to about 60 per cent attack accuracy for HTTPS.
The researchers caution that their experiments were conducted in an environment where the traffic quality was pure and that real world network conditions may lead to different results. But they remain dubious about QUIC’s capacity for privacy.
“The superior transmission performance of the QUIC protocol brings opportunities for speeding up the Internet, but its security risks bring uncertainties,” the boffins conclude. “…The vulnerability of QUIC on early traffic poses a significant challenge to the privacy and confidentiality guaranteed.” ®