GET /out: Automated Discovery of Application-Layer Censorship Evasion Strategies
GET /out: Automated Discovery of Application-Layer Censorship Evasion Strategies
复制标题
DOI：
--
复制
发表时间：
2022
期刊：
影响因子：
--
通讯作者：
Michael Harrity;Kevin Bock;Frederick Sell;Dave Levin
中科院分区：
文献类型：
--
作者：
Michael Harrity;Kevin Bock;Frederick Sell;Dave Levin
审查军备竞赛最近经历了一个转变，这要归功于最近的努力，表明可以以自动化的方式发现逃避审查的新方法。然而，所有这些先前的自动化工作都是通过操纵TCP/IP报头来进行的;虽然令人印象深刻，但部署这些报头已被证明具有挑战性，因为报头修改通常需要比审查阻止应用程序更大的权限。在这方面的工作中，应用层在很大程度上还没有被开发。这不是没有原因的：应用程序消息的空间比TCP/IP报头大得多，结构化程度也低得多。在本文中，我们提出了第一种技术，完全在应用层自动发现新的审查规避技术。我们提出了一个通用的解决方案，并将其具体应用于中国，印度和哈萨克斯坦的HTTP和DNS审查。我们的自动化技术共发现了77种用于HTTP的独特规避策略和9种用于DNS的独特规避策略，所有这些策略都只需要应用程序层修改，使它们更容易融入应用程序和部署。我们分析了这些策略，并揭示了审查机构的内部运作。我们发现，应用层策略的成功在很大程度上取决于目标服务器的类型和版本。令人惊讶的是，我们的一大类规避策略利用了审查器比流行的应用服务器更符合RFC的实例。我们已经公开了我们的代码。
The censorship arms race has recently gone through a transformation, thanks to recent efforts showing that new ways to evade censorship can be discovered in an automated fashion. However, all of these prior automated efforts operate by manipulating TCP/IP headers; while impressive, deploying these have proven challenging, as header modifications often require greater privileges than are available to censorship cir-cumvention apps. In that line of work, the application layer has gone largely unexplored. This is not without reason: the space of application messages is much larger and far less structured than TCP/IP headers. In this paper, we present the first techniques to automate the discovery of new censorship evasion techniques purely in the application layer. We present a general solution and apply it specifically to HTTP and DNS censorship in China, India, and Kazakhstan. Our automated techniques discovered a total of 77 unique evasion strategies for HTTP and 9 for DNS, all of which require only application-layer modifications, making them easier to incorporate into apps and deploy. We analyze these strategies and shed new light into the inner workings of the censors. We find that the success of application-layer strategies can depend heavily on the type and version of the destination server. Surprisingly, a large class of our evasion strategies exploit instances in which censors are more RFC-compliant than popular application servers. We have made our code publicly available.