Internet Filtering in China in 2004 2005 

However, the filtering system can be bypassed   inserting an ampersand (&) into the HPPT GET 

request, such as  search?&q=cache , allowed access to Google's cache.

176

  We could not access Web sites 

with sensitive keywords, such as  falun,  in their URLs.  The filtering mechanism appears specifically 

designed to target Google's cache, since caches of other popular search engines, such as Yahoo!, worked 

properly.

177

Although China no longer blocks Google entirely, a Chinese user will have a very different 

experience when using the search engine for some queries due to the state's filtering practices.  Accessing 

Google's cache is a well known method of ad hoc circumvention of Internet censorship, and China's 

filtering mechanism seems designed specifically to close this loophole. 

I. Filtering by Chinese Search Engines Baidu and Yisou 

In July 2004, Reporters Sans Frontieres admonished Google and Yahoo! as complicit in China's 

filtering practices based on the companies' holdings in two domestic Chinese search engines, Baidu.com 

and Yisou.com.

178

  We researched RSF's claims and confirmed that Baidu and Yisou filter by keyword and 

remove certain search results from their lists, but found that some keyword searches were blocked by 

China's gateway filtering and not the search engines themselves. 

We searched Baidu and Yisou for various sensitive keywords, such as  free Tibet  and  falun.   By 

refining our searches to look specifically within URLs as well as for page content, we concluded that the 

search engines index sensitive sites, but do not list them among search results.  We posit that the search 

engine  crawlers  that compile results may be able to index content despite China's filtering   perhaps by 

operating from a remote location outside China   since the crawlers did index some sensitive sites. We 

also found that some cached versions of sensitive sites were sporadically available, leading us to conclude 

that filtering occurs  upstream,  at the Internet infrastructure level.     

Interestingly, our Baidu and Yisou testing provided important insight into the mechanics of 

China's Web filtering.  When a user requests a banned keyword, the filtering system terminates that user's 

connection to the destination server by sending a TCP RST (reset) packet to the user, followed by 

advertising a TCP ZeroWindow size.

179

  This technique uses TCP's flow control feature to prevent the 

user's computer from transmitting additional data to the destination server (such as the Baidu.com search 

engine).  This disconnection persisted for prolonged periods despite multiple attempts to reconnect. 

We confirmed partially Reporters Sans Frontieres' claims that the search engines Baidu and 

Yisou, with which Google and Yahoo! have investment relationships, filter the Web content they return 

when users search for certain sensitive keywords.  However, this is only part of a set of complex, 

176

 See the enumeration report documenting our test results at 

http://www.opennetinitiative.net/bulletins/006/googlecacheservers mod.html. 

177

 See the enumeration report documenting our test results at 

http://www.opennetinitiative.net/bulletins/006/othersearchengines.html. 

178

 Reporters Sans Frontieres, Google   Yahoo Market Battle Threatens Freedom of Expression, at 

http://www.rsf.org/article.php3?id_article=11031 (July 26, 2004). 

179

 See OpenNet Initiative, Probing Chinese Search Engine Filtering, at 

http://www.opennetinitiative.net/bulletins/005/#res (Aug. 19, 2004); see generally Von Welch, A User's Guide to TCP 

Windows, at http://www.ncsa.uiuc.edu/People/vwelch/net_perf/tcp_windows.html (last updated June 19, 1996). 

49