- From: Poul-Henning Kamp <phk@phk.freebsd.dk>
- Date: Tue, 15 Jul 2014 11:40:29 +0000
- To: Mark Nottingham <mnot@mnot.net>
- cc: HTTP Working Group <ietf-http-wg@w3.org>
In message <A5DB7CA0-D6DC-4E21-911A-A9FDEAA8CC52@mnot.net>, Mark Nottingham wri tes: >> A new version of I-D, draft-nottingham-http-proxy-problem-01.txt In 2.4 I think you should mention that content filtering is explicitly mandated by law in various jurisdictions. That doesn't make it any less controversial of course, quite the contrary, but mentioning it would make it clear that there some level of support will happen, no matter what. Under 5.4 you should probably add "... unless made illegal by a legitimate authority" to clarify the extent of our reach. I think it is unwise to lump header-modifications with "2.4 Content Modifications". Modifying an Accept-Bla header, in order to make filtering easier, is not nearly as controversial as tacking advertising onto webpages. Otherwise it's a good summary of the situation. I would like to propose a section 6.8: 6.8 content-privacy with (some) visible meta-data A lot of filtering proxies care only about restricting access to particular sites or to particular paths under those sites. Today they handle privacy protected access one of two ways. Simpler proxies will resolve the CONNECT argument and compare it to a white- or black-list of IP numbers, and allow or deny access whole-sale to filtered IP numbers. This causes rather dramatic over-reach when one user on a web portal posts content which causes the entire portal to be blacklisted. More comprehensive proxies deploy man-in-the-middle certificates to its clients (via out-of-band methods) and demasks all privacy protected transactions in order to be able to inspect them, gaining as a side effect also the ability to modify them. Very little actual privacy would be lost if a minimum of meta-data, specifically the URL less the query part, were transmitted in the clear. The IP numbers are already in the clear with TLS, and giving the quality of traffic fingerprinting already available, expanding this visibility to reveal the FQDN implies a very small loss of actual privacy. Revealing also the path component of the URL, but not the query part, likewise would amount to very little actual loss of privacy, but would allow filtering proxies to ban only example.com/user/evil rather than all of example.com. The proposed distinction between "envelope" and "contents" has been recognized in legislation for centuries and would therefore provide a well-known and -understood tool for tailoring legislation more narrowly, than the currently very broad strokes used with respect to privacy on the internet. (The proposed scheme would also have other benefits outside the scope of this document, mainly for incoming load-balancers and reverse proxies on the server side.) -- Poul-Henning Kamp | UNIX since Zilog Zeus 3.20 phk@FreeBSD.ORG | TCP/IP since RFC 956 FreeBSD committer | BSD since 4.3-tahoe Never attribute to malice what can adequately be explained by incompetence.
Received on Tuesday, 15 July 2014 11:40:53 UTC