> Amos Jeffries wrote: >>> So of course the problem is proventia corrupting the HTTP headers and >>> we will raise an issue about that with IBM. >>> >>> But for the time being: is there a chance to make squid more >>> "tolerant" about those kind of problems? Without surprize I did not >>> find any fitting config options :-) >>> >> Not nearly as easy as it will be for IBM to issue a fix for it. Or even >> to replace the box with free software that works well. > > Hehe, sure, no objections, it is just the world being far from perfect :-) > >> Not also without opening some potential data-injection and cache >> poisoning flaws into Squid. >> >> Consider what happens with: >> >> HTTP/1.1 200 OK >> Bwahaha: " >> Cache-Control: private >> >> ...something you really did not want public... >> . >> >> vs: >> >> HTTP/1.1 200 OK >> Content-Type: "fu >> bar: tender: and: wine" >> Cache-Control: private > > Hmm, reading the specs for HTTP message headers [1] I think this could > be done without imposing security issues. As per specification your last > example would read correctly: > > -------CUT------- > HTTP/1.1 200 OK > Content-Type: "fu > bar: tender: and: wine" > Cache-Control: private > -------CUT------- > > note the leading whitespace.
I know. Both of what I pointed are two cases of the same brokenness. Squid handles the one is can interpret 'safely'[1] (second), but at expense of dropping the first set when no termination can be found at all and its clearly very unsafe to make assumptions. [1] for some vague value of safe the paranoid in me gets very edgy about as-is. Amos
