Common Vulnerabilities and Exposures assigned an identifier CVE-2009-0801 to the following vulnerability: Name: CVE-2009-0801 URL: http://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2009-0801 Assigned: 20090304 Reference: CERT-VN:VU#435052 Reference: URL: http://www.kb.cert.org/vuls/id/435052 Reference: BID:33858 Reference: URL: http://www.securityfocus.com/bid/33858 Squid, when transparent interception mode is enabled, uses the HTTP Host header to determine the remote endpoint, which allows remote attackers to bypass access controls for Flash, Java, Silverlight, and probably other technologies, and possibly communicate with restricted intranet sites, via a crafted web page that causes a client to send HTTP requests with a modified Host header.
Some additional references: http://www.theregister.co.uk/2009/02/23/serious_proxy_server_flaw/ http://marc.info/?l=squid-dev&m=123542836103750&w=4 And some additional notes: This seems to have low impact due to a user having to visit a malicious untrusted site to load active content, or to load malicious active content from a trusted site. It only affects proxies in transparent mode and same-origin policies in browsers should prevent an attacker from re-using auth credentials like cookies, and does NOT affect proxies in reverse mode (active content being flash or javascript). As well, this should only be able to affect regular HTTP connections since HTTPS does not use Host headers. There are some good comments on theregister (#8 and #9) that discuss why this may not be overly serious in real-world use. Also, there is nothing yet from the squid devs.
As in most matters there is a tradeoff between efficienty, simplicity and "security" in this matter. Imho (as an individual) the main security flaw is allowing untrusted "applets" direct network access bypassing the HTTP stack of the browser they run within. Having a browser plugin implement their own HTTP stack is...
From what I can see, there is still no upstream fix for this.
This is being tracked upstream here: http://bugs.squid-cache.org/show_bug.cgi?id=3243
Created attachment 515548 [details] Upstream patch
Statement: (none)
Upstream has corrected this in 3.2.0.11: http://www.squid-cache.org/Advisories/SQUID-2011_1.txt Regarding older squid versions, the advisory states: "Due to the design of request handling processes in older Squid patches and fixed releases for older versions of Squid are not being provided at this time."
Upstream bug also indicates that the fix is not backportable to pre-3.2 versions, as it depends on other internal changes done during the 3.2 development. SQUID-2011:1 is less clear, as it says upstream does not provide any backports to older version as this time, not saying if it's planned for the future. As for patch in comment #6, it seems incomplete and missing all the hostHeaderVerify() code. Henrik, do you know of any good place that describes what checks exactly are now performed by Squid? Based on my reading so far, it seems the core is the check of client TCP connection destination IP / port against HTTP header info for consistency. Though I don't think that info is always available to Squid (e.g. when connections are dnat-ed to proxy from gateway host). Any pointer or corrections are appreciated.
(In reply to comment #6) > Created attachment 515548 [details] > Upstream patch This seems to be: http://bazaar.launchpad.net/~squid/squid/3.2/revision/11253 While these are also part of the fix: http://bazaar.launchpad.net/~squid/squid/3.2/revision/11252 http://bazaar.launchpad.net/~squid/squid/3.2/revision/11266 http://bazaar.launchpad.net/~squid/squid/3.2/revision/11267 Release notes change with some brief info: http://bazaar.launchpad.net/~squid/squid/3.2/revision/11302
I have not kept a close eye on these changs. If you need some clarification I think the best is to post your question to squid-dev. As you correctly note this won't work at all if you DNAT externally to the Squid box as you then overwrite the original destination IP:port irrecoverably. Such setups is considered broken and have always been. Better to use policy routing and do the IP interception on the Squid box preserving data properly (i.e. WCCP, route maps, policy routes etc on the gateway..)
Fedora 16+ have 3.2.0.16, which contains the fix.