For now, I want to try and resolve the issues with net\_ssl\_test, and run more tests. However, in the meantime, I thought I would document the issue here.
-- Brian May
+
+> Openid's security model does not rely on the openid consumer (ie,
+> ikiwiki) performing any sanity checking of the openid server. All the
+> security authentication goes on between your web browser and the openid
+> server. This may involve ssl, or not.
+>
+>> Note that I'm not an openid expert, and the above may need to be taken
+>> with a grain of salt. I also can make no general statements about openid
+>> being secure. ;-) --[[Joey]]
+>
+> For example, my openid is "http://joey.kitenet.net/". If I log in with
+> this openid, ikiwiki connects to that http url to determine what openid
+> server it uses, and then redirects my browser to the server
+> (https://www.myopenid.com/server), which validates the user and redirects
+> the browser back to ikiwiki with a flag set indicating that the openid
+> was validated. At no point does ikiwiki need to verify that the https url
+> is good.
+> --[[Joey]]
+
+>> Ok, so I guess the worst that could happen when ikiwiki talks to the http
+>> address is that it gets intercepted, and ikiwiki gets the wrong address.
+>> ikiwiki will then redirect the browser to the wrong address. An attacker could
+>> trick ikiwiki to redirect to their site which always validates the user
+>> and then redirects back to ikiwiki. The legitimate user may not even notice.
+>> That doesn't so seem secure to me...
+
+>> All the attacker needs is access to the network somewhere between ikiwiki
+>> and http://joey.kitenet.net/ or the ability to inject false DNS host names
+>> for use by ikiwiki and the rest is simple.
+
+>> -- Brian May
+
+>>> I guess that the place to add SSL cert checking would be in either
+>>> [[!cpan LWPx::ParanoidAgent]] or [[!cpan Net::OpenID::Consumer]]. Adding
+>>> it to ikiwiki itself, which is just a user of those libraries, doesn't
+>>> seem right.
+>>>
+>>> It's not particularly clear to me how a SSL cert can usefully be
+>>> checked at this level, where there is no way to do anything but
+>>> succeed, or fail; and where the extent of the check that can be done is
+>>> that the SSL cert is issued by a trusted party and matches the domain name
+>>> of the site being connected to. I also don't personally think that SSL
+>>> certs are the right fix for DNS poisoning issues. --[[Joey]]
+
+I was a bit vague myself on the details on openid. So I looked up the standard.
+I was surprised to note that they have already considered these issues, in
+section 15.1.2, <http://openid.net/specs/openid-authentication-2_0.html#anchor41>.
+
+It says:
+
+"Using SSL with certificates signed by a trusted authority prevents these kinds of
+attacks by verifying the results of the DNS look-up against the certificate. Once
+the validity of the certificate has been established, tampering is not possible.
+Impersonating an SSL server requires forging or stealing a certificate, which is
+significantly harder than the network based attacks."
+
+With regards to implementation, I am surprised that the libraries don't seem to
+do this checking, already, and by default. Unfortunately, I am not sure how to test
+this adequately, see [[!debbug 466055]]. -- Brian May
+
+---
+
+I think [[!cpan Crypt::SSLeay]] already supports checking the certificate. The trick
+is to get [[!cpan LWP::UserAgent]], which is used by [[!cpan LWPx::ParanoidAgent]] to
+enable this checking.
+
+I think the trick is to set one of the the following environment variables before retrieving
+the data:
+
+$ENV{HTTPS\_CA\_DIR} = "/etc/ssl/certs/";
+$ENV{HTTPS\_CA\_FILE} = "/etc/ssl/certs/file.pem";
+
+Unfortunately I get weird results if the certificate verification fails, see [[!debbug 503440]].
+It still seems to work though, regardless.
+
+-- Brian May