rhafer@suse.de wrote:
Am Freitag 18 Februar 2011, 13:17:17 schrieb andrew.findlay@skills-1st.co.uk:
On Fri, Feb 18, 2011 at 12:30:25AM +0000, hyc@symas.com wrote:
Used to work - since when, what release, what else has changed since then?
Unfortunately I cannot tell you exactly when this changed. In any case, the change only affects a different bug which was masking the problem that I now see.
I do know that 2.3.32 as shipped with SLES 10.3 masks the problem by not checking the server certificate properly. So does 2.4.12 as shipped with OpenSuSE 11.1. Both will allow ldapsearch -ZZ to connect to *any* TLS-capable server if they do *not* have access to the CA certificate.
2.4.24 built on OpenSuSE 11.3 (i.e. using OpenSSL 1.0) correctly refuses to connect if there is no CA cert.
Yes, older openSUSE releases used a bad (i.e. less secure) default for the TLS_REQCERT setting. That's why you never ran into the problem before :/
Sigh... Apparently nobody reads the part of the doc that says "there is no good reason to change this from the libldap default."
All versions that I have tested (certainly back to 2.3.32) incorrectly fail to connect when the URL is ldap://localhost:1389/ and a CA cert is provided.
Yes, AFAIK libldap's behavior of trying to figure out the machines real hostname when connecting to localhost and using that hostname to verify the server's certificate is there since quite some time already. It will only verify against "localhost" when it completely fails to get the machines' real name. I always considered that a feature more than a bug.
Agreed. Server certs are supposed to carry the host's FQDN. "localhost" is just a magic alias; in a cert it shouldn't be anywhere other than a subjectAltName.
I'll note that I just tested some localhost certs a few days ago and
they were
fine, and the cert verification code hasn't changed in quite a long
time.
(E.g., ITS#6711 the test setup there uses localhost with no problem.)
But if I see it correctly it sets "TLSVerifyClient allow" on the master and "tls_reqcert=never" on the consumer. So it doesn't really verify the certificates. (I might have overlooked something, took only a brief glance at it).
Ah, that sounds familiar.
Closing this ITS. It's not a regression and not a bug, just a long established feature.