Don't try to look up and close outgoing connections to a given server
when aborting incoming connections from that server due to certificate
verification errors. The ejabberd_s2s:find_connection/2 call actually
created one or more *new* connections if less than 'max_s2s_connections'
connections were found. Then, no more than one of those possibly new
connections were stopped by the ejabberd_s2s_out:stop_connection/1 call.
It's not really necessary to bother with outgoing connections at all,
here.
Prior to this commit, ejabberd handled certificate authentication for
incoming s2s connections like this:
1. Verify the certificate without checking the host name. On failure,
behave according to 's2s_use_starttls'. On success:
2. Offer SASL EXTERNAL.
3. If the remote server chooses SASL EXTERNAL, compare the authorization
identity against the certificate host name(s). On failure, abort the
connection unconditionally.
ejabberd now does this instead:
1. Verify the certificate and compare the certificate host name(s)
against the 'from' attribute of the stream header. On failure,
behave according to 's2s_use_starttls'. On success:
2. Offer SASL EXTERNAL.
3. If the remote server chooses SASL EXTERNAL, ignore the authorization
identity (if any) and consider the peer authenticated.
The old behavior was suggested by previous versions of XEP-0178, the new
behavior is suggested by the current version 1.1.
As the version string is auto-generated from the git-describe(1) output,
the configure script may need to be regenerated even if configure.ac
wasn't modified.
Don't log a "configuration problem" message if "extauth_cache: false" is
explicitly specified, as that's a valid configuration setting as per the
documentation.
As the session manager handles messages sent to unavailable resources
just like messages sent to bare JIDs, mod_carboncopy must do that, too.
That is, forward them only to those carbon-copy-enabled resources that
don't have a top priority, in order to avoid duplicates.
