cxf-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Polar Humenn <>
Subject Re: HttpConduitTest failed when Jetty upgraded to 6.1.3
Date Fri, 18 May 2007 14:27:56 GMT

I can take care of upgrading to Jetty 6.1.3. But I want to do a couple 
of things along the way.

I want to fix the SSL Configuration, (is anybody adverse to this?) and I 
also want to be able to programatically configure destination, which is 
presently impossible for Jetty SSL Destination (I filed JIRA CXF-661).

Fred and I also want to be able to configure these with Keystore or 
KeyManager objects without having to indirectly specify it through file 
name, as currently it must be done. For Spring config keystore or 
keymanagers may be supplied with beans.

There is also a point I'd like to do. Jetty has been factored out to 
separate it from the client side, but the server stuff is still 
dependent on the client libraries. I'd like to be able to separate them 
and make them independent.

Does anybody have objections to this?


Willem Jiang wrote:
> Hi Polar,
> Please see my comments in the mail.
> Polar Humenn wrote:
>> Aside from the incompatibilities with a new version of Jetty, CXF SSL 
>> configuration has a lot of shortcomings. I've been burned by the same 
>> thing.
>> First of all, as Eoghan might say, I tore my hair out, before I found 
>> out that if your authentication certificate/key is in a PKCS12 
>> format, then your TrustStore has to be a PEM certificate, not a 
>> truststore, i.e. PCKS12 or Java Key Store (JKS) format. This must 
>> have made sense to someone about not mixing PCKS12 with JKS or 
>> others, but makes no logical sense to me. So, Willem, if the 
>> truststore is *just* a certificate, then there is no password 
>> possible. Furthermore, when you use this approach you are only 
>> allowed to have *1* certificate in the truststore.
> I don't mean not to set the TrustStroePassword to be null.
> Current Jetty SslSocketConnector don't support to set the 
> TrustStroePassword to be null.
> Now I don't find a way to walk around it. It just may take some time 
> to convince Jetty guys to change the code.
>> So, the basic upshot is that if you initialize a PCKS12 
>> key/certificate, then you are only allowed to have only 1 trust 
>> point.  Most browsers have something like 100.
> We can support the TrustStroePassword with null value in CXF , right.
>> So, this approach is really unacceptable. I would like to revamp SSL 
>> configuration and the whole approach to include:
>> 1. Different types Keystores for the the authentication private 
>> keys/certificates/chains.
>> 2. Different types of password protected TrustStores with multiple 
>> trust points.
>>     2a. I don't really care about passwords on TrustStores, as they 
>> are mainly for
>>           for integrity protection (somebody doesn't slip on in 
>> there), but the user
>>           should have the option of forgoing that, say, just a 
>> concatenated list of
>>           of PEM certificates.
>> 3. Remove the restrictions between the keystore type and the 
>> truststore type.
>> Thinking quickly, this will add two elements to the SSL Policy 
>> configuration for both Client and Server sides.
>> <TrustStoreType>
>> <TrustStorePassword>
> Let's add the TrueStorePassword to the SSL Policy configuration. :)
>> Cheers,
>> -Polar
>> Willem Jiang wrote:
>>> Hi
>>> I found the HttpConduitTest failed in Systest when I upgraded the 
>>> Jetty version from 6.1.2rc0 to 6.1.3.
>>> I checked the Jetty's SslSocketConnector change log, and found that 
>>> the errors are related with the different trustManager
>>> setting on the Server and Client side. In another words,CXF now does 
>>> not support to load the cert with password.
>>> Current CXF JettySslConnectorFactory doesn't do any trustManager 
>>> setting, and the jetty will set the trustManagers to null,
>>> if there is no setting for the _truststore.
>>> But after Jetty 6.1.2rc5, the TrustManager will be initiated whether 
>>> you do the trustManager setting or not.
>>> [*Server side*]
>>> Here is the Jetty SslSocketConnector TrustManagers Code, the 
>>> trustStore load the  with a _trustPassword which can't be null.
>>> >>> after 6.1.2rc5
>>>        if (_truststore==null)
>>>        {
>>>            _truststore=_keystore;
>>>            _truststoreType=_keystoreType;
>>>        }
>>> >>>>
>>>       ......
>>>       TrustManager[] trustManagers = null;
>>>       if (_truststore != null)
>>>        {
>>>            KeyStore trustStore = KeyStore.getInstance(_truststoreType);
>>> trustStore.load(Resource.newResource(_truststore).getInputStream(), 
>>> _trustPassword.toString().toCharArray());
>>>                      TrustManagerFactory trustManagerFactory = 
>>> TrustManagerFactory.getInstance(_sslTrustManagerFactoryAlgorithm);
>>>            trustManagerFactory.init(trustStore);
>>>            trustManagers = trustManagerFactory.getTrustManagers();
>>>        }
>>> [*Client side*]
>>> CXF SSLUtil is responsible for the creation of  the TrustManager,  
>>> but it just load the cert with null password.
>>> protected static TrustManager[] getTrustStoreManagers( ...
>>>           KeyStore trustedCertStore = 
>>> KeyStore.getInstance(trustStoreType);
>>>  ......               trustedCertStore.load(new 
>>> FileInputStream(trustStoreLocation), null);
>>>  ......
>>> I went through The SSLClientPolicy and SSLServerPolicy , there is no 
>>> attribute for the TrustStorePassword.
>>> I also check the KeyStore.loadload(InputStream stream, char[] 
>>> password) API
>>> *the password used to check the integrity of  the keystore, the 
>>> password used to unlock the keystore,  or <code>null</code> *
>>> This issue can be solved from two side.
>>> One is let Jetty SslSocketConnector support calling the 
>>> trustStore.load with the password to be null.
>>> The other is we still need CXF SSL{Client|Server}Policy support 
>>> TrustStorePassword attribute.
>>> IMO, we need to add the TrustStorePassword attribute to the 
>>> SSL{Client|Server}Policy.
>>> Any thoughts?
>>> Cheers,
>>> Willem.
> Cheers,
> Willem.

View raw message