Minor Description
The current versions of both the Java broker and the C++ broker support EXTERNAL authentication mechanism / SSL Client Authentication. Nevertheless the implementations are not fully identical. One of the differences is the way the usernames are constructed for the authenticated connections.
The C++ broker takes the CN of the certificate (+ eventually the domain components / DCs) and creates a username as <CN>@<DC1>.<DC2>.<DC3>....<DCN>. In case there are no DCs, only the CN is used. On the other hand, the Java broker takes the full distinguished name as the username.
Example 1:
DN: CN=person
C++ username: person
Java username: CN=person
Example 2:
DN: CN=person,DC=example,DC=com
C++ username: person@example.com
Java username: CN=person,DC=example,DC=com
Example 3:
DN: CN=person,DC=example,DC=com,O=My Company Ltd,L=Newbury,ST=Berkshire,C=GB
C++ username: person@example.com
Java username: CN=person,DC=example,DC=com,O=My Company Ltd,L=Newbury,ST=Berkshire,C=GB
This difference between C++ and Java broker makes it more complicated to migrate between the different brokers. Also, the particular implementation in Java can make it a bit more complicated to use the SSL Client Authentication in some cases.
Therefore I implemented an enhancement which adds a possibility to the Java broker to construct the usernames in the same way the C++ broker does. I added a new configuration attribute "useCNAsUsername" for the ExternalAuthenticationProvider / Manager which allows switching between the current DN based usernames and the CN based usernames. By default, the old behaviour is used. The setting is passed from the ExternalAuthenticationManager to the ExternalSaslServer, which (in case the useCNAsUsername is switched ON) constructs the new username and returns it using the UsernamePrinciple. I also added some unit tests to cover the new functionality.
It would be great if someone can have a look at the attached patch and eventually provide me with some review comments. One of the areas where I'm not entirely sure about the patch is the way it handles the configuration parameter from the JSON configuration. It is also no entirely clear to me what is the strategy for naming the parameters - it seems to be kind of a mixture of short names, long names, lower-case only, camel-case etc.
I didn't added any new systests to cover this feature - I have not really an idea how to implement them. Especially how to get the authenticated username once connected. If you point me in some direction, I can have a look at it.