The mapping between the CORBA object and the related object you must implement is done through the use of C structures (all named with a prefix of Gda_SRV_). These structures contain several members that are needed for this purpose, so, as said on the generated source code files, you can add any needed member to these structures but respecting the given order and without removing any of the predefined fields. (All this mapping is done in the gda-srv.[ch] generated files)
The thing is as simple as that: the functions you implement are called with some parameters (Gda_SRV_* objects and other stuff). You perform any needed operation and fill in the structures' fields following some directives. Then, the "real" CORBA implementation converts this data to any needed format and send it back to the CORBA client (in this case, probably an application using the GDA client library).