Herwig:

First, lets correct one thing. You are not weird. I think you meant to say that at this time, you are confronted with a weird situation. But dont feel bad because Ive been weird my whole life! :-)

Now, for the serious business:

For R1 to be able to establish a BGP multihop sesssion with R4, it must have a route to the address that R1 is trying to peer with on R4. R4 is R1's BGP "next-hop," and vise versa. Also, R2 and R3 must be able to forward traffic back and forth between those neighbor addresses. This way, R1 and R4 will be able to exchange BGP information and forward user traffic to each other.

Lastly, R2 and R3 would have to have a default pointing to R4 and they would have to have routes for the inside networks, as well. So, you are right.

By the way, the situation you describe is one that shows the importance of synchronization, where the BGP table must be in synch with the IGP RIB. This is especially true if R4 is going to pass traffic fron another AS to R1, and then R1 is going to pass traffic to a third AS. In other words, R1 and R4 are part of a transient AS.

HTH

Victor