RP & GFS

Assuming you are talking about Red Hat GFS, and RecoverPoint/SE, /EX or /CL.  RecoverPoint is a block-level solution.  Data is replicated by replicating any writes to the protected luns.  The writes are split down multiple paths - one path goes to the RecoverPoint appliance which handles the replication of the writes, and the other path goes to the target LUN. 

Those splits happen at the array level, assuming you are using a recent version of RecoverPoint. When any of the hosts in the cluster send a write I/O to the array to a protected LUN, a splitter in the array handles the work of sending the I/O to the RP appliance.  It doesn't care which host is performing the write - it's just yet another initiator sending a write to the target.

What is replicated is a "crash consistent copy."  You create a consistency group within RecoverPoint that contains all the target luns you want kept consistent within themselves.  RecoverPoint ensures that all writes to the LUN's in the consistency group maintain their write order.  Thus, although your copy isn't necessary synchronized with an application or filesystem checkpoint, it's is consistent in the same way they would be consistent after a crash of the array, the SAN, etc.

The OS of the host that is attempting to use the replicated copy will perform normal filesystem cleanup that occurs after a crash, then mark the filesystem clean.