There are many reasons why a connection between server and client may get disconnected.

Green:
1.  admin does 'umount'

Red:
2. n/w problem
3. client segfault

We should have a clear way to differentiate 'Green' and 'Red' reasons for client disconnect.

Best way to implement it is through 'notify()' framework, as notify() is cluster aware, and a given notify event gets passed on to all the translators in the graph by default.

Implement a new mop, which can pass on just these notify 'events' to server protocol from client protocol.

This is considered critical as its needed to understand the health of a client machine (which may be outside glusterfs management nodes) from server nodes.