Ceph » CephFS

Client::_unmount will unmap and tear down all of the open Fh objects before returning. Programs that use the lowlevel interfaces however may hold pointers to these objects, and can issue calls against them. In particular, we may end up calling ceph_ll_close on the Fh after the unmount has returned, which can lead to doing a double put on the object.

Fix ceph_ll_close to be a no-op when the cmount is unmounted. We may need to do similar treatment for other calls as well, but that should at least fix the immediate problem I've seen in ganesha.