If an nfs export of gfs does not set 'fsid' in /etc/exports on the server, the client mount fails with -ESTALE. eg: server: # cat /etc/exports /mnt/gfs <world>(rw,wdelay,no_root_squash,sync) client # mount server:/mnt/gfs /mnt/nfs mount: stale NFS file handle when I add fsid=1 to /etc/exports, i.e /mnt/gfs <world>(rw,wdelay,no_root_squash,fsid=1,sync) client mount succeeds. Should fsid be a mandatory param for NFS over GFS?
Yes, it needs to be. We can't create filehandles from device numbers like local filesystems can, so we have this system of having to specify it. It ought to be documented though, so if its missing from the docs, then please turn this into a doc bug. Potentially we might be able to create an ID by using the UUID or some other unique identifier in the future. The only problem being that we can never be sure that it will not be a duplicate of an existing one. The only really safe way is to get the user to specify it.