b=20989 Don't search for inode in nfs get_dentry

Getting inode separately without some sort of locking enabled a situation where
several threads tried to populate same inode with different lsm data at the same
time leading to lsm pointer change (and leak).

i=johann
i=bzzz