Current situation: nfsd_open() calls break_lease() but break_lease has only inode pointer, indication of requested access--no notion of owner identity.
Requirements:
Possible solutions:
NOTE: obstacle to write deleg implementation currently: initial checks for d_count and i_count prevent granting *any* write leases at this point.... FIX!
If we continue to count on d_count and i_count to identify other writers, then we also need to coalesce nfsd's readers and writers; currently it keeps a struct file per stateid.
If we kept a single open file for the entire server, would we prevent the filesystem/security subsystem from performing appropriate security checks at open time? *Must* get these sorts of questions about file scope answered:
Question to research: is current self-conflicting semantics correct for file leases?