The NFS lock routines (nlmsvc_lock, nlmsvc_testlock, and maybe others)
are being called with file_lock structures that contain uninitialized
list_head structs. (this as of 2.4.0-test7).

The "next" links should be circular chains, but are instead NULL.
This causes problems later when they are used.  I don't know where they
should be initialized, but a INIT_LIST_HEAD() should be done when these
file_lock structs are created.  For example, fl_wait.task_list not
being set shows itself when an NFS SETLK is followed by a GETLK.
The result is a dereference of the NULL pointer in locks_free_lock
when it tries the list_empty().

I opened another defect (#24) on a similar problem where
nlmsvc_lock calls nlmsvc_create_block to create a nlm_block struct
and then failed to init these circular chains.  The result was that
SETLKW operations resulted in the NULL pointer being used if the
request blocked.  This, however, now appears to be a much more
prevelant problem that is happening more than reported in defect #24.