count (*not* node count) and exceeds the partition's node limit for a job.

Minor code clean-up in node_mgr.c
Change comment in slurmd/req.c

Minor code clean-up in node_mgr.c

   but semaphore doesn't.
 o shm.c : added new _shm_sane() call that attempts to ensure shared memory is
   not improperly locked on attach.
 o slurmd.c : move more initialization code into slurmd_init()
 o slurmd.c : _reconfigure() now calls _read_config()

to "*" suffix (e.g. "IDLE*").

The job would fail if submitted with "immediate" flag.

immediate flag when partition configuration prevents immediate initiation.

 o chdir to /tmp to avoid filling spooldir with corefiles (just in case)

 o reinitialize buffers on _log_init()

   attach

  log_set_argv0() to avoid reinitializing the log mutex.
  This avoids spurious debug messages on job stderr when default stderr
  logging is high (a rare but annoying bug)

 o Do not allow attach to tasks if tasks have exited

Added security_2_2.c