Also fix a wrong type use in sizeof calculation fed to xmalloc.

Also fix a wrong type use in sizeof calculation fed to xmalloc.

Also fix a wrong type use in sizeof calculation fed to xmalloc.

Also fix a wrong type use in sizeof calculation fed to xmalloc.

Also fix a wrong type use in sizeof calculation fed to xmalloc.

From the comment block in checkpoint_poe.c:

  This is based upon checkpoint support of poe in the 2005 time frame for
  the ASCI Purple computer. It does not work with current versions of POE.
  From Gary Mincher (IBM, Sept 6 2012): "Checkpoint/restart on Linux is only
  supported for user-space parallel jobs with a maximum of 512 tasks that are
  run on Power 775 nodes, but jobs that use a resource manager or scheduler
  other than LoadLeveler are not supported."
N

Also change signature of io_thread_start() and cleanup the
error handling in the one calling location.

All are detached threads, both the slurm_attr_init and
pthread_attr_setdetachstate calls are in _cancel_jobs
and apply throughout.

Which fixes yet another missing slurm_attr_destroy.

Change function signature as the only error path out of here
was if the thread creation failed. Cleanup code in fed_mgr.c
trying to manage that event. (The one other calling path just
ignored the error anyways.)

And another missing slurm_attr_destroy call.

Interestingly enough there was a missing slurm_attr_destroy() call here.

While here fix two function signatures, as the only error
path was if pthread_create failed, and the calling locations
ignored the return code anyways.

Bug 4163.

Set to fatal on pthread_create failure, as this is the LCD of all
existing behaviors, and as such if something were stopping thread
creation from succeeding you could expect some code path to trip
that eventually. Rather than gamble on when/if that is triggered,
just immediately die.

Bug 4163.

CID 44819.

A second PMI2_Init() within the same step is invalid, and cannot succeed.

Return an error code back to the client end, and close the fd to force the
step to terminate immediately.

Due to a bug in our libpmi code, just returning a cmd=response_to_init with
an appropriate rc number will not tear down the connection properly, so
send back something else that will trigger the error path.

Bug 3520.

Need to make a local copy of the user_name field, otherwise
you'll xfree() the cache's version and eventually crash or
fatal() slurmctld.