Now that the x11_forwarding code is running under seteuid/setegid,
and not using _become_user in the stepd, the environment is not set
automatically so xauth was trying to alter /root/.Xauthority instead
of the user's. (That restriction is in place so hostkey auth can work
safely.)

While here, move the xauth cleanup into the x11 forwarding shutdown
thread and out of the stepd's mgr.c.

Caught by -Werror=unused-result.

x11 flags come in the cred here, not in the msg.

For reasons that aren't quite clear, this causes quite a few failures
in the regression suite.

As the X11 race has been solved elsewhere, reverting this is the safest
option for now.

This reverts commit 34b67180332cfec6f4a72cc79c670bd73395c3b7.

This fixes a weird issue where new X11 applications will not be
able to start rendering on screen until an input even is forced
on the other X11 connections, e.g. by moving your mouse over them.

This has something to do with complications of the non-blocking I/O
in libssh2, but I'm at a loss to better describe why this make it
work correctly.

Shut down in a slightly nicer fashion by closing the listening
port and ssh connection properly.

The prolog is marked complete as soon as job_manager starts.
However, this is well before the extern step setup has actually
had time to complete, which leads to a race condition in the x11
code where the display has not been established.

Adding this locking keeps _handle_x11_display from returning
until the display number has been set.

Make this consistent with x11_remove_xauth.

Used to cleanup the magic cookie added earlier by x11_set_xauth.

Determine if this node has x11 forwarding enabled based
on the node index, and only setup forwarding or alter the DISPLAY
variable if appropriate.

The batch step for a job in the only batch step,
so simplify this info message. And switch to %u for uid.

Throw an error if enabled but the build had it disabled.

Reject jobs with X11 flags set if the support is not enabled,
and add an appropriate error code for this case.

- Do not re-launch the extern step if !first_run.
- Move before _remove_job_running_prolog() to make sure the
  extern step has at least a slight head start before steps can
  launch. Note that this only ensures the stepd has started, but
  not that the stepd is finished it's own internal setup - further
  work in slurmstepd would be required to remove this race completely.

Bug 4264.

Return an error if something doesn't work out.

Need to split this back out to have a chance to encapsulate the
extern step launch in _rpc_prolog as well.

This is a repeat of a block later on, at a point when
rc could potentialy be something other than SLURM_SUCCESS.

Ensures that slurmctld can centrally manage availability of the flag.

Of course, x11 forwarding needs to be setup as part of the prolog,
but this will avoid expensive calls to the extern stepd to lookup the
display if the job definitely does not have x11 forwarding enabled.

This commit is here to get the Makefile changes in place ahead
of the implementation. Only build these if the native X11
forwarding is enabled (which implies libssh2 is functional as well).

Coverity CID 44985

Coverity CID 44974

Coverity CID 44801

Coverity CID 44781

Bug 4138

Avoid using configured SallocDefaultCommand and failing test

Required for test to make sure the time limit is not adjusted by
  the backfill scheduler

Bug 4276