The code used to call g_main_loop_quit() from the main thread, without
having any guarantee that runaway_killer_thread_func() has even entered
its g_main_loop_run().  If a main loop is not running,
g_main_loop_quit() has no effect.

This could occasionally be reproduced in
test-polkitbackendjsauthority.c, which is creating several very
short-lived PolkitBackendJSAuthority instances.  Real polkitd should not
generally be affected, because it is using a single instance running for
the life of the process ~ for the uptime of the system, enough time to
enter the runaway_killer_thread main loop.

To fix this, use g_idle_source_new () to make sure g_main_loop_quit ()
is called from within the running main loop.

Also, simplify the initialization of runaway_killer_thread by moving the
creation of rkt_context and rkt_loop into the main thread; this makes
the condition variable and its associated mutex completely unnecessary.

Finally, only destroy rkt_timeout_pending_mutex _after_ the thread
terminates; before, we were certain that rkt_source was destroyed by
that time, but AFAICS that does not ensure that the rkt_on_timeout ()
callback has already terminated.

https://bugs.freedesktop.org/show_bug.cgi?id=95513



Signed-off-by: Miloslav Trmač <mitr@redhat.com>