Clean up VMM resource on server shutdown

After the integration of #639, the CI tests report lingering VMM
resources after conclusion of the run.

Author: pfmooney

bin/propolis-server/src/lib/vm/mod.rs

@@ -872,12 +872,15 @@ impl VmController {
 impl Drop for VmController {
     fn drop(&mut self) {
         info!(self.log, "Dropping VM controller");
-        let instance = self
+        let machine = self
             .vm_objects
             .machine
             .take()
             .expect("VM controller should have an instance at drop");
-        drop(instance);
+
+        // Destroy the underlying kernel VMM resource
+        let hdl = machine.destroy();
+        let _ = hdl.destroy();
 
         // A fully-initialized controller is kept alive in part by its worker
         // thread, which owns the sender side of the controller's state-change

lib/propolis/src/vmm/machine.rs

@@ -96,7 +96,12 @@ impl Machine {
 }
 impl Drop for Machine {
     fn drop(&mut self) {
-        let _ = self.do_destroy();
+        // If consumer has not opted to own the fate of the contained VmmHdl
+        // through an explicit call to Machine::destroy(), then clean up the
+        // kernel VMM resource backing it automatically.
+        if let Some(hdl) = self.do_destroy() {
+            let _ = hdl.destroy();
+        }
     }
 }