Merge branch 'maint-0.4.5' into release-0.4.5

changes/bug40515

+  o Minor bugfixes (controller, path bias):
+    - When a circuit's path is specified, in full or in part, from the
+      controller API, do not count that circuit towards our path-bias
+      calculations. (Doing so was incorrect, since we cannot tell whether
+      the controller is selecting relays randomly.)  Resolves a "Bug"
+      warning. Fixes bug 40515; bugfix on 0.2.4.10-alpha.

src/core/or/origin_circuit_st.h

@@ -182,6 +182,12 @@ struct origin_circuit_t {
    */
   unsigned first_hop_from_controller : 1;
 
+  /**
+   * If true, this circuit's path has been chosen, in full or in part,
+   * by the controller API, and it's okay to ignore checks that we'd
+   * usually do on the path as whole. */
+  unsigned int any_hop_from_controller : 1;
+
   /**
    * Tristate variable to guard against pathbias miscounting
    * due to circuit purpose transitions changing the decision

src/feature/client/circpathbias.c

@@ -363,6 +363,17 @@ pathbias_should_count(origin_circuit_t *circ)
     return 0;
   }
 
+  /* Ignore circuits where the controller helped choose the path.  When
+   * this happens, we can't be sure whether the path was chosen randomly
+   * or not. */
+  if (circ->any_hop_from_controller) {
+    /* (In this case, we _don't_ check to see if shouldcount is changing,
+     * since it's possible that an already-created circuit later gets extended
+     * by the controller. */
+    circ->pathbias_shouldcount = PATHBIAS_SHOULDCOUNT_IGNORED;
+    return 0;
+  }
+
   /* Completely ignore one hop circuits */
   if (circ->build_state->onehop_tunnel ||
       circ->build_state->desired_path_len == 1) {

src/feature/control/control_cmd.c

@@ -822,6 +822,8 @@ handle_control_extendcircuit(control_connection_t *conn,
     circ->first_hop_from_controller = 1;
   }
 
+  circ->any_hop_from_controller = 1;
+
   /* now circ refers to something that is ready to be extended */
   first_node = zero_circ;
   SMARTLIST_FOREACH(nodes, const node_t *, node,