s3:winbind: Only ever handle one event after a select call

While handling an fd event, the situation with other fds can change. I've just
seen a winbind stuck in the accept() call on the privileged pipe. I can only
imagine this happen because under high load we first handled other requests and
meanwhile the client on the privileged pipe went away.

diff --git a/source/winbindd/winbindd.c b/source/winbindd/winbindd.c
index 9005e6754d911357ab91e7025db547876bf6ba09..1d618e263174d4ea365b82c96e8eafea1e817a8f 100644 (file)
--- a/source/winbindd/winbindd.c
+++ b/source/winbindd/winbindd.c
@@ -847,7 +847,9 @@ static void process_loop(void)
 
        message_dispatch(winbind_messaging_context());
 
-       run_events(winbind_event_context(), 0, NULL, NULL);
+       if (run_events(winbind_event_context(), 0, NULL, NULL)) {
+               return;
+       }
 
        /* refresh the trusted domain cache */
 
@@ -927,8 +929,10 @@ static void process_loop(void)
                        flags |= EVENT_FD_READ;
                if (FD_ISSET(ev->fd, &w_fds))
                        flags |= EVENT_FD_WRITE;
-               if (flags)
+               if (flags) {
                        ev->handler(ev, flags);
+                       return;
+               }
                ev = next;
        }