+ *
+ * Callers should not inspect "errno" on failure, but rather pass in a
+ * "failure_errno" parameter, on failure the "errno" will indicate the
+ * type of failure encountered, but not necessarily one that came from
+ * a syscall. We might have faked it up.