(In reply to Luis Chamberlain from comment #23) > Also, if the culprit is libtirpc function __rpc_dtbsize() shouldn't we be > able to proof of concept this easily? And wouldn't this issue be more > visible and easier to reproduce elsewhere? See previous comments, it's due to the fact __rpc_dtbsize() is called with rlimits inherited from PID1. The rlimits of started service are initialized *after* the call to initgroups(3).