On Mon, Jul 11, 2022 at 06:06:21PM -0500, Eric W. Biederman wrote: > Tycho Andersen <tycho@tycho.pizza> writes: > It is not different enough to change the semantics. What I am aiming > for is having a dedicated flag indicating a task will exit, that > fatal_signal_pending can check. And I intend to make that flag one way > so that once it is set it will never be cleared. Ok - how far out is that? I'd like to try to convince Miklos to land the fuse part of this fix now, but without the "look at shared signals too" patch, that fix is useless. I'm not married to my patch, but I would like to get this fixed somehow soon. > The other thing I have played with that might be relevant was removing > the explicit wait in zap_pid_ns_processes and simply not allowing wait > to reap the pid namespace init until all it's children had been reaped. > Essentially how we deal with the thread group leader for ordinary > processes. Does that sound like it might help in the fuse case? No, the problem is that the wait code doesn't know to look in the right place, so waiting later still won't help. Tycho
