On Mon, Jun 22, 2020 at 03:01:55PM +0200, Oleg Nesterov wrote:
> On 06/22, Christian Brauner wrote:
> >
> > It is a supported case however unlikely. I just tried to answer
> > Dominique's specific question pointing out that even in that unlikely
> > case sighand_struct is stable.
>
> I too tried to say this, but apparently just added more confusion ;)
>
> > Just as an fyi, CLONE_SIGHAND with CLONE_VM but without CLONE_THREAD is
> > actually used quite a bit, e.g. in newlib, in stress-ng, and in criu.
>
> OK,
>
> > you'd want CLONE_VM which enforces
> > CLONE_SIGHAND so that would be another use-case afaict.
>
> Cough no ;) CLONE_SIGHAND requires CLONE_VM, not vice versa.
Oh, you're right. I was thinking of ksys_unshare() here. :)
/*
* If unsharing vm, must also unshare signal handlers.
*/
if (unshare_flags & CLONE_VM)
unshare_flags |= CLONE_SIGHAND;
Thanks!
Christian
