Re: [PATCH 1/4] ns: prepare time namespace for clone3()

From: Christian Brauner
Date: Wed Mar 18 2020 - 07:57:45 EST


On Wed, Mar 18, 2020 at 02:28:14PM +0300, Cyrill Gorcunov wrote:
> On Wed, Mar 18, 2020 at 12:17:26PM +0100, Christian Brauner wrote:
> > On Wed, Mar 18, 2020 at 01:57:47PM +0300, Cyrill Gorcunov wrote:
> > > On Tue, Mar 17, 2020 at 09:30:41AM +0100, Adrian Reber wrote:
> > > ...
> > > > +/*
> > > > + * This structure is used to set the time namespace offset
> > > > + * via /proc as well as via clone3().
> > > > + */
> > > > +struct set_timens_offset {
> > > > + int clockid;
> > > > + struct timespec64 val;
> > > > +};
> > > > +
> > >
> > > I'm sorry, I didn't follow this series much so can't comment right now.
> > > Still this structure made me a bit wonder -- the @val seems to be 8 byte
> > > aligned and I guess we have a useless pad between these members. If so
> > > can we swap them? Or it is already part of api?
> >
> > It's not part of the api yet. We're still arguing about how and what we
> > want to pass down.
>
> I see, thanks for info! So I think we could swap the members.

Yeah, _if_ we go with a version of this struct we definitely need to
ensure proper padding/alignment!