Re: [PATCH 0/4] allow multiple kthreadd's

From: Trond Myklebust
Date: Fri May 01 2020 - 15:05:52 EST


On Fri, 2020-05-01 at 14:49 -0400, J. Bruce Fields wrote:
> On Fri, May 01, 2020 at 02:21:54PM -0400, Tejun Heo wrote:
> > Hello,
> >
> > On Fri, May 01, 2020 at 10:59:24AM -0700, Linus Torvalds wrote:
> > > Which kind of makes me want to point a finger at Tejun. But it's
> > > been
> > > mostly PeterZ touching this file lately..
> >
> > Looks fine to me too. I don't quite understand the usecase tho. It
> > looks
> > like all it's being used for is to tag some kthreads as belonging
> > to the
> > same group.
>
> Pretty much.
>

Wen running an instance of knfsd from inside a container, you want to
be able to have the knfsd kthreads be parented to the container init
process so that they get killed off when the container is killed.

Right now, we can easily leak those kernel threads simply by killing
the container.

> > Can't that be done with kthread_data()?
>
> Huh, maybe so, thanks.
>
> I need to check this from generic file locking code that could be run
> by
> any task--but I assume there's an easy way I can check if I'm a
> kthread
> before calling kthread_data(current).
>
> I do expect to expose a delegation interface for userspace servers
> eventually too. But we could do the tgid check for them and still
> use
> kthread_data() for nfsd. That might work.
>
> --b.
>
--
Trond Myklebust
Linux NFS client maintainer, Hammerspace
trond.myklebust@xxxxxxxxxxxxxxx