Re: [PATCH v2 3/3] vhost/vsock: use netns of process that opens the vhost-vsock-netns device

[Bobby Eshleman]

On Fri, Mar 28, 2025 at 05:19:44PM +0100, Stefano Garzarella wrote:
> On Fri, Mar 28, 2025 at 09:07:22AM -0700, Bobby Eshleman wrote:
> > On Thu, Mar 27, 2025 at 10:14:59AM +0100, Stefano Garzarella wrote:
> > > On Tue, Mar 25, 2025 at 05:11:49PM -0700, Bobby Eshleman wrote:
> > > > On Fri, Mar 21, 2025 at 11:02:34AM +0100, Stefano Garzarella wrote:
> > > > > On Thu, Mar 20, 2025 at 02:05:38PM -0700, Bobby Eshleman wrote:
> > > > > > On Thu, Mar 20, 2025 at 10:08:02AM +0100, Stefano Garzarella wrote:
> > > > > > > On Wed, Mar 19, 2025 at 10:09:44PM +0100, Paolo Abeni wrote:
> > > > > > > > On 3/12/25 9:59 PM, Bobby Eshleman wrote:
> > > > > > > > > @@ -753,6 +783,8 @@ static int vhost_vsock_dev_release(struct inode *inode, struct file *file)
> > > > > > > > >  	virtio_vsock_skb_queue_purge(&vsock->send_pkt_queue);
> > > > > > > > >
> > > > > > > > >  	vhost_dev_cleanup(&vsock->dev);
> > > > > > > > > +	if (vsock->net)
> > > > > > > > > +		put_net(vsock->net);
> > > > > > > >
> > > > > > > > put_net() is a deprecated API, you should use put_net_track() instead.
> > > > > > > >
> > > > > > > > >  	kfree(vsock->dev.vqs);
> > > > > > > > >  	vhost_vsock_free(vsock);
> > > > > > > > >  	return 0;
> > > > > > > >
> > > > > > > > Also series introducing new features should also include the related
> > > > > > > > self-tests.
> > > > > > >
> > > > > > > Yes, I was thinking about testing as well, but to test this I think we need
> > > > > > > to run QEMU with Linux in it, is this feasible in self-tests?
> > > > > > >
> > > > > > > We should start looking at that, because for now I have my own ansible
> > > > > > > script that runs tests (tools/testing/vsock/vsock_test) in nested VMs to
> > > > > > > test both host (vhost-vsock) and guest (virtio-vsock).
> > > > > > >
> > > > > >
> > > > > > Maybe as a baseline we could follow the model of
> > > > > > tools/testing/selftests/bpf/vmtest.sh and start by reusing your
> > > > > > vsock_test parameters from your Ansible script?
> > > > >
> > > > > Yeah, my playbooks are here:
> > > > > https://github.com/stefano-garzarella/ansible-vsock
> > > > >
> > > > > Note: they are heavily customized on my env, I wrote some notes on how to
> > > > > change various wired path.
> > > > >
> > > > > >
> > > > > > I don't mind writing the patches.
> > > > >
> > > > > That would be great and very much appreciated.
> > > > > Maybe you can do it in a separate series and then here add just the
> > > > > configuration we need.
> > > > >
> > > > > Thanks,
> > > > > Stefano
> > > > >
> > > >
> > > > Hey Stefano,
> > > >
> > > > I noticed that bpf/vmtest.sh uses images hosted from libbpf's CI/CD. I
> > > > wonder if you have any thoughts on a good repo we may use to pull our
> > > > qcow image(s)? Or a preferred way to host some images, if no repo
> > > > exists?
> > > 
> > > Good question!
> > > 
> > > I created this group/repo mainily to keep trak of work, not sure if we can
> > > reuse: https://gitlab.com/vsock/
> > > 
> > > I can add you there if you need to create new repo, etc.
> > > 
> > > But I'm also open to other solutions.
> > > 
> > 
> > Sounds good to me. I also was considering using virtme-ng, which would
> > avoid the need, at the cost of the dependency. What are your thoughts on
> > that route?
> 
> I just saw that Paolo had proposed the same, but his response was off-list
> by mistake!
> 
> So I would say it is an explorable path. I have no experience with it, but
> it looks like it could do the job!

Sounds good! I'm currently prototyping with it, so save for unforeseen
issues that will likely be what v1 uses.

Thanks,
Bobby