Date:   Tue, 16 Aug 2022 08:29:04 +0000
From:   Bobby Eshleman <bobbyeshleman@...il.com>
To:     Jakub Kicinski <kuba@...nel.org>
Cc:     Bobby Eshleman <bobby.eshleman@...il.com>,
        "Michael S. Tsirkin" <mst@...hat.com>,
        Bobby Eshleman <bobby.eshleman@...edance.com>,
        Cong Wang <cong.wang@...edance.com>,
        Jiang Wang <jiang.wang@...edance.com>,
        Stefan Hajnoczi <stefanha@...hat.com>,
        Stefano Garzarella <sgarzare@...hat.com>,
        Jason Wang <jasowang@...hat.com>,
        "David S. Miller" <davem@...emloft.net>,
        Eric Dumazet <edumazet@...gle.com>,
        Paolo Abeni <pabeni@...hat.com>, kvm@...r.kernel.org,
        virtualization@...ts.linux-foundation.org, netdev@...r.kernel.org,
        linux-kernel@...r.kernel.org,
        Toke Høiland-Jørgensen <toke@...e.dk>
Subject: Re: [PATCH 3/6] vsock: add netdev to vhost/virtio vsock

On Tue, Aug 16, 2022 at 04:07:55PM -0700, Jakub Kicinski wrote:
> On Tue, 16 Aug 2022 07:02:33 +0000 Bobby Eshleman wrote:
> > > From a cursory look (and Documentation/ would be nice..) it feels
> > > very wrong to me. Do you know of any uses of a netdev which would 
> > > be semantically similar to what you're doing? Treating netdevs as
> > > buildings blocks for arbitrary message passing solutions is something 
> > > I dislike quite strongly.  
> > 
> > The big difference between vsock and "arbitrary message passing" is that
> > vsock is actually constrained by the virtio device that backs it (made
> > up of virtqueues and the underlying protocol). That virtqueue pair is
> > acting like the queues on a physical NIC, so it actually makes sense to
> > manage the queuing of vsock's device like we would manage the queueing
> > of a real device.
> > 
> > Still, I concede that ignoring the netdev state is a probably bad idea.
> > 
> > That said, I also think that using packet scheduling in vsock is a good
> > idea, and that ideally we can reuse Linux's already robust library of
> > packet scheduling algorithms by introducing qdisc somehow.
> 
> We've been burnt in the past by people doing the "let me just pick
> these useful pieces out of netdev" thing. Makes life hard both for
> maintainers and users trying to make sense of the interfaces.
> 
> What comes to mind if you're just after queuing is that we already
> bastardized the CoDel implementation (include/net/codel_impl.h).
> If CoDel is good enough for you maybe that's the easiest way?
> Although I suspect that you're after fairness not early drops.
> Wireless folks use CoDel as a second layer queuing. (CC: Toke)
> 

That is certainly interesting to me. Sitting next to "codel_impl.h" is
"include/net/fq_impl.h", and it looks like it may solve the datagram
flooding issue. The downside to this approach is the baking of a
specific policy into vsock... which I don't exactly love either.

I'm not seeing too many other of these qdisc bastardizations in
include/net, are there any others that you are aware of?

> > > Could you recommend where I can learn more about vsocks?  
> > 
> > I think the spec is probably the best place to start[1].
> > 
> > [1]: https://docs.oasis-open.org/virtio/virtio/v1.2/virtio-v1.2.html
> 
> Eh, I was hoping it was a side channel of an existing virtio_net 
> which is not the case. Given the zero-config requirement IDK if 
> we'll be able to fit this into netdev semantics :(

It's certainly possible that it may not fit :/ I feel that it partially
depends on what we mean by zero-config. Is it "no config required to
have a working socket" or is it "no config required, but also no
tuning/policy/etc... supported"?

Best,
Bobby

Powered by blists - more mailing lists