Re: [PATCH net-next 0/6] vsock: add local transport support

[Stefano Garzarella <sgarzare@redhat.com>]

On Thu, Nov 21, 2019 at 02:45:32PM +0000, Jorgen Hansen wrote:
> > From: Stefano Garzarella [mailto:sgarzare@redhat.com]
> > Sent: Tuesday, November 19, 2019 12:01 PM
> > This series introduces a new transport (vsock_loopback) to handle
> > local communication.
> > This could be useful to test vsock core itself and to allow developers
> > to test their applications without launching a VM.
> > 
> > Before this series, vmci and virtio transports allowed this behavior,
> > but only in the guest.
> > We are moving the loopback handling in a new transport, because it
> > might be useful to provide this feature also in the host or when
> > no H2G/G2H transports (hyperv, virtio, vmci) are loaded.
> > 
> > The user can use the loopback with the new VMADDR_CID_LOCAL (that
> > replaces VMADDR_CID_RESERVED) in any condition.
> > Otherwise, if the G2H transport is loaded, it can also use the guest
> > local CID as previously supported by vmci and virtio transports.
> > If G2H transport is not loaded, the user can also use VMADDR_CID_HOST
> > for local communication.
> > 
> > Patch 1 is a cleanup to build virtio_transport_common without virtio
> > Patch 2 adds the new VMADDR_CID_LOCAL, replacing
> > VMADDR_CID_RESERVED
> > Patch 3 adds a new feature flag to register a loopback transport
> > Patch 4 adds the new vsock_loopback transport based on the loopback
> >         implementation of virtio_transport
> > Patch 5 implements the logic to use the local transport for loopback
> >         communication
> > Patch 6 removes the loopback from virtio_transport
> > 
> > @Jorgen: Do you think it might be a problem to replace
> > VMADDR_CID_RESERVED with VMADDR_CID_LOCAL?
> 
> No, that should be fine. It has never allowed for use with stream sockets in
> AF_VSOCK. The only potential use would be for datagram sockets, but that
> side appears to be unaffected by your changes, since loopback is only
> introduced for SOCK_STREAM.
> 

Yes, datagram sockets are not affected.

Thanks for the clarification,
Stefano