Linux Virtualization - Nov 2019 - [PATCH net-next 5/6] vsock: use local transport when it is loaded

On Tue, Nov 19, 2019 at 12:01:20PM +0100, Stefano Garzarella
wrote:
> @@ -420,9 +436,10 @@ int vsock_assign_transport(struct vsock_sock *vsk,
struct vsock_sock *psk)
>  		new_transport = transport_dgram;
>  		break;
>  	case SOCK_STREAM:
> -		if (remote_cid <= VMADDR_CID_HOST ||
> -		    (transport_g2h &&
> -		     remote_cid == transport_g2h->get_local_cid()))
> +		if (vsock_use_local_transport(remote_cid))
> +			new_transport = transport_local;
> +		else if (remote_cid == VMADDR_CID_HOST ||
> +			 remote_cid == VMADDR_CID_HYPERVISOR)
>  			new_transport = transport_g2h;
>  		else
>  			new_transport = transport_h2g;
We used to send VMADDR_CID_RESERVED to the host.  Now we send
VMADDR_CID_RESERVED (LOCAL) to the guest when there is no
transport_local loaded?

If this is correct, is there a justification for this change?  It seems
safest to retain existing behavior.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 488 bytes
Desc: not available
URL:
<http://lists.linuxfoundation.org/pipermail/virtualization/attachments/20191121/233aa21b/attachment.sig>

On Thu, Nov 21, 2019 at 09:46:14AM +0000, Stefan Hajnoczi
wrote:
> On Tue, Nov 19, 2019 at 12:01:20PM +0100, Stefano Garzarella wrote:
> > @@ -420,9 +436,10 @@ int vsock_assign_transport(struct vsock_sock
*vsk, struct vsock_sock *psk)
> >  		new_transport = transport_dgram;
> >  		break;
> >  	case SOCK_STREAM:
> > -		if (remote_cid <= VMADDR_CID_HOST ||
> > -		    (transport_g2h &&
> > -		     remote_cid == transport_g2h->get_local_cid()))
> > +		if (vsock_use_local_transport(remote_cid))
> > +			new_transport = transport_local;
> > +		else if (remote_cid == VMADDR_CID_HOST ||
> > +			 remote_cid == VMADDR_CID_HYPERVISOR)
> >  			new_transport = transport_g2h;
> >  		else
> >  			new_transport = transport_h2g;
> 
> We used to send VMADDR_CID_RESERVED to the host.  Now we send
> VMADDR_CID_RESERVED (LOCAL) to the guest when there is no
> transport_local loaded?
> 
> If this is correct, is there a justification for this change?  It seems
> safest to retain existing behavior.
You're right, I'll revert this change in v2.

Thanks,
Stefano