Re: [PATCH net-next 3/6] vsock: add local transport support in the vsock core

[Stefano Garzarella]

On Thu, Nov 21, 2019 at 03:53:47PM +0000, Jorgen Hansen wrote:
> > From: Stefano Garzarella [mailto:sgarzare@xxxxxxxxxx]
> > Sent: Thursday, November 21, 2019 4:22 PM
> > 
> > On Thu, Nov 21, 2019 at 03:04:18PM +0000, Jorgen Hansen wrote:
> > > > From: Stefano Garzarella [mailto:sgarzare@xxxxxxxxxx]
> > > > Sent: Tuesday, November 19, 2019 12:01 PM
> > > > To: netdev@xxxxxxxxxxxxxxx
> > > >
> > > > This patch allows to register a transport able to handle
> > > > local communication (loopback).
> > > >
> > > > Signed-off-by: Stefano Garzarella <sgarzare@xxxxxxxxxx>
> > > > ---
> > > >  include/net/af_vsock.h   |  2 ++
> > > >  net/vmw_vsock/af_vsock.c | 17 ++++++++++++++++-
> > > >  2 files changed, 18 insertions(+), 1 deletion(-)
> > > >
> > > > diff --git a/include/net/af_vsock.h b/include/net/af_vsock.h
> > > > index 4206dc6d813f..b1c717286993 100644
> > > > --- a/include/net/af_vsock.h
> > > > +++ b/include/net/af_vsock.h
> > > > @@ -98,6 +98,8 @@ struct vsock_transport_send_notify_data {
> > > >  #define VSOCK_TRANSPORT_F_G2H		0x00000002
> > > >  /* Transport provides DGRAM communication */
> > > >  #define VSOCK_TRANSPORT_F_DGRAM		0x00000004
> > > > +/* Transport provides local (loopback) communication */
> > > > +#define VSOCK_TRANSPORT_F_LOCAL		0x00000008
> > > >
> > > >  struct vsock_transport {
> > > >  	struct module *module;
> > > > diff --git a/net/vmw_vsock/af_vsock.c b/net/vmw_vsock/af_vsock.c
> > > > index cc8659838bf2..c9e5bad59dc1 100644
> > > > --- a/net/vmw_vsock/af_vsock.c
> > > > +++ b/net/vmw_vsock/af_vsock.c
> > > > @@ -136,6 +136,8 @@ static const struct vsock_transport
> > *transport_h2g;
> > > >  static const struct vsock_transport *transport_g2h;
> > > >  /* Transport used for DGRAM communication */
> > > >  static const struct vsock_transport *transport_dgram;
> > > > +/* Transport used for local communication */
> > > > +static const struct vsock_transport *transport_local;
> > > >  static DEFINE_MUTEX(vsock_register_mutex);
> > > >
> > > >  /**** UTILS ****/
> > > > @@ -2130,7 +2132,7 @@
> > EXPORT_SYMBOL_GPL(vsock_core_get_transport);
> > > >
> > > >  int vsock_core_register(const struct vsock_transport *t, int features)
> > > >  {
> > > > -	const struct vsock_transport *t_h2g, *t_g2h, *t_dgram;
> > > > +	const struct vsock_transport *t_h2g, *t_g2h, *t_dgram, *t_local;
> > > >  	int err = mutex_lock_interruptible(&vsock_register_mutex);
> > > >
> > > >  	if (err)
> > > > @@ -2139,6 +2141,7 @@ int vsock_core_register(const struct
> > > > vsock_transport *t, int features)
> > > >  	t_h2g = transport_h2g;
> > > >  	t_g2h = transport_g2h;
> > > >  	t_dgram = transport_dgram;
> > > > +	t_local = transport_local;
> > > >
> > > >  	if (features & VSOCK_TRANSPORT_F_H2G) {
> > > >  		if (t_h2g) {
> > > > @@ -2164,9 +2167,18 @@ int vsock_core_register(const struct
> > > > vsock_transport *t, int features)
> > > >  		t_dgram = t;
> > > >  	}
> > > >
> > > > +	if (features & VSOCK_TRANSPORT_F_LOCAL) {
> > > > +		if (t_local) {
> > > > +			err = -EBUSY;
> > > > +			goto err_busy;
> > > > +		}
> > > > +		t_local = t;
> > > > +	}
> > > > +
> > > >  	transport_h2g = t_h2g;
> > > >  	transport_g2h = t_g2h;
> > > >  	transport_dgram = t_dgram;
> > > > +	transport_local = t_local;
> > > >
> > > >  err_busy:
> > > >  	mutex_unlock(&vsock_register_mutex);
> > > > @@ -2187,6 +2199,9 @@ void vsock_core_unregister(const struct
> > > > vsock_transport *t)
> > > >  	if (transport_dgram == t)
> > > >  		transport_dgram = NULL;
> > > >
> > > > +	if (transport_local == t)
> > > > +		transport_local = NULL;
> > > > +
> > > >  	mutex_unlock(&vsock_register_mutex);
> > > >  }
> > > >  EXPORT_SYMBOL_GPL(vsock_core_unregister);
> > > > --
> > > > 2.21.0
> > >
> > > Having loopback support as a separate transport fits nicely, but do we need
> > to support
> > > different variants of loopback? It could just be built in.
> > 
> > I agree with you, indeed initially I developed it as built in, but
> > DEPMOD found a cyclic dependency because vsock_transport use
> > virtio_transport_common that use vsock, so if I include vsock_transport
> > in the vsock module, DEPMOD is not happy.
> > 
> > I don't know how to break this cyclic dependency, do you have any ideas?
> 
> One way to view this would be that the loopback transport and the support
> it uses from virtio_transport_common are independent of virtio as such,
> and could be part of  the af_vsock module if loopback is configured. So
> in a way, the virtio g2h and h2g transports would be extensions of the
> built in loopback transport. But that brings in quite a bit of code so
> it could be better to just keep it as is.

Great idea!

Stefan already suggested (as a long-term goal) to rename the generic
functionality in virtio_transport_common.c

Maybe I can do both in another series later on, since it requires enough
changes.

Thanks,
Stefano