Re: [PATCH 08/18] xen/pvcalls: implement connect command

[Stefano Stabellini]

On Mon, 15 May 2017, Boris Ostrovsky wrote:
> On 05/15/2017 04:36 PM, Stefano Stabellini wrote:
> > Allocate a socket. Keep track of socket <-> ring mappings with a new data
> > structure, called sock_mapping. Implement the connect command by calling
> > inet_stream_connect, and mapping the new indexes page and data ring.
> > Associate the socket to an ioworker randomly.
> > 
> > When an active socket is closed (sk_state_change), set in_error to
> > -ENOTCONN and notify the other end, as specified by the protocol.
> > 
> > sk_data_ready will be implemented later.
> > 
> > Signed-off-by: Stefano Stabellini <stefano@xxxxxxxxxxx>
> > CC: boris.ostrovsky@xxxxxxxxxx
> > CC: jgross@xxxxxxxx
> > ---
> >  drivers/xen/pvcalls-back.c | 145
> > +++++++++++++++++++++++++++++++++++++++++++++
> >  1 file changed, 145 insertions(+)
> > 
> > diff --git a/drivers/xen/pvcalls-back.c b/drivers/xen/pvcalls-back.c
> > index 2eae096..9ac1cf2 100644
> > --- a/drivers/xen/pvcalls-back.c
> > +++ b/drivers/xen/pvcalls-back.c
> > @@ -63,6 +63,29 @@ struct pvcalls_back_priv {
> >  	struct work_struct register_work;
> >  };
> > 
> > +struct sock_mapping {
> > +	struct list_head list;
> > +	struct list_head queue;
> 
> Since you have two lists it would be helpful if names were a bit more
> descriptive.
> 
> (and comments for at least some fields would be welcome too)

Yeah, you are right. list is used to add sock_mapping to
priv->socket_mappings, the per frontend list of active sockets. queue is
used to add sock_mapping to the ioworker list. I'll add a comment.


> > +	struct pvcalls_back_priv *priv;
> > +	struct socket *sock;
> > +	int data_worker;
> > +	uint64_t id;
> > +	grant_ref_t ref;
> > +	struct pvcalls_data_intf *ring;
> > +	void *bytes;
> > +	struct pvcalls_data data;
> > +	uint32_t ring_order;
> > +	int irq;
> > +	atomic_t read;
> > +	atomic_t write;
> > +	atomic_t release;
> > +	void (*saved_data_ready)(struct sock *sk);
> > +};
> > +
> > +static irqreturn_t pvcalls_back_conn_event(int irq, void *sock_map);
> > +static int pvcalls_back_release_active(struct xenbus_device *dev,
> > +				       struct pvcalls_back_priv *priv,
> > +				       struct sock_mapping *map);
> >  static void pvcalls_back_ioworker(struct work_struct *work)
> >  {
> >  }
> > @@ -97,9 +120,126 @@ static int pvcalls_back_socket(struct xenbus_device
> > *dev,
> >  	return 1;
> >  }
> > 
> > +static void pvcalls_sk_state_change(struct sock *sock)
> > +{
> > +	struct sock_mapping *map = sock->sk_user_data;
> > +	struct pvcalls_data_intf *intf;
> > +
> > +	if (map == NULL)
> > +		return;
> > +
> > +	intf = map->ring;
> > +	intf->in_error = -ENOTCONN;
> > +	notify_remote_via_irq(map->irq);
> > +}
> > +
> > +static void pvcalls_sk_data_ready(struct sock *sock)
> > +{
> > +}
> > +
> >  static int pvcalls_back_connect(struct xenbus_device *dev,
> >  				struct xen_pvcalls_request *req)
> >  {
> > +	struct pvcalls_back_priv *priv;
> > +	int ret;
> > +	struct socket *sock;
> > +	struct sock_mapping *map = NULL;
> > +	void *page;
> > +	struct xen_pvcalls_response *rsp;
> > +
> > +	if (dev == NULL)
> > +		return 0;
> > +	priv = dev_get_drvdata(&dev->dev);
> > +
> > +	map = kzalloc(sizeof(*map), GFP_KERNEL);
> > +	if (map == NULL) {
> > +		ret = -ENOMEM;
> > +		goto out;
> > +	}
> > +	ret = sock_create(AF_INET, SOCK_STREAM, 0, &sock);
> > +	if (ret < 0) {
> > +		kfree(map);
> > +		goto out;
> > +	}
> > +	INIT_LIST_HEAD(&map->queue);
> > +	map->data_worker = get_random_int() %
> > pvcalls_back_global.nr_ioworkers;
> > +
> > +	map->priv = priv;
> > +	map->sock = sock;
> > +	map->id = req->u.connect.id;
> > +	map->ref = req->u.connect.ref;
> > +
> > +	ret = xenbus_map_ring_valloc(dev, &req->u.connect.ref, 1, &page);
> > +	if (ret < 0) {
> > +		sock_release(map->sock);
> > +		kfree(map);
> > +		goto out;
> > +	}
> > +	map->ring = page;
> > +	map->ring_order = map->ring->ring_order;
> > +	/* first read the order, then map the data ring */
> > +	virt_rmb();
> 
> 
> Not sure I understand what the barrier is for here. I don't think compiler
> will reorder ring_order access with the call.

It's to avoid using the live version of ring_order to map the data ring
pages (the other end could be changing that value at any time). We want
to be sure that the compiler doesn't optimize out map->ring_order and
use map->ring->ring_order instead.


> > +	if (map->ring_order > MAX_RING_ORDER) {
> > +		ret = -EFAULT;
> > +		goto out;
> > +	}
> 
> If the barrier is indeed needed this check belongs before it.

I don't think so, see above.


> 
> 
> > +	ret = xenbus_map_ring_valloc(dev, map->ring->ref,
> > +				     (1 << map->ring_order), &page);
> > +	if (ret < 0) {
> > +		sock_release(map->sock);
> > +		xenbus_unmap_ring_vfree(dev, map->ring);
> > +		kfree(map);
> > +		goto out;
> > +	}
> > +	map->bytes = page;
> > +
> > +	ret = bind_interdomain_evtchn_to_irqhandler(priv->dev->otherend_id,
> > +						    req->u.connect.evtchn,
> > +						    pvcalls_back_conn_event,
> > +						    0,
> > +						    "pvcalls-backend",
> > +						    map);
> > +	if (ret < 0) {
> > +		sock_release(map->sock);
> > +		kfree(map);
> > +		goto out;
> > +	}
> > +	map->irq = ret;
> > +
> > +	map->data.in = map->bytes;
> > +	map->data.out = map->bytes + XEN_FLEX_RING_SIZE(map->ring_order);
> > +
> > +	down_write(&priv->pvcallss_lock);
> > +	list_add_tail(&map->list, &priv->socket_mappings);
> > +	up_write(&priv->pvcallss_lock);
> > +
> > +	ret = inet_stream_connect(sock, (struct sockaddr
> > *)&req->u.connect.addr,
> > +				  req->u.connect.len, req->u.connect.flags);
> > +	if (ret < 0) {
> > +		pvcalls_back_release_active(dev, priv, map);
> > +	} else {
> > +		lock_sock(sock->sk);
> > +		map->saved_data_ready = sock->sk->sk_data_ready;
> > +		sock->sk->sk_user_data = map;
> > +		sock->sk->sk_data_ready = pvcalls_sk_data_ready;
> > +		sock->sk->sk_state_change = pvcalls_sk_state_change;
> > +		release_sock(sock->sk);
> > +	}
> > +
> > +out:
> > +	rsp = RING_GET_RESPONSE(&priv->ring, priv->ring.rsp_prod_pvt++);
> > +	rsp->req_id = req->req_id;
> > +	rsp->cmd = req->cmd;
> > +	rsp->u.connect.id = req->u.connect.id;
> > +	rsp->ret = ret;
> > +
> > +	return 1;
> > +}
> > +
> > +static int pvcalls_back_release_active(struct xenbus_device *dev,
> > +				       struct pvcalls_back_priv *priv,
> > +				       struct sock_mapping *map)
> > +{
> >  	return 0;
> >  }
> > 
> > @@ -215,6 +355,11 @@ static irqreturn_t pvcalls_back_event(int irq, void
> > *dev_id)
> >  	return IRQ_HANDLED;
> >  }
> > 
> > +static irqreturn_t pvcalls_back_conn_event(int irq, void *sock_map)
> > +{
> > +	return IRQ_HANDLED;
> > +}
> > +
> >  static int backend_connect(struct xenbus_device *dev)
> >  {
> >  	int err, evtchn;
> > 
>