[patch net-next v2 1/2] rtnetlink: allow to register ops without ops->setup set

So far, it is assumed that ops->setup is filled up. But there might be
case that ops might make sense even without ->setup. In that case,
forbid to newlink and dellink.

This allows to register simple rtnl link ops containing only ->kind.
That allows consistent way of passing device kind (either device-kind or
slave-kind) to userspace.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
---

v1->v2: included comments from Eric fixing default_device_exit_batch,
	not checking setup in dellink and added hopefully clearer description
	and comment.

 net/core/rtnetlink.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c
index 1063996..27acaf7 100644
--- a/net/core/rtnetlink.c
+++ b/net/core/rtnetlink.c

@@ -299,7 +299,12 @@ int __rtnl_link_register(struct rtnl_link_ops *ops)
 	if (rtnl_link_ops_get(ops->kind))
 		return -EEXIST;
 
-	if (!ops->dellink)
+	/* The check for setup is here because if ops
+	 * does not have that filled up, it is not possible
+	 * to use the ops for creating device. So do not
+	 * fill up dellink as well. That disables rtnl_dellink.
+	 */
+	if (ops->setup && !ops->dellink)
 		ops->dellink = unregister_netdevice_queue;
 
 	list_add_tail(&ops->list, &link_ops);

@@ -1777,7 +1782,7 @@ static int rtnl_dellink(struct sk_buff *skb, struct nlmsghdr *nlh)
 		return -ENODEV;
 
 	ops = dev->rtnl_link_ops;
-	if (!ops)
+	if (!ops || !ops->dellink)
 		return -EOPNOTSUPP;
 
 	ops->dellink(dev, &list_kill);

@@ -2038,6 +2043,9 @@ replay:
 			return -EOPNOTSUPP;
 		}
 
+		if (!ops->setup)
+			return -EOPNOTSUPP;
+
 		if (!ifname[0])
 			snprintf(ifname, IFNAMSIZ, "%s%%d", ops->kind);
 

-- 
1.9.0

This stub now allows userspace to see IFLA_INFO_KIND for ovs master and
IFLA_INFO_SLAVE_KIND for slave.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>
---
 net/openvswitch/datapath.c           |  9 ++++++++-
 net/openvswitch/vport-internal_dev.c | 16 ++++++++++++++++
 net/openvswitch/vport-internal_dev.h |  2 ++
 3 files changed, 26 insertions(+), 1 deletion(-)

diff --git a/net/openvswitch/datapath.c b/net/openvswitch/datapath.c
index 0d407bc..fe95b6c 100644
--- a/net/openvswitch/datapath.c
+++ b/net/openvswitch/datapath.c

@@ -2054,10 +2054,14 @@ static int __init dp_init(void)
 
 	pr_info("Open vSwitch switching datapath\n");
 
-	err = ovs_flow_init();
+	err = ovs_internal_dev_rtnl_link_register();
 	if (err)
 		goto error;
 
+	err = ovs_flow_init();
+	if (err)
+		goto error_unreg_rtnl_link;
+
 	err = ovs_vport_init();
 	if (err)
 		goto error_flow_exit;

@@ -2084,6 +2088,8 @@ error_vport_exit:
 	ovs_vport_exit();
 error_flow_exit:
 	ovs_flow_exit();
+error_unreg_rtnl_link:
+	ovs_internal_dev_rtnl_link_unregister();
 error:
 	return err;
 }

@@ -2096,6 +2102,7 @@ static void dp_cleanup(void)
 	rcu_barrier();
 	ovs_vport_exit();
 	ovs_flow_exit();
+	ovs_internal_dev_rtnl_link_unregister();
 }
 
 module_init(dp_init);

diff --git a/net/openvswitch/vport-internal_dev.c b/net/openvswitch/vport-internal_dev.c
index 789af92..295471a 100644
--- a/net/openvswitch/vport-internal_dev.c
+++ b/net/openvswitch/vport-internal_dev.c

@@ -26,6 +26,7 @@
 
 #include <net/dst.h>
 #include <net/xfrm.h>
+#include <net/rtnetlink.h>
 
 #include "datapath.h"
 #include "vport-internal_dev.h"

@@ -121,6 +122,10 @@ static const struct net_device_ops internal_dev_netdev_ops = {
 	.ndo_get_stats64 = internal_dev_get_stats,
 };
 
+static struct rtnl_link_ops internal_dev_link_ops __read_mostly = {
+	.kind = "openvswitch",
+};
+
 static void do_setup(struct net_device *netdev)
 {
 	ether_setup(netdev);

@@ -131,6 +136,7 @@ static void do_setup(struct net_device *netdev)
 	netdev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 	netdev->destructor = internal_dev_destructor;
 	netdev->ethtool_ops = &internal_dev_ethtool_ops;
+	netdev->rtnl_link_ops = &internal_dev_link_ops;
 	netdev->tx_queue_len = 0;
 
 	netdev->features = NETIF_F_LLTX | NETIF_F_SG | NETIF_F_FRAGLIST |

@@ -248,3 +254,13 @@ struct vport *ovs_internal_dev_get_vport(struct net_device *netdev)
 
 	return internal_dev_priv(netdev)->vport;
 }
+
+int ovs_internal_dev_rtnl_link_register(void)
+{
+	return rtnl_link_register(&internal_dev_link_ops);
+}
+
+void ovs_internal_dev_rtnl_link_unregister(void)
+{
+	rtnl_link_unregister(&internal_dev_link_ops);
+}

diff --git a/net/openvswitch/vport-internal_dev.h b/net/openvswitch/vport-internal_dev.h
index 9a7d30e..1b179a1 100644
--- a/net/openvswitch/vport-internal_dev.h
+++ b/net/openvswitch/vport-internal_dev.h

@@ -24,5 +24,7 @@
 
 int ovs_is_internal_dev(const struct net_device *);
 struct vport *ovs_internal_dev_get_vport(struct net_device *);
+int ovs_internal_dev_rtnl_link_register(void);
+void ovs_internal_dev_rtnl_link_unregister(void);
 
 #endif /* vport-internal_dev.h */

-- 
1.9.0

Jiri Pirko [off-list ref] writes:

So far, it is assumed that ops->setup is filled up. But there might be
case that ops might make sense even without ->setup. In that case,
forbid to newlink and dellink.

This allows to register simple rtnl link ops containing only ->kind.
That allows consistent way of passing device kind (either device-kind or
slave-kind) to userspace.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>

Nacked-by: "Eric W. Biederman" [off-list ref]

This is absolutely unmaintainable.  You can't even write the patch
correctly when specific problems are pointed out I can't imagine anyone
else will be able to cope.

As it happens default_device_exit_batch is more broken with this version
of your patch.

On Wed, 25 Jun 2014 18:58:22 +0200
Jiri Pirko [off-list ref] wrote:

So far, it is assumed that ops->setup is filled up. But there might be
case that ops might make sense even without ->setup. In that case,
forbid to newlink and dellink.

This allows to register simple rtnl link ops containing only ->kind.
That allows consistent way of passing device kind (either device-kind or
slave-kind) to userspace.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>

Can't we fix these kind of devices to all create/delete. At least
allow delete.

Wed, Jun 25, 2014 at 08:45:00PM CEST, stephen@networkplumber.org wrote:

On Wed, 25 Jun 2014 18:58:22 +0200
Jiri Pirko [off-list ref] wrote:

quoted

So far, it is assumed that ops->setup is filled up. But there might be
case that ops might make sense even without ->setup. In that case,
forbid to newlink and dellink.

This allows to register simple rtnl link ops containing only ->kind.
That allows consistent way of passing device kind (either device-kind or
slave-kind) to userspace.

Signed-off-by: Jiri Pirko <jiri@resnulli.us>

Can't we fix these kind of devices to all create/delete. At least
allow delete.

Well it is not that easy to create openvswitch device. To do that you
have to have info specified in vport_params (see ovs_vport_alloc).

Delete could be implemented. But looking at the code, it could be a bit
tricky. But would that make sense to implement del and don't implement
add? I think it's better to leave both out.