diff mbox

[net-next-2.6,1/4] net: extend netlink interface to handle generic slave management

Message ID 20110211152125.GA2763@psychotron.brq.redhat.com
State Changes Requested, archived
Delegated to: David Miller
Headers show

Commit Message

Jiri Pirko Feb. 11, 2011, 3:21 p.m. UTC 
Drivers like bridge and bonding uses their own way to manipulate with
underlink devices. This is an attempt to introduce common interface using
netlink.

Signed-off-by: Jiri Pirko <jpirko@redhat.com>
---
 include/linux/if_link.h   |   13 +++++++
 include/linux/netdevice.h |   22 ++++++++++++
 net/core/rtnetlink.c      |   81 +++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 116 insertions(+), 0 deletions(-)

Comments

Patrick McHardy Feb. 11, 2011, 3:48 p.m. UTC | #1 
On 11.02.2011 16:21, Jiri Pirko wrote:
> Drivers like bridge and bonding uses their own way to manipulate with
> underlink devices. This is an attempt to introduce common interface using
> netlink.

Thanks for working on this, this has been on my TODO list for a
long time.

> --- a/include/linux/if_link.h
> +++ b/include/linux/if_link.h
> @@ -136,6 +136,9 @@ enum {
>  	IFLA_PORT_SELF,
>  	IFLA_AF_SPEC,
>  	IFLA_GROUP,		/* Group the device belongs to */
> +	IFLA_SLAVE_LIST,
> +	IFLA_SLAVE_ADD,
> +	IFLA_SLAVE_DEL,

I don't like this very much though, the attributes usually contain
data, not commands. We already have NEWLINK, DELLINK etc. on the
top level, the combinations of NEWLINK/NLM_F_CREAT and SLAVE_DEL
or DELLINK and SLAVE_ADD and so on simply don't make sense.

We usually also try to keep the interface symetrical in both
directions (a NEWLINK message from the kernel is identical to a
NEWLINK message from userspace, a DELLINK message as well besides
containing additional information), so using different attributes
for dumping slaves than for adding them seems wrong. If we can
dump all slaves in one message, it should also be possible to
enslave multiple devices using the same message.

What I originally had planned to support enslaving devices is to
make use of the IFLA_MASTER attribute. The IFLA_MASTER attribute
would contain the bond or bridge ifindex and the IFLA_IFNAME
attribute or ifindex would specify the slave device. All operations
would be performed on the slave device as usual, if the IFLA_MASTER
attribute is present we'd additionally call a master specific
callback for enslaving or releasing slave devices. Besides allowing
to keep messages symetrical, an additional benefit is that it would
be possible to create and enslave a device in a single step.


--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Jiri Pirko Feb. 11, 2011, 5:40 p.m. UTC | #2 
Fri, Feb 11, 2011 at 04:48:19PM CET, kaber@trash.net wrote:
>On 11.02.2011 16:21, Jiri Pirko wrote:
>> Drivers like bridge and bonding uses their own way to manipulate with
>> underlink devices. This is an attempt to introduce common interface using
>> netlink.
>
>Thanks for working on this, this has been on my TODO list for a
>long time.
>
>> --- a/include/linux/if_link.h
>> +++ b/include/linux/if_link.h
>> @@ -136,6 +136,9 @@ enum {
>>  	IFLA_PORT_SELF,
>>  	IFLA_AF_SPEC,
>>  	IFLA_GROUP,		/* Group the device belongs to */
>> +	IFLA_SLAVE_LIST,
>> +	IFLA_SLAVE_ADD,
>> +	IFLA_SLAVE_DEL,
>
>I don't like this very much though, the attributes usually contain
>data, not commands. We already have NEWLINK, DELLINK etc. on the
>top level, the combinations of NEWLINK/NLM_F_CREAT and SLAVE_DEL
>or DELLINK and SLAVE_ADD and so on simply don't make sense.
>
>We usually also try to keep the interface symetrical in both
>directions (a NEWLINK message from the kernel is identical to a
>NEWLINK message from userspace, a DELLINK message as well besides
>containing additional information), so using different attributes
>for dumping slaves than for adding them seems wrong. If we can
>dump all slaves in one message, it should also be possible to
>enslave multiple devices using the same message.
>
>What I originally had planned to support enslaving devices is to
>make use of the IFLA_MASTER attribute. The IFLA_MASTER attribute
>would contain the bond or bridge ifindex and the IFLA_IFNAME
>attribute or ifindex would specify the slave device. All operations
>would be performed on the slave device as usual, if the IFLA_MASTER
>attribute is present we'd additionally call a master specific
>callback for enslaving or releasing slave devices. Besides allowing
>to keep messages symetrical, an additional benefit is that it would
>be possible to create and enslave a device in a single step.
>

Yes, that makes sense. I'm going to respin the patchset soon.

Jirka
>
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/include/linux/if_link.h b/include/linux/if_link.h
index f4a2e6b..48a5f95 100644
--- a/include/linux/if_link.h
+++ b/include/linux/if_link.h
@@ -136,6 +136,9 @@  enum {
 	IFLA_PORT_SELF,
 	IFLA_AF_SPEC,
 	IFLA_GROUP,		/* Group the device belongs to */
+	IFLA_SLAVE_LIST,
+	IFLA_SLAVE_ADD,
+	IFLA_SLAVE_DEL,
 	__IFLA_MAX
 };
 
@@ -379,4 +382,14 @@  struct ifla_port_vsi {
 	__u8 pad[3];
 };
 
+/* Slave devices management section */
+
+enum {
+	IFLA_SLAVE_UNSPEC,
+	IFLA_SLAVE_DEV,
+	__IFLA_SLAVE_MAX,
+};
+
+#define IFLA_SLAVE_MAX (__IFLA_SLAVE_MAX - 1)
+
 #endif /* _LINUX_IF_LINK_H */
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index c7d7074..844cb85 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -783,6 +783,21 @@  struct netdev_tc_txq {
  *	Set hardware filter for RFS.  rxq_index is the target queue index;
  *	flow_id is a flow ID to be passed to rps_may_expire_flow() later.
  *	Return the filter ID on success, or a negative error code.
+ *
+ *	Slave management functions (for bridge, bonding, etc). User should
+ *	make sure that slave list doesn't change within rtnl_lock.
+ * int (*ndo_add_slave)(struct net_device *dev, struct net_device *slave_dev);
+ *	Called to make another netdev an underlink.
+ *
+ * int (*ndo_del_slave)(struct net_device *dev, struct net_device *slave_dev);
+ *	Called to release previously enslaved netdev.
+ *
+ * int (*ndo_get_slave_count)(const struct net_device *dev);
+ *	Called to get number of enslaved devices.
+ *
+ * struct net_device * (*ndo_get_slave)(const struct net_device *dev,
+ *					int slave_index);
+ *	Called to get slave device by it's index.
  */
 #define HAVE_NET_DEVICE_OPS
 struct net_device_ops {
@@ -862,6 +877,13 @@  struct net_device_ops {
 						     u16 rxq_index,
 						     u32 flow_id);
 #endif
+	int			(*ndo_add_slave)(struct net_device *dev,
+						 struct net_device *slave_dev);
+	int			(*ndo_del_slave)(struct net_device *dev,
+						 struct net_device *slave_dev);
+	int			(*ndo_get_slave_count)(const struct net_device *dev);
+	struct net_device *	(*ndo_get_slave)(const struct net_device *dev,
+						 int slave_index);
 };
 
 /*
diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c
index da0fe45..1402a3f 100644
--- a/net/core/rtnetlink.c
+++ b/net/core/rtnetlink.c
@@ -739,6 +739,20 @@  static size_t rtnl_port_size(const struct net_device *dev)
 		return port_self_size;
 }
 
+static size_t rtnl_slave_list_size(const struct net_device *dev)
+{
+	size_t slave_size = nla_total_size(sizeof(struct nlattr)) +
+			    nla_total_size(4);
+	size_t slave_list_size = nla_total_size(sizeof(struct nlattr));
+	int slave_count;
+
+	if (!dev->netdev_ops->ndo_get_slave_count ||
+	    !dev->netdev_ops->ndo_get_slave)
+		return 0;
+	slave_count = dev->netdev_ops->ndo_get_slave_count(dev);
+	return slave_list_size + slave_count * slave_size;
+}
+
 static noinline size_t if_nlmsg_size(const struct net_device *dev)
 {
 	return NLMSG_ALIGN(sizeof(struct ifinfomsg))
@@ -760,6 +774,7 @@  static noinline size_t if_nlmsg_size(const struct net_device *dev)
 	       + nla_total_size(4) /* IFLA_NUM_VF */
 	       + rtnl_vfinfo_size(dev) /* IFLA_VFINFO_LIST */
 	       + rtnl_port_size(dev) /* IFLA_VF_PORTS + IFLA_PORT_SELF */
+	       + rtnl_slave_list_size(dev) /* IFLA_SLAVE_LIST */
 	       + rtnl_link_get_size(dev) /* IFLA_LINKINFO */
 	       + rtnl_link_get_af_size(dev); /* IFLA_AF_SPEC */
 }
@@ -839,6 +854,39 @@  static int rtnl_port_fill(struct sk_buff *skb, struct net_device *dev)
 	return 0;
 }
 
+static int rtnl_slave_list_fill(struct sk_buff *skb, struct net_device *dev)
+{
+	int slave_count;
+	int i;
+	struct nlattr *slave_list;
+
+	if (!dev->netdev_ops->ndo_get_slave_count ||
+	    !dev->netdev_ops->ndo_get_slave)
+		return 0;
+	slave_count = dev->netdev_ops->ndo_get_slave_count(dev);
+	slave_list = nla_nest_start(skb, IFLA_SLAVE_LIST);
+	if (!slave_list)
+		return -EMSGSIZE;
+	for (i = 0; i < slave_count; i++) {
+		struct net_device *slave_dev;
+
+		slave_dev = dev->netdev_ops->ndo_get_slave(dev, i);
+		if (!slave_dev) {
+			nla_nest_cancel(skb, slave_list);
+			goto nla_put_failure;
+		}
+		NLA_PUT_U32(skb, IFLA_SLAVE_DEV, slave_dev->ifindex);
+	}
+	nla_nest_end(skb, slave_list);
+
+	return 0;
+
+nla_put_failure:
+	nla_nest_cancel(skb, slave_list);
+	return -EMSGSIZE;
+
+}
+
 static int rtnl_fill_ifinfo(struct sk_buff *skb, struct net_device *dev,
 			    int type, u32 pid, u32 seq, u32 change,
 			    unsigned int flags)
@@ -953,6 +1001,9 @@  static int rtnl_fill_ifinfo(struct sk_buff *skb, struct net_device *dev,
 	if (rtnl_port_fill(skb, dev))
 		goto nla_put_failure;
 
+	if (rtnl_slave_list_fill(skb, dev))
+		goto nla_put_failure;
+
 	if (dev->rtnl_link_ops) {
 		if (rtnl_link_fill(skb, dev) < 0)
 			goto nla_put_failure;
@@ -1047,6 +1098,8 @@  const struct nla_policy ifla_policy[IFLA_MAX+1] = {
 	[IFLA_VF_PORTS]		= { .type = NLA_NESTED },
 	[IFLA_PORT_SELF]	= { .type = NLA_NESTED },
 	[IFLA_AF_SPEC]		= { .type = NLA_NESTED },
+	[IFLA_SLAVE_ADD]	= { .type = NLA_U32 },
+	[IFLA_SLAVE_DEL]	= { .type = NLA_U32 },
 };
 EXPORT_SYMBOL(ifla_policy);
 
@@ -1392,6 +1445,34 @@  static int do_setlink(struct net_device *dev, struct ifinfomsg *ifm,
 			modified = 1;
 		}
 	}
+
+	if (tb[IFLA_SLAVE_ADD]) {
+		int ifindex = nla_get_u32(tb[IFLA_SLAVE_ADD]);
+		struct net_device *slave_dev;
+
+		err = -EOPNOTSUPP;
+		if (ops->ndo_add_slave) {
+			slave_dev = __dev_get_by_index(dev_net(dev), ifindex);
+			err = ops->ndo_add_slave(dev, slave_dev);
+		}
+		if (err < 0)
+			goto errout;
+		modified = 1;
+	}
+
+	if (tb[IFLA_SLAVE_DEL]) {
+		int ifindex = nla_get_u32(tb[IFLA_SLAVE_DEL]);
+		struct net_device *slave_dev;
+
+		err = -EOPNOTSUPP;
+		if (ops->ndo_del_slave) {
+			slave_dev = __dev_get_by_index(dev_net(dev), ifindex);
+			err = ops->ndo_del_slave(dev, slave_dev);
+		}
+		if (err < 0)
+			goto errout;
+		modified = 1;
+	}
 	err = 0;
 
 errout: