udp_tunnel: add central NIC RX port offload infrastructure

Cater to devices which:
 (a) may want to sleep in the callbacks;
 (b) only have IPv4 support;
 (c) need all the programming to happen while the netdev is up.

Drivers attach UDP tunnel offload info struct to their netdevs,
where they declare how many UDP ports of various tunnel types
they support. Core takes care of tracking which ports to offload.

Use a fixed-size array since this matches what almost all drivers
do, and avoids a complexity and uncertainty around memory allocations
in an atomic context.

Make sure that tunnel drivers don't try to replay the ports when
new NIC netdev is registered. Automatic replays would mess up
reference counting, and will be removed completely once all drivers
are converted.

v4:
 - use a #define NULL to avoid build issues with CONFIG_INET=n.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/geneve.c

@@ -1796,9 +1796,11 @@ static int geneve_netdevice_event(struct notifier_block *unused,
 	    event == NETDEV_UDP_TUNNEL_DROP_INFO) {
 		geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
 	} else if (event == NETDEV_UNREGISTER) {
-		geneve_offload_rx_ports(dev, false);
+		if (!dev->udp_tunnel_nic_info)
+			geneve_offload_rx_ports(dev, false);
 	} else if (event == NETDEV_REGISTER) {
-		geneve_offload_rx_ports(dev, true);
+		if (!dev->udp_tunnel_nic_info)
+			geneve_offload_rx_ports(dev, true);
 	}
 
 	return NOTIFY_DONE;

drivers/net/vxlan.c

@@ -4477,10 +4477,12 @@ static int vxlan_netdevice_event(struct notifier_block *unused,
 	struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
 
 	if (event == NETDEV_UNREGISTER) {
-		vxlan_offload_rx_ports(dev, false);
+		if (!dev->udp_tunnel_nic_info)
+			vxlan_offload_rx_ports(dev, false);
 		vxlan_handle_lowerdev_unregister(vn, dev);
 	} else if (event == NETDEV_REGISTER) {
-		vxlan_offload_rx_ports(dev, true);
+		if (!dev->udp_tunnel_nic_info)
+			vxlan_offload_rx_ports(dev, true);
 	} else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
 		   event == NETDEV_UDP_TUNNEL_DROP_INFO) {
 		vxlan_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);

include/linux/netdevice.h

@@ -65,6 +65,8 @@ struct wpan_dev;
 struct mpls_dev;
 /* UDP Tunnel offloads */
 struct udp_tunnel_info;
+struct udp_tunnel_nic_info;
+struct udp_tunnel_nic;
 struct bpf_prog;
 struct xdp_buff;
 
@@ -1836,6 +1838,10 @@ enum netdev_priv_flags {
  *
  *	@macsec_ops:    MACsec offloading ops
  *
+ *	@udp_tunnel_nic_info:	static structure describing the UDP tunnel
+ *				offload capabilities of the device
+ *	@udp_tunnel_nic:	UDP tunnel offload state
+ *
  *	FIXME: cleanup struct net_device such that network protocol info
  *	moves out.
  */
@@ -2134,6 +2140,8 @@ struct net_device {
 	/* MACsec management functions */
 	const struct macsec_ops *macsec_ops;
 #endif
+	const struct udp_tunnel_nic_info	*udp_tunnel_nic_info;
+	struct udp_tunnel_nic	*udp_tunnel_nic;
 };
 #define to_net_dev(d) container_of(d, struct net_device, dev)
 

include/net/udp_tunnel.h

@@ -115,6 +115,7 @@ struct udp_tunnel_info {
 	unsigned short type;
 	sa_family_t sa_family;
 	__be16 port;
+	u8 hw_priv;
 };
 
 /* Notify network devices of offloadable types */
@@ -181,4 +182,140 @@ static inline void udp_tunnel_encap_enable(struct socket *sock)
 		udp_encap_enable();
 }
 
+#define UDP_TUNNEL_NIC_MAX_TABLES	4
+
+enum udp_tunnel_nic_info_flags {
+	/* Device callbacks may sleep */
+	UDP_TUNNEL_NIC_INFO_MAY_SLEEP	= BIT(0),
+	/* Device only supports offloads when it's open, all ports
+	 * will be removed before close and re-added after open.
+	 */
+	UDP_TUNNEL_NIC_INFO_OPEN_ONLY	= BIT(1),
+	/* Device supports only IPv4 tunnels */
+	UDP_TUNNEL_NIC_INFO_IPV4_ONLY	= BIT(2),
+};
+
+/**
+ * struct udp_tunnel_nic_info - driver UDP tunnel offload information
+ * @set_port:	callback for adding a new port
+ * @unset_port:	callback for removing a port
+ * @sync_table:	callback for syncing the entire port table at once
+ * @flags:	device flags from enum udp_tunnel_nic_info_flags
+ * @tables:	UDP port tables this device has
+ * @tables.n_entries:		number of entries in this table
+ * @tables.tunnel_types:	types of tunnels this table accepts
+ *
+ * Drivers are expected to provide either @set_port and @unset_port callbacks
+ * or the @sync_table callback. Callbacks are invoked with rtnl lock held.
+ *
+ * Known limitations:
+ *  - UDP tunnel port notifications are fundamentally best-effort -
+ *    it is likely the driver will both see skbs which use a UDP tunnel port,
+ *    while not being a tunneled skb, and tunnel skbs from other ports -
+ *    drivers should only use these ports for non-critical RX-side offloads,
+ *    e.g. the checksum offload;
+ *  - none of the devices care about the socket family at present, so we don't
+ *    track it. Please extend this code if you care.
+ */
+struct udp_tunnel_nic_info {
+	/* one-by-one */
+	int (*set_port)(struct net_device *dev,
+			unsigned int table, unsigned int entry,
+			struct udp_tunnel_info *ti);
+	int (*unset_port)(struct net_device *dev,
+			  unsigned int table, unsigned int entry,
+			  struct udp_tunnel_info *ti);
+
+	/* all at once */
+	int (*sync_table)(struct net_device *dev, unsigned int table);
+
+	unsigned int flags;
+
+	struct udp_tunnel_nic_table_info {
+		unsigned int n_entries;
+		unsigned int tunnel_types;
+	} tables[UDP_TUNNEL_NIC_MAX_TABLES];
+};
+
+/* UDP tunnel module dependencies
+ *
+ * Tunnel drivers are expected to have a hard dependency on the udp_tunnel
+ * module. NIC drivers are not, they just attach their
+ * struct udp_tunnel_nic_info to the netdev and wait for callbacks to come.
+ * Loading a tunnel driver will cause the udp_tunnel module to be loaded
+ * and only then will all the required state structures be allocated.
+ * Since we want a weak dependency from the drivers and the core to udp_tunnel
+ * we call things through the following stubs.
+ */
+struct udp_tunnel_nic_ops {
+	void (*get_port)(struct net_device *dev, unsigned int table,
+			 unsigned int idx, struct udp_tunnel_info *ti);
+	void (*set_port_priv)(struct net_device *dev, unsigned int table,
+			      unsigned int idx, u8 priv);
+	void (*add_port)(struct net_device *dev, struct udp_tunnel_info *ti);
+	void (*del_port)(struct net_device *dev, struct udp_tunnel_info *ti);
+	void (*reset_ntf)(struct net_device *dev);
+};
+
+#ifdef CONFIG_INET
+extern const struct udp_tunnel_nic_ops *udp_tunnel_nic_ops;
+#else
+#define udp_tunnel_nic_ops	((struct udp_tunnel_nic_ops *)NULL)
+#endif
+
+static inline void
+udp_tunnel_nic_get_port(struct net_device *dev, unsigned int table,
+			unsigned int idx, struct udp_tunnel_info *ti)
+{
+	/* This helper is used from .sync_table, we indicate empty entries
+	 * by zero'ed @ti. Drivers which need to know the details of a port
+	 * when it gets deleted should use the .set_port / .unset_port
+	 * callbacks.
+	 * Zero out here, otherwise !CONFIG_INET causes uninitilized warnings.
+	 */
+	memset(ti, 0, sizeof(*ti));
+
+	if (udp_tunnel_nic_ops)
+		udp_tunnel_nic_ops->get_port(dev, table, idx, ti);
+}
+
+static inline void
+udp_tunnel_nic_set_port_priv(struct net_device *dev, unsigned int table,
+			     unsigned int idx, u8 priv)
+{
+	if (udp_tunnel_nic_ops)
+		udp_tunnel_nic_ops->set_port_priv(dev, table, idx, priv);
+}
+
+static inline void
+udp_tunnel_nic_add_port(struct net_device *dev, struct udp_tunnel_info *ti)
+{
+	if (udp_tunnel_nic_ops)
+		udp_tunnel_nic_ops->add_port(dev, ti);
+}
+
+static inline void
+udp_tunnel_nic_del_port(struct net_device *dev, struct udp_tunnel_info *ti)
+{
+	if (udp_tunnel_nic_ops)
+		udp_tunnel_nic_ops->del_port(dev, ti);
+}
+
+/**
+ * udp_tunnel_nic_reset_ntf() - device-originating reset notification
+ * @dev: network interface device structure
+ *
+ * Called by the driver to inform the core that the entire UDP tunnel port
+ * state has been lost, usually due to device reset. Core will assume device
+ * forgot all the ports and issue .set_port and .sync_table callbacks as
+ * necessary.
+ *
+ * This function must be called with rtnl lock held, and will issue all
+ * the callbacks before returning.
+ */
+static inline void udp_tunnel_nic_reset_ntf(struct net_device *dev)
+{
+	if (udp_tunnel_nic_ops)
+		udp_tunnel_nic_ops->reset_ntf(dev);
+}
 #endif

net/ipv4/Makefile

@@ -14,7 +14,7 @@ obj-y     := route.o inetpeer.o protocol.o \
 	     udp_offload.o arp.o icmp.o devinet.o af_inet.o igmp.o \
 	     fib_frontend.o fib_semantics.o fib_trie.o fib_notifier.o \
 	     inet_fragment.o ping.o ip_tunnel_core.o gre_offload.o \
-	     metrics.o netlink.o nexthop.o
+	     metrics.o netlink.o nexthop.o udp_tunnel_stub.o
 
 obj-$(CONFIG_BPFILTER) += bpfilter/
 
@@ -29,6 +29,7 @@ gre-y := gre_demux.o
 obj-$(CONFIG_NET_FOU) += fou.o
 obj-$(CONFIG_NET_IPGRE_DEMUX) += gre.o
 obj-$(CONFIG_NET_IPGRE) += ip_gre.o
+udp_tunnel-y := udp_tunnel_core.o udp_tunnel_nic.o
 obj-$(CONFIG_NET_UDP_TUNNEL) += udp_tunnel.o
 obj-$(CONFIG_NET_IPVTI) += ip_vti.o
 obj-$(CONFIG_SYN_COOKIES) += syncookies.o