xref: /linux/include/net/udp_tunnel.h (revision 8e3ed5440b0c305dcd1d5fa7419bd8066d22ef42)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __NET_UDP_TUNNEL_H
3 #define __NET_UDP_TUNNEL_H
4 
5 #include <net/ip_tunnels.h>
6 #include <net/udp.h>
7 
8 #if IS_ENABLED(CONFIG_IPV6)
9 #include <net/ipv6.h>
10 #include <net/ipv6_stubs.h>
11 #endif
12 
13 struct udp_port_cfg {
14 	u8			family;
15 
16 	/* Used only for kernel-created sockets */
17 	union {
18 		struct in_addr		local_ip;
19 #if IS_ENABLED(CONFIG_IPV6)
20 		struct in6_addr		local_ip6;
21 #endif
22 	};
23 
24 	union {
25 		struct in_addr		peer_ip;
26 #if IS_ENABLED(CONFIG_IPV6)
27 		struct in6_addr		peer_ip6;
28 #endif
29 	};
30 
31 	__be16			local_udp_port;
32 	__be16			peer_udp_port;
33 	int			bind_ifindex;
34 	unsigned int		use_udp_checksums:1,
35 				use_udp6_tx_checksums:1,
36 				use_udp6_rx_checksums:1,
37 				ipv6_v6only:1;
38 };
39 
40 int udp_sock_create4(struct net *net, struct udp_port_cfg *cfg,
41 		     struct socket **sockp);
42 
43 #if IS_ENABLED(CONFIG_IPV6)
44 int udp_sock_create6(struct net *net, struct udp_port_cfg *cfg,
45 		     struct socket **sockp);
46 #else
47 static inline int udp_sock_create6(struct net *net, struct udp_port_cfg *cfg,
48 				   struct socket **sockp)
49 {
50 	return 0;
51 }
52 #endif
53 
54 static inline int udp_sock_create(struct net *net,
55 				  struct udp_port_cfg *cfg,
56 				  struct socket **sockp)
57 {
58 	if (cfg->family == AF_INET)
59 		return udp_sock_create4(net, cfg, sockp);
60 
61 	if (cfg->family == AF_INET6)
62 		return udp_sock_create6(net, cfg, sockp);
63 
64 	return -EPFNOSUPPORT;
65 }
66 
67 typedef int (*udp_tunnel_encap_rcv_t)(struct sock *sk, struct sk_buff *skb);
68 typedef int (*udp_tunnel_encap_err_lookup_t)(struct sock *sk,
69 					     struct sk_buff *skb);
70 typedef void (*udp_tunnel_encap_err_rcv_t)(struct sock *sk,
71 					   struct sk_buff *skb, int err,
72 					   __be16 port, u32 info, u8 *payload);
73 typedef void (*udp_tunnel_encap_destroy_t)(struct sock *sk);
74 typedef struct sk_buff *(*udp_tunnel_gro_receive_t)(struct sock *sk,
75 						    struct list_head *head,
76 						    struct sk_buff *skb);
77 typedef int (*udp_tunnel_gro_complete_t)(struct sock *sk, struct sk_buff *skb,
78 					 int nhoff);
79 
80 struct udp_tunnel_sock_cfg {
81 	void *sk_user_data;     /* user data used by encap_rcv call back */
82 	/* Used for setting up udp_sock fields, see udp.h for details */
83 	__u8  encap_type;
84 	udp_tunnel_encap_rcv_t encap_rcv;
85 	udp_tunnel_encap_err_lookup_t encap_err_lookup;
86 	udp_tunnel_encap_err_rcv_t encap_err_rcv;
87 	udp_tunnel_encap_destroy_t encap_destroy;
88 	udp_tunnel_gro_receive_t gro_receive;
89 	udp_tunnel_gro_complete_t gro_complete;
90 };
91 
92 /* Setup the given (UDP) sock to receive UDP encapsulated packets */
93 void setup_udp_tunnel_sock(struct net *net, struct socket *sock,
94 			   struct udp_tunnel_sock_cfg *sock_cfg);
95 
96 /* -- List of parsable UDP tunnel types --
97  *
98  * Adding to this list will result in serious debate.  The main issue is
99  * that this list is essentially a list of workarounds for either poorly
100  * designed tunnels, or poorly designed device offloads.
101  *
102  * The parsing supported via these types should really be used for Rx
103  * traffic only as the network stack will have already inserted offsets for
104  * the location of the headers in the skb.  In addition any ports that are
105  * pushed should be kept within the namespace without leaking to other
106  * devices such as VFs or other ports on the same device.
107  *
108  * It is strongly encouraged to use CHECKSUM_COMPLETE for Rx to avoid the
109  * need to use this for Rx checksum offload.  It should not be necessary to
110  * call this function to perform Tx offloads on outgoing traffic.
111  */
112 enum udp_parsable_tunnel_type {
113 	UDP_TUNNEL_TYPE_VXLAN	  = BIT(0), /* RFC 7348 */
114 	UDP_TUNNEL_TYPE_GENEVE	  = BIT(1), /* draft-ietf-nvo3-geneve */
115 	UDP_TUNNEL_TYPE_VXLAN_GPE = BIT(2), /* draft-ietf-nvo3-vxlan-gpe */
116 };
117 
118 struct udp_tunnel_info {
119 	unsigned short type;
120 	sa_family_t sa_family;
121 	__be16 port;
122 	u8 hw_priv;
123 };
124 
125 /* Notify network devices of offloadable types */
126 void udp_tunnel_push_rx_port(struct net_device *dev, struct socket *sock,
127 			     unsigned short type);
128 void udp_tunnel_drop_rx_port(struct net_device *dev, struct socket *sock,
129 			     unsigned short type);
130 void udp_tunnel_notify_add_rx_port(struct socket *sock, unsigned short type);
131 void udp_tunnel_notify_del_rx_port(struct socket *sock, unsigned short type);
132 
133 static inline void udp_tunnel_get_rx_info(struct net_device *dev)
134 {
135 	ASSERT_RTNL();
136 	if (!(dev->features & NETIF_F_RX_UDP_TUNNEL_PORT))
137 		return;
138 	call_netdevice_notifiers(NETDEV_UDP_TUNNEL_PUSH_INFO, dev);
139 }
140 
141 static inline void udp_tunnel_drop_rx_info(struct net_device *dev)
142 {
143 	ASSERT_RTNL();
144 	if (!(dev->features & NETIF_F_RX_UDP_TUNNEL_PORT))
145 		return;
146 	call_netdevice_notifiers(NETDEV_UDP_TUNNEL_DROP_INFO, dev);
147 }
148 
149 /* Transmit the skb using UDP encapsulation. */
150 void udp_tunnel_xmit_skb(struct rtable *rt, struct sock *sk, struct sk_buff *skb,
151 			 __be32 src, __be32 dst, __u8 tos, __u8 ttl,
152 			 __be16 df, __be16 src_port, __be16 dst_port,
153 			 bool xnet, bool nocheck);
154 
155 int udp_tunnel6_xmit_skb(struct dst_entry *dst, struct sock *sk,
156 			 struct sk_buff *skb,
157 			 struct net_device *dev,
158 			 const struct in6_addr *saddr,
159 			 const struct in6_addr *daddr,
160 			 __u8 prio, __u8 ttl, __be32 label,
161 			 __be16 src_port, __be16 dst_port, bool nocheck);
162 
163 void udp_tunnel_sock_release(struct socket *sock);
164 
165 struct rtable *udp_tunnel_dst_lookup(struct sk_buff *skb,
166 				     struct net_device *dev,
167 				     struct net *net, int oif,
168 				     __be32 *saddr,
169 				     const struct ip_tunnel_key *key,
170 				     __be16 sport, __be16 dport, u8 tos,
171 				     struct dst_cache *dst_cache);
172 struct dst_entry *udp_tunnel6_dst_lookup(struct sk_buff *skb,
173 					 struct net_device *dev,
174 					 struct net *net,
175 					 struct socket *sock, int oif,
176 					 struct in6_addr *saddr,
177 					 const struct ip_tunnel_key *key,
178 					 __be16 sport, __be16 dport, u8 dsfield,
179 					 struct dst_cache *dst_cache);
180 
181 struct metadata_dst *udp_tun_rx_dst(struct sk_buff *skb, unsigned short family,
182 				    __be16 flags, __be64 tunnel_id,
183 				    int md_size);
184 
185 #ifdef CONFIG_INET
186 static inline int udp_tunnel_handle_offloads(struct sk_buff *skb, bool udp_csum)
187 {
188 	int type = udp_csum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
189 
190 	return iptunnel_handle_offloads(skb, type);
191 }
192 #endif
193 
194 static inline void udp_tunnel_encap_enable(struct sock *sk)
195 {
196 	if (udp_test_and_set_bit(ENCAP_ENABLED, sk))
197 		return;
198 
199 #if IS_ENABLED(CONFIG_IPV6)
200 	if (READ_ONCE(sk->sk_family) == PF_INET6)
201 		ipv6_stub->udpv6_encap_enable();
202 #endif
203 	udp_encap_enable();
204 }
205 
206 #define UDP_TUNNEL_NIC_MAX_TABLES	4
207 
208 enum udp_tunnel_nic_info_flags {
209 	/* Device callbacks may sleep */
210 	UDP_TUNNEL_NIC_INFO_MAY_SLEEP	= BIT(0),
211 	/* Device only supports offloads when it's open, all ports
212 	 * will be removed before close and re-added after open.
213 	 */
214 	UDP_TUNNEL_NIC_INFO_OPEN_ONLY	= BIT(1),
215 	/* Device supports only IPv4 tunnels */
216 	UDP_TUNNEL_NIC_INFO_IPV4_ONLY	= BIT(2),
217 	/* Device has hard-coded the IANA VXLAN port (4789) as VXLAN.
218 	 * This port must not be counted towards n_entries of any table.
219 	 * Driver will not receive any callback associated with port 4789.
220 	 */
221 	UDP_TUNNEL_NIC_INFO_STATIC_IANA_VXLAN	= BIT(3),
222 };
223 
224 struct udp_tunnel_nic;
225 
226 #define UDP_TUNNEL_NIC_MAX_SHARING_DEVICES	(U16_MAX / 2)
227 
228 struct udp_tunnel_nic_shared {
229 	struct udp_tunnel_nic *udp_tunnel_nic_info;
230 
231 	struct list_head devices;
232 };
233 
234 struct udp_tunnel_nic_shared_node {
235 	struct net_device *dev;
236 	struct list_head list;
237 };
238 
239 /**
240  * struct udp_tunnel_nic_info - driver UDP tunnel offload information
241  * @set_port:	callback for adding a new port
242  * @unset_port:	callback for removing a port
243  * @sync_table:	callback for syncing the entire port table at once
244  * @shared:	reference to device global state (optional)
245  * @flags:	device flags from enum udp_tunnel_nic_info_flags
246  * @tables:	UDP port tables this device has
247  * @tables.n_entries:		number of entries in this table
248  * @tables.tunnel_types:	types of tunnels this table accepts
249  *
250  * Drivers are expected to provide either @set_port and @unset_port callbacks
251  * or the @sync_table callback. Callbacks are invoked with rtnl lock held.
252  *
253  * Devices which (misguidedly) share the UDP tunnel port table across multiple
254  * netdevs should allocate an instance of struct udp_tunnel_nic_shared and
255  * point @shared at it.
256  * There must never be more than %UDP_TUNNEL_NIC_MAX_SHARING_DEVICES devices
257  * sharing a table.
258  *
259  * Known limitations:
260  *  - UDP tunnel port notifications are fundamentally best-effort -
261  *    it is likely the driver will both see skbs which use a UDP tunnel port,
262  *    while not being a tunneled skb, and tunnel skbs from other ports -
263  *    drivers should only use these ports for non-critical RX-side offloads,
264  *    e.g. the checksum offload;
265  *  - none of the devices care about the socket family at present, so we don't
266  *    track it. Please extend this code if you care.
267  */
268 struct udp_tunnel_nic_info {
269 	/* one-by-one */
270 	int (*set_port)(struct net_device *dev,
271 			unsigned int table, unsigned int entry,
272 			struct udp_tunnel_info *ti);
273 	int (*unset_port)(struct net_device *dev,
274 			  unsigned int table, unsigned int entry,
275 			  struct udp_tunnel_info *ti);
276 
277 	/* all at once */
278 	int (*sync_table)(struct net_device *dev, unsigned int table);
279 
280 	struct udp_tunnel_nic_shared *shared;
281 
282 	unsigned int flags;
283 
284 	struct udp_tunnel_nic_table_info {
285 		unsigned int n_entries;
286 		unsigned int tunnel_types;
287 	} tables[UDP_TUNNEL_NIC_MAX_TABLES];
288 };
289 
290 /* UDP tunnel module dependencies
291  *
292  * Tunnel drivers are expected to have a hard dependency on the udp_tunnel
293  * module. NIC drivers are not, they just attach their
294  * struct udp_tunnel_nic_info to the netdev and wait for callbacks to come.
295  * Loading a tunnel driver will cause the udp_tunnel module to be loaded
296  * and only then will all the required state structures be allocated.
297  * Since we want a weak dependency from the drivers and the core to udp_tunnel
298  * we call things through the following stubs.
299  */
300 struct udp_tunnel_nic_ops {
301 	void (*get_port)(struct net_device *dev, unsigned int table,
302 			 unsigned int idx, struct udp_tunnel_info *ti);
303 	void (*set_port_priv)(struct net_device *dev, unsigned int table,
304 			      unsigned int idx, u8 priv);
305 	void (*add_port)(struct net_device *dev, struct udp_tunnel_info *ti);
306 	void (*del_port)(struct net_device *dev, struct udp_tunnel_info *ti);
307 	void (*reset_ntf)(struct net_device *dev);
308 
309 	size_t (*dump_size)(struct net_device *dev, unsigned int table);
310 	int (*dump_write)(struct net_device *dev, unsigned int table,
311 			  struct sk_buff *skb);
312 };
313 
314 #ifdef CONFIG_INET
315 extern const struct udp_tunnel_nic_ops *udp_tunnel_nic_ops;
316 #else
317 #define udp_tunnel_nic_ops	((struct udp_tunnel_nic_ops *)NULL)
318 #endif
319 
320 static inline void
321 udp_tunnel_nic_get_port(struct net_device *dev, unsigned int table,
322 			unsigned int idx, struct udp_tunnel_info *ti)
323 {
324 	/* This helper is used from .sync_table, we indicate empty entries
325 	 * by zero'ed @ti. Drivers which need to know the details of a port
326 	 * when it gets deleted should use the .set_port / .unset_port
327 	 * callbacks.
328 	 * Zero out here, otherwise !CONFIG_INET causes uninitilized warnings.
329 	 */
330 	memset(ti, 0, sizeof(*ti));
331 
332 	if (udp_tunnel_nic_ops)
333 		udp_tunnel_nic_ops->get_port(dev, table, idx, ti);
334 }
335 
336 static inline void
337 udp_tunnel_nic_set_port_priv(struct net_device *dev, unsigned int table,
338 			     unsigned int idx, u8 priv)
339 {
340 	if (udp_tunnel_nic_ops)
341 		udp_tunnel_nic_ops->set_port_priv(dev, table, idx, priv);
342 }
343 
344 static inline void
345 udp_tunnel_nic_add_port(struct net_device *dev, struct udp_tunnel_info *ti)
346 {
347 	if (!(dev->features & NETIF_F_RX_UDP_TUNNEL_PORT))
348 		return;
349 	if (udp_tunnel_nic_ops)
350 		udp_tunnel_nic_ops->add_port(dev, ti);
351 }
352 
353 static inline void
354 udp_tunnel_nic_del_port(struct net_device *dev, struct udp_tunnel_info *ti)
355 {
356 	if (!(dev->features & NETIF_F_RX_UDP_TUNNEL_PORT))
357 		return;
358 	if (udp_tunnel_nic_ops)
359 		udp_tunnel_nic_ops->del_port(dev, ti);
360 }
361 
362 /**
363  * udp_tunnel_nic_reset_ntf() - device-originating reset notification
364  * @dev: network interface device structure
365  *
366  * Called by the driver to inform the core that the entire UDP tunnel port
367  * state has been lost, usually due to device reset. Core will assume device
368  * forgot all the ports and issue .set_port and .sync_table callbacks as
369  * necessary.
370  *
371  * This function must be called with rtnl lock held, and will issue all
372  * the callbacks before returning.
373  */
374 static inline void udp_tunnel_nic_reset_ntf(struct net_device *dev)
375 {
376 	if (udp_tunnel_nic_ops)
377 		udp_tunnel_nic_ops->reset_ntf(dev);
378 }
379 
380 static inline size_t
381 udp_tunnel_nic_dump_size(struct net_device *dev, unsigned int table)
382 {
383 	if (!udp_tunnel_nic_ops)
384 		return 0;
385 	return udp_tunnel_nic_ops->dump_size(dev, table);
386 }
387 
388 static inline int
389 udp_tunnel_nic_dump_write(struct net_device *dev, unsigned int table,
390 			  struct sk_buff *skb)
391 {
392 	if (!udp_tunnel_nic_ops)
393 		return 0;
394 	return udp_tunnel_nic_ops->dump_write(dev, table, skb);
395 }
396 #endif
397