xref: /linux/include/net/udp_tunnel.h (revision dfecb0c5af3b07ebfa84be63a7a21bfc9e29a872)
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 #endif
11 
12 #define UDP_TUNNEL_PARTIAL_FEATURES	NETIF_F_GSO_ENCAP_ALL
13 #define UDP_TUNNEL_STRIPPED_GSO_TYPES	((UDP_TUNNEL_PARTIAL_FEATURES |	\
14 					  NETIF_F_GSO_PARTIAL) >>	\
15 					 NETIF_F_GSO_SHIFT)
16 
17 struct udp_port_cfg {
18 	u8			family;
19 
20 	/* Used only for kernel-created sockets */
21 	union {
22 		struct in_addr		local_ip;
23 #if IS_ENABLED(CONFIG_IPV6)
24 		struct in6_addr		local_ip6;
25 #endif
26 	};
27 
28 	union {
29 		struct in_addr		peer_ip;
30 #if IS_ENABLED(CONFIG_IPV6)
31 		struct in6_addr		peer_ip6;
32 #endif
33 	};
34 
35 	__be16			local_udp_port;
36 	__be16			peer_udp_port;
37 	int			bind_ifindex;
38 	unsigned int		use_udp_checksums:1,
39 				use_udp6_tx_checksums:1,
40 				use_udp6_rx_checksums:1,
41 				ipv6_v6only:1;
42 };
43 
44 int udp_sock_create4(struct net *net, struct udp_port_cfg *cfg,
45 		     struct socket **sockp);
46 
47 #if IS_ENABLED(CONFIG_IPV6)
48 int udp_sock_create6(struct net *net, struct udp_port_cfg *cfg,
49 		     struct socket **sockp);
50 #else
51 static inline int udp_sock_create6(struct net *net, struct udp_port_cfg *cfg,
52 				   struct socket **sockp)
53 {
54 	return -EPFNOSUPPORT;
55 }
56 #endif
57 
58 static inline int udp_sock_create(struct net *net,
59 				  struct udp_port_cfg *cfg,
60 				  struct socket **sockp)
61 {
62 	if (cfg->family == AF_INET)
63 		return udp_sock_create4(net, cfg, sockp);
64 
65 	if (cfg->family == AF_INET6)
66 		return udp_sock_create6(net, cfg, sockp);
67 
68 	return -EPFNOSUPPORT;
69 }
70 
71 typedef int (*udp_tunnel_encap_rcv_t)(struct sock *sk, struct sk_buff *skb);
72 typedef int (*udp_tunnel_encap_err_lookup_t)(struct sock *sk,
73 					     struct sk_buff *skb);
74 typedef void (*udp_tunnel_encap_err_rcv_t)(struct sock *sk,
75 					   struct sk_buff *skb, int err,
76 					   __be16 port, u32 info, u8 *payload);
77 typedef void (*udp_tunnel_encap_destroy_t)(struct sock *sk);
78 typedef struct sk_buff *(*udp_tunnel_gro_receive_t)(struct sock *sk,
79 						    struct list_head *head,
80 						    struct sk_buff *skb);
81 typedef int (*udp_tunnel_gro_complete_t)(struct sock *sk, struct sk_buff *skb,
82 					 int nhoff);
83 
84 struct udp_tunnel_sock_cfg {
85 	void *sk_user_data;     /* user data used by encap_rcv call back */
86 	/* Used for setting up udp_sock fields, see udp.h for details */
87 	__u8  encap_type;
88 	udp_tunnel_encap_rcv_t encap_rcv;
89 	udp_tunnel_encap_err_lookup_t encap_err_lookup;
90 	udp_tunnel_encap_err_rcv_t encap_err_rcv;
91 	udp_tunnel_encap_destroy_t encap_destroy;
92 	udp_tunnel_gro_receive_t gro_receive;
93 	udp_tunnel_gro_complete_t gro_complete;
94 };
95 
96 /* Setup the given (UDP) sock to receive UDP encapsulated packets */
97 void setup_udp_tunnel_sock(struct net *net, struct socket *sock,
98 			   struct udp_tunnel_sock_cfg *sock_cfg);
99 
100 /* -- List of parsable UDP tunnel types --
101  *
102  * Adding to this list will result in serious debate.  The main issue is
103  * that this list is essentially a list of workarounds for either poorly
104  * designed tunnels, or poorly designed device offloads.
105  *
106  * The parsing supported via these types should really be used for Rx
107  * traffic only as the network stack will have already inserted offsets for
108  * the location of the headers in the skb.  In addition any ports that are
109  * pushed should be kept within the namespace without leaking to other
110  * devices such as VFs or other ports on the same device.
111  *
112  * It is strongly encouraged to use CHECKSUM_COMPLETE for Rx to avoid the
113  * need to use this for Rx checksum offload.  It should not be necessary to
114  * call this function to perform Tx offloads on outgoing traffic.
115  */
116 enum udp_parsable_tunnel_type {
117 	UDP_TUNNEL_TYPE_VXLAN	  = BIT(0), /* RFC 7348 */
118 	UDP_TUNNEL_TYPE_GENEVE	  = BIT(1), /* draft-ietf-nvo3-geneve */
119 	UDP_TUNNEL_TYPE_VXLAN_GPE = BIT(2), /* draft-ietf-nvo3-vxlan-gpe */
120 };
121 
122 struct udp_tunnel_info {
123 	unsigned short type;
124 	sa_family_t sa_family;
125 	__be16 port;
126 	u8 hw_priv;
127 };
128 
129 /* Notify network devices of offloadable types */
130 void udp_tunnel_push_rx_port(struct net_device *dev, struct socket *sock,
131 			     unsigned short type);
132 void udp_tunnel_drop_rx_port(struct net_device *dev, struct socket *sock,
133 			     unsigned short type);
134 void udp_tunnel_notify_add_rx_port(struct socket *sock, unsigned short type);
135 void udp_tunnel_notify_del_rx_port(struct socket *sock, unsigned short type);
136 
137 /* Transmit the skb using UDP encapsulation. */
138 void udp_tunnel_xmit_skb(struct rtable *rt, struct sock *sk, struct sk_buff *skb,
139 			 __be32 src, __be32 dst, __u8 tos, __u8 ttl,
140 			 __be16 df, __be16 src_port, __be16 dst_port,
141 			 bool xnet, bool nocheck, u16 ipcb_flags);
142 
143 void udp_tunnel6_xmit_skb(struct dst_entry *dst, struct sock *sk,
144 			  struct sk_buff *skb,
145 			  struct net_device *dev,
146 			  const struct in6_addr *saddr,
147 			  const struct in6_addr *daddr,
148 			  __u8 prio, __u8 ttl, __be32 label,
149 			  __be16 src_port, __be16 dst_port, bool nocheck,
150 			  u16 ip6cb_flags);
151 
152 static inline bool udp_tunnel_handle_partial(struct sk_buff *skb)
153 {
154 	bool double_encap = !!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL);
155 
156 	/*
157 	 * If the skb went through partial segmentation, lower devices
158 	 * will not need to offload the related features - except for
159 	 * UDP_TUNNEL, that will be re-added by the later
160 	 * udp_tunnel_handle_offloads().
161 	 */
162 	if (double_encap)
163 		skb_shinfo(skb)->gso_type &= ~UDP_TUNNEL_STRIPPED_GSO_TYPES;
164 	return double_encap;
165 }
166 
167 static inline void udp_tunnel_set_inner_protocol(struct sk_buff *skb,
168 						 bool double_encap,
169 						 __be16 inner_proto)
170 {
171 	/*
172 	 * The inner protocol has been set by the nested tunnel, don't
173 	 * overraid it.
174 	 */
175 	if (!double_encap)
176 		skb_set_inner_protocol(skb, inner_proto);
177 }
178 
179 void udp_tunnel_sock_release(struct socket *sock);
180 
181 struct rtable *udp_tunnel_dst_lookup(struct sk_buff *skb,
182 				     struct net_device *dev,
183 				     struct net *net, int oif,
184 				     __be32 *saddr,
185 				     const struct ip_tunnel_key *key,
186 				     __be16 sport, __be16 dport, u8 tos,
187 				     struct dst_cache *dst_cache);
188 struct dst_entry *udp_tunnel6_dst_lookup(struct sk_buff *skb,
189 					 struct net_device *dev,
190 					 struct net *net,
191 					 struct socket *sock, int oif,
192 					 struct in6_addr *saddr,
193 					 const struct ip_tunnel_key *key,
194 					 __be16 sport, __be16 dport, u8 dsfield,
195 					 struct dst_cache *dst_cache);
196 
197 struct metadata_dst *udp_tun_rx_dst(struct sk_buff *skb, unsigned short family,
198 				    const unsigned long *flags,
199 				    __be64 tunnel_id, int md_size);
200 
201 #ifdef CONFIG_INET
202 static inline int udp_tunnel_handle_offloads(struct sk_buff *skb, bool udp_csum)
203 {
204 	int type = udp_csum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
205 
206 	return iptunnel_handle_offloads(skb, type);
207 }
208 #endif
209 
210 #if IS_ENABLED(CONFIG_NET_UDP_TUNNEL)
211 void udp_tunnel_update_gro_lookup(struct net *net, struct sock *sk, bool add);
212 void udp_tunnel_update_gro_rcv(struct sock *sk, bool add);
213 #else
214 static inline void udp_tunnel_update_gro_lookup(struct net *net,
215 						struct sock *sk, bool add) {}
216 static inline void udp_tunnel_update_gro_rcv(struct sock *sk, bool add) {}
217 #endif
218 
219 static inline void udp_tunnel_cleanup_gro(struct sock *sk)
220 {
221 	udp_tunnel_update_gro_rcv(sk, false);
222 	udp_tunnel_update_gro_lookup(sock_net(sk), sk, false);
223 }
224 
225 static inline void udp_tunnel_encap_enable(struct sock *sk)
226 {
227 	if (udp_test_and_set_bit(ENCAP_ENABLED, sk))
228 		return;
229 
230 #if IS_ENABLED(CONFIG_IPV6)
231 	if (READ_ONCE(sk->sk_family) == PF_INET6)
232 		udpv6_encap_enable();
233 #endif
234 	udp_encap_enable();
235 }
236 
237 #define UDP_TUNNEL_NIC_MAX_TABLES	4
238 
239 enum udp_tunnel_nic_info_flags {
240 	/* Device only supports offloads when it's open, all ports
241 	 * will be removed before close and re-added after open.
242 	 */
243 	UDP_TUNNEL_NIC_INFO_OPEN_ONLY	= BIT(0),
244 	/* Device supports only IPv4 tunnels */
245 	UDP_TUNNEL_NIC_INFO_IPV4_ONLY	= BIT(1),
246 	/* Device has hard-coded the IANA VXLAN port (4789) as VXLAN.
247 	 * This port must not be counted towards n_entries of any table.
248 	 * Driver will not receive any callback associated with port 4789.
249 	 */
250 	UDP_TUNNEL_NIC_INFO_STATIC_IANA_VXLAN	= BIT(2),
251 };
252 
253 struct udp_tunnel_nic;
254 
255 #define UDP_TUNNEL_NIC_MAX_SHARING_DEVICES	(U16_MAX / 2)
256 
257 struct udp_tunnel_nic_shared {
258 	struct udp_tunnel_nic *udp_tunnel_nic_info;
259 
260 	struct list_head devices;
261 };
262 
263 struct udp_tunnel_nic_shared_node {
264 	struct net_device *dev;
265 	struct list_head list;
266 };
267 
268 /**
269  * struct udp_tunnel_nic_info - driver UDP tunnel offload information
270  * @set_port:	callback for adding a new port
271  * @unset_port:	callback for removing a port
272  * @sync_table:	callback for syncing the entire port table at once
273  * @shared:	reference to device global state (optional)
274  * @flags:	device flags from enum udp_tunnel_nic_info_flags
275  * @tables:	UDP port tables this device has
276  * @tables.n_entries:		number of entries in this table
277  * @tables.tunnel_types:	types of tunnels this table accepts
278  *
279  * Drivers are expected to provide either @set_port and @unset_port callbacks
280  * or the @sync_table callback. Callbacks are invoked with rtnl lock held.
281  *
282  * Devices which (misguidedly) share the UDP tunnel port table across multiple
283  * netdevs should allocate an instance of struct udp_tunnel_nic_shared and
284  * point @shared at it.
285  * There must never be more than %UDP_TUNNEL_NIC_MAX_SHARING_DEVICES devices
286  * sharing a table.
287  *
288  * Known limitations:
289  *  - UDP tunnel port notifications are fundamentally best-effort -
290  *    it is likely the driver will both see skbs which use a UDP tunnel port,
291  *    while not being a tunneled skb, and tunnel skbs from other ports -
292  *    drivers should only use these ports for non-critical RX-side offloads,
293  *    e.g. the checksum offload;
294  *  - none of the devices care about the socket family at present, so we don't
295  *    track it. Please extend this code if you care.
296  */
297 struct udp_tunnel_nic_info {
298 	/* one-by-one */
299 	int (*set_port)(struct net_device *dev,
300 			unsigned int table, unsigned int entry,
301 			struct udp_tunnel_info *ti);
302 	int (*unset_port)(struct net_device *dev,
303 			  unsigned int table, unsigned int entry,
304 			  struct udp_tunnel_info *ti);
305 
306 	/* all at once */
307 	int (*sync_table)(struct net_device *dev, unsigned int table);
308 
309 	struct udp_tunnel_nic_shared *shared;
310 
311 	unsigned int flags;
312 
313 	struct udp_tunnel_nic_table_info {
314 		unsigned int n_entries;
315 		unsigned int tunnel_types;
316 	} tables[UDP_TUNNEL_NIC_MAX_TABLES];
317 };
318 
319 /* UDP tunnel module dependencies
320  *
321  * Tunnel drivers are expected to have a hard dependency on the udp_tunnel
322  * module. NIC drivers are not, they just attach their
323  * struct udp_tunnel_nic_info to the netdev and wait for callbacks to come.
324  * Loading a tunnel driver will cause the udp_tunnel module to be loaded
325  * and only then will all the required state structures be allocated.
326  * Since we want a weak dependency from the drivers and the core to udp_tunnel
327  * we call things through the following stubs.
328  */
329 struct udp_tunnel_nic_ops {
330 	void (*get_port)(struct net_device *dev, unsigned int table,
331 			 unsigned int idx, struct udp_tunnel_info *ti);
332 	void (*set_port_priv)(struct net_device *dev, unsigned int table,
333 			      unsigned int idx, u8 priv);
334 	void (*add_port)(struct net_device *dev, struct udp_tunnel_info *ti);
335 	void (*del_port)(struct net_device *dev, struct udp_tunnel_info *ti);
336 	void (*reset_ntf)(struct net_device *dev);
337 
338 	size_t (*dump_size)(struct net_device *dev, unsigned int table);
339 	int (*dump_write)(struct net_device *dev, unsigned int table,
340 			  struct sk_buff *skb);
341 	void (*assert_locked)(struct net_device *dev);
342 	void (*lock)(struct net_device *dev);
343 	void (*unlock)(struct net_device *dev);
344 };
345 
346 #ifdef CONFIG_INET
347 extern const struct udp_tunnel_nic_ops *udp_tunnel_nic_ops;
348 #else
349 #define udp_tunnel_nic_ops	((struct udp_tunnel_nic_ops *)NULL)
350 #endif
351 
352 static inline void
353 udp_tunnel_nic_get_port(struct net_device *dev, unsigned int table,
354 			unsigned int idx, struct udp_tunnel_info *ti)
355 {
356 	/* This helper is used from .sync_table, we indicate empty entries
357 	 * by zero'ed @ti. Drivers which need to know the details of a port
358 	 * when it gets deleted should use the .set_port / .unset_port
359 	 * callbacks.
360 	 * Zero out here, otherwise !CONFIG_INET causes uninitilized warnings.
361 	 */
362 	memset(ti, 0, sizeof(*ti));
363 
364 	if (udp_tunnel_nic_ops)
365 		udp_tunnel_nic_ops->get_port(dev, table, idx, ti);
366 }
367 
368 static inline void
369 udp_tunnel_nic_set_port_priv(struct net_device *dev, unsigned int table,
370 			     unsigned int idx, u8 priv)
371 {
372 	if (udp_tunnel_nic_ops) {
373 		udp_tunnel_nic_ops->assert_locked(dev);
374 		udp_tunnel_nic_ops->set_port_priv(dev, table, idx, priv);
375 	}
376 }
377 
378 static inline void udp_tunnel_nic_assert_locked(struct net_device *dev)
379 {
380 	if (udp_tunnel_nic_ops)
381 		udp_tunnel_nic_ops->assert_locked(dev);
382 }
383 
384 static inline void udp_tunnel_nic_lock(struct net_device *dev)
385 {
386 	if (udp_tunnel_nic_ops)
387 		udp_tunnel_nic_ops->lock(dev);
388 }
389 
390 static inline void udp_tunnel_nic_unlock(struct net_device *dev)
391 {
392 	if (udp_tunnel_nic_ops)
393 		udp_tunnel_nic_ops->unlock(dev);
394 }
395 
396 static inline void
397 udp_tunnel_nic_add_port(struct net_device *dev, struct udp_tunnel_info *ti)
398 {
399 	if (!(dev->features & NETIF_F_RX_UDP_TUNNEL_PORT))
400 		return;
401 	if (udp_tunnel_nic_ops)
402 		udp_tunnel_nic_ops->add_port(dev, ti);
403 }
404 
405 static inline void
406 udp_tunnel_nic_del_port(struct net_device *dev, struct udp_tunnel_info *ti)
407 {
408 	if (!(dev->features & NETIF_F_RX_UDP_TUNNEL_PORT))
409 		return;
410 	if (udp_tunnel_nic_ops)
411 		udp_tunnel_nic_ops->del_port(dev, ti);
412 }
413 
414 /**
415  * udp_tunnel_nic_reset_ntf() - device-originating reset notification
416  * @dev: network interface device structure
417  *
418  * Called by the driver to inform the core that the entire UDP tunnel port
419  * state has been lost, usually due to device reset. Core will assume device
420  * forgot all the ports and issue .set_port and .sync_table callbacks as
421  * necessary.
422  *
423  * This function must be called with rtnl lock held, and will issue all
424  * the callbacks before returning.
425  */
426 static inline void udp_tunnel_nic_reset_ntf(struct net_device *dev)
427 {
428 	if (udp_tunnel_nic_ops)
429 		udp_tunnel_nic_ops->reset_ntf(dev);
430 }
431 
432 static inline size_t
433 udp_tunnel_nic_dump_size(struct net_device *dev, unsigned int table)
434 {
435 	size_t ret;
436 
437 	if (!udp_tunnel_nic_ops)
438 		return 0;
439 
440 	udp_tunnel_nic_ops->lock(dev);
441 	ret = udp_tunnel_nic_ops->dump_size(dev, table);
442 	udp_tunnel_nic_ops->unlock(dev);
443 
444 	return ret;
445 }
446 
447 static inline int
448 udp_tunnel_nic_dump_write(struct net_device *dev, unsigned int table,
449 			  struct sk_buff *skb)
450 {
451 	int ret;
452 
453 	if (!udp_tunnel_nic_ops)
454 		return 0;
455 
456 	udp_tunnel_nic_ops->lock(dev);
457 	ret = udp_tunnel_nic_ops->dump_write(dev, table, skb);
458 	udp_tunnel_nic_ops->unlock(dev);
459 
460 	return ret;
461 }
462 
463 static inline void udp_tunnel_get_rx_info(struct net_device *dev)
464 {
465 	ASSERT_RTNL();
466 	if (!(dev->features & NETIF_F_RX_UDP_TUNNEL_PORT))
467 		return;
468 	udp_tunnel_nic_assert_locked(dev);
469 	call_netdevice_notifiers(NETDEV_UDP_TUNNEL_PUSH_INFO, dev);
470 }
471 
472 static inline void udp_tunnel_drop_rx_info(struct net_device *dev)
473 {
474 	ASSERT_RTNL();
475 	if (!(dev->features & NETIF_F_RX_UDP_TUNNEL_PORT))
476 		return;
477 	udp_tunnel_nic_assert_locked(dev);
478 	call_netdevice_notifiers(NETDEV_UDP_TUNNEL_DROP_INFO, dev);
479 }
480 
481 #endif
482