udp_tunnel.h

  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
 13struct 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
 40int udp_sock_create4(struct net *net, struct udp_port_cfg *cfg,
 41		     struct socket **sockp);
 42
 43#if IS_ENABLED(CONFIG_IPV6)
 44int udp_sock_create6(struct net *net, struct udp_port_cfg *cfg,
 45		     struct socket **sockp);
 46#else
 47static inline int udp_sock_create6(struct net *net, struct udp_port_cfg *cfg,
 48				   struct socket **sockp)
 49{
 50	return -EPFNOSUPPORT;
 51}
 52#endif
 53
 54static 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
 67typedef int (*udp_tunnel_encap_rcv_t)(struct sock *sk, struct sk_buff *skb);
 68typedef int (*udp_tunnel_encap_err_lookup_t)(struct sock *sk,
 69					     struct sk_buff *skb);
 70typedef void (*udp_tunnel_encap_err_rcv_t)(struct sock *sk,
 71					   struct sk_buff *skb, int err,
 72					   __be16 port, u32 info, u8 *payload);
 73typedef void (*udp_tunnel_encap_destroy_t)(struct sock *sk);
 74typedef struct sk_buff *(*udp_tunnel_gro_receive_t)(struct sock *sk,
 75						    struct list_head *head,
 76						    struct sk_buff *skb);
 77typedef int (*udp_tunnel_gro_complete_t)(struct sock *sk, struct sk_buff *skb,
 78					 int nhoff);
 79
 80struct 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 */
 93void 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 */
112enum 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
118struct 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 */
126void udp_tunnel_push_rx_port(struct net_device *dev, struct socket *sock,
127			     unsigned short type);
128void udp_tunnel_drop_rx_port(struct net_device *dev, struct socket *sock,
129			     unsigned short type);
130void udp_tunnel_notify_add_rx_port(struct socket *sock, unsigned short type);
131void udp_tunnel_notify_del_rx_port(struct socket *sock, unsigned short type);
132
133/* Transmit the skb using UDP encapsulation. */
134void udp_tunnel_xmit_skb(struct rtable *rt, struct sock *sk, struct sk_buff *skb,
135			 __be32 src, __be32 dst, __u8 tos, __u8 ttl,
136			 __be16 df, __be16 src_port, __be16 dst_port,
137			 bool xnet, bool nocheck, u16 ipcb_flags);
138
139void udp_tunnel6_xmit_skb(struct dst_entry *dst, struct sock *sk,
140			  struct sk_buff *skb,
141			  struct net_device *dev,
142			  const struct in6_addr *saddr,
143			  const struct in6_addr *daddr,
144			  __u8 prio, __u8 ttl, __be32 label,
145			  __be16 src_port, __be16 dst_port, bool nocheck,
146			  u16 ip6cb_flags);
147
148void udp_tunnel_sock_release(struct socket *sock);
149
150struct rtable *udp_tunnel_dst_lookup(struct sk_buff *skb,
151				     struct net_device *dev,
152				     struct net *net, int oif,
153				     __be32 *saddr,
154				     const struct ip_tunnel_key *key,
155				     __be16 sport, __be16 dport, u8 tos,
156				     struct dst_cache *dst_cache);
157struct dst_entry *udp_tunnel6_dst_lookup(struct sk_buff *skb,
158					 struct net_device *dev,
159					 struct net *net,
160					 struct socket *sock, int oif,
161					 struct in6_addr *saddr,
162					 const struct ip_tunnel_key *key,
163					 __be16 sport, __be16 dport, u8 dsfield,
164					 struct dst_cache *dst_cache);
165
166struct metadata_dst *udp_tun_rx_dst(struct sk_buff *skb, unsigned short family,
167				    const unsigned long *flags,
168				    __be64 tunnel_id, int md_size);
169
170#ifdef CONFIG_INET
171static inline int udp_tunnel_handle_offloads(struct sk_buff *skb, bool udp_csum)
172{
173	int type = udp_csum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
174
175	return iptunnel_handle_offloads(skb, type);
176}
177#endif
178
179#if IS_ENABLED(CONFIG_NET_UDP_TUNNEL)
180void udp_tunnel_update_gro_lookup(struct net *net, struct sock *sk, bool add);
181void udp_tunnel_update_gro_rcv(struct sock *sk, bool add);
182#else
183static inline void udp_tunnel_update_gro_lookup(struct net *net,
184						struct sock *sk, bool add) {}
185static inline void udp_tunnel_update_gro_rcv(struct sock *sk, bool add) {}
186#endif
187
188static inline void udp_tunnel_cleanup_gro(struct sock *sk)
189{
190	udp_tunnel_update_gro_rcv(sk, false);
191	udp_tunnel_update_gro_lookup(sock_net(sk), sk, false);
192}
193
194static 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
208enum udp_tunnel_nic_info_flags {
209	/* Device only supports offloads when it's open, all ports
210	 * will be removed before close and re-added after open.
211	 */
212	UDP_TUNNEL_NIC_INFO_OPEN_ONLY	= BIT(0),
213	/* Device supports only IPv4 tunnels */
214	UDP_TUNNEL_NIC_INFO_IPV4_ONLY	= BIT(1),
215	/* Device has hard-coded the IANA VXLAN port (4789) as VXLAN.
216	 * This port must not be counted towards n_entries of any table.
217	 * Driver will not receive any callback associated with port 4789.
218	 */
219	UDP_TUNNEL_NIC_INFO_STATIC_IANA_VXLAN	= BIT(2),
220};
221
222struct udp_tunnel_nic;
223
224#define UDP_TUNNEL_NIC_MAX_SHARING_DEVICES	(U16_MAX / 2)
225
226struct udp_tunnel_nic_shared {
227	struct udp_tunnel_nic *udp_tunnel_nic_info;
228
229	struct list_head devices;
230};
231
232struct udp_tunnel_nic_shared_node {
233	struct net_device *dev;
234	struct list_head list;
235};
236
237/**
238 * struct udp_tunnel_nic_info - driver UDP tunnel offload information
239 * @set_port:	callback for adding a new port
240 * @unset_port:	callback for removing a port
241 * @sync_table:	callback for syncing the entire port table at once
242 * @shared:	reference to device global state (optional)
243 * @flags:	device flags from enum udp_tunnel_nic_info_flags
244 * @tables:	UDP port tables this device has
245 * @tables.n_entries:		number of entries in this table
246 * @tables.tunnel_types:	types of tunnels this table accepts
247 *
248 * Drivers are expected to provide either @set_port and @unset_port callbacks
249 * or the @sync_table callback. Callbacks are invoked with rtnl lock held.
250 *
251 * Devices which (misguidedly) share the UDP tunnel port table across multiple
252 * netdevs should allocate an instance of struct udp_tunnel_nic_shared and
253 * point @shared at it.
254 * There must never be more than %UDP_TUNNEL_NIC_MAX_SHARING_DEVICES devices
255 * sharing a table.
256 *
257 * Known limitations:
258 *  - UDP tunnel port notifications are fundamentally best-effort -
259 *    it is likely the driver will both see skbs which use a UDP tunnel port,
260 *    while not being a tunneled skb, and tunnel skbs from other ports -
261 *    drivers should only use these ports for non-critical RX-side offloads,
262 *    e.g. the checksum offload;
263 *  - none of the devices care about the socket family at present, so we don't
264 *    track it. Please extend this code if you care.
265 */
266struct udp_tunnel_nic_info {
267	/* one-by-one */
268	int (*set_port)(struct net_device *dev,
269			unsigned int table, unsigned int entry,
270			struct udp_tunnel_info *ti);
271	int (*unset_port)(struct net_device *dev,
272			  unsigned int table, unsigned int entry,
273			  struct udp_tunnel_info *ti);
274
275	/* all at once */
276	int (*sync_table)(struct net_device *dev, unsigned int table);
277
278	struct udp_tunnel_nic_shared *shared;
279
280	unsigned int flags;
281
282	struct udp_tunnel_nic_table_info {
283		unsigned int n_entries;
284		unsigned int tunnel_types;
285	} tables[UDP_TUNNEL_NIC_MAX_TABLES];
286};
287
288/* UDP tunnel module dependencies
289 *
290 * Tunnel drivers are expected to have a hard dependency on the udp_tunnel
291 * module. NIC drivers are not, they just attach their
292 * struct udp_tunnel_nic_info to the netdev and wait for callbacks to come.
293 * Loading a tunnel driver will cause the udp_tunnel module to be loaded
294 * and only then will all the required state structures be allocated.
295 * Since we want a weak dependency from the drivers and the core to udp_tunnel
296 * we call things through the following stubs.
297 */
298struct udp_tunnel_nic_ops {
299	void (*get_port)(struct net_device *dev, unsigned int table,
300			 unsigned int idx, struct udp_tunnel_info *ti);
301	void (*set_port_priv)(struct net_device *dev, unsigned int table,
302			      unsigned int idx, u8 priv);
303	void (*add_port)(struct net_device *dev, struct udp_tunnel_info *ti);
304	void (*del_port)(struct net_device *dev, struct udp_tunnel_info *ti);
305	void (*reset_ntf)(struct net_device *dev);
306
307	size_t (*dump_size)(struct net_device *dev, unsigned int table);
308	int (*dump_write)(struct net_device *dev, unsigned int table,
309			  struct sk_buff *skb);
310	void (*assert_locked)(struct net_device *dev);
311	void (*lock)(struct net_device *dev);
312	void (*unlock)(struct net_device *dev);
313};
314
315#ifdef CONFIG_INET
316extern const struct udp_tunnel_nic_ops *udp_tunnel_nic_ops;
317#else
318#define udp_tunnel_nic_ops	((struct udp_tunnel_nic_ops *)NULL)
319#endif
320
321static inline void
322udp_tunnel_nic_get_port(struct net_device *dev, unsigned int table,
323			unsigned int idx, struct udp_tunnel_info *ti)
324{
325	/* This helper is used from .sync_table, we indicate empty entries
326	 * by zero'ed @ti. Drivers which need to know the details of a port
327	 * when it gets deleted should use the .set_port / .unset_port
328	 * callbacks.
329	 * Zero out here, otherwise !CONFIG_INET causes uninitilized warnings.
330	 */
331	memset(ti, 0, sizeof(*ti));
332
333	if (udp_tunnel_nic_ops)
334		udp_tunnel_nic_ops->get_port(dev, table, idx, ti);
335}
336
337static inline void
338udp_tunnel_nic_set_port_priv(struct net_device *dev, unsigned int table,
339			     unsigned int idx, u8 priv)
340{
341	if (udp_tunnel_nic_ops) {
342		udp_tunnel_nic_ops->assert_locked(dev);
343		udp_tunnel_nic_ops->set_port_priv(dev, table, idx, priv);
344	}
345}
346
347static inline void udp_tunnel_nic_assert_locked(struct net_device *dev)
348{
349	if (udp_tunnel_nic_ops)
350		udp_tunnel_nic_ops->assert_locked(dev);
351}
352
353static inline void udp_tunnel_nic_lock(struct net_device *dev)
354{
355	if (udp_tunnel_nic_ops)
356		udp_tunnel_nic_ops->lock(dev);
357}
358
359static inline void udp_tunnel_nic_unlock(struct net_device *dev)
360{
361	if (udp_tunnel_nic_ops)
362		udp_tunnel_nic_ops->unlock(dev);
363}
364
365static inline void
366udp_tunnel_nic_add_port(struct net_device *dev, struct udp_tunnel_info *ti)
367{
368	if (!(dev->features & NETIF_F_RX_UDP_TUNNEL_PORT))
369		return;
370	if (udp_tunnel_nic_ops)
371		udp_tunnel_nic_ops->add_port(dev, ti);
372}
373
374static inline void
375udp_tunnel_nic_del_port(struct net_device *dev, struct udp_tunnel_info *ti)
376{
377	if (!(dev->features & NETIF_F_RX_UDP_TUNNEL_PORT))
378		return;
379	if (udp_tunnel_nic_ops)
380		udp_tunnel_nic_ops->del_port(dev, ti);
381}
382
383/**
384 * udp_tunnel_nic_reset_ntf() - device-originating reset notification
385 * @dev: network interface device structure
386 *
387 * Called by the driver to inform the core that the entire UDP tunnel port
388 * state has been lost, usually due to device reset. Core will assume device
389 * forgot all the ports and issue .set_port and .sync_table callbacks as
390 * necessary.
391 *
392 * This function must be called with rtnl lock held, and will issue all
393 * the callbacks before returning.
394 */
395static inline void udp_tunnel_nic_reset_ntf(struct net_device *dev)
396{
397	if (udp_tunnel_nic_ops)
398		udp_tunnel_nic_ops->reset_ntf(dev);
399}
400
401static inline size_t
402udp_tunnel_nic_dump_size(struct net_device *dev, unsigned int table)
403{
404	size_t ret;
405
406	if (!udp_tunnel_nic_ops)
407		return 0;
408
409	udp_tunnel_nic_ops->lock(dev);
410	ret = udp_tunnel_nic_ops->dump_size(dev, table);
411	udp_tunnel_nic_ops->unlock(dev);
412
413	return ret;
414}
415
416static inline int
417udp_tunnel_nic_dump_write(struct net_device *dev, unsigned int table,
418			  struct sk_buff *skb)
419{
420	int ret;
421
422	if (!udp_tunnel_nic_ops)
423		return 0;
424
425	udp_tunnel_nic_ops->lock(dev);
426	ret = udp_tunnel_nic_ops->dump_write(dev, table, skb);
427	udp_tunnel_nic_ops->unlock(dev);
428
429	return ret;
430}
431
432static inline void udp_tunnel_get_rx_info(struct net_device *dev)
433{
434	ASSERT_RTNL();
435	if (!(dev->features & NETIF_F_RX_UDP_TUNNEL_PORT))
436		return;
437	udp_tunnel_nic_assert_locked(dev);
438	call_netdevice_notifiers(NETDEV_UDP_TUNNEL_PUSH_INFO, dev);
439}
440
441static inline void udp_tunnel_drop_rx_info(struct net_device *dev)
442{
443	ASSERT_RTNL();
444	if (!(dev->features & NETIF_F_RX_UDP_TUNNEL_PORT))
445		return;
446	udp_tunnel_nic_assert_locked(dev);
447	call_netdevice_notifiers(NETDEV_UDP_TUNNEL_DROP_INFO, dev);
448}
449
450#endif