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