1213277Sgonzo/* SPDX-License-Identifier: GPL-2.0 */ 2213277Sgonzo#ifndef __NET_UDP_TUNNEL_H 3213277Sgonzo#define __NET_UDP_TUNNEL_H 4213277Sgonzo 5213277Sgonzo#include <net/ip_tunnels.h> 6213277Sgonzo#include <net/udp.h> 7213277Sgonzo 8213277Sgonzo#if IS_ENABLED(CONFIG_IPV6) 9213277Sgonzo#include <net/ipv6.h> 10213277Sgonzo#include <net/ipv6_stubs.h> 11213277Sgonzo#endif 12213277Sgonzo 13213277Sgonzostruct udp_port_cfg { 14213277Sgonzo u8 family; 15213277Sgonzo 16213277Sgonzo /* Used only for kernel-created sockets */ 17213277Sgonzo union { 18213277Sgonzo struct in_addr local_ip; 19213277Sgonzo#if IS_ENABLED(CONFIG_IPV6) 20213277Sgonzo struct in6_addr local_ip6; 21213277Sgonzo#endif 22213277Sgonzo }; 23213277Sgonzo 24213277Sgonzo union { 25213277Sgonzo struct in_addr peer_ip; 26213277Sgonzo#if IS_ENABLED(CONFIG_IPV6) 27213237Sgonzo struct in6_addr peer_ip6; 28213237Sgonzo#endif 29213237Sgonzo }; 30213237Sgonzo 31213237Sgonzo __be16 local_udp_port; 32213237Sgonzo __be16 peer_udp_port; 33213237Sgonzo int bind_ifindex; 34274670Sloos unsigned int use_udp_checksums:1, 35213237Sgonzo use_udp6_tx_checksums:1, 36213237Sgonzo use_udp6_rx_checksums:1, 37213237Sgonzo ipv6_v6only:1; 38213237Sgonzo}; 39213237Sgonzo 40274670Sloosint udp_sock_create4(struct net *net, struct udp_port_cfg *cfg, 41274670Sloos struct socket **sockp); 42213237Sgonzo 43279761Sloos#if IS_ENABLED(CONFIG_IPV6) 44213237Sgonzoint udp_sock_create6(struct net *net, struct udp_port_cfg *cfg, 45213237Sgonzo struct socket **sockp); 46213237Sgonzo#else 47213237Sgonzostatic inline int udp_sock_create6(struct net *net, struct udp_port_cfg *cfg, 48213237Sgonzo struct socket **sockp) 49213237Sgonzo{ 50213237Sgonzo return 0; 51213237Sgonzo} 52213237Sgonzo#endif 53213237Sgonzo 54213237Sgonzostatic inline int udp_sock_create(struct net *net, 55213237Sgonzo struct udp_port_cfg *cfg, 56213237Sgonzo struct socket **sockp) 57213237Sgonzo{ 58213237Sgonzo if (cfg->family == AF_INET) 59213237Sgonzo return udp_sock_create4(net, cfg, sockp); 60213237Sgonzo 61213237Sgonzo if (cfg->family == AF_INET6) 62213237Sgonzo return udp_sock_create6(net, cfg, sockp); 63213237Sgonzo 64213237Sgonzo return -EPFNOSUPPORT; 65213237Sgonzo} 66213237Sgonzo 67213237Sgonzotypedef int (*udp_tunnel_encap_rcv_t)(struct sock *sk, struct sk_buff *skb); 68213237Sgonzotypedef int (*udp_tunnel_encap_err_lookup_t)(struct sock *sk, 69213237Sgonzo struct sk_buff *skb); 70213237Sgonzotypedef void (*udp_tunnel_encap_err_rcv_t)(struct sock *sk, 71213237Sgonzo struct sk_buff *skb, int err, 72213237Sgonzo __be16 port, u32 info, u8 *payload); 73213237Sgonzotypedef void (*udp_tunnel_encap_destroy_t)(struct sock *sk); 74213237Sgonzotypedef struct sk_buff *(*udp_tunnel_gro_receive_t)(struct sock *sk, 75213237Sgonzo struct list_head *head, 76213237Sgonzo struct sk_buff *skb); 77213237Sgonzotypedef int (*udp_tunnel_gro_complete_t)(struct sock *sk, struct sk_buff *skb, 78213237Sgonzo int nhoff); 79213237Sgonzo 80213237Sgonzostruct udp_tunnel_sock_cfg { 81312006Sloos void *sk_user_data; /* user data used by encap_rcv call back */ 82312006Sloos /* Used for setting up udp_sock fields, see udp.h for details */ 83312006Sloos __u8 encap_type; 84213237Sgonzo udp_tunnel_encap_rcv_t encap_rcv; 85312006Sloos udp_tunnel_encap_err_lookup_t encap_err_lookup; 86213237Sgonzo udp_tunnel_encap_err_rcv_t encap_err_rcv; 87213237Sgonzo udp_tunnel_encap_destroy_t encap_destroy; 88213237Sgonzo udp_tunnel_gro_receive_t gro_receive; 89312006Sloos udp_tunnel_gro_complete_t gro_complete; 90312006Sloos}; 91312006Sloos 92312006Sloos/* Setup the given (UDP) sock to receive UDP encapsulated packets */ 93312006Sloosvoid setup_udp_tunnel_sock(struct net *net, struct socket *sock, 94312006Sloos struct udp_tunnel_sock_cfg *sock_cfg); 95312006Sloos 96312006Sloos/* -- List of parsable UDP tunnel types -- 97213237Sgonzo * 98213237Sgonzo * Adding to this list will result in serious debate. The main issue is 99213237Sgonzo * that this list is essentially a list of workarounds for either poorly 100213237Sgonzo * designed tunnels, or poorly designed device offloads. 101213237Sgonzo * 102213237Sgonzo * The parsing supported via these types should really be used for Rx 103213237Sgonzo * traffic only as the network stack will have already inserted offsets for 104213237Sgonzo * the location of the headers in the skb. In addition any ports that are 105213237Sgonzo * pushed should be kept within the namespace without leaking to other 106213237Sgonzo * devices such as VFs or other ports on the same device. 107213237Sgonzo * 108213237Sgonzo * It is strongly encouraged to use CHECKSUM_COMPLETE for Rx to avoid the 109213237Sgonzo * need to use this for Rx checksum offload. It should not be necessary to 110233767Sgonzo * call this function to perform Tx offloads on outgoing traffic. 111213237Sgonzo */ 112213237Sgonzoenum udp_parsable_tunnel_type { 113213237Sgonzo UDP_TUNNEL_TYPE_VXLAN = BIT(0), /* RFC 7348 */ 114213237Sgonzo UDP_TUNNEL_TYPE_GENEVE = BIT(1), /* draft-ietf-nvo3-geneve */ 115213237Sgonzo UDP_TUNNEL_TYPE_VXLAN_GPE = BIT(2), /* draft-ietf-nvo3-vxlan-gpe */ 116213237Sgonzo}; 117213237Sgonzo 118213237Sgonzostruct udp_tunnel_info { 119213237Sgonzo unsigned short type; 120213237Sgonzo sa_family_t sa_family; 121213237Sgonzo __be16 port; 122213237Sgonzo u8 hw_priv; 123279761Sloos}; 124213237Sgonzo 125213237Sgonzo/* Notify network devices of offloadable types */ 126213237Sgonzovoid udp_tunnel_push_rx_port(struct net_device *dev, struct socket *sock, 127213237Sgonzo unsigned short type); 128324755Sianvoid udp_tunnel_drop_rx_port(struct net_device *dev, struct socket *sock, 129324755Sian unsigned short type); 130274670Sloosvoid udp_tunnel_notify_add_rx_port(struct socket *sock, unsigned short type); 131213237Sgonzovoid udp_tunnel_notify_del_rx_port(struct socket *sock, unsigned short type); 132279761Sloos 133279761Sloosstatic inline void udp_tunnel_get_rx_info(struct net_device *dev) 134279761Sloos{ 135213237Sgonzo ASSERT_RTNL(); 136213237Sgonzo if (!(dev->features & NETIF_F_RX_UDP_TUNNEL_PORT)) 137213237Sgonzo return; 138213237Sgonzo call_netdevice_notifiers(NETDEV_UDP_TUNNEL_PUSH_INFO, dev); 139213237Sgonzo} 140213237Sgonzo 141213237Sgonzostatic inline void udp_tunnel_drop_rx_info(struct net_device *dev) 142213237Sgonzo{ 143213237Sgonzo ASSERT_RTNL(); 144213237Sgonzo if (!(dev->features & NETIF_F_RX_UDP_TUNNEL_PORT)) 145213237Sgonzo return; 146213237Sgonzo call_netdevice_notifiers(NETDEV_UDP_TUNNEL_DROP_INFO, dev); 147213237Sgonzo} 148213237Sgonzo 149213237Sgonzo/* Transmit the skb using UDP encapsulation. */ 150279761Sloosvoid udp_tunnel_xmit_skb(struct rtable *rt, struct sock *sk, struct sk_buff *skb, 151213237Sgonzo __be32 src, __be32 dst, __u8 tos, __u8 ttl, 152213237Sgonzo __be16 df, __be16 src_port, __be16 dst_port, 153213237Sgonzo bool xnet, bool nocheck); 154213237Sgonzo 155213237Sgonzoint udp_tunnel6_xmit_skb(struct dst_entry *dst, struct sock *sk, 156274670Sloos struct sk_buff *skb, 157274670Sloos struct net_device *dev, 158274670Sloos const struct in6_addr *saddr, 159274670Sloos const struct in6_addr *daddr, 160274670Sloos __u8 prio, __u8 ttl, __be32 label, 161274670Sloos __be16 src_port, __be16 dst_port, bool nocheck); 162213237Sgonzo 163213237Sgonzovoid udp_tunnel_sock_release(struct socket *sock); 164213237Sgonzo 165213237Sgonzostruct rtable *udp_tunnel_dst_lookup(struct sk_buff *skb, 166213237Sgonzo struct net_device *dev, 167213237Sgonzo struct net *net, int oif, 168213237Sgonzo __be32 *saddr, 169213237Sgonzo const struct ip_tunnel_key *key, 170213237Sgonzo __be16 sport, __be16 dport, u8 tos, 171213237Sgonzo struct dst_cache *dst_cache); 172213237Sgonzostruct dst_entry *udp_tunnel6_dst_lookup(struct sk_buff *skb, 173213237Sgonzo struct net_device *dev, 174213237Sgonzo struct net *net, 175213237Sgonzo struct socket *sock, int oif, 176213237Sgonzo struct in6_addr *saddr, 177213237Sgonzo const struct ip_tunnel_key *key, 178213237Sgonzo __be16 sport, __be16 dport, u8 dsfield, 179213237Sgonzo struct dst_cache *dst_cache); 180213237Sgonzo 181213237Sgonzostruct metadata_dst *udp_tun_rx_dst(struct sk_buff *skb, unsigned short family, 182213237Sgonzo __be16 flags, __be64 tunnel_id, 183213237Sgonzo int md_size); 184279761Sloos 185279761Sloos#ifdef CONFIG_INET 186279761Sloosstatic inline int udp_tunnel_handle_offloads(struct sk_buff *skb, bool udp_csum) 187279761Sloos{ 188279761Sloos int type = udp_csum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL; 189279761Sloos 190324755Sian return iptunnel_handle_offloads(skb, type); 191324755Sian} 192324755Sian#endif 193324755Sian 194324755Sianstatic inline void udp_tunnel_encap_enable(struct sock *sk) 195324755Sian{ 196324755Sian if (udp_test_and_set_bit(ENCAP_ENABLED, sk)) 197324755Sian return; 198324755Sian 199324755Sian#if IS_ENABLED(CONFIG_IPV6) 200213237Sgonzo if (READ_ONCE(sk->sk_family) == PF_INET6) 201213237Sgonzo ipv6_stub->udpv6_encap_enable(); 202213237Sgonzo#endif 203213237Sgonzo udp_encap_enable(); 204213237Sgonzo} 205213237Sgonzo 206213237Sgonzo#define UDP_TUNNEL_NIC_MAX_TABLES 4 207213237Sgonzo 208213237Sgonzoenum udp_tunnel_nic_info_flags { 209213237Sgonzo /* Device callbacks may sleep */ 210213237Sgonzo UDP_TUNNEL_NIC_INFO_MAY_SLEEP = BIT(0), 211213237Sgonzo /* Device only supports offloads when it's open, all ports 212213237Sgonzo * will be removed before close and re-added after open. 213213237Sgonzo */ 214213237Sgonzo UDP_TUNNEL_NIC_INFO_OPEN_ONLY = BIT(1), 215213237Sgonzo /* Device supports only IPv4 tunnels */ 216213237Sgonzo UDP_TUNNEL_NIC_INFO_IPV4_ONLY = BIT(2), 217299396Sgonzo /* Device has hard-coded the IANA VXLAN port (4789) as VXLAN. 218213237Sgonzo * This port must not be counted towards n_entries of any table. 219213237Sgonzo * Driver will not receive any callback associated with port 4789. 220215142Sthompsa */ 221213237Sgonzo UDP_TUNNEL_NIC_INFO_STATIC_IANA_VXLAN = BIT(3), 222213237Sgonzo}; 223213237Sgonzo 224213237Sgonzostruct udp_tunnel_nic; 225213237Sgonzo 226213237Sgonzo#define UDP_TUNNEL_NIC_MAX_SHARING_DEVICES (U16_MAX / 2) 227213237Sgonzo 228213237Sgonzostruct udp_tunnel_nic_shared { 229213237Sgonzo struct udp_tunnel_nic *udp_tunnel_nic_info; 230 231 struct list_head devices; 232}; 233 234struct 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 */ 268struct 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 */ 300struct 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 315extern 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 320static inline void 321udp_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 336static inline void 337udp_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 344static inline void 345udp_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 353static inline void 354udp_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 */ 374static 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 380static inline size_t 381udp_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 388static inline int 389udp_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