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_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 133static 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 141static 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. */ 150void 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 155int udp_tunnel6_xmit_skb(struct dst_entry *dst, struct sock *sk, 156 struct sk_buff *skb, 157 struct net_device *dev, struct in6_addr *saddr, 158 struct in6_addr *daddr, 159 __u8 prio, __u8 ttl, __be32 label, 160 __be16 src_port, __be16 dst_port, bool nocheck); 161 162void udp_tunnel_sock_release(struct socket *sock); 163 164struct metadata_dst *udp_tun_rx_dst(struct sk_buff *skb, unsigned short family, 165 __be16 flags, __be64 tunnel_id, 166 int md_size); 167 168#ifdef CONFIG_INET 169static inline int udp_tunnel_handle_offloads(struct sk_buff *skb, bool udp_csum) 170{ 171 int type = udp_csum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL; 172 173 return iptunnel_handle_offloads(skb, type); 174} 175#endif 176 177static inline void udp_tunnel_encap_enable(struct sock *sk) 178{ 179 if (udp_test_and_set_bit(ENCAP_ENABLED, sk)) 180 return; 181 182#if IS_ENABLED(CONFIG_IPV6) 183 if (READ_ONCE(sk->sk_family) == PF_INET6) 184 ipv6_stub->udpv6_encap_enable(); 185#endif 186 udp_encap_enable(); 187} 188 189#define UDP_TUNNEL_NIC_MAX_TABLES 4 190 191enum udp_tunnel_nic_info_flags { 192 /* Device callbacks may sleep */ 193 UDP_TUNNEL_NIC_INFO_MAY_SLEEP = BIT(0), 194 /* Device only supports offloads when it's open, all ports 195 * will be removed before close and re-added after open. 196 */ 197 UDP_TUNNEL_NIC_INFO_OPEN_ONLY = BIT(1), 198 /* Device supports only IPv4 tunnels */ 199 UDP_TUNNEL_NIC_INFO_IPV4_ONLY = BIT(2), 200 /* Device has hard-coded the IANA VXLAN port (4789) as VXLAN. 201 * This port must not be counted towards n_entries of any table. 202 * Driver will not receive any callback associated with port 4789. 203 */ 204 UDP_TUNNEL_NIC_INFO_STATIC_IANA_VXLAN = BIT(3), 205}; 206 207struct udp_tunnel_nic; 208 209#define UDP_TUNNEL_NIC_MAX_SHARING_DEVICES (U16_MAX / 2) 210 211struct udp_tunnel_nic_shared { 212 struct udp_tunnel_nic *udp_tunnel_nic_info; 213 214 struct list_head devices; 215}; 216 217struct udp_tunnel_nic_shared_node { 218 struct net_device *dev; 219 struct list_head list; 220}; 221 222/** 223 * struct udp_tunnel_nic_info - driver UDP tunnel offload information 224 * @set_port: callback for adding a new port 225 * @unset_port: callback for removing a port 226 * @sync_table: callback for syncing the entire port table at once 227 * @shared: reference to device global state (optional) 228 * @flags: device flags from enum udp_tunnel_nic_info_flags 229 * @tables: UDP port tables this device has 230 * @tables.n_entries: number of entries in this table 231 * @tables.tunnel_types: types of tunnels this table accepts 232 * 233 * Drivers are expected to provide either @set_port and @unset_port callbacks 234 * or the @sync_table callback. Callbacks are invoked with rtnl lock held. 235 * 236 * Devices which (misguidedly) share the UDP tunnel port table across multiple 237 * netdevs should allocate an instance of struct udp_tunnel_nic_shared and 238 * point @shared at it. 239 * There must never be more than %UDP_TUNNEL_NIC_MAX_SHARING_DEVICES devices 240 * sharing a table. 241 * 242 * Known limitations: 243 * - UDP tunnel port notifications are fundamentally best-effort - 244 * it is likely the driver will both see skbs which use a UDP tunnel port, 245 * while not being a tunneled skb, and tunnel skbs from other ports - 246 * drivers should only use these ports for non-critical RX-side offloads, 247 * e.g. the checksum offload; 248 * - none of the devices care about the socket family at present, so we don't 249 * track it. Please extend this code if you care. 250 */ 251struct udp_tunnel_nic_info { 252 /* one-by-one */ 253 int (*set_port)(struct net_device *dev, 254 unsigned int table, unsigned int entry, 255 struct udp_tunnel_info *ti); 256 int (*unset_port)(struct net_device *dev, 257 unsigned int table, unsigned int entry, 258 struct udp_tunnel_info *ti); 259 260 /* all at once */ 261 int (*sync_table)(struct net_device *dev, unsigned int table); 262 263 struct udp_tunnel_nic_shared *shared; 264 265 unsigned int flags; 266 267 struct udp_tunnel_nic_table_info { 268 unsigned int n_entries; 269 unsigned int tunnel_types; 270 } tables[UDP_TUNNEL_NIC_MAX_TABLES]; 271}; 272 273/* UDP tunnel module dependencies 274 * 275 * Tunnel drivers are expected to have a hard dependency on the udp_tunnel 276 * module. NIC drivers are not, they just attach their 277 * struct udp_tunnel_nic_info to the netdev and wait for callbacks to come. 278 * Loading a tunnel driver will cause the udp_tunnel module to be loaded 279 * and only then will all the required state structures be allocated. 280 * Since we want a weak dependency from the drivers and the core to udp_tunnel 281 * we call things through the following stubs. 282 */ 283struct udp_tunnel_nic_ops { 284 void (*get_port)(struct net_device *dev, unsigned int table, 285 unsigned int idx, struct udp_tunnel_info *ti); 286 void (*set_port_priv)(struct net_device *dev, unsigned int table, 287 unsigned int idx, u8 priv); 288 void (*add_port)(struct net_device *dev, struct udp_tunnel_info *ti); 289 void (*del_port)(struct net_device *dev, struct udp_tunnel_info *ti); 290 void (*reset_ntf)(struct net_device *dev); 291 292 size_t (*dump_size)(struct net_device *dev, unsigned int table); 293 int (*dump_write)(struct net_device *dev, unsigned int table, 294 struct sk_buff *skb); 295}; 296 297#ifdef CONFIG_INET 298extern const struct udp_tunnel_nic_ops *udp_tunnel_nic_ops; 299#else 300#define udp_tunnel_nic_ops ((struct udp_tunnel_nic_ops *)NULL) 301#endif 302 303static inline void 304udp_tunnel_nic_get_port(struct net_device *dev, unsigned int table, 305 unsigned int idx, struct udp_tunnel_info *ti) 306{ 307 /* This helper is used from .sync_table, we indicate empty entries 308 * by zero'ed @ti. Drivers which need to know the details of a port 309 * when it gets deleted should use the .set_port / .unset_port 310 * callbacks. 311 * Zero out here, otherwise !CONFIG_INET causes uninitilized warnings. 312 */ 313 memset(ti, 0, sizeof(*ti)); 314 315 if (udp_tunnel_nic_ops) 316 udp_tunnel_nic_ops->get_port(dev, table, idx, ti); 317} 318 319static inline void 320udp_tunnel_nic_set_port_priv(struct net_device *dev, unsigned int table, 321 unsigned int idx, u8 priv) 322{ 323 if (udp_tunnel_nic_ops) 324 udp_tunnel_nic_ops->set_port_priv(dev, table, idx, priv); 325} 326 327static inline void 328udp_tunnel_nic_add_port(struct net_device *dev, struct udp_tunnel_info *ti) 329{ 330 if (!(dev->features & NETIF_F_RX_UDP_TUNNEL_PORT)) 331 return; 332 if (udp_tunnel_nic_ops) 333 udp_tunnel_nic_ops->add_port(dev, ti); 334} 335 336static inline void 337udp_tunnel_nic_del_port(struct net_device *dev, struct udp_tunnel_info *ti) 338{ 339 if (!(dev->features & NETIF_F_RX_UDP_TUNNEL_PORT)) 340 return; 341 if (udp_tunnel_nic_ops) 342 udp_tunnel_nic_ops->del_port(dev, ti); 343} 344 345/** 346 * udp_tunnel_nic_reset_ntf() - device-originating reset notification 347 * @dev: network interface device structure 348 * 349 * Called by the driver to inform the core that the entire UDP tunnel port 350 * state has been lost, usually due to device reset. Core will assume device 351 * forgot all the ports and issue .set_port and .sync_table callbacks as 352 * necessary. 353 * 354 * This function must be called with rtnl lock held, and will issue all 355 * the callbacks before returning. 356 */ 357static inline void udp_tunnel_nic_reset_ntf(struct net_device *dev) 358{ 359 if (udp_tunnel_nic_ops) 360 udp_tunnel_nic_ops->reset_ntf(dev); 361} 362 363static inline size_t 364udp_tunnel_nic_dump_size(struct net_device *dev, unsigned int table) 365{ 366 if (!udp_tunnel_nic_ops) 367 return 0; 368 return udp_tunnel_nic_ops->dump_size(dev, table); 369} 370 371static inline int 372udp_tunnel_nic_dump_write(struct net_device *dev, unsigned int table, 373 struct sk_buff *skb) 374{ 375 if (!udp_tunnel_nic_ops) 376 return 0; 377 return udp_tunnel_nic_ops->dump_write(dev, table, skb); 378} 379#endif 380