1 #ifndef __NET_IP_TUNNELS_H 2 #define __NET_IP_TUNNELS_H 1 3 4 #include <linux/if_tunnel.h> 5 #include <linux/netdevice.h> 6 #include <linux/skbuff.h> 7 #include <linux/socket.h> 8 #include <linux/types.h> 9 #include <linux/u64_stats_sync.h> 10 #include <net/dsfield.h> 11 #include <net/gro_cells.h> 12 #include <net/inet_ecn.h> 13 #include <net/netns/generic.h> 14 #include <net/rtnetlink.h> 15 #include <net/lwtunnel.h> 16 #include <net/dst_cache.h> 17 18 #if IS_ENABLED(CONFIG_IPV6) 19 #include <net/ipv6.h> 20 #include <net/ip6_fib.h> 21 #include <net/ip6_route.h> 22 #endif 23 24 /* Keep error state on tunnel for 30 sec */ 25 #define IPTUNNEL_ERR_TIMEO (30*HZ) 26 27 /* Used to memset ip_tunnel padding. */ 28 #define IP_TUNNEL_KEY_SIZE offsetofend(struct ip_tunnel_key, tp_dst) 29 30 /* Used to memset ipv4 address padding. */ 31 #define IP_TUNNEL_KEY_IPV4_PAD offsetofend(struct ip_tunnel_key, u.ipv4.dst) 32 #define IP_TUNNEL_KEY_IPV4_PAD_LEN \ 33 (FIELD_SIZEOF(struct ip_tunnel_key, u) - \ 34 FIELD_SIZEOF(struct ip_tunnel_key, u.ipv4)) 35 36 struct ip_tunnel_key { 37 __be64 tun_id; 38 union { 39 struct { 40 __be32 src; 41 __be32 dst; 42 } ipv4; 43 struct { 44 struct in6_addr src; 45 struct in6_addr dst; 46 } ipv6; 47 } u; 48 __be16 tun_flags; 49 u8 tos; /* TOS for IPv4, TC for IPv6 */ 50 u8 ttl; /* TTL for IPv4, HL for IPv6 */ 51 __be32 label; /* Flow Label for IPv6 */ 52 __be16 tp_src; 53 __be16 tp_dst; 54 }; 55 56 /* Flags for ip_tunnel_info mode. */ 57 #define IP_TUNNEL_INFO_TX 0x01 /* represents tx tunnel parameters */ 58 #define IP_TUNNEL_INFO_IPV6 0x02 /* key contains IPv6 addresses */ 59 60 struct ip_tunnel_info { 61 struct ip_tunnel_key key; 62 #ifdef CONFIG_DST_CACHE 63 struct dst_cache dst_cache; 64 #endif 65 u8 options_len; 66 u8 mode; 67 }; 68 69 /* 6rd prefix/relay information */ 70 #ifdef CONFIG_IPV6_SIT_6RD 71 struct ip_tunnel_6rd_parm { 72 struct in6_addr prefix; 73 __be32 relay_prefix; 74 u16 prefixlen; 75 u16 relay_prefixlen; 76 }; 77 #endif 78 79 struct ip_tunnel_encap { 80 u16 type; 81 u16 flags; 82 __be16 sport; 83 __be16 dport; 84 }; 85 86 struct ip_tunnel_prl_entry { 87 struct ip_tunnel_prl_entry __rcu *next; 88 __be32 addr; 89 u16 flags; 90 struct rcu_head rcu_head; 91 }; 92 93 struct metadata_dst; 94 95 struct ip_tunnel { 96 struct ip_tunnel __rcu *next; 97 struct hlist_node hash_node; 98 struct net_device *dev; 99 struct net *net; /* netns for packet i/o */ 100 101 int err_count; /* Number of arrived ICMP errors */ 102 unsigned long err_time; /* Time when the last ICMP error 103 * arrived */ 104 105 /* These four fields used only by GRE */ 106 u32 i_seqno; /* The last seen seqno */ 107 u32 o_seqno; /* The last output seqno */ 108 int tun_hlen; /* Precalculated header length */ 109 int mlink; 110 111 struct dst_cache dst_cache; 112 113 struct ip_tunnel_parm parms; 114 115 int encap_hlen; /* Encap header length (FOU,GUE) */ 116 struct ip_tunnel_encap encap; 117 118 int hlen; /* tun_hlen + encap_hlen */ 119 120 /* for SIT */ 121 #ifdef CONFIG_IPV6_SIT_6RD 122 struct ip_tunnel_6rd_parm ip6rd; 123 #endif 124 struct ip_tunnel_prl_entry __rcu *prl; /* potential router list */ 125 unsigned int prl_count; /* # of entries in PRL */ 126 int ip_tnl_net_id; 127 struct gro_cells gro_cells; 128 bool collect_md; 129 }; 130 131 #define TUNNEL_CSUM __cpu_to_be16(0x01) 132 #define TUNNEL_ROUTING __cpu_to_be16(0x02) 133 #define TUNNEL_KEY __cpu_to_be16(0x04) 134 #define TUNNEL_SEQ __cpu_to_be16(0x08) 135 #define TUNNEL_STRICT __cpu_to_be16(0x10) 136 #define TUNNEL_REC __cpu_to_be16(0x20) 137 #define TUNNEL_VERSION __cpu_to_be16(0x40) 138 #define TUNNEL_NO_KEY __cpu_to_be16(0x80) 139 #define TUNNEL_DONT_FRAGMENT __cpu_to_be16(0x0100) 140 #define TUNNEL_OAM __cpu_to_be16(0x0200) 141 #define TUNNEL_CRIT_OPT __cpu_to_be16(0x0400) 142 #define TUNNEL_GENEVE_OPT __cpu_to_be16(0x0800) 143 #define TUNNEL_VXLAN_OPT __cpu_to_be16(0x1000) 144 #define TUNNEL_NOCACHE __cpu_to_be16(0x2000) 145 146 #define TUNNEL_OPTIONS_PRESENT (TUNNEL_GENEVE_OPT | TUNNEL_VXLAN_OPT) 147 148 struct tnl_ptk_info { 149 __be16 flags; 150 __be16 proto; 151 __be32 key; 152 __be32 seq; 153 }; 154 155 #define PACKET_RCVD 0 156 #define PACKET_REJECT 1 157 158 #define IP_TNL_HASH_BITS 7 159 #define IP_TNL_HASH_SIZE (1 << IP_TNL_HASH_BITS) 160 161 struct ip_tunnel_net { 162 struct net_device *fb_tunnel_dev; 163 struct hlist_head tunnels[IP_TNL_HASH_SIZE]; 164 struct ip_tunnel __rcu *collect_md_tun; 165 }; 166 167 struct ip_tunnel_encap_ops { 168 size_t (*encap_hlen)(struct ip_tunnel_encap *e); 169 int (*build_header)(struct sk_buff *skb, struct ip_tunnel_encap *e, 170 u8 *protocol, struct flowi4 *fl4); 171 }; 172 173 #define MAX_IPTUN_ENCAP_OPS 8 174 175 extern const struct ip_tunnel_encap_ops __rcu * 176 iptun_encaps[MAX_IPTUN_ENCAP_OPS]; 177 178 int ip_tunnel_encap_add_ops(const struct ip_tunnel_encap_ops *op, 179 unsigned int num); 180 int ip_tunnel_encap_del_ops(const struct ip_tunnel_encap_ops *op, 181 unsigned int num); 182 183 static inline void ip_tunnel_key_init(struct ip_tunnel_key *key, 184 __be32 saddr, __be32 daddr, 185 u8 tos, u8 ttl, __be32 label, 186 __be16 tp_src, __be16 tp_dst, 187 __be64 tun_id, __be16 tun_flags) 188 { 189 key->tun_id = tun_id; 190 key->u.ipv4.src = saddr; 191 key->u.ipv4.dst = daddr; 192 memset((unsigned char *)key + IP_TUNNEL_KEY_IPV4_PAD, 193 0, IP_TUNNEL_KEY_IPV4_PAD_LEN); 194 key->tos = tos; 195 key->ttl = ttl; 196 key->label = label; 197 key->tun_flags = tun_flags; 198 199 /* For the tunnel types on the top of IPsec, the tp_src and tp_dst of 200 * the upper tunnel are used. 201 * E.g: GRE over IPSEC, the tp_src and tp_port are zero. 202 */ 203 key->tp_src = tp_src; 204 key->tp_dst = tp_dst; 205 206 /* Clear struct padding. */ 207 if (sizeof(*key) != IP_TUNNEL_KEY_SIZE) 208 memset((unsigned char *)key + IP_TUNNEL_KEY_SIZE, 209 0, sizeof(*key) - IP_TUNNEL_KEY_SIZE); 210 } 211 212 static inline bool 213 ip_tunnel_dst_cache_usable(const struct sk_buff *skb, 214 const struct ip_tunnel_info *info) 215 { 216 if (skb->mark) 217 return false; 218 if (!info) 219 return true; 220 if (info->key.tun_flags & TUNNEL_NOCACHE) 221 return false; 222 223 return true; 224 } 225 226 static inline unsigned short ip_tunnel_info_af(const struct ip_tunnel_info 227 *tun_info) 228 { 229 return tun_info->mode & IP_TUNNEL_INFO_IPV6 ? AF_INET6 : AF_INET; 230 } 231 232 #ifdef CONFIG_INET 233 234 int ip_tunnel_init(struct net_device *dev); 235 void ip_tunnel_uninit(struct net_device *dev); 236 void ip_tunnel_dellink(struct net_device *dev, struct list_head *head); 237 struct net *ip_tunnel_get_link_net(const struct net_device *dev); 238 int ip_tunnel_get_iflink(const struct net_device *dev); 239 int ip_tunnel_init_net(struct net *net, int ip_tnl_net_id, 240 struct rtnl_link_ops *ops, char *devname); 241 242 void ip_tunnel_delete_net(struct ip_tunnel_net *itn, struct rtnl_link_ops *ops); 243 244 void ip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev, 245 const struct iphdr *tnl_params, const u8 protocol); 246 int ip_tunnel_ioctl(struct net_device *dev, struct ip_tunnel_parm *p, int cmd); 247 int ip_tunnel_encap(struct sk_buff *skb, struct ip_tunnel *t, 248 u8 *protocol, struct flowi4 *fl4); 249 int __ip_tunnel_change_mtu(struct net_device *dev, int new_mtu, bool strict); 250 int ip_tunnel_change_mtu(struct net_device *dev, int new_mtu); 251 252 struct rtnl_link_stats64 *ip_tunnel_get_stats64(struct net_device *dev, 253 struct rtnl_link_stats64 *tot); 254 struct ip_tunnel *ip_tunnel_lookup(struct ip_tunnel_net *itn, 255 int link, __be16 flags, 256 __be32 remote, __be32 local, 257 __be32 key); 258 259 int ip_tunnel_rcv(struct ip_tunnel *tunnel, struct sk_buff *skb, 260 const struct tnl_ptk_info *tpi, struct metadata_dst *tun_dst, 261 bool log_ecn_error); 262 int ip_tunnel_changelink(struct net_device *dev, struct nlattr *tb[], 263 struct ip_tunnel_parm *p); 264 int ip_tunnel_newlink(struct net_device *dev, struct nlattr *tb[], 265 struct ip_tunnel_parm *p); 266 void ip_tunnel_setup(struct net_device *dev, int net_id); 267 int ip_tunnel_encap_setup(struct ip_tunnel *t, 268 struct ip_tunnel_encap *ipencap); 269 270 /* Extract dsfield from inner protocol */ 271 static inline u8 ip_tunnel_get_dsfield(const struct iphdr *iph, 272 const struct sk_buff *skb) 273 { 274 if (skb->protocol == htons(ETH_P_IP)) 275 return iph->tos; 276 else if (skb->protocol == htons(ETH_P_IPV6)) 277 return ipv6_get_dsfield((const struct ipv6hdr *)iph); 278 else 279 return 0; 280 } 281 282 /* Propogate ECN bits out */ 283 static inline u8 ip_tunnel_ecn_encap(u8 tos, const struct iphdr *iph, 284 const struct sk_buff *skb) 285 { 286 u8 inner = ip_tunnel_get_dsfield(iph, skb); 287 288 return INET_ECN_encapsulate(tos, inner); 289 } 290 291 int iptunnel_pull_header(struct sk_buff *skb, int hdr_len, __be16 inner_proto, 292 bool xnet); 293 void iptunnel_xmit(struct sock *sk, struct rtable *rt, struct sk_buff *skb, 294 __be32 src, __be32 dst, u8 proto, 295 u8 tos, u8 ttl, __be16 df, bool xnet); 296 struct metadata_dst *iptunnel_metadata_reply(struct metadata_dst *md, 297 gfp_t flags); 298 299 struct sk_buff *iptunnel_handle_offloads(struct sk_buff *skb, int gso_type_mask); 300 301 static inline void iptunnel_xmit_stats(struct net_device *dev, int pkt_len) 302 { 303 if (pkt_len > 0) { 304 struct pcpu_sw_netstats *tstats = get_cpu_ptr(dev->tstats); 305 306 u64_stats_update_begin(&tstats->syncp); 307 tstats->tx_bytes += pkt_len; 308 tstats->tx_packets++; 309 u64_stats_update_end(&tstats->syncp); 310 put_cpu_ptr(tstats); 311 } else { 312 struct net_device_stats *err_stats = &dev->stats; 313 314 if (pkt_len < 0) { 315 err_stats->tx_errors++; 316 err_stats->tx_aborted_errors++; 317 } else { 318 err_stats->tx_dropped++; 319 } 320 } 321 } 322 323 static inline void *ip_tunnel_info_opts(struct ip_tunnel_info *info) 324 { 325 return info + 1; 326 } 327 328 static inline void ip_tunnel_info_opts_get(void *to, 329 const struct ip_tunnel_info *info) 330 { 331 memcpy(to, info + 1, info->options_len); 332 } 333 334 static inline void ip_tunnel_info_opts_set(struct ip_tunnel_info *info, 335 const void *from, int len) 336 { 337 memcpy(ip_tunnel_info_opts(info), from, len); 338 info->options_len = len; 339 } 340 341 static inline struct ip_tunnel_info *lwt_tun_info(struct lwtunnel_state *lwtstate) 342 { 343 return (struct ip_tunnel_info *)lwtstate->data; 344 } 345 346 extern struct static_key ip_tunnel_metadata_cnt; 347 348 /* Returns > 0 if metadata should be collected */ 349 static inline int ip_tunnel_collect_metadata(void) 350 { 351 return static_key_false(&ip_tunnel_metadata_cnt); 352 } 353 354 void __init ip_tunnel_core_init(void); 355 356 void ip_tunnel_need_metadata(void); 357 void ip_tunnel_unneed_metadata(void); 358 359 #else /* CONFIG_INET */ 360 361 static inline struct ip_tunnel_info *lwt_tun_info(struct lwtunnel_state *lwtstate) 362 { 363 return NULL; 364 } 365 366 static inline void ip_tunnel_need_metadata(void) 367 { 368 } 369 370 static inline void ip_tunnel_unneed_metadata(void) 371 { 372 } 373 374 static inline void ip_tunnel_info_opts_get(void *to, 375 const struct ip_tunnel_info *info) 376 { 377 } 378 379 static inline void ip_tunnel_info_opts_set(struct ip_tunnel_info *info, 380 const void *from, int len) 381 { 382 info->options_len = 0; 383 } 384 385 #endif /* CONFIG_INET */ 386 387 #endif /* __NET_IP_TUNNELS_H */ 388