1 #ifndef __NET_VXLAN_H 2 #define __NET_VXLAN_H 1 3 4 #include <linux/ip.h> 5 #include <linux/ipv6.h> 6 #include <linux/if_vlan.h> 7 #include <linux/skbuff.h> 8 #include <linux/netdevice.h> 9 #include <linux/udp.h> 10 #include <net/dst_metadata.h> 11 12 /* VXLAN protocol (RFC 7348) header: 13 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 14 * |R|R|R|R|I|R|R|R| Reserved | 15 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 16 * | VXLAN Network Identifier (VNI) | Reserved | 17 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 18 * 19 * I = VXLAN Network Identifier (VNI) present. 20 */ 21 struct vxlanhdr { 22 __be32 vx_flags; 23 __be32 vx_vni; 24 }; 25 26 /* VXLAN header flags. */ 27 #define VXLAN_HF_VNI BIT(27) 28 29 #define VXLAN_N_VID (1u << 24) 30 #define VXLAN_VID_MASK (VXLAN_N_VID - 1) 31 #define VXLAN_VNI_MASK (VXLAN_VID_MASK << 8) 32 #define VXLAN_HLEN (sizeof(struct udphdr) + sizeof(struct vxlanhdr)) 33 34 #define VNI_HASH_BITS 10 35 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS) 36 #define FDB_HASH_BITS 8 37 #define FDB_HASH_SIZE (1<<FDB_HASH_BITS) 38 39 /* Remote checksum offload for VXLAN (VXLAN_F_REMCSUM_[RT]X): 40 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 41 * |R|R|R|R|I|R|R|R|R|R|C| Reserved | 42 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 43 * | VXLAN Network Identifier (VNI) |O| Csum start | 44 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 45 * 46 * C = Remote checksum offload bit. When set indicates that the 47 * remote checksum offload data is present. 48 * 49 * O = Offset bit. Indicates the checksum offset relative to 50 * checksum start. 51 * 52 * Csum start = Checksum start divided by two. 53 * 54 * http://tools.ietf.org/html/draft-herbert-vxlan-rco 55 */ 56 57 /* VXLAN-RCO header flags. */ 58 #define VXLAN_HF_RCO BIT(21) 59 60 /* Remote checksum offload header option */ 61 #define VXLAN_RCO_MASK 0x7f /* Last byte of vni field */ 62 #define VXLAN_RCO_UDP 0x80 /* Indicate UDP RCO (TCP when not set *) */ 63 #define VXLAN_RCO_SHIFT 1 /* Left shift of start */ 64 #define VXLAN_RCO_SHIFT_MASK ((1 << VXLAN_RCO_SHIFT) - 1) 65 #define VXLAN_MAX_REMCSUM_START (VXLAN_RCO_MASK << VXLAN_RCO_SHIFT) 66 67 /* 68 * VXLAN Group Based Policy Extension (VXLAN_F_GBP): 69 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 70 * |G|R|R|R|I|R|R|R|R|D|R|R|A|R|R|R| Group Policy ID | 71 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 72 * | VXLAN Network Identifier (VNI) | Reserved | 73 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 74 * 75 * G = Group Policy ID present. 76 * 77 * D = Don't Learn bit. When set, this bit indicates that the egress 78 * VTEP MUST NOT learn the source address of the encapsulated frame. 79 * 80 * A = Indicates that the group policy has already been applied to 81 * this packet. Policies MUST NOT be applied by devices when the 82 * A bit is set. 83 * 84 * https://tools.ietf.org/html/draft-smith-vxlan-group-policy 85 */ 86 struct vxlanhdr_gbp { 87 u8 vx_flags; 88 #ifdef __LITTLE_ENDIAN_BITFIELD 89 u8 reserved_flags1:3, 90 policy_applied:1, 91 reserved_flags2:2, 92 dont_learn:1, 93 reserved_flags3:1; 94 #elif defined(__BIG_ENDIAN_BITFIELD) 95 u8 reserved_flags1:1, 96 dont_learn:1, 97 reserved_flags2:2, 98 policy_applied:1, 99 reserved_flags3:3; 100 #else 101 #error "Please fix <asm/byteorder.h>" 102 #endif 103 __be16 policy_id; 104 __be32 vx_vni; 105 }; 106 107 /* VXLAN-GBP header flags. */ 108 #define VXLAN_HF_GBP BIT(31) 109 110 #define VXLAN_GBP_USED_BITS (VXLAN_HF_GBP | 0xFFFFFF) 111 112 /* skb->mark mapping 113 * 114 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 115 * |R|R|R|R|R|R|R|R|R|D|R|R|A|R|R|R| Group Policy ID | 116 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 117 */ 118 #define VXLAN_GBP_DONT_LEARN (BIT(6) << 16) 119 #define VXLAN_GBP_POLICY_APPLIED (BIT(3) << 16) 120 #define VXLAN_GBP_ID_MASK (0xFFFF) 121 122 struct vxlan_metadata { 123 u32 gbp; 124 }; 125 126 /* per UDP socket information */ 127 struct vxlan_sock { 128 struct hlist_node hlist; 129 struct work_struct del_work; 130 struct socket *sock; 131 struct rcu_head rcu; 132 struct hlist_head vni_list[VNI_HASH_SIZE]; 133 atomic_t refcnt; 134 struct udp_offload udp_offloads; 135 u32 flags; 136 }; 137 138 union vxlan_addr { 139 struct sockaddr_in sin; 140 struct sockaddr_in6 sin6; 141 struct sockaddr sa; 142 }; 143 144 struct vxlan_rdst { 145 union vxlan_addr remote_ip; 146 __be16 remote_port; 147 u32 remote_vni; 148 u32 remote_ifindex; 149 struct list_head list; 150 struct rcu_head rcu; 151 }; 152 153 struct vxlan_config { 154 union vxlan_addr remote_ip; 155 union vxlan_addr saddr; 156 u32 vni; 157 int remote_ifindex; 158 int mtu; 159 __be16 dst_port; 160 u16 port_min; 161 u16 port_max; 162 u8 tos; 163 u8 ttl; 164 u32 flags; 165 unsigned long age_interval; 166 unsigned int addrmax; 167 bool no_share; 168 }; 169 170 /* Pseudo network device */ 171 struct vxlan_dev { 172 struct hlist_node hlist; /* vni hash table */ 173 struct list_head next; /* vxlan's per namespace list */ 174 struct vxlan_sock *vn4_sock; /* listening socket for IPv4 */ 175 #if IS_ENABLED(CONFIG_IPV6) 176 struct vxlan_sock *vn6_sock; /* listening socket for IPv6 */ 177 #endif 178 struct net_device *dev; 179 struct net *net; /* netns for packet i/o */ 180 struct vxlan_rdst default_dst; /* default destination */ 181 u32 flags; /* VXLAN_F_* in vxlan.h */ 182 183 struct timer_list age_timer; 184 spinlock_t hash_lock; 185 unsigned int addrcnt; 186 struct gro_cells gro_cells; 187 188 struct vxlan_config cfg; 189 190 struct hlist_head fdb_head[FDB_HASH_SIZE]; 191 }; 192 193 #define VXLAN_F_LEARN 0x01 194 #define VXLAN_F_PROXY 0x02 195 #define VXLAN_F_RSC 0x04 196 #define VXLAN_F_L2MISS 0x08 197 #define VXLAN_F_L3MISS 0x10 198 #define VXLAN_F_IPV6 0x20 199 #define VXLAN_F_UDP_CSUM 0x40 200 #define VXLAN_F_UDP_ZERO_CSUM6_TX 0x80 201 #define VXLAN_F_UDP_ZERO_CSUM6_RX 0x100 202 #define VXLAN_F_REMCSUM_TX 0x200 203 #define VXLAN_F_REMCSUM_RX 0x400 204 #define VXLAN_F_GBP 0x800 205 #define VXLAN_F_REMCSUM_NOPARTIAL 0x1000 206 #define VXLAN_F_COLLECT_METADATA 0x2000 207 208 /* Flags that are used in the receive path. These flags must match in 209 * order for a socket to be shareable 210 */ 211 #define VXLAN_F_RCV_FLAGS (VXLAN_F_GBP | \ 212 VXLAN_F_UDP_ZERO_CSUM6_RX | \ 213 VXLAN_F_REMCSUM_RX | \ 214 VXLAN_F_REMCSUM_NOPARTIAL | \ 215 VXLAN_F_COLLECT_METADATA) 216 217 struct net_device *vxlan_dev_create(struct net *net, const char *name, 218 u8 name_assign_type, struct vxlan_config *conf); 219 220 static inline __be16 vxlan_dev_dst_port(struct vxlan_dev *vxlan, 221 unsigned short family) 222 { 223 #if IS_ENABLED(CONFIG_IPV6) 224 if (family == AF_INET6) 225 return inet_sk(vxlan->vn6_sock->sock->sk)->inet_sport; 226 #endif 227 return inet_sk(vxlan->vn4_sock->sock->sk)->inet_sport; 228 } 229 230 static inline netdev_features_t vxlan_features_check(struct sk_buff *skb, 231 netdev_features_t features) 232 { 233 u8 l4_hdr = 0; 234 235 if (!skb->encapsulation) 236 return features; 237 238 switch (vlan_get_protocol(skb)) { 239 case htons(ETH_P_IP): 240 l4_hdr = ip_hdr(skb)->protocol; 241 break; 242 case htons(ETH_P_IPV6): 243 l4_hdr = ipv6_hdr(skb)->nexthdr; 244 break; 245 default: 246 return features;; 247 } 248 249 if ((l4_hdr == IPPROTO_UDP) && 250 (skb->inner_protocol_type != ENCAP_TYPE_ETHER || 251 skb->inner_protocol != htons(ETH_P_TEB) || 252 (skb_inner_mac_header(skb) - skb_transport_header(skb) != 253 sizeof(struct udphdr) + sizeof(struct vxlanhdr)))) 254 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 255 256 return features; 257 } 258 259 /* IP header + UDP + VXLAN + Ethernet header */ 260 #define VXLAN_HEADROOM (20 + 8 + 8 + 14) 261 /* IPv6 header + UDP + VXLAN + Ethernet header */ 262 #define VXLAN6_HEADROOM (40 + 8 + 8 + 14) 263 264 #if IS_ENABLED(CONFIG_VXLAN) 265 void vxlan_get_rx_port(struct net_device *netdev); 266 #else 267 static inline void vxlan_get_rx_port(struct net_device *netdev) 268 { 269 } 270 #endif 271 272 static inline unsigned short vxlan_get_sk_family(struct vxlan_sock *vs) 273 { 274 return vs->sock->sk->sk_family; 275 } 276 277 #endif 278