1 // SPDX-License-Identifier: GPL-2.0 2 3 /* In-place tunneling */ 4 5 #include <stdbool.h> 6 #include <string.h> 7 8 #include <linux/stddef.h> 9 #include <linux/bpf.h> 10 #include <linux/if_ether.h> 11 #include <linux/in.h> 12 #include <linux/ip.h> 13 #include <linux/ipv6.h> 14 #include <linux/mpls.h> 15 #include <linux/tcp.h> 16 #include <linux/udp.h> 17 #include <linux/pkt_cls.h> 18 #include <linux/types.h> 19 20 #include <bpf/bpf_endian.h> 21 #include <bpf/bpf_helpers.h> 22 23 static const int cfg_port = 8000; 24 25 static const int cfg_udp_src = 20000; 26 27 #define L2_PAD_SZ (sizeof(struct vxlanhdr) + ETH_HLEN) 28 29 #define UDP_PORT 5555 30 #define MPLS_OVER_UDP_PORT 6635 31 #define ETH_OVER_UDP_PORT 7777 32 #define VXLAN_UDP_PORT 8472 33 34 #define EXTPROTO_VXLAN 0x1 35 36 #define VXLAN_N_VID (1u << 24) 37 #define VXLAN_VNI_MASK bpf_htonl((VXLAN_N_VID - 1) << 8) 38 #define VXLAN_FLAGS 0x8 39 #define VXLAN_VNI 1 40 41 /* MPLS label 1000 with S bit (last label) set and ttl of 255. */ 42 static const __u32 mpls_label = __bpf_constant_htonl(1000 << 12 | 43 MPLS_LS_S_MASK | 0xff); 44 45 struct vxlanhdr { 46 __be32 vx_flags; 47 __be32 vx_vni; 48 } __attribute__((packed)); 49 50 struct gre_hdr { 51 __be16 flags; 52 __be16 protocol; 53 } __attribute__((packed)); 54 55 union l4hdr { 56 struct udphdr udp; 57 struct gre_hdr gre; 58 }; 59 60 struct v4hdr { 61 struct iphdr ip; 62 union l4hdr l4hdr; 63 __u8 pad[L2_PAD_SZ]; /* space for L2 header / vxlan header ... */ 64 } __attribute__((packed)); 65 66 struct v6hdr { 67 struct ipv6hdr ip; 68 union l4hdr l4hdr; 69 __u8 pad[L2_PAD_SZ]; /* space for L2 header / vxlan header ... */ 70 } __attribute__((packed)); 71 72 static __always_inline void set_ipv4_csum(struct iphdr *iph) 73 { 74 __u16 *iph16 = (__u16 *)iph; 75 __u32 csum; 76 int i; 77 78 iph->check = 0; 79 80 #pragma clang loop unroll(full) 81 for (i = 0, csum = 0; i < sizeof(*iph) >> 1; i++) 82 csum += *iph16++; 83 84 iph->check = ~((csum & 0xffff) + (csum >> 16)); 85 } 86 87 static __always_inline int __encap_ipv4(struct __sk_buff *skb, __u8 encap_proto, 88 __u16 l2_proto, __u16 ext_proto) 89 { 90 __u16 udp_dst = UDP_PORT; 91 struct iphdr iph_inner; 92 struct v4hdr h_outer; 93 struct tcphdr tcph; 94 int olen, l2_len; 95 __u8 *l2_hdr = NULL; 96 int tcp_off; 97 __u64 flags; 98 99 /* Most tests encapsulate a packet into a tunnel with the same 100 * network protocol, and derive the outer header fields from 101 * the inner header. 102 * 103 * The 6in4 case tests different inner and outer protocols. As 104 * the inner is ipv6, but the outer expects an ipv4 header as 105 * input, manually build a struct iphdr based on the ipv6hdr. 106 */ 107 if (encap_proto == IPPROTO_IPV6) { 108 const __u32 saddr = (192 << 24) | (168 << 16) | (1 << 8) | 1; 109 const __u32 daddr = (192 << 24) | (168 << 16) | (1 << 8) | 2; 110 struct ipv6hdr iph6_inner; 111 112 /* Read the IPv6 header */ 113 if (bpf_skb_load_bytes(skb, ETH_HLEN, &iph6_inner, 114 sizeof(iph6_inner)) < 0) 115 return TC_ACT_OK; 116 117 /* Derive the IPv4 header fields from the IPv6 header */ 118 memset(&iph_inner, 0, sizeof(iph_inner)); 119 iph_inner.version = 4; 120 iph_inner.ihl = 5; 121 iph_inner.tot_len = bpf_htons(sizeof(iph6_inner) + 122 bpf_ntohs(iph6_inner.payload_len)); 123 iph_inner.ttl = iph6_inner.hop_limit - 1; 124 iph_inner.protocol = iph6_inner.nexthdr; 125 iph_inner.saddr = __bpf_constant_htonl(saddr); 126 iph_inner.daddr = __bpf_constant_htonl(daddr); 127 128 tcp_off = sizeof(iph6_inner); 129 } else { 130 if (bpf_skb_load_bytes(skb, ETH_HLEN, &iph_inner, 131 sizeof(iph_inner)) < 0) 132 return TC_ACT_OK; 133 134 tcp_off = sizeof(iph_inner); 135 } 136 137 /* filter only packets we want */ 138 if (iph_inner.ihl != 5 || iph_inner.protocol != IPPROTO_TCP) 139 return TC_ACT_OK; 140 141 if (bpf_skb_load_bytes(skb, ETH_HLEN + tcp_off, 142 &tcph, sizeof(tcph)) < 0) 143 return TC_ACT_OK; 144 145 if (tcph.dest != __bpf_constant_htons(cfg_port)) 146 return TC_ACT_OK; 147 148 olen = sizeof(h_outer.ip); 149 l2_len = 0; 150 151 flags = BPF_F_ADJ_ROOM_FIXED_GSO | BPF_F_ADJ_ROOM_ENCAP_L3_IPV4; 152 153 switch (l2_proto) { 154 case ETH_P_MPLS_UC: 155 l2_len = sizeof(mpls_label); 156 udp_dst = MPLS_OVER_UDP_PORT; 157 break; 158 case ETH_P_TEB: 159 l2_len = ETH_HLEN; 160 if (ext_proto & EXTPROTO_VXLAN) { 161 udp_dst = VXLAN_UDP_PORT; 162 l2_len += sizeof(struct vxlanhdr); 163 } else 164 udp_dst = ETH_OVER_UDP_PORT; 165 break; 166 } 167 flags |= BPF_F_ADJ_ROOM_ENCAP_L2(l2_len); 168 169 switch (encap_proto) { 170 case IPPROTO_GRE: 171 flags |= BPF_F_ADJ_ROOM_ENCAP_L4_GRE; 172 olen += sizeof(h_outer.l4hdr.gre); 173 h_outer.l4hdr.gre.protocol = bpf_htons(l2_proto); 174 h_outer.l4hdr.gre.flags = 0; 175 break; 176 case IPPROTO_UDP: 177 flags |= BPF_F_ADJ_ROOM_ENCAP_L4_UDP; 178 olen += sizeof(h_outer.l4hdr.udp); 179 h_outer.l4hdr.udp.source = __bpf_constant_htons(cfg_udp_src); 180 h_outer.l4hdr.udp.dest = bpf_htons(udp_dst); 181 h_outer.l4hdr.udp.check = 0; 182 h_outer.l4hdr.udp.len = bpf_htons(bpf_ntohs(iph_inner.tot_len) + 183 sizeof(h_outer.l4hdr.udp) + 184 l2_len); 185 break; 186 case IPPROTO_IPIP: 187 case IPPROTO_IPV6: 188 break; 189 default: 190 return TC_ACT_OK; 191 } 192 193 /* add L2 encap (if specified) */ 194 l2_hdr = (__u8 *)&h_outer + olen; 195 switch (l2_proto) { 196 case ETH_P_MPLS_UC: 197 *(__u32 *)l2_hdr = mpls_label; 198 break; 199 case ETH_P_TEB: 200 flags |= BPF_F_ADJ_ROOM_ENCAP_L2_ETH; 201 202 if (ext_proto & EXTPROTO_VXLAN) { 203 struct vxlanhdr *vxlan_hdr = (struct vxlanhdr *)l2_hdr; 204 205 vxlan_hdr->vx_flags = VXLAN_FLAGS; 206 vxlan_hdr->vx_vni = bpf_htonl((VXLAN_VNI & VXLAN_VNI_MASK) << 8); 207 208 l2_hdr += sizeof(struct vxlanhdr); 209 } 210 211 if (bpf_skb_load_bytes(skb, 0, l2_hdr, ETH_HLEN)) 212 return TC_ACT_SHOT; 213 214 break; 215 } 216 olen += l2_len; 217 218 /* add room between mac and network header */ 219 if (bpf_skb_adjust_room(skb, olen, BPF_ADJ_ROOM_MAC, flags)) 220 return TC_ACT_SHOT; 221 222 /* prepare new outer network header */ 223 h_outer.ip = iph_inner; 224 h_outer.ip.tot_len = bpf_htons(olen + 225 bpf_ntohs(h_outer.ip.tot_len)); 226 h_outer.ip.protocol = encap_proto; 227 228 set_ipv4_csum((void *)&h_outer.ip); 229 230 /* store new outer network header */ 231 if (bpf_skb_store_bytes(skb, ETH_HLEN, &h_outer, olen, 232 BPF_F_INVALIDATE_HASH) < 0) 233 return TC_ACT_SHOT; 234 235 /* if changing outer proto type, update eth->h_proto */ 236 if (encap_proto == IPPROTO_IPV6) { 237 struct ethhdr eth; 238 239 if (bpf_skb_load_bytes(skb, 0, ð, sizeof(eth)) < 0) 240 return TC_ACT_SHOT; 241 eth.h_proto = bpf_htons(ETH_P_IP); 242 if (bpf_skb_store_bytes(skb, 0, ð, sizeof(eth), 0) < 0) 243 return TC_ACT_SHOT; 244 } 245 246 return TC_ACT_OK; 247 } 248 249 static __always_inline int encap_ipv4(struct __sk_buff *skb, __u8 encap_proto, 250 __u16 l2_proto) 251 { 252 return __encap_ipv4(skb, encap_proto, l2_proto, 0); 253 } 254 255 static __always_inline int __encap_ipv6(struct __sk_buff *skb, __u8 encap_proto, 256 __u16 l2_proto, __u16 ext_proto) 257 { 258 __u16 udp_dst = UDP_PORT; 259 struct ipv6hdr iph_inner; 260 struct v6hdr h_outer; 261 struct tcphdr tcph; 262 int olen, l2_len; 263 __u8 *l2_hdr = NULL; 264 __u16 tot_len; 265 __u64 flags; 266 267 if (bpf_skb_load_bytes(skb, ETH_HLEN, &iph_inner, 268 sizeof(iph_inner)) < 0) 269 return TC_ACT_OK; 270 271 /* filter only packets we want */ 272 if (bpf_skb_load_bytes(skb, ETH_HLEN + sizeof(iph_inner), 273 &tcph, sizeof(tcph)) < 0) 274 return TC_ACT_OK; 275 276 if (tcph.dest != __bpf_constant_htons(cfg_port)) 277 return TC_ACT_OK; 278 279 olen = sizeof(h_outer.ip); 280 l2_len = 0; 281 282 flags = BPF_F_ADJ_ROOM_FIXED_GSO | BPF_F_ADJ_ROOM_ENCAP_L3_IPV6; 283 284 switch (l2_proto) { 285 case ETH_P_MPLS_UC: 286 l2_len = sizeof(mpls_label); 287 udp_dst = MPLS_OVER_UDP_PORT; 288 break; 289 case ETH_P_TEB: 290 l2_len = ETH_HLEN; 291 if (ext_proto & EXTPROTO_VXLAN) { 292 udp_dst = VXLAN_UDP_PORT; 293 l2_len += sizeof(struct vxlanhdr); 294 } else 295 udp_dst = ETH_OVER_UDP_PORT; 296 break; 297 } 298 flags |= BPF_F_ADJ_ROOM_ENCAP_L2(l2_len); 299 300 switch (encap_proto) { 301 case IPPROTO_GRE: 302 flags |= BPF_F_ADJ_ROOM_ENCAP_L4_GRE; 303 olen += sizeof(h_outer.l4hdr.gre); 304 h_outer.l4hdr.gre.protocol = bpf_htons(l2_proto); 305 h_outer.l4hdr.gre.flags = 0; 306 break; 307 case IPPROTO_UDP: 308 flags |= BPF_F_ADJ_ROOM_ENCAP_L4_UDP; 309 olen += sizeof(h_outer.l4hdr.udp); 310 h_outer.l4hdr.udp.source = __bpf_constant_htons(cfg_udp_src); 311 h_outer.l4hdr.udp.dest = bpf_htons(udp_dst); 312 tot_len = bpf_ntohs(iph_inner.payload_len) + sizeof(iph_inner) + 313 sizeof(h_outer.l4hdr.udp) + l2_len; 314 h_outer.l4hdr.udp.check = 0; 315 h_outer.l4hdr.udp.len = bpf_htons(tot_len); 316 break; 317 case IPPROTO_IPV6: 318 break; 319 default: 320 return TC_ACT_OK; 321 } 322 323 /* add L2 encap (if specified) */ 324 l2_hdr = (__u8 *)&h_outer + olen; 325 switch (l2_proto) { 326 case ETH_P_MPLS_UC: 327 *(__u32 *)l2_hdr = mpls_label; 328 break; 329 case ETH_P_TEB: 330 flags |= BPF_F_ADJ_ROOM_ENCAP_L2_ETH; 331 332 if (ext_proto & EXTPROTO_VXLAN) { 333 struct vxlanhdr *vxlan_hdr = (struct vxlanhdr *)l2_hdr; 334 335 vxlan_hdr->vx_flags = VXLAN_FLAGS; 336 vxlan_hdr->vx_vni = bpf_htonl((VXLAN_VNI & VXLAN_VNI_MASK) << 8); 337 338 l2_hdr += sizeof(struct vxlanhdr); 339 } 340 341 if (bpf_skb_load_bytes(skb, 0, l2_hdr, ETH_HLEN)) 342 return TC_ACT_SHOT; 343 break; 344 } 345 olen += l2_len; 346 347 /* add room between mac and network header */ 348 if (bpf_skb_adjust_room(skb, olen, BPF_ADJ_ROOM_MAC, flags)) 349 return TC_ACT_SHOT; 350 351 /* prepare new outer network header */ 352 h_outer.ip = iph_inner; 353 h_outer.ip.payload_len = bpf_htons(olen + 354 bpf_ntohs(h_outer.ip.payload_len)); 355 356 h_outer.ip.nexthdr = encap_proto; 357 358 /* store new outer network header */ 359 if (bpf_skb_store_bytes(skb, ETH_HLEN, &h_outer, olen, 360 BPF_F_INVALIDATE_HASH) < 0) 361 return TC_ACT_SHOT; 362 363 return TC_ACT_OK; 364 } 365 366 static __always_inline int encap_ipv6(struct __sk_buff *skb, __u8 encap_proto, 367 __u16 l2_proto) 368 { 369 return __encap_ipv6(skb, encap_proto, l2_proto, 0); 370 } 371 372 SEC("encap_ipip_none") 373 int __encap_ipip_none(struct __sk_buff *skb) 374 { 375 if (skb->protocol == __bpf_constant_htons(ETH_P_IP)) 376 return encap_ipv4(skb, IPPROTO_IPIP, ETH_P_IP); 377 else 378 return TC_ACT_OK; 379 } 380 381 SEC("encap_gre_none") 382 int __encap_gre_none(struct __sk_buff *skb) 383 { 384 if (skb->protocol == __bpf_constant_htons(ETH_P_IP)) 385 return encap_ipv4(skb, IPPROTO_GRE, ETH_P_IP); 386 else 387 return TC_ACT_OK; 388 } 389 390 SEC("encap_gre_mpls") 391 int __encap_gre_mpls(struct __sk_buff *skb) 392 { 393 if (skb->protocol == __bpf_constant_htons(ETH_P_IP)) 394 return encap_ipv4(skb, IPPROTO_GRE, ETH_P_MPLS_UC); 395 else 396 return TC_ACT_OK; 397 } 398 399 SEC("encap_gre_eth") 400 int __encap_gre_eth(struct __sk_buff *skb) 401 { 402 if (skb->protocol == __bpf_constant_htons(ETH_P_IP)) 403 return encap_ipv4(skb, IPPROTO_GRE, ETH_P_TEB); 404 else 405 return TC_ACT_OK; 406 } 407 408 SEC("encap_udp_none") 409 int __encap_udp_none(struct __sk_buff *skb) 410 { 411 if (skb->protocol == __bpf_constant_htons(ETH_P_IP)) 412 return encap_ipv4(skb, IPPROTO_UDP, ETH_P_IP); 413 else 414 return TC_ACT_OK; 415 } 416 417 SEC("encap_udp_mpls") 418 int __encap_udp_mpls(struct __sk_buff *skb) 419 { 420 if (skb->protocol == __bpf_constant_htons(ETH_P_IP)) 421 return encap_ipv4(skb, IPPROTO_UDP, ETH_P_MPLS_UC); 422 else 423 return TC_ACT_OK; 424 } 425 426 SEC("encap_udp_eth") 427 int __encap_udp_eth(struct __sk_buff *skb) 428 { 429 if (skb->protocol == __bpf_constant_htons(ETH_P_IP)) 430 return encap_ipv4(skb, IPPROTO_UDP, ETH_P_TEB); 431 else 432 return TC_ACT_OK; 433 } 434 435 SEC("encap_vxlan_eth") 436 int __encap_vxlan_eth(struct __sk_buff *skb) 437 { 438 if (skb->protocol == __bpf_constant_htons(ETH_P_IP)) 439 return __encap_ipv4(skb, IPPROTO_UDP, 440 ETH_P_TEB, 441 EXTPROTO_VXLAN); 442 else 443 return TC_ACT_OK; 444 } 445 446 SEC("encap_sit_none") 447 int __encap_sit_none(struct __sk_buff *skb) 448 { 449 if (skb->protocol == __bpf_constant_htons(ETH_P_IPV6)) 450 return encap_ipv4(skb, IPPROTO_IPV6, ETH_P_IP); 451 else 452 return TC_ACT_OK; 453 } 454 455 SEC("encap_ip6tnl_none") 456 int __encap_ip6tnl_none(struct __sk_buff *skb) 457 { 458 if (skb->protocol == __bpf_constant_htons(ETH_P_IPV6)) 459 return encap_ipv6(skb, IPPROTO_IPV6, ETH_P_IPV6); 460 else 461 return TC_ACT_OK; 462 } 463 464 SEC("encap_ip6gre_none") 465 int __encap_ip6gre_none(struct __sk_buff *skb) 466 { 467 if (skb->protocol == __bpf_constant_htons(ETH_P_IPV6)) 468 return encap_ipv6(skb, IPPROTO_GRE, ETH_P_IPV6); 469 else 470 return TC_ACT_OK; 471 } 472 473 SEC("encap_ip6gre_mpls") 474 int __encap_ip6gre_mpls(struct __sk_buff *skb) 475 { 476 if (skb->protocol == __bpf_constant_htons(ETH_P_IPV6)) 477 return encap_ipv6(skb, IPPROTO_GRE, ETH_P_MPLS_UC); 478 else 479 return TC_ACT_OK; 480 } 481 482 SEC("encap_ip6gre_eth") 483 int __encap_ip6gre_eth(struct __sk_buff *skb) 484 { 485 if (skb->protocol == __bpf_constant_htons(ETH_P_IPV6)) 486 return encap_ipv6(skb, IPPROTO_GRE, ETH_P_TEB); 487 else 488 return TC_ACT_OK; 489 } 490 491 SEC("encap_ip6udp_none") 492 int __encap_ip6udp_none(struct __sk_buff *skb) 493 { 494 if (skb->protocol == __bpf_constant_htons(ETH_P_IPV6)) 495 return encap_ipv6(skb, IPPROTO_UDP, ETH_P_IPV6); 496 else 497 return TC_ACT_OK; 498 } 499 500 SEC("encap_ip6udp_mpls") 501 int __encap_ip6udp_mpls(struct __sk_buff *skb) 502 { 503 if (skb->protocol == __bpf_constant_htons(ETH_P_IPV6)) 504 return encap_ipv6(skb, IPPROTO_UDP, ETH_P_MPLS_UC); 505 else 506 return TC_ACT_OK; 507 } 508 509 SEC("encap_ip6udp_eth") 510 int __encap_ip6udp_eth(struct __sk_buff *skb) 511 { 512 if (skb->protocol == __bpf_constant_htons(ETH_P_IPV6)) 513 return encap_ipv6(skb, IPPROTO_UDP, ETH_P_TEB); 514 else 515 return TC_ACT_OK; 516 } 517 518 SEC("encap_ip6vxlan_eth") 519 int __encap_ip6vxlan_eth(struct __sk_buff *skb) 520 { 521 if (skb->protocol == __bpf_constant_htons(ETH_P_IPV6)) 522 return __encap_ipv6(skb, IPPROTO_UDP, 523 ETH_P_TEB, 524 EXTPROTO_VXLAN); 525 else 526 return TC_ACT_OK; 527 } 528 529 static int decap_internal(struct __sk_buff *skb, int off, int len, char proto) 530 { 531 char buf[sizeof(struct v6hdr)]; 532 struct gre_hdr greh; 533 struct udphdr udph; 534 int olen = len; 535 536 switch (proto) { 537 case IPPROTO_IPIP: 538 case IPPROTO_IPV6: 539 break; 540 case IPPROTO_GRE: 541 olen += sizeof(struct gre_hdr); 542 if (bpf_skb_load_bytes(skb, off + len, &greh, sizeof(greh)) < 0) 543 return TC_ACT_OK; 544 switch (bpf_ntohs(greh.protocol)) { 545 case ETH_P_MPLS_UC: 546 olen += sizeof(mpls_label); 547 break; 548 case ETH_P_TEB: 549 olen += ETH_HLEN; 550 break; 551 } 552 break; 553 case IPPROTO_UDP: 554 olen += sizeof(struct udphdr); 555 if (bpf_skb_load_bytes(skb, off + len, &udph, sizeof(udph)) < 0) 556 return TC_ACT_OK; 557 switch (bpf_ntohs(udph.dest)) { 558 case MPLS_OVER_UDP_PORT: 559 olen += sizeof(mpls_label); 560 break; 561 case ETH_OVER_UDP_PORT: 562 olen += ETH_HLEN; 563 break; 564 case VXLAN_UDP_PORT: 565 olen += ETH_HLEN + sizeof(struct vxlanhdr); 566 break; 567 } 568 break; 569 default: 570 return TC_ACT_OK; 571 } 572 573 if (bpf_skb_adjust_room(skb, -olen, BPF_ADJ_ROOM_MAC, 574 BPF_F_ADJ_ROOM_FIXED_GSO)) 575 return TC_ACT_SHOT; 576 577 return TC_ACT_OK; 578 } 579 580 static int decap_ipv4(struct __sk_buff *skb) 581 { 582 struct iphdr iph_outer; 583 584 if (bpf_skb_load_bytes(skb, ETH_HLEN, &iph_outer, 585 sizeof(iph_outer)) < 0) 586 return TC_ACT_OK; 587 588 if (iph_outer.ihl != 5) 589 return TC_ACT_OK; 590 591 return decap_internal(skb, ETH_HLEN, sizeof(iph_outer), 592 iph_outer.protocol); 593 } 594 595 static int decap_ipv6(struct __sk_buff *skb) 596 { 597 struct ipv6hdr iph_outer; 598 599 if (bpf_skb_load_bytes(skb, ETH_HLEN, &iph_outer, 600 sizeof(iph_outer)) < 0) 601 return TC_ACT_OK; 602 603 return decap_internal(skb, ETH_HLEN, sizeof(iph_outer), 604 iph_outer.nexthdr); 605 } 606 607 SEC("decap") 608 int decap_f(struct __sk_buff *skb) 609 { 610 switch (skb->protocol) { 611 case __bpf_constant_htons(ETH_P_IP): 612 return decap_ipv4(skb); 613 case __bpf_constant_htons(ETH_P_IPV6): 614 return decap_ipv6(skb); 615 default: 616 /* does not match, ignore */ 617 return TC_ACT_OK; 618 } 619 } 620 621 char __license[] SEC("license") = "GPL"; 622