1 /* Copyright (c) 2017 Facebook 2 * 3 * This program is free software; you can redistribute it and/or 4 * modify it under the terms of version 2 of the GNU General Public 5 * License as published by the Free Software Foundation. 6 */ 7 #include <stddef.h> 8 #include <stdbool.h> 9 #include <string.h> 10 #include <linux/pkt_cls.h> 11 #include <linux/bpf.h> 12 #include <linux/in.h> 13 #include <linux/if_ether.h> 14 #include <linux/ip.h> 15 #include <linux/ipv6.h> 16 #include <linux/icmp.h> 17 #include <linux/icmpv6.h> 18 #include <linux/tcp.h> 19 #include <linux/udp.h> 20 #include "bpf_helpers.h" 21 #include "test_iptunnel_common.h" 22 #include "bpf_endian.h" 23 24 int _version SEC("version") = 1; 25 26 static inline __u32 rol32(__u32 word, unsigned int shift) 27 { 28 return (word << shift) | (word >> ((-shift) & 31)); 29 } 30 31 /* copy paste of jhash from kernel sources to make sure llvm 32 * can compile it into valid sequence of bpf instructions 33 */ 34 #define __jhash_mix(a, b, c) \ 35 { \ 36 a -= c; a ^= rol32(c, 4); c += b; \ 37 b -= a; b ^= rol32(a, 6); a += c; \ 38 c -= b; c ^= rol32(b, 8); b += a; \ 39 a -= c; a ^= rol32(c, 16); c += b; \ 40 b -= a; b ^= rol32(a, 19); a += c; \ 41 c -= b; c ^= rol32(b, 4); b += a; \ 42 } 43 44 #define __jhash_final(a, b, c) \ 45 { \ 46 c ^= b; c -= rol32(b, 14); \ 47 a ^= c; a -= rol32(c, 11); \ 48 b ^= a; b -= rol32(a, 25); \ 49 c ^= b; c -= rol32(b, 16); \ 50 a ^= c; a -= rol32(c, 4); \ 51 b ^= a; b -= rol32(a, 14); \ 52 c ^= b; c -= rol32(b, 24); \ 53 } 54 55 #define JHASH_INITVAL 0xdeadbeef 56 57 typedef unsigned int u32; 58 59 static inline u32 jhash(const void *key, u32 length, u32 initval) 60 { 61 u32 a, b, c; 62 const unsigned char *k = key; 63 64 a = b = c = JHASH_INITVAL + length + initval; 65 66 while (length > 12) { 67 a += *(u32 *)(k); 68 b += *(u32 *)(k + 4); 69 c += *(u32 *)(k + 8); 70 __jhash_mix(a, b, c); 71 length -= 12; 72 k += 12; 73 } 74 switch (length) { 75 case 12: c += (u32)k[11]<<24; 76 case 11: c += (u32)k[10]<<16; 77 case 10: c += (u32)k[9]<<8; 78 case 9: c += k[8]; 79 case 8: b += (u32)k[7]<<24; 80 case 7: b += (u32)k[6]<<16; 81 case 6: b += (u32)k[5]<<8; 82 case 5: b += k[4]; 83 case 4: a += (u32)k[3]<<24; 84 case 3: a += (u32)k[2]<<16; 85 case 2: a += (u32)k[1]<<8; 86 case 1: a += k[0]; 87 __jhash_final(a, b, c); 88 case 0: /* Nothing left to add */ 89 break; 90 } 91 92 return c; 93 } 94 95 static inline u32 __jhash_nwords(u32 a, u32 b, u32 c, u32 initval) 96 { 97 a += initval; 98 b += initval; 99 c += initval; 100 __jhash_final(a, b, c); 101 return c; 102 } 103 104 static inline u32 jhash_2words(u32 a, u32 b, u32 initval) 105 { 106 return __jhash_nwords(a, b, 0, initval + JHASH_INITVAL + (2 << 2)); 107 } 108 109 #define PCKT_FRAGMENTED 65343 110 #define IPV4_HDR_LEN_NO_OPT 20 111 #define IPV4_PLUS_ICMP_HDR 28 112 #define IPV6_PLUS_ICMP_HDR 48 113 #define RING_SIZE 2 114 #define MAX_VIPS 12 115 #define MAX_REALS 5 116 #define CTL_MAP_SIZE 16 117 #define CH_RINGS_SIZE (MAX_VIPS * RING_SIZE) 118 #define F_IPV6 (1 << 0) 119 #define F_HASH_NO_SRC_PORT (1 << 0) 120 #define F_ICMP (1 << 0) 121 #define F_SYN_SET (1 << 1) 122 123 struct packet_description { 124 union { 125 __be32 src; 126 __be32 srcv6[4]; 127 }; 128 union { 129 __be32 dst; 130 __be32 dstv6[4]; 131 }; 132 union { 133 __u32 ports; 134 __u16 port16[2]; 135 }; 136 __u8 proto; 137 __u8 flags; 138 }; 139 140 struct ctl_value { 141 union { 142 __u64 value; 143 __u32 ifindex; 144 __u8 mac[6]; 145 }; 146 }; 147 148 struct vip_meta { 149 __u32 flags; 150 __u32 vip_num; 151 }; 152 153 struct real_definition { 154 union { 155 __be32 dst; 156 __be32 dstv6[4]; 157 }; 158 __u8 flags; 159 }; 160 161 struct vip_stats { 162 __u64 bytes; 163 __u64 pkts; 164 }; 165 166 struct eth_hdr { 167 unsigned char eth_dest[ETH_ALEN]; 168 unsigned char eth_source[ETH_ALEN]; 169 unsigned short eth_proto; 170 }; 171 172 struct { 173 __u32 type; 174 __u32 max_entries; 175 struct vip *key; 176 struct vip_meta *value; 177 } vip_map SEC(".maps") = { 178 .type = BPF_MAP_TYPE_HASH, 179 .max_entries = MAX_VIPS, 180 }; 181 182 struct { 183 __u32 type; 184 __u32 max_entries; 185 __u32 *key; 186 __u32 *value; 187 } ch_rings SEC(".maps") = { 188 .type = BPF_MAP_TYPE_ARRAY, 189 .max_entries = CH_RINGS_SIZE, 190 }; 191 192 struct { 193 __u32 type; 194 __u32 max_entries; 195 __u32 *key; 196 struct real_definition *value; 197 } reals SEC(".maps") = { 198 .type = BPF_MAP_TYPE_ARRAY, 199 .max_entries = MAX_REALS, 200 }; 201 202 struct { 203 __u32 type; 204 __u32 max_entries; 205 __u32 *key; 206 struct vip_stats *value; 207 } stats SEC(".maps") = { 208 .type = BPF_MAP_TYPE_PERCPU_ARRAY, 209 .max_entries = MAX_VIPS, 210 }; 211 212 struct { 213 __u32 type; 214 __u32 max_entries; 215 __u32 *key; 216 struct ctl_value *value; 217 } ctl_array SEC(".maps") = { 218 .type = BPF_MAP_TYPE_ARRAY, 219 .max_entries = CTL_MAP_SIZE, 220 }; 221 222 static __always_inline __u32 get_packet_hash(struct packet_description *pckt, 223 bool ipv6) 224 { 225 if (ipv6) 226 return jhash_2words(jhash(pckt->srcv6, 16, MAX_VIPS), 227 pckt->ports, CH_RINGS_SIZE); 228 else 229 return jhash_2words(pckt->src, pckt->ports, CH_RINGS_SIZE); 230 } 231 232 static __always_inline bool get_packet_dst(struct real_definition **real, 233 struct packet_description *pckt, 234 struct vip_meta *vip_info, 235 bool is_ipv6) 236 { 237 __u32 hash = get_packet_hash(pckt, is_ipv6) % RING_SIZE; 238 __u32 key = RING_SIZE * vip_info->vip_num + hash; 239 __u32 *real_pos; 240 241 real_pos = bpf_map_lookup_elem(&ch_rings, &key); 242 if (!real_pos) 243 return false; 244 key = *real_pos; 245 *real = bpf_map_lookup_elem(&reals, &key); 246 if (!(*real)) 247 return false; 248 return true; 249 } 250 251 static __always_inline int parse_icmpv6(void *data, void *data_end, __u64 off, 252 struct packet_description *pckt) 253 { 254 struct icmp6hdr *icmp_hdr; 255 struct ipv6hdr *ip6h; 256 257 icmp_hdr = data + off; 258 if (icmp_hdr + 1 > data_end) 259 return TC_ACT_SHOT; 260 if (icmp_hdr->icmp6_type != ICMPV6_PKT_TOOBIG) 261 return TC_ACT_OK; 262 off += sizeof(struct icmp6hdr); 263 ip6h = data + off; 264 if (ip6h + 1 > data_end) 265 return TC_ACT_SHOT; 266 pckt->proto = ip6h->nexthdr; 267 pckt->flags |= F_ICMP; 268 memcpy(pckt->srcv6, ip6h->daddr.s6_addr32, 16); 269 memcpy(pckt->dstv6, ip6h->saddr.s6_addr32, 16); 270 return TC_ACT_UNSPEC; 271 } 272 273 static __always_inline int parse_icmp(void *data, void *data_end, __u64 off, 274 struct packet_description *pckt) 275 { 276 struct icmphdr *icmp_hdr; 277 struct iphdr *iph; 278 279 icmp_hdr = data + off; 280 if (icmp_hdr + 1 > data_end) 281 return TC_ACT_SHOT; 282 if (icmp_hdr->type != ICMP_DEST_UNREACH || 283 icmp_hdr->code != ICMP_FRAG_NEEDED) 284 return TC_ACT_OK; 285 off += sizeof(struct icmphdr); 286 iph = data + off; 287 if (iph + 1 > data_end) 288 return TC_ACT_SHOT; 289 if (iph->ihl != 5) 290 return TC_ACT_SHOT; 291 pckt->proto = iph->protocol; 292 pckt->flags |= F_ICMP; 293 pckt->src = iph->daddr; 294 pckt->dst = iph->saddr; 295 return TC_ACT_UNSPEC; 296 } 297 298 static __always_inline bool parse_udp(void *data, __u64 off, void *data_end, 299 struct packet_description *pckt) 300 { 301 struct udphdr *udp; 302 udp = data + off; 303 304 if (udp + 1 > data_end) 305 return false; 306 307 if (!(pckt->flags & F_ICMP)) { 308 pckt->port16[0] = udp->source; 309 pckt->port16[1] = udp->dest; 310 } else { 311 pckt->port16[0] = udp->dest; 312 pckt->port16[1] = udp->source; 313 } 314 return true; 315 } 316 317 static __always_inline bool parse_tcp(void *data, __u64 off, void *data_end, 318 struct packet_description *pckt) 319 { 320 struct tcphdr *tcp; 321 322 tcp = data + off; 323 if (tcp + 1 > data_end) 324 return false; 325 326 if (tcp->syn) 327 pckt->flags |= F_SYN_SET; 328 329 if (!(pckt->flags & F_ICMP)) { 330 pckt->port16[0] = tcp->source; 331 pckt->port16[1] = tcp->dest; 332 } else { 333 pckt->port16[0] = tcp->dest; 334 pckt->port16[1] = tcp->source; 335 } 336 return true; 337 } 338 339 static __always_inline int process_packet(void *data, __u64 off, void *data_end, 340 bool is_ipv6, struct __sk_buff *skb) 341 { 342 void *pkt_start = (void *)(long)skb->data; 343 struct packet_description pckt = {}; 344 struct eth_hdr *eth = pkt_start; 345 struct bpf_tunnel_key tkey = {}; 346 struct vip_stats *data_stats; 347 struct real_definition *dst; 348 struct vip_meta *vip_info; 349 struct ctl_value *cval; 350 __u32 v4_intf_pos = 1; 351 __u32 v6_intf_pos = 2; 352 struct ipv6hdr *ip6h; 353 struct vip vip = {}; 354 struct iphdr *iph; 355 int tun_flag = 0; 356 __u16 pkt_bytes; 357 __u64 iph_len; 358 __u32 ifindex; 359 __u8 protocol; 360 __u32 vip_num; 361 int action; 362 363 tkey.tunnel_ttl = 64; 364 if (is_ipv6) { 365 ip6h = data + off; 366 if (ip6h + 1 > data_end) 367 return TC_ACT_SHOT; 368 369 iph_len = sizeof(struct ipv6hdr); 370 protocol = ip6h->nexthdr; 371 pckt.proto = protocol; 372 pkt_bytes = bpf_ntohs(ip6h->payload_len); 373 off += iph_len; 374 if (protocol == IPPROTO_FRAGMENT) { 375 return TC_ACT_SHOT; 376 } else if (protocol == IPPROTO_ICMPV6) { 377 action = parse_icmpv6(data, data_end, off, &pckt); 378 if (action >= 0) 379 return action; 380 off += IPV6_PLUS_ICMP_HDR; 381 } else { 382 memcpy(pckt.srcv6, ip6h->saddr.s6_addr32, 16); 383 memcpy(pckt.dstv6, ip6h->daddr.s6_addr32, 16); 384 } 385 } else { 386 iph = data + off; 387 if (iph + 1 > data_end) 388 return TC_ACT_SHOT; 389 if (iph->ihl != 5) 390 return TC_ACT_SHOT; 391 392 protocol = iph->protocol; 393 pckt.proto = protocol; 394 pkt_bytes = bpf_ntohs(iph->tot_len); 395 off += IPV4_HDR_LEN_NO_OPT; 396 397 if (iph->frag_off & PCKT_FRAGMENTED) 398 return TC_ACT_SHOT; 399 if (protocol == IPPROTO_ICMP) { 400 action = parse_icmp(data, data_end, off, &pckt); 401 if (action >= 0) 402 return action; 403 off += IPV4_PLUS_ICMP_HDR; 404 } else { 405 pckt.src = iph->saddr; 406 pckt.dst = iph->daddr; 407 } 408 } 409 protocol = pckt.proto; 410 411 if (protocol == IPPROTO_TCP) { 412 if (!parse_tcp(data, off, data_end, &pckt)) 413 return TC_ACT_SHOT; 414 } else if (protocol == IPPROTO_UDP) { 415 if (!parse_udp(data, off, data_end, &pckt)) 416 return TC_ACT_SHOT; 417 } else { 418 return TC_ACT_SHOT; 419 } 420 421 if (is_ipv6) 422 memcpy(vip.daddr.v6, pckt.dstv6, 16); 423 else 424 vip.daddr.v4 = pckt.dst; 425 426 vip.dport = pckt.port16[1]; 427 vip.protocol = pckt.proto; 428 vip_info = bpf_map_lookup_elem(&vip_map, &vip); 429 if (!vip_info) { 430 vip.dport = 0; 431 vip_info = bpf_map_lookup_elem(&vip_map, &vip); 432 if (!vip_info) 433 return TC_ACT_SHOT; 434 pckt.port16[1] = 0; 435 } 436 437 if (vip_info->flags & F_HASH_NO_SRC_PORT) 438 pckt.port16[0] = 0; 439 440 if (!get_packet_dst(&dst, &pckt, vip_info, is_ipv6)) 441 return TC_ACT_SHOT; 442 443 if (dst->flags & F_IPV6) { 444 cval = bpf_map_lookup_elem(&ctl_array, &v6_intf_pos); 445 if (!cval) 446 return TC_ACT_SHOT; 447 ifindex = cval->ifindex; 448 memcpy(tkey.remote_ipv6, dst->dstv6, 16); 449 tun_flag = BPF_F_TUNINFO_IPV6; 450 } else { 451 cval = bpf_map_lookup_elem(&ctl_array, &v4_intf_pos); 452 if (!cval) 453 return TC_ACT_SHOT; 454 ifindex = cval->ifindex; 455 tkey.remote_ipv4 = dst->dst; 456 } 457 vip_num = vip_info->vip_num; 458 data_stats = bpf_map_lookup_elem(&stats, &vip_num); 459 if (!data_stats) 460 return TC_ACT_SHOT; 461 data_stats->pkts++; 462 data_stats->bytes += pkt_bytes; 463 bpf_skb_set_tunnel_key(skb, &tkey, sizeof(tkey), tun_flag); 464 *(u32 *)eth->eth_dest = tkey.remote_ipv4; 465 return bpf_redirect(ifindex, 0); 466 } 467 468 SEC("l4lb-demo") 469 int balancer_ingress(struct __sk_buff *ctx) 470 { 471 void *data_end = (void *)(long)ctx->data_end; 472 void *data = (void *)(long)ctx->data; 473 struct eth_hdr *eth = data; 474 __u32 eth_proto; 475 __u32 nh_off; 476 477 nh_off = sizeof(struct eth_hdr); 478 if (data + nh_off > data_end) 479 return TC_ACT_SHOT; 480 eth_proto = eth->eth_proto; 481 if (eth_proto == bpf_htons(ETH_P_IP)) 482 return process_packet(data, nh_off, data_end, false, ctx); 483 else if (eth_proto == bpf_htons(ETH_P_IPV6)) 484 return process_packet(data, nh_off, data_end, true, ctx); 485 else 486 return TC_ACT_SHOT; 487 } 488 char _license[] SEC("license") = "GPL"; 489