1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * ip_vs_proto_udp.c: UDP load balancing support for IPVS 4 * 5 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org> 6 * Julian Anastasov <ja@ssi.bg> 7 * 8 * Changes: Hans Schillstrom <hans.schillstrom@ericsson.com> 9 * Network name space (netns) aware. 10 */ 11 12 #define KMSG_COMPONENT "IPVS" 13 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt 14 15 #include <linux/in.h> 16 #include <linux/ip.h> 17 #include <linux/kernel.h> 18 #include <linux/netfilter.h> 19 #include <linux/netfilter_ipv4.h> 20 #include <linux/udp.h> 21 #include <linux/indirect_call_wrapper.h> 22 23 #include <net/ip_vs.h> 24 #include <net/ip.h> 25 #include <net/ip6_checksum.h> 26 27 static int 28 udp_csum_check(int af, struct sk_buff *skb, struct ip_vs_protocol *pp); 29 30 static int 31 udp_conn_schedule(struct netns_ipvs *ipvs, int af, struct sk_buff *skb, 32 struct ip_vs_proto_data *pd, 33 int *verdict, struct ip_vs_conn **cpp, 34 struct ip_vs_iphdr *iph) 35 { 36 struct ip_vs_service *svc; 37 struct udphdr _udph, *uh; 38 __be16 _ports[2], *ports = NULL; 39 40 if (likely(!ip_vs_iph_icmp(iph))) { 41 /* IPv6 fragments, only first fragment will hit this */ 42 uh = skb_header_pointer(skb, iph->len, sizeof(_udph), &_udph); 43 if (uh) 44 ports = &uh->source; 45 } else { 46 ports = skb_header_pointer( 47 skb, iph->len, sizeof(_ports), &_ports); 48 } 49 50 if (!ports) { 51 *verdict = NF_DROP; 52 return 0; 53 } 54 55 if (likely(!ip_vs_iph_inverse(iph))) 56 svc = ip_vs_service_find(ipvs, af, skb->mark, iph->protocol, 57 &iph->daddr, ports[1]); 58 else 59 svc = ip_vs_service_find(ipvs, af, skb->mark, iph->protocol, 60 &iph->saddr, ports[0]); 61 62 if (svc) { 63 int ignored; 64 65 if (ip_vs_todrop(ipvs)) { 66 /* 67 * It seems that we are very loaded. 68 * We have to drop this packet :( 69 */ 70 *verdict = NF_DROP; 71 return 0; 72 } 73 74 /* 75 * Let the virtual server select a real server for the 76 * incoming connection, and create a connection entry. 77 */ 78 *cpp = ip_vs_schedule(svc, skb, pd, &ignored, iph); 79 if (!*cpp && ignored <= 0) { 80 if (!ignored) 81 *verdict = ip_vs_leave(svc, skb, pd, iph); 82 else 83 *verdict = NF_DROP; 84 return 0; 85 } 86 } 87 /* NF_ACCEPT */ 88 return 1; 89 } 90 91 92 static inline void 93 udp_fast_csum_update(int af, struct udphdr *uhdr, 94 const union nf_inet_addr *oldip, 95 const union nf_inet_addr *newip, 96 __be16 oldport, __be16 newport) 97 { 98 #ifdef CONFIG_IP_VS_IPV6 99 if (af == AF_INET6) 100 uhdr->check = 101 csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6, 102 ip_vs_check_diff2(oldport, newport, 103 ~csum_unfold(uhdr->check)))); 104 else 105 #endif 106 uhdr->check = 107 csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip, 108 ip_vs_check_diff2(oldport, newport, 109 ~csum_unfold(uhdr->check)))); 110 if (!uhdr->check) 111 uhdr->check = CSUM_MANGLED_0; 112 } 113 114 static inline void 115 udp_partial_csum_update(int af, struct udphdr *uhdr, 116 const union nf_inet_addr *oldip, 117 const union nf_inet_addr *newip, 118 __be16 oldlen, __be16 newlen) 119 { 120 #ifdef CONFIG_IP_VS_IPV6 121 if (af == AF_INET6) 122 uhdr->check = 123 ~csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6, 124 ip_vs_check_diff2(oldlen, newlen, 125 csum_unfold(uhdr->check)))); 126 else 127 #endif 128 uhdr->check = 129 ~csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip, 130 ip_vs_check_diff2(oldlen, newlen, 131 csum_unfold(uhdr->check)))); 132 } 133 134 135 INDIRECT_CALLABLE_SCOPE int 136 udp_snat_handler(struct sk_buff *skb, struct ip_vs_protocol *pp, 137 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph) 138 { 139 struct udphdr *udph; 140 unsigned int udphoff = iph->len; 141 bool payload_csum = false; 142 int oldlen; 143 144 #ifdef CONFIG_IP_VS_IPV6 145 if (cp->af == AF_INET6 && iph->fragoffs) 146 return 1; 147 #endif 148 oldlen = skb->len - udphoff; 149 150 /* csum_check requires unshared skb */ 151 if (!skb_make_writable(skb, udphoff+sizeof(*udph))) 152 return 0; 153 154 if (unlikely(cp->app != NULL)) { 155 int ret; 156 157 /* Some checks before mangling */ 158 if (!udp_csum_check(cp->af, skb, pp)) 159 return 0; 160 161 /* 162 * Call application helper if needed 163 */ 164 if (!(ret = ip_vs_app_pkt_out(cp, skb, iph))) 165 return 0; 166 /* ret=2: csum update is needed after payload mangling */ 167 if (ret == 1) 168 oldlen = skb->len - udphoff; 169 else 170 payload_csum = true; 171 } 172 173 udph = (void *)skb_network_header(skb) + udphoff; 174 udph->source = cp->vport; 175 176 /* 177 * Adjust UDP checksums 178 */ 179 if (skb->ip_summed == CHECKSUM_PARTIAL) { 180 udp_partial_csum_update(cp->af, udph, &cp->daddr, &cp->vaddr, 181 htons(oldlen), 182 htons(skb->len - udphoff)); 183 } else if (!payload_csum && (udph->check != 0)) { 184 /* Only port and addr are changed, do fast csum update */ 185 udp_fast_csum_update(cp->af, udph, &cp->daddr, &cp->vaddr, 186 cp->dport, cp->vport); 187 if (skb->ip_summed == CHECKSUM_COMPLETE) 188 skb->ip_summed = cp->app ? 189 CHECKSUM_UNNECESSARY : CHECKSUM_NONE; 190 } else { 191 /* full checksum calculation */ 192 udph->check = 0; 193 skb->csum = skb_checksum(skb, udphoff, skb->len - udphoff, 0); 194 #ifdef CONFIG_IP_VS_IPV6 195 if (cp->af == AF_INET6) 196 udph->check = csum_ipv6_magic(&cp->vaddr.in6, 197 &cp->caddr.in6, 198 skb->len - udphoff, 199 cp->protocol, skb->csum); 200 else 201 #endif 202 udph->check = csum_tcpudp_magic(cp->vaddr.ip, 203 cp->caddr.ip, 204 skb->len - udphoff, 205 cp->protocol, 206 skb->csum); 207 if (udph->check == 0) 208 udph->check = CSUM_MANGLED_0; 209 skb->ip_summed = CHECKSUM_UNNECESSARY; 210 IP_VS_DBG(11, "O-pkt: %s O-csum=%d (+%zd)\n", 211 pp->name, udph->check, 212 (char*)&(udph->check) - (char*)udph); 213 } 214 return 1; 215 } 216 217 218 static int 219 udp_dnat_handler(struct sk_buff *skb, struct ip_vs_protocol *pp, 220 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph) 221 { 222 struct udphdr *udph; 223 unsigned int udphoff = iph->len; 224 bool payload_csum = false; 225 int oldlen; 226 227 #ifdef CONFIG_IP_VS_IPV6 228 if (cp->af == AF_INET6 && iph->fragoffs) 229 return 1; 230 #endif 231 oldlen = skb->len - udphoff; 232 233 /* csum_check requires unshared skb */ 234 if (!skb_make_writable(skb, udphoff+sizeof(*udph))) 235 return 0; 236 237 if (unlikely(cp->app != NULL)) { 238 int ret; 239 240 /* Some checks before mangling */ 241 if (!udp_csum_check(cp->af, skb, pp)) 242 return 0; 243 244 /* 245 * Attempt ip_vs_app call. 246 * It will fix ip_vs_conn 247 */ 248 if (!(ret = ip_vs_app_pkt_in(cp, skb, iph))) 249 return 0; 250 /* ret=2: csum update is needed after payload mangling */ 251 if (ret == 1) 252 oldlen = skb->len - udphoff; 253 else 254 payload_csum = true; 255 } 256 257 udph = (void *)skb_network_header(skb) + udphoff; 258 udph->dest = cp->dport; 259 260 /* 261 * Adjust UDP checksums 262 */ 263 if (skb->ip_summed == CHECKSUM_PARTIAL) { 264 udp_partial_csum_update(cp->af, udph, &cp->vaddr, &cp->daddr, 265 htons(oldlen), 266 htons(skb->len - udphoff)); 267 } else if (!payload_csum && (udph->check != 0)) { 268 /* Only port and addr are changed, do fast csum update */ 269 udp_fast_csum_update(cp->af, udph, &cp->vaddr, &cp->daddr, 270 cp->vport, cp->dport); 271 if (skb->ip_summed == CHECKSUM_COMPLETE) 272 skb->ip_summed = cp->app ? 273 CHECKSUM_UNNECESSARY : CHECKSUM_NONE; 274 } else { 275 /* full checksum calculation */ 276 udph->check = 0; 277 skb->csum = skb_checksum(skb, udphoff, skb->len - udphoff, 0); 278 #ifdef CONFIG_IP_VS_IPV6 279 if (cp->af == AF_INET6) 280 udph->check = csum_ipv6_magic(&cp->caddr.in6, 281 &cp->daddr.in6, 282 skb->len - udphoff, 283 cp->protocol, skb->csum); 284 else 285 #endif 286 udph->check = csum_tcpudp_magic(cp->caddr.ip, 287 cp->daddr.ip, 288 skb->len - udphoff, 289 cp->protocol, 290 skb->csum); 291 if (udph->check == 0) 292 udph->check = CSUM_MANGLED_0; 293 skb->ip_summed = CHECKSUM_UNNECESSARY; 294 } 295 return 1; 296 } 297 298 299 static int 300 udp_csum_check(int af, struct sk_buff *skb, struct ip_vs_protocol *pp) 301 { 302 struct udphdr _udph, *uh; 303 unsigned int udphoff; 304 305 #ifdef CONFIG_IP_VS_IPV6 306 if (af == AF_INET6) 307 udphoff = sizeof(struct ipv6hdr); 308 else 309 #endif 310 udphoff = ip_hdrlen(skb); 311 312 uh = skb_header_pointer(skb, udphoff, sizeof(_udph), &_udph); 313 if (uh == NULL) 314 return 0; 315 316 if (uh->check != 0) { 317 switch (skb->ip_summed) { 318 case CHECKSUM_NONE: 319 skb->csum = skb_checksum(skb, udphoff, 320 skb->len - udphoff, 0); 321 /* fall through */ 322 case CHECKSUM_COMPLETE: 323 #ifdef CONFIG_IP_VS_IPV6 324 if (af == AF_INET6) { 325 if (csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 326 &ipv6_hdr(skb)->daddr, 327 skb->len - udphoff, 328 ipv6_hdr(skb)->nexthdr, 329 skb->csum)) { 330 IP_VS_DBG_RL_PKT(0, af, pp, skb, 0, 331 "Failed checksum for"); 332 return 0; 333 } 334 } else 335 #endif 336 if (csum_tcpudp_magic(ip_hdr(skb)->saddr, 337 ip_hdr(skb)->daddr, 338 skb->len - udphoff, 339 ip_hdr(skb)->protocol, 340 skb->csum)) { 341 IP_VS_DBG_RL_PKT(0, af, pp, skb, 0, 342 "Failed checksum for"); 343 return 0; 344 } 345 break; 346 default: 347 /* No need to checksum. */ 348 break; 349 } 350 } 351 return 1; 352 } 353 354 static inline __u16 udp_app_hashkey(__be16 port) 355 { 356 return (((__force u16)port >> UDP_APP_TAB_BITS) ^ (__force u16)port) 357 & UDP_APP_TAB_MASK; 358 } 359 360 361 static int udp_register_app(struct netns_ipvs *ipvs, struct ip_vs_app *inc) 362 { 363 struct ip_vs_app *i; 364 __u16 hash; 365 __be16 port = inc->port; 366 int ret = 0; 367 struct ip_vs_proto_data *pd = ip_vs_proto_data_get(ipvs, IPPROTO_UDP); 368 369 hash = udp_app_hashkey(port); 370 371 list_for_each_entry(i, &ipvs->udp_apps[hash], p_list) { 372 if (i->port == port) { 373 ret = -EEXIST; 374 goto out; 375 } 376 } 377 list_add_rcu(&inc->p_list, &ipvs->udp_apps[hash]); 378 atomic_inc(&pd->appcnt); 379 380 out: 381 return ret; 382 } 383 384 385 static void 386 udp_unregister_app(struct netns_ipvs *ipvs, struct ip_vs_app *inc) 387 { 388 struct ip_vs_proto_data *pd = ip_vs_proto_data_get(ipvs, IPPROTO_UDP); 389 390 atomic_dec(&pd->appcnt); 391 list_del_rcu(&inc->p_list); 392 } 393 394 395 static int udp_app_conn_bind(struct ip_vs_conn *cp) 396 { 397 struct netns_ipvs *ipvs = cp->ipvs; 398 int hash; 399 struct ip_vs_app *inc; 400 int result = 0; 401 402 /* Default binding: bind app only for NAT */ 403 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) 404 return 0; 405 406 /* Lookup application incarnations and bind the right one */ 407 hash = udp_app_hashkey(cp->vport); 408 409 list_for_each_entry_rcu(inc, &ipvs->udp_apps[hash], p_list) { 410 if (inc->port == cp->vport) { 411 if (unlikely(!ip_vs_app_inc_get(inc))) 412 break; 413 414 IP_VS_DBG_BUF(9, "%s(): Binding conn %s:%u->" 415 "%s:%u to app %s on port %u\n", 416 __func__, 417 IP_VS_DBG_ADDR(cp->af, &cp->caddr), 418 ntohs(cp->cport), 419 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), 420 ntohs(cp->vport), 421 inc->name, ntohs(inc->port)); 422 423 cp->app = inc; 424 if (inc->init_conn) 425 result = inc->init_conn(inc, cp); 426 break; 427 } 428 } 429 430 return result; 431 } 432 433 434 static const int udp_timeouts[IP_VS_UDP_S_LAST+1] = { 435 [IP_VS_UDP_S_NORMAL] = 5*60*HZ, 436 [IP_VS_UDP_S_LAST] = 2*HZ, 437 }; 438 439 static const char *const udp_state_name_table[IP_VS_UDP_S_LAST+1] = { 440 [IP_VS_UDP_S_NORMAL] = "UDP", 441 [IP_VS_UDP_S_LAST] = "BUG!", 442 }; 443 444 static const char * udp_state_name(int state) 445 { 446 if (state >= IP_VS_UDP_S_LAST) 447 return "ERR!"; 448 return udp_state_name_table[state] ? udp_state_name_table[state] : "?"; 449 } 450 451 static void 452 udp_state_transition(struct ip_vs_conn *cp, int direction, 453 const struct sk_buff *skb, 454 struct ip_vs_proto_data *pd) 455 { 456 if (unlikely(!pd)) { 457 pr_err("UDP no ns data\n"); 458 return; 459 } 460 461 cp->timeout = pd->timeout_table[IP_VS_UDP_S_NORMAL]; 462 if (direction == IP_VS_DIR_OUTPUT) 463 ip_vs_control_assure_ct(cp); 464 } 465 466 static int __udp_init(struct netns_ipvs *ipvs, struct ip_vs_proto_data *pd) 467 { 468 ip_vs_init_hash_table(ipvs->udp_apps, UDP_APP_TAB_SIZE); 469 pd->timeout_table = ip_vs_create_timeout_table((int *)udp_timeouts, 470 sizeof(udp_timeouts)); 471 if (!pd->timeout_table) 472 return -ENOMEM; 473 return 0; 474 } 475 476 static void __udp_exit(struct netns_ipvs *ipvs, struct ip_vs_proto_data *pd) 477 { 478 kfree(pd->timeout_table); 479 } 480 481 482 struct ip_vs_protocol ip_vs_protocol_udp = { 483 .name = "UDP", 484 .protocol = IPPROTO_UDP, 485 .num_states = IP_VS_UDP_S_LAST, 486 .dont_defrag = 0, 487 .init = NULL, 488 .exit = NULL, 489 .init_netns = __udp_init, 490 .exit_netns = __udp_exit, 491 .conn_schedule = udp_conn_schedule, 492 .conn_in_get = ip_vs_conn_in_get_proto, 493 .conn_out_get = ip_vs_conn_out_get_proto, 494 .snat_handler = udp_snat_handler, 495 .dnat_handler = udp_dnat_handler, 496 .state_transition = udp_state_transition, 497 .state_name = udp_state_name, 498 .register_app = udp_register_app, 499 .unregister_app = udp_unregister_app, 500 .app_conn_bind = udp_app_conn_bind, 501 .debug_packet = ip_vs_tcpudp_debug_packet, 502 .timeout_change = NULL, 503 }; 504