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