1 /* 2 * ip_vs_proto_tcp.c: TCP 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: Hans Schillstrom <hans.schillstrom@ericsson.com> 13 * 14 * Network name space (netns) aware. 15 * Global data moved to netns i.e struct netns_ipvs 16 * tcp_timeouts table has copy per netns in a hash table per 17 * protocol ip_vs_proto_data and is handled by netns 18 */ 19 20 #define KMSG_COMPONENT "IPVS" 21 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt 22 23 #include <linux/kernel.h> 24 #include <linux/ip.h> 25 #include <linux/tcp.h> /* for tcphdr */ 26 #include <net/ip.h> 27 #include <net/tcp.h> /* for csum_tcpudp_magic */ 28 #include <net/ip6_checksum.h> 29 #include <linux/netfilter.h> 30 #include <linux/netfilter_ipv4.h> 31 32 #include <net/ip_vs.h> 33 34 static int 35 tcp_conn_schedule(struct netns_ipvs *ipvs, int af, struct sk_buff *skb, 36 struct ip_vs_proto_data *pd, 37 int *verdict, struct ip_vs_conn **cpp, 38 struct ip_vs_iphdr *iph) 39 { 40 struct ip_vs_service *svc; 41 struct tcphdr _tcph, *th; 42 __be16 _ports[2], *ports = NULL; 43 44 /* In the event of icmp, we're only guaranteed to have the first 8 45 * bytes of the transport header, so we only check the rest of the 46 * TCP packet for non-ICMP packets 47 */ 48 if (likely(!ip_vs_iph_icmp(iph))) { 49 th = skb_header_pointer(skb, iph->len, sizeof(_tcph), &_tcph); 50 if (th) { 51 if (th->rst || !(sysctl_sloppy_tcp(ipvs) || th->syn)) 52 return 1; 53 ports = &th->source; 54 } 55 } else { 56 ports = skb_header_pointer( 57 skb, iph->len, sizeof(_ports), &_ports); 58 } 59 60 if (!ports) { 61 *verdict = NF_DROP; 62 return 0; 63 } 64 65 /* No !th->ack check to allow scheduling on SYN+ACK for Active FTP */ 66 67 if (likely(!ip_vs_iph_inverse(iph))) 68 svc = ip_vs_service_find(ipvs, af, skb->mark, iph->protocol, 69 &iph->daddr, ports[1]); 70 else 71 svc = ip_vs_service_find(ipvs, af, skb->mark, iph->protocol, 72 &iph->saddr, ports[0]); 73 74 if (svc) { 75 int ignored; 76 77 if (ip_vs_todrop(ipvs)) { 78 /* 79 * It seems that we are very loaded. 80 * We have to drop this packet :( 81 */ 82 *verdict = NF_DROP; 83 return 0; 84 } 85 86 /* 87 * Let the virtual server select a real server for the 88 * incoming connection, and create a connection entry. 89 */ 90 *cpp = ip_vs_schedule(svc, skb, pd, &ignored, iph); 91 if (!*cpp && ignored <= 0) { 92 if (!ignored) 93 *verdict = ip_vs_leave(svc, skb, pd, iph); 94 else 95 *verdict = NF_DROP; 96 return 0; 97 } 98 } 99 /* NF_ACCEPT */ 100 return 1; 101 } 102 103 104 static inline void 105 tcp_fast_csum_update(int af, struct tcphdr *tcph, 106 const union nf_inet_addr *oldip, 107 const union nf_inet_addr *newip, 108 __be16 oldport, __be16 newport) 109 { 110 #ifdef CONFIG_IP_VS_IPV6 111 if (af == AF_INET6) 112 tcph->check = 113 csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6, 114 ip_vs_check_diff2(oldport, newport, 115 ~csum_unfold(tcph->check)))); 116 else 117 #endif 118 tcph->check = 119 csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip, 120 ip_vs_check_diff2(oldport, newport, 121 ~csum_unfold(tcph->check)))); 122 } 123 124 125 static inline void 126 tcp_partial_csum_update(int af, struct tcphdr *tcph, 127 const union nf_inet_addr *oldip, 128 const union nf_inet_addr *newip, 129 __be16 oldlen, __be16 newlen) 130 { 131 #ifdef CONFIG_IP_VS_IPV6 132 if (af == AF_INET6) 133 tcph->check = 134 ~csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6, 135 ip_vs_check_diff2(oldlen, newlen, 136 csum_unfold(tcph->check)))); 137 else 138 #endif 139 tcph->check = 140 ~csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip, 141 ip_vs_check_diff2(oldlen, newlen, 142 csum_unfold(tcph->check)))); 143 } 144 145 146 static int 147 tcp_snat_handler(struct sk_buff *skb, struct ip_vs_protocol *pp, 148 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph) 149 { 150 struct tcphdr *tcph; 151 unsigned int tcphoff = iph->len; 152 int oldlen; 153 int payload_csum = 0; 154 155 #ifdef CONFIG_IP_VS_IPV6 156 if (cp->af == AF_INET6 && iph->fragoffs) 157 return 1; 158 #endif 159 oldlen = skb->len - tcphoff; 160 161 /* csum_check requires unshared skb */ 162 if (!skb_make_writable(skb, tcphoff+sizeof(*tcph))) 163 return 0; 164 165 if (unlikely(cp->app != NULL)) { 166 int ret; 167 168 /* Some checks before mangling */ 169 if (pp->csum_check && !pp->csum_check(cp->af, skb, pp)) 170 return 0; 171 172 /* Call application helper if needed */ 173 if (!(ret = ip_vs_app_pkt_out(cp, skb, iph))) 174 return 0; 175 /* ret=2: csum update is needed after payload mangling */ 176 if (ret == 1) 177 oldlen = skb->len - tcphoff; 178 else 179 payload_csum = 1; 180 } 181 182 tcph = (void *)skb_network_header(skb) + tcphoff; 183 tcph->source = cp->vport; 184 185 /* Adjust TCP checksums */ 186 if (skb->ip_summed == CHECKSUM_PARTIAL) { 187 tcp_partial_csum_update(cp->af, tcph, &cp->daddr, &cp->vaddr, 188 htons(oldlen), 189 htons(skb->len - tcphoff)); 190 } else if (!payload_csum) { 191 /* Only port and addr are changed, do fast csum update */ 192 tcp_fast_csum_update(cp->af, tcph, &cp->daddr, &cp->vaddr, 193 cp->dport, cp->vport); 194 if (skb->ip_summed == CHECKSUM_COMPLETE) 195 skb->ip_summed = (cp->app && pp->csum_check) ? 196 CHECKSUM_UNNECESSARY : CHECKSUM_NONE; 197 } else { 198 /* full checksum calculation */ 199 tcph->check = 0; 200 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0); 201 #ifdef CONFIG_IP_VS_IPV6 202 if (cp->af == AF_INET6) 203 tcph->check = csum_ipv6_magic(&cp->vaddr.in6, 204 &cp->caddr.in6, 205 skb->len - tcphoff, 206 cp->protocol, skb->csum); 207 else 208 #endif 209 tcph->check = csum_tcpudp_magic(cp->vaddr.ip, 210 cp->caddr.ip, 211 skb->len - tcphoff, 212 cp->protocol, 213 skb->csum); 214 skb->ip_summed = CHECKSUM_UNNECESSARY; 215 216 IP_VS_DBG(11, "O-pkt: %s O-csum=%d (+%zd)\n", 217 pp->name, tcph->check, 218 (char*)&(tcph->check) - (char*)tcph); 219 } 220 return 1; 221 } 222 223 224 static int 225 tcp_dnat_handler(struct sk_buff *skb, struct ip_vs_protocol *pp, 226 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph) 227 { 228 struct tcphdr *tcph; 229 unsigned int tcphoff = iph->len; 230 int oldlen; 231 int payload_csum = 0; 232 233 #ifdef CONFIG_IP_VS_IPV6 234 if (cp->af == AF_INET6 && iph->fragoffs) 235 return 1; 236 #endif 237 oldlen = skb->len - tcphoff; 238 239 /* csum_check requires unshared skb */ 240 if (!skb_make_writable(skb, tcphoff+sizeof(*tcph))) 241 return 0; 242 243 if (unlikely(cp->app != NULL)) { 244 int ret; 245 246 /* Some checks before mangling */ 247 if (pp->csum_check && !pp->csum_check(cp->af, skb, pp)) 248 return 0; 249 250 /* 251 * Attempt ip_vs_app call. 252 * It will fix ip_vs_conn and iph ack_seq stuff 253 */ 254 if (!(ret = ip_vs_app_pkt_in(cp, skb, iph))) 255 return 0; 256 /* ret=2: csum update is needed after payload mangling */ 257 if (ret == 1) 258 oldlen = skb->len - tcphoff; 259 else 260 payload_csum = 1; 261 } 262 263 tcph = (void *)skb_network_header(skb) + tcphoff; 264 tcph->dest = cp->dport; 265 266 /* 267 * Adjust TCP checksums 268 */ 269 if (skb->ip_summed == CHECKSUM_PARTIAL) { 270 tcp_partial_csum_update(cp->af, tcph, &cp->vaddr, &cp->daddr, 271 htons(oldlen), 272 htons(skb->len - tcphoff)); 273 } else if (!payload_csum) { 274 /* Only port and addr are changed, do fast csum update */ 275 tcp_fast_csum_update(cp->af, tcph, &cp->vaddr, &cp->daddr, 276 cp->vport, cp->dport); 277 if (skb->ip_summed == CHECKSUM_COMPLETE) 278 skb->ip_summed = (cp->app && pp->csum_check) ? 279 CHECKSUM_UNNECESSARY : CHECKSUM_NONE; 280 } else { 281 /* full checksum calculation */ 282 tcph->check = 0; 283 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0); 284 #ifdef CONFIG_IP_VS_IPV6 285 if (cp->af == AF_INET6) 286 tcph->check = csum_ipv6_magic(&cp->caddr.in6, 287 &cp->daddr.in6, 288 skb->len - tcphoff, 289 cp->protocol, skb->csum); 290 else 291 #endif 292 tcph->check = csum_tcpudp_magic(cp->caddr.ip, 293 cp->daddr.ip, 294 skb->len - tcphoff, 295 cp->protocol, 296 skb->csum); 297 skb->ip_summed = CHECKSUM_UNNECESSARY; 298 } 299 return 1; 300 } 301 302 303 static int 304 tcp_csum_check(int af, struct sk_buff *skb, struct ip_vs_protocol *pp) 305 { 306 unsigned int tcphoff; 307 308 #ifdef CONFIG_IP_VS_IPV6 309 if (af == AF_INET6) 310 tcphoff = sizeof(struct ipv6hdr); 311 else 312 #endif 313 tcphoff = ip_hdrlen(skb); 314 315 switch (skb->ip_summed) { 316 case CHECKSUM_NONE: 317 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0); 318 /* fall through */ 319 case CHECKSUM_COMPLETE: 320 #ifdef CONFIG_IP_VS_IPV6 321 if (af == AF_INET6) { 322 if (csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 323 &ipv6_hdr(skb)->daddr, 324 skb->len - tcphoff, 325 ipv6_hdr(skb)->nexthdr, 326 skb->csum)) { 327 IP_VS_DBG_RL_PKT(0, af, pp, skb, 0, 328 "Failed checksum for"); 329 return 0; 330 } 331 } else 332 #endif 333 if (csum_tcpudp_magic(ip_hdr(skb)->saddr, 334 ip_hdr(skb)->daddr, 335 skb->len - tcphoff, 336 ip_hdr(skb)->protocol, 337 skb->csum)) { 338 IP_VS_DBG_RL_PKT(0, af, pp, skb, 0, 339 "Failed checksum for"); 340 return 0; 341 } 342 break; 343 default: 344 /* No need to checksum. */ 345 break; 346 } 347 348 return 1; 349 } 350 351 352 #define TCP_DIR_INPUT 0 353 #define TCP_DIR_OUTPUT 4 354 #define TCP_DIR_INPUT_ONLY 8 355 356 static const int tcp_state_off[IP_VS_DIR_LAST] = { 357 [IP_VS_DIR_INPUT] = TCP_DIR_INPUT, 358 [IP_VS_DIR_OUTPUT] = TCP_DIR_OUTPUT, 359 [IP_VS_DIR_INPUT_ONLY] = TCP_DIR_INPUT_ONLY, 360 }; 361 362 /* 363 * Timeout table[state] 364 */ 365 static const int tcp_timeouts[IP_VS_TCP_S_LAST+1] = { 366 [IP_VS_TCP_S_NONE] = 2*HZ, 367 [IP_VS_TCP_S_ESTABLISHED] = 15*60*HZ, 368 [IP_VS_TCP_S_SYN_SENT] = 2*60*HZ, 369 [IP_VS_TCP_S_SYN_RECV] = 1*60*HZ, 370 [IP_VS_TCP_S_FIN_WAIT] = 2*60*HZ, 371 [IP_VS_TCP_S_TIME_WAIT] = 2*60*HZ, 372 [IP_VS_TCP_S_CLOSE] = 10*HZ, 373 [IP_VS_TCP_S_CLOSE_WAIT] = 60*HZ, 374 [IP_VS_TCP_S_LAST_ACK] = 30*HZ, 375 [IP_VS_TCP_S_LISTEN] = 2*60*HZ, 376 [IP_VS_TCP_S_SYNACK] = 120*HZ, 377 [IP_VS_TCP_S_LAST] = 2*HZ, 378 }; 379 380 static const char *const tcp_state_name_table[IP_VS_TCP_S_LAST+1] = { 381 [IP_VS_TCP_S_NONE] = "NONE", 382 [IP_VS_TCP_S_ESTABLISHED] = "ESTABLISHED", 383 [IP_VS_TCP_S_SYN_SENT] = "SYN_SENT", 384 [IP_VS_TCP_S_SYN_RECV] = "SYN_RECV", 385 [IP_VS_TCP_S_FIN_WAIT] = "FIN_WAIT", 386 [IP_VS_TCP_S_TIME_WAIT] = "TIME_WAIT", 387 [IP_VS_TCP_S_CLOSE] = "CLOSE", 388 [IP_VS_TCP_S_CLOSE_WAIT] = "CLOSE_WAIT", 389 [IP_VS_TCP_S_LAST_ACK] = "LAST_ACK", 390 [IP_VS_TCP_S_LISTEN] = "LISTEN", 391 [IP_VS_TCP_S_SYNACK] = "SYNACK", 392 [IP_VS_TCP_S_LAST] = "BUG!", 393 }; 394 395 static const bool tcp_state_active_table[IP_VS_TCP_S_LAST] = { 396 [IP_VS_TCP_S_NONE] = false, 397 [IP_VS_TCP_S_ESTABLISHED] = true, 398 [IP_VS_TCP_S_SYN_SENT] = true, 399 [IP_VS_TCP_S_SYN_RECV] = true, 400 [IP_VS_TCP_S_FIN_WAIT] = false, 401 [IP_VS_TCP_S_TIME_WAIT] = false, 402 [IP_VS_TCP_S_CLOSE] = false, 403 [IP_VS_TCP_S_CLOSE_WAIT] = false, 404 [IP_VS_TCP_S_LAST_ACK] = false, 405 [IP_VS_TCP_S_LISTEN] = false, 406 [IP_VS_TCP_S_SYNACK] = true, 407 }; 408 409 #define sNO IP_VS_TCP_S_NONE 410 #define sES IP_VS_TCP_S_ESTABLISHED 411 #define sSS IP_VS_TCP_S_SYN_SENT 412 #define sSR IP_VS_TCP_S_SYN_RECV 413 #define sFW IP_VS_TCP_S_FIN_WAIT 414 #define sTW IP_VS_TCP_S_TIME_WAIT 415 #define sCL IP_VS_TCP_S_CLOSE 416 #define sCW IP_VS_TCP_S_CLOSE_WAIT 417 #define sLA IP_VS_TCP_S_LAST_ACK 418 #define sLI IP_VS_TCP_S_LISTEN 419 #define sSA IP_VS_TCP_S_SYNACK 420 421 struct tcp_states_t { 422 int next_state[IP_VS_TCP_S_LAST]; 423 }; 424 425 static const char * tcp_state_name(int state) 426 { 427 if (state >= IP_VS_TCP_S_LAST) 428 return "ERR!"; 429 return tcp_state_name_table[state] ? tcp_state_name_table[state] : "?"; 430 } 431 432 static bool tcp_state_active(int state) 433 { 434 if (state >= IP_VS_TCP_S_LAST) 435 return false; 436 return tcp_state_active_table[state]; 437 } 438 439 static struct tcp_states_t tcp_states[] = { 440 /* INPUT */ 441 /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */ 442 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSR }}, 443 /*fin*/ {{sCL, sCW, sSS, sTW, sTW, sTW, sCL, sCW, sLA, sLI, sTW }}, 444 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }}, 445 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sSR }}, 446 447 /* OUTPUT */ 448 /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */ 449 /*syn*/ {{sSS, sES, sSS, sSR, sSS, sSS, sSS, sSS, sSS, sLI, sSR }}, 450 /*fin*/ {{sTW, sFW, sSS, sTW, sFW, sTW, sCL, sTW, sLA, sLI, sTW }}, 451 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sLA, sES, sES }}, 452 /*rst*/ {{sCL, sCL, sSS, sCL, sCL, sTW, sCL, sCL, sCL, sCL, sCL }}, 453 454 /* INPUT-ONLY */ 455 /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */ 456 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSR }}, 457 /*fin*/ {{sCL, sFW, sSS, sTW, sFW, sTW, sCL, sCW, sLA, sLI, sTW }}, 458 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }}, 459 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }}, 460 }; 461 462 static struct tcp_states_t tcp_states_dos[] = { 463 /* INPUT */ 464 /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */ 465 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSA }}, 466 /*fin*/ {{sCL, sCW, sSS, sTW, sTW, sTW, sCL, sCW, sLA, sLI, sSA }}, 467 /*ack*/ {{sES, sES, sSS, sSR, sFW, sTW, sCL, sCW, sCL, sLI, sSA }}, 468 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }}, 469 470 /* OUTPUT */ 471 /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */ 472 /*syn*/ {{sSS, sES, sSS, sSA, sSS, sSS, sSS, sSS, sSS, sLI, sSA }}, 473 /*fin*/ {{sTW, sFW, sSS, sTW, sFW, sTW, sCL, sTW, sLA, sLI, sTW }}, 474 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sLA, sES, sES }}, 475 /*rst*/ {{sCL, sCL, sSS, sCL, sCL, sTW, sCL, sCL, sCL, sCL, sCL }}, 476 477 /* INPUT-ONLY */ 478 /* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */ 479 /*syn*/ {{sSA, sES, sES, sSR, sSA, sSA, sSA, sSA, sSA, sSA, sSA }}, 480 /*fin*/ {{sCL, sFW, sSS, sTW, sFW, sTW, sCL, sCW, sLA, sLI, sTW }}, 481 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }}, 482 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }}, 483 }; 484 485 static void tcp_timeout_change(struct ip_vs_proto_data *pd, int flags) 486 { 487 int on = (flags & 1); /* secure_tcp */ 488 489 /* 490 ** FIXME: change secure_tcp to independent sysctl var 491 ** or make it per-service or per-app because it is valid 492 ** for most if not for all of the applications. Something 493 ** like "capabilities" (flags) for each object. 494 */ 495 pd->tcp_state_table = (on ? tcp_states_dos : tcp_states); 496 } 497 498 static inline int tcp_state_idx(struct tcphdr *th) 499 { 500 if (th->rst) 501 return 3; 502 if (th->syn) 503 return 0; 504 if (th->fin) 505 return 1; 506 if (th->ack) 507 return 2; 508 return -1; 509 } 510 511 static inline void 512 set_tcp_state(struct ip_vs_proto_data *pd, struct ip_vs_conn *cp, 513 int direction, struct tcphdr *th) 514 { 515 int state_idx; 516 int new_state = IP_VS_TCP_S_CLOSE; 517 int state_off = tcp_state_off[direction]; 518 519 /* 520 * Update state offset to INPUT_ONLY if necessary 521 * or delete NO_OUTPUT flag if output packet detected 522 */ 523 if (cp->flags & IP_VS_CONN_F_NOOUTPUT) { 524 if (state_off == TCP_DIR_OUTPUT) 525 cp->flags &= ~IP_VS_CONN_F_NOOUTPUT; 526 else 527 state_off = TCP_DIR_INPUT_ONLY; 528 } 529 530 if ((state_idx = tcp_state_idx(th)) < 0) { 531 IP_VS_DBG(8, "tcp_state_idx=%d!!!\n", state_idx); 532 goto tcp_state_out; 533 } 534 535 new_state = 536 pd->tcp_state_table[state_off+state_idx].next_state[cp->state]; 537 538 tcp_state_out: 539 if (new_state != cp->state) { 540 struct ip_vs_dest *dest = cp->dest; 541 542 IP_VS_DBG_BUF(8, "%s %s [%c%c%c%c] %s:%d->" 543 "%s:%d state: %s->%s conn->refcnt:%d\n", 544 pd->pp->name, 545 ((state_off == TCP_DIR_OUTPUT) ? 546 "output " : "input "), 547 th->syn ? 'S' : '.', 548 th->fin ? 'F' : '.', 549 th->ack ? 'A' : '.', 550 th->rst ? 'R' : '.', 551 IP_VS_DBG_ADDR(cp->daf, &cp->daddr), 552 ntohs(cp->dport), 553 IP_VS_DBG_ADDR(cp->af, &cp->caddr), 554 ntohs(cp->cport), 555 tcp_state_name(cp->state), 556 tcp_state_name(new_state), 557 refcount_read(&cp->refcnt)); 558 559 if (dest) { 560 if (!(cp->flags & IP_VS_CONN_F_INACTIVE) && 561 !tcp_state_active(new_state)) { 562 atomic_dec(&dest->activeconns); 563 atomic_inc(&dest->inactconns); 564 cp->flags |= IP_VS_CONN_F_INACTIVE; 565 } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) && 566 tcp_state_active(new_state)) { 567 atomic_inc(&dest->activeconns); 568 atomic_dec(&dest->inactconns); 569 cp->flags &= ~IP_VS_CONN_F_INACTIVE; 570 } 571 } 572 } 573 574 if (likely(pd)) 575 cp->timeout = pd->timeout_table[cp->state = new_state]; 576 else /* What to do ? */ 577 cp->timeout = tcp_timeouts[cp->state = new_state]; 578 } 579 580 /* 581 * Handle state transitions 582 */ 583 static void 584 tcp_state_transition(struct ip_vs_conn *cp, int direction, 585 const struct sk_buff *skb, 586 struct ip_vs_proto_data *pd) 587 { 588 struct tcphdr _tcph, *th; 589 590 #ifdef CONFIG_IP_VS_IPV6 591 int ihl = cp->af == AF_INET ? ip_hdrlen(skb) : sizeof(struct ipv6hdr); 592 #else 593 int ihl = ip_hdrlen(skb); 594 #endif 595 596 th = skb_header_pointer(skb, ihl, sizeof(_tcph), &_tcph); 597 if (th == NULL) 598 return; 599 600 spin_lock_bh(&cp->lock); 601 set_tcp_state(pd, cp, direction, th); 602 spin_unlock_bh(&cp->lock); 603 } 604 605 static inline __u16 tcp_app_hashkey(__be16 port) 606 { 607 return (((__force u16)port >> TCP_APP_TAB_BITS) ^ (__force u16)port) 608 & TCP_APP_TAB_MASK; 609 } 610 611 612 static int tcp_register_app(struct netns_ipvs *ipvs, struct ip_vs_app *inc) 613 { 614 struct ip_vs_app *i; 615 __u16 hash; 616 __be16 port = inc->port; 617 int ret = 0; 618 struct ip_vs_proto_data *pd = ip_vs_proto_data_get(ipvs, IPPROTO_TCP); 619 620 hash = tcp_app_hashkey(port); 621 622 list_for_each_entry(i, &ipvs->tcp_apps[hash], p_list) { 623 if (i->port == port) { 624 ret = -EEXIST; 625 goto out; 626 } 627 } 628 list_add_rcu(&inc->p_list, &ipvs->tcp_apps[hash]); 629 atomic_inc(&pd->appcnt); 630 631 out: 632 return ret; 633 } 634 635 636 static void 637 tcp_unregister_app(struct netns_ipvs *ipvs, struct ip_vs_app *inc) 638 { 639 struct ip_vs_proto_data *pd = ip_vs_proto_data_get(ipvs, IPPROTO_TCP); 640 641 atomic_dec(&pd->appcnt); 642 list_del_rcu(&inc->p_list); 643 } 644 645 646 static int 647 tcp_app_conn_bind(struct ip_vs_conn *cp) 648 { 649 struct netns_ipvs *ipvs = cp->ipvs; 650 int hash; 651 struct ip_vs_app *inc; 652 int result = 0; 653 654 /* Default binding: bind app only for NAT */ 655 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) 656 return 0; 657 658 /* Lookup application incarnations and bind the right one */ 659 hash = tcp_app_hashkey(cp->vport); 660 661 list_for_each_entry_rcu(inc, &ipvs->tcp_apps[hash], p_list) { 662 if (inc->port == cp->vport) { 663 if (unlikely(!ip_vs_app_inc_get(inc))) 664 break; 665 666 IP_VS_DBG_BUF(9, "%s(): Binding conn %s:%u->" 667 "%s:%u to app %s on port %u\n", 668 __func__, 669 IP_VS_DBG_ADDR(cp->af, &cp->caddr), 670 ntohs(cp->cport), 671 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), 672 ntohs(cp->vport), 673 inc->name, ntohs(inc->port)); 674 675 cp->app = inc; 676 if (inc->init_conn) 677 result = inc->init_conn(inc, cp); 678 break; 679 } 680 } 681 682 return result; 683 } 684 685 686 /* 687 * Set LISTEN timeout. (ip_vs_conn_put will setup timer) 688 */ 689 void ip_vs_tcp_conn_listen(struct ip_vs_conn *cp) 690 { 691 struct ip_vs_proto_data *pd = ip_vs_proto_data_get(cp->ipvs, IPPROTO_TCP); 692 693 spin_lock_bh(&cp->lock); 694 cp->state = IP_VS_TCP_S_LISTEN; 695 cp->timeout = (pd ? pd->timeout_table[IP_VS_TCP_S_LISTEN] 696 : tcp_timeouts[IP_VS_TCP_S_LISTEN]); 697 spin_unlock_bh(&cp->lock); 698 } 699 700 /* --------------------------------------------- 701 * timeouts is netns related now. 702 * --------------------------------------------- 703 */ 704 static int __ip_vs_tcp_init(struct netns_ipvs *ipvs, struct ip_vs_proto_data *pd) 705 { 706 ip_vs_init_hash_table(ipvs->tcp_apps, TCP_APP_TAB_SIZE); 707 pd->timeout_table = ip_vs_create_timeout_table((int *)tcp_timeouts, 708 sizeof(tcp_timeouts)); 709 if (!pd->timeout_table) 710 return -ENOMEM; 711 pd->tcp_state_table = tcp_states; 712 return 0; 713 } 714 715 static void __ip_vs_tcp_exit(struct netns_ipvs *ipvs, struct ip_vs_proto_data *pd) 716 { 717 kfree(pd->timeout_table); 718 } 719 720 721 struct ip_vs_protocol ip_vs_protocol_tcp = { 722 .name = "TCP", 723 .protocol = IPPROTO_TCP, 724 .num_states = IP_VS_TCP_S_LAST, 725 .dont_defrag = 0, 726 .init = NULL, 727 .exit = NULL, 728 .init_netns = __ip_vs_tcp_init, 729 .exit_netns = __ip_vs_tcp_exit, 730 .register_app = tcp_register_app, 731 .unregister_app = tcp_unregister_app, 732 .conn_schedule = tcp_conn_schedule, 733 .conn_in_get = ip_vs_conn_in_get_proto, 734 .conn_out_get = ip_vs_conn_out_get_proto, 735 .snat_handler = tcp_snat_handler, 736 .dnat_handler = tcp_dnat_handler, 737 .csum_check = tcp_csum_check, 738 .state_name = tcp_state_name, 739 .state_transition = tcp_state_transition, 740 .app_conn_bind = tcp_app_conn_bind, 741 .debug_packet = ip_vs_tcpudp_debug_packet, 742 .timeout_change = tcp_timeout_change, 743 }; 744