1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * IPVS An implementation of the IP virtual server support for the 4 * LINUX operating system. IPVS is now implemented as a module 5 * over the Netfilter framework. IPVS can be used to build a 6 * high-performance and highly available server based on a 7 * cluster of servers. 8 * 9 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org> 10 * Peter Kese <peter.kese@ijs.si> 11 * Julian Anastasov <ja@ssi.bg> 12 * 13 * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese, 14 * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms 15 * and others. 16 * 17 * Changes: 18 * Paul `Rusty' Russell properly handle non-linear skbs 19 * Harald Welte don't use nfcache 20 */ 21 22 #define KMSG_COMPONENT "IPVS" 23 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt 24 25 #include <linux/module.h> 26 #include <linux/kernel.h> 27 #include <linux/ip.h> 28 #include <linux/tcp.h> 29 #include <linux/sctp.h> 30 #include <linux/icmp.h> 31 #include <linux/slab.h> 32 33 #include <net/ip.h> 34 #include <net/tcp.h> 35 #include <net/udp.h> 36 #include <net/icmp.h> /* for icmp_send */ 37 #include <net/gue.h> 38 #include <net/gre.h> 39 #include <net/route.h> 40 #include <net/ip6_checksum.h> 41 #include <net/netns/generic.h> /* net_generic() */ 42 43 #include <linux/netfilter.h> 44 #include <linux/netfilter_ipv4.h> 45 46 #ifdef CONFIG_IP_VS_IPV6 47 #include <net/ipv6.h> 48 #include <linux/netfilter_ipv6.h> 49 #include <net/ip6_route.h> 50 #endif 51 52 #include <net/ip_vs.h> 53 #include <linux/indirect_call_wrapper.h> 54 55 56 EXPORT_SYMBOL(register_ip_vs_scheduler); 57 EXPORT_SYMBOL(unregister_ip_vs_scheduler); 58 EXPORT_SYMBOL(ip_vs_proto_name); 59 EXPORT_SYMBOL(ip_vs_conn_new); 60 EXPORT_SYMBOL(ip_vs_conn_in_get); 61 EXPORT_SYMBOL(ip_vs_conn_out_get); 62 #ifdef CONFIG_IP_VS_PROTO_TCP 63 EXPORT_SYMBOL(ip_vs_tcp_conn_listen); 64 #endif 65 EXPORT_SYMBOL(ip_vs_conn_put); 66 #ifdef CONFIG_IP_VS_DEBUG 67 EXPORT_SYMBOL(ip_vs_get_debug_level); 68 #endif 69 EXPORT_SYMBOL(ip_vs_new_conn_out); 70 71 #if defined(CONFIG_IP_VS_PROTO_TCP) && defined(CONFIG_IP_VS_PROTO_UDP) 72 #define SNAT_CALL(f, ...) \ 73 INDIRECT_CALL_2(f, tcp_snat_handler, udp_snat_handler, __VA_ARGS__) 74 #elif defined(CONFIG_IP_VS_PROTO_TCP) 75 #define SNAT_CALL(f, ...) INDIRECT_CALL_1(f, tcp_snat_handler, __VA_ARGS__) 76 #elif defined(CONFIG_IP_VS_PROTO_UDP) 77 #define SNAT_CALL(f, ...) INDIRECT_CALL_1(f, udp_snat_handler, __VA_ARGS__) 78 #else 79 #define SNAT_CALL(f, ...) f(__VA_ARGS__) 80 #endif 81 82 static unsigned int ip_vs_net_id __read_mostly; 83 /* netns cnt used for uniqueness */ 84 static atomic_t ipvs_netns_cnt = ATOMIC_INIT(0); 85 86 /* ID used in ICMP lookups */ 87 #define icmp_id(icmph) (((icmph)->un).echo.id) 88 #define icmpv6_id(icmph) (icmph->icmp6_dataun.u_echo.identifier) 89 90 const char *ip_vs_proto_name(unsigned int proto) 91 { 92 static char buf[20]; 93 94 switch (proto) { 95 case IPPROTO_IP: 96 return "IP"; 97 case IPPROTO_UDP: 98 return "UDP"; 99 case IPPROTO_TCP: 100 return "TCP"; 101 case IPPROTO_SCTP: 102 return "SCTP"; 103 case IPPROTO_ICMP: 104 return "ICMP"; 105 #ifdef CONFIG_IP_VS_IPV6 106 case IPPROTO_ICMPV6: 107 return "ICMPv6"; 108 #endif 109 default: 110 sprintf(buf, "IP_%u", proto); 111 return buf; 112 } 113 } 114 115 void ip_vs_init_hash_table(struct list_head *table, int rows) 116 { 117 while (--rows >= 0) 118 INIT_LIST_HEAD(&table[rows]); 119 } 120 121 static inline void 122 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb) 123 { 124 struct ip_vs_dest *dest = cp->dest; 125 struct netns_ipvs *ipvs = cp->ipvs; 126 127 if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) { 128 struct ip_vs_cpu_stats *s; 129 struct ip_vs_service *svc; 130 131 local_bh_disable(); 132 133 s = this_cpu_ptr(dest->stats.cpustats); 134 u64_stats_update_begin(&s->syncp); 135 u64_stats_inc(&s->cnt.inpkts); 136 u64_stats_add(&s->cnt.inbytes, skb->len); 137 u64_stats_update_end(&s->syncp); 138 139 svc = rcu_dereference(dest->svc); 140 s = this_cpu_ptr(svc->stats.cpustats); 141 u64_stats_update_begin(&s->syncp); 142 u64_stats_inc(&s->cnt.inpkts); 143 u64_stats_add(&s->cnt.inbytes, skb->len); 144 u64_stats_update_end(&s->syncp); 145 146 s = this_cpu_ptr(ipvs->tot_stats->s.cpustats); 147 u64_stats_update_begin(&s->syncp); 148 u64_stats_inc(&s->cnt.inpkts); 149 u64_stats_add(&s->cnt.inbytes, skb->len); 150 u64_stats_update_end(&s->syncp); 151 152 local_bh_enable(); 153 } 154 } 155 156 157 static inline void 158 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb) 159 { 160 struct ip_vs_dest *dest = cp->dest; 161 struct netns_ipvs *ipvs = cp->ipvs; 162 163 if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) { 164 struct ip_vs_cpu_stats *s; 165 struct ip_vs_service *svc; 166 167 local_bh_disable(); 168 169 s = this_cpu_ptr(dest->stats.cpustats); 170 u64_stats_update_begin(&s->syncp); 171 u64_stats_inc(&s->cnt.outpkts); 172 u64_stats_add(&s->cnt.outbytes, skb->len); 173 u64_stats_update_end(&s->syncp); 174 175 svc = rcu_dereference(dest->svc); 176 s = this_cpu_ptr(svc->stats.cpustats); 177 u64_stats_update_begin(&s->syncp); 178 u64_stats_inc(&s->cnt.outpkts); 179 u64_stats_add(&s->cnt.outbytes, skb->len); 180 u64_stats_update_end(&s->syncp); 181 182 s = this_cpu_ptr(ipvs->tot_stats->s.cpustats); 183 u64_stats_update_begin(&s->syncp); 184 u64_stats_inc(&s->cnt.outpkts); 185 u64_stats_add(&s->cnt.outbytes, skb->len); 186 u64_stats_update_end(&s->syncp); 187 188 local_bh_enable(); 189 } 190 } 191 192 193 static inline void 194 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc) 195 { 196 struct netns_ipvs *ipvs = svc->ipvs; 197 struct ip_vs_cpu_stats *s; 198 199 local_bh_disable(); 200 201 s = this_cpu_ptr(cp->dest->stats.cpustats); 202 u64_stats_update_begin(&s->syncp); 203 u64_stats_inc(&s->cnt.conns); 204 u64_stats_update_end(&s->syncp); 205 206 s = this_cpu_ptr(svc->stats.cpustats); 207 u64_stats_update_begin(&s->syncp); 208 u64_stats_inc(&s->cnt.conns); 209 u64_stats_update_end(&s->syncp); 210 211 s = this_cpu_ptr(ipvs->tot_stats->s.cpustats); 212 u64_stats_update_begin(&s->syncp); 213 u64_stats_inc(&s->cnt.conns); 214 u64_stats_update_end(&s->syncp); 215 216 local_bh_enable(); 217 } 218 219 220 static inline void 221 ip_vs_set_state(struct ip_vs_conn *cp, int direction, 222 const struct sk_buff *skb, 223 struct ip_vs_proto_data *pd) 224 { 225 if (likely(pd->pp->state_transition)) 226 pd->pp->state_transition(cp, direction, skb, pd); 227 } 228 229 static inline int 230 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc, 231 struct sk_buff *skb, int protocol, 232 const union nf_inet_addr *caddr, __be16 cport, 233 const union nf_inet_addr *vaddr, __be16 vport, 234 struct ip_vs_conn_param *p) 235 { 236 ip_vs_conn_fill_param(svc->ipvs, svc->af, protocol, caddr, cport, vaddr, 237 vport, p); 238 p->pe = rcu_dereference(svc->pe); 239 if (p->pe && p->pe->fill_param) 240 return p->pe->fill_param(p, skb); 241 242 return 0; 243 } 244 245 /* 246 * IPVS persistent scheduling function 247 * It creates a connection entry according to its template if exists, 248 * or selects a server and creates a connection entry plus a template. 249 * Locking: we are svc user (svc->refcnt), so we hold all dests too 250 * Protocols supported: TCP, UDP 251 */ 252 static struct ip_vs_conn * 253 ip_vs_sched_persist(struct ip_vs_service *svc, 254 struct sk_buff *skb, __be16 src_port, __be16 dst_port, 255 int *ignored, struct ip_vs_iphdr *iph) 256 { 257 struct ip_vs_conn *cp = NULL; 258 struct ip_vs_dest *dest; 259 struct ip_vs_conn *ct; 260 __be16 dport = 0; /* destination port to forward */ 261 unsigned int flags; 262 struct ip_vs_conn_param param; 263 const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) }; 264 union nf_inet_addr snet; /* source network of the client, 265 after masking */ 266 const union nf_inet_addr *src_addr, *dst_addr; 267 268 if (likely(!ip_vs_iph_inverse(iph))) { 269 src_addr = &iph->saddr; 270 dst_addr = &iph->daddr; 271 } else { 272 src_addr = &iph->daddr; 273 dst_addr = &iph->saddr; 274 } 275 276 277 /* Mask saddr with the netmask to adjust template granularity */ 278 #ifdef CONFIG_IP_VS_IPV6 279 if (svc->af == AF_INET6) 280 ipv6_addr_prefix(&snet.in6, &src_addr->in6, 281 (__force __u32) svc->netmask); 282 else 283 #endif 284 snet.ip = src_addr->ip & svc->netmask; 285 286 IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u " 287 "mnet %s\n", 288 IP_VS_DBG_ADDR(svc->af, src_addr), ntohs(src_port), 289 IP_VS_DBG_ADDR(svc->af, dst_addr), ntohs(dst_port), 290 IP_VS_DBG_ADDR(svc->af, &snet)); 291 292 /* 293 * As far as we know, FTP is a very complicated network protocol, and 294 * it uses control connection and data connections. For active FTP, 295 * FTP server initialize data connection to the client, its source port 296 * is often 20. For passive FTP, FTP server tells the clients the port 297 * that it passively listens to, and the client issues the data 298 * connection. In the tunneling or direct routing mode, the load 299 * balancer is on the client-to-server half of connection, the port 300 * number is unknown to the load balancer. So, a conn template like 301 * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP 302 * service, and a template like <caddr, 0, vaddr, vport, daddr, dport> 303 * is created for other persistent services. 304 */ 305 { 306 int protocol = iph->protocol; 307 const union nf_inet_addr *vaddr = dst_addr; 308 __be16 vport = 0; 309 310 if (dst_port == svc->port) { 311 /* non-FTP template: 312 * <protocol, caddr, 0, vaddr, vport, daddr, dport> 313 * FTP template: 314 * <protocol, caddr, 0, vaddr, 0, daddr, 0> 315 */ 316 if (svc->port != FTPPORT) 317 vport = dst_port; 318 } else { 319 /* Note: persistent fwmark-based services and 320 * persistent port zero service are handled here. 321 * fwmark template: 322 * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0> 323 * port zero template: 324 * <protocol,caddr,0,vaddr,0,daddr,0> 325 */ 326 if (svc->fwmark) { 327 protocol = IPPROTO_IP; 328 vaddr = &fwmark; 329 } 330 } 331 /* return *ignored = -1 so NF_DROP can be used */ 332 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0, 333 vaddr, vport, ¶m) < 0) { 334 *ignored = -1; 335 return NULL; 336 } 337 } 338 339 /* Check if a template already exists */ 340 ct = ip_vs_ct_in_get(¶m); 341 if (!ct || !ip_vs_check_template(ct, NULL)) { 342 struct ip_vs_scheduler *sched; 343 344 /* 345 * No template found or the dest of the connection 346 * template is not available. 347 * return *ignored=0 i.e. ICMP and NF_DROP 348 */ 349 sched = rcu_dereference(svc->scheduler); 350 if (sched) { 351 /* read svc->sched_data after svc->scheduler */ 352 smp_rmb(); 353 dest = sched->schedule(svc, skb, iph); 354 } else { 355 dest = NULL; 356 } 357 if (!dest) { 358 IP_VS_DBG(1, "p-schedule: no dest found.\n"); 359 kfree(param.pe_data); 360 *ignored = 0; 361 return NULL; 362 } 363 364 if (dst_port == svc->port && svc->port != FTPPORT) 365 dport = dest->port; 366 367 /* Create a template 368 * This adds param.pe_data to the template, 369 * and thus param.pe_data will be destroyed 370 * when the template expires */ 371 ct = ip_vs_conn_new(¶m, dest->af, &dest->addr, dport, 372 IP_VS_CONN_F_TEMPLATE, dest, skb->mark); 373 if (ct == NULL) { 374 kfree(param.pe_data); 375 *ignored = -1; 376 return NULL; 377 } 378 379 ct->timeout = svc->timeout; 380 } else { 381 /* set destination with the found template */ 382 dest = ct->dest; 383 kfree(param.pe_data); 384 } 385 386 dport = dst_port; 387 if (dport == svc->port && dest->port) 388 dport = dest->port; 389 390 flags = (svc->flags & IP_VS_SVC_F_ONEPACKET 391 && iph->protocol == IPPROTO_UDP) ? 392 IP_VS_CONN_F_ONE_PACKET : 0; 393 394 /* 395 * Create a new connection according to the template 396 */ 397 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol, src_addr, 398 src_port, dst_addr, dst_port, ¶m); 399 400 cp = ip_vs_conn_new(¶m, dest->af, &dest->addr, dport, flags, dest, 401 skb->mark); 402 if (cp == NULL) { 403 ip_vs_conn_put(ct); 404 *ignored = -1; 405 return NULL; 406 } 407 408 /* 409 * Add its control 410 */ 411 ip_vs_control_add(cp, ct); 412 ip_vs_conn_put(ct); 413 414 ip_vs_conn_stats(cp, svc); 415 return cp; 416 } 417 418 419 /* 420 * IPVS main scheduling function 421 * It selects a server according to the virtual service, and 422 * creates a connection entry. 423 * Protocols supported: TCP, UDP 424 * 425 * Usage of *ignored 426 * 427 * 1 : protocol tried to schedule (eg. on SYN), found svc but the 428 * svc/scheduler decides that this packet should be accepted with 429 * NF_ACCEPT because it must not be scheduled. 430 * 431 * 0 : scheduler can not find destination, so try bypass or 432 * return ICMP and then NF_DROP (ip_vs_leave). 433 * 434 * -1 : scheduler tried to schedule but fatal error occurred, eg. 435 * ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param 436 * failure such as missing Call-ID, ENOMEM on skb_linearize 437 * or pe_data. In this case we should return NF_DROP without 438 * any attempts to send ICMP with ip_vs_leave. 439 */ 440 struct ip_vs_conn * 441 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb, 442 struct ip_vs_proto_data *pd, int *ignored, 443 struct ip_vs_iphdr *iph) 444 { 445 struct ip_vs_protocol *pp = pd->pp; 446 struct ip_vs_conn *cp = NULL; 447 struct ip_vs_scheduler *sched; 448 struct ip_vs_dest *dest; 449 __be16 _ports[2], *pptr, cport, vport; 450 const void *caddr, *vaddr; 451 unsigned int flags; 452 453 *ignored = 1; 454 /* 455 * IPv6 frags, only the first hit here. 456 */ 457 pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports); 458 if (pptr == NULL) 459 return NULL; 460 461 if (likely(!ip_vs_iph_inverse(iph))) { 462 cport = pptr[0]; 463 caddr = &iph->saddr; 464 vport = pptr[1]; 465 vaddr = &iph->daddr; 466 } else { 467 cport = pptr[1]; 468 caddr = &iph->daddr; 469 vport = pptr[0]; 470 vaddr = &iph->saddr; 471 } 472 473 /* 474 * FTPDATA needs this check when using local real server. 475 * Never schedule Active FTPDATA connections from real server. 476 * For LVS-NAT they must be already created. For other methods 477 * with persistence the connection is created on SYN+ACK. 478 */ 479 if (cport == FTPDATA) { 480 IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off, 481 "Not scheduling FTPDATA"); 482 return NULL; 483 } 484 485 /* 486 * Do not schedule replies from local real server. 487 */ 488 if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK)) { 489 iph->hdr_flags ^= IP_VS_HDR_INVERSE; 490 cp = INDIRECT_CALL_1(pp->conn_in_get, 491 ip_vs_conn_in_get_proto, svc->ipvs, 492 svc->af, skb, iph); 493 iph->hdr_flags ^= IP_VS_HDR_INVERSE; 494 495 if (cp) { 496 IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off, 497 "Not scheduling reply for existing" 498 " connection"); 499 __ip_vs_conn_put(cp); 500 return NULL; 501 } 502 } 503 504 /* 505 * Persistent service 506 */ 507 if (svc->flags & IP_VS_SVC_F_PERSISTENT) 508 return ip_vs_sched_persist(svc, skb, cport, vport, ignored, 509 iph); 510 511 *ignored = 0; 512 513 /* 514 * Non-persistent service 515 */ 516 if (!svc->fwmark && vport != svc->port) { 517 if (!svc->port) 518 pr_err("Schedule: port zero only supported " 519 "in persistent services, " 520 "check your ipvs configuration\n"); 521 return NULL; 522 } 523 524 sched = rcu_dereference(svc->scheduler); 525 if (sched) { 526 /* read svc->sched_data after svc->scheduler */ 527 smp_rmb(); 528 dest = sched->schedule(svc, skb, iph); 529 } else { 530 dest = NULL; 531 } 532 if (dest == NULL) { 533 IP_VS_DBG(1, "Schedule: no dest found.\n"); 534 return NULL; 535 } 536 537 flags = (svc->flags & IP_VS_SVC_F_ONEPACKET 538 && iph->protocol == IPPROTO_UDP) ? 539 IP_VS_CONN_F_ONE_PACKET : 0; 540 541 /* 542 * Create a connection entry. 543 */ 544 { 545 struct ip_vs_conn_param p; 546 547 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol, 548 caddr, cport, vaddr, vport, &p); 549 cp = ip_vs_conn_new(&p, dest->af, &dest->addr, 550 dest->port ? dest->port : vport, 551 flags, dest, skb->mark); 552 if (!cp) { 553 *ignored = -1; 554 return NULL; 555 } 556 } 557 558 IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u " 559 "d:%s:%u conn->flags:%X conn->refcnt:%d\n", 560 ip_vs_fwd_tag(cp), 561 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport), 562 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport), 563 IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport), 564 cp->flags, refcount_read(&cp->refcnt)); 565 566 ip_vs_conn_stats(cp, svc); 567 return cp; 568 } 569 570 static inline int ip_vs_addr_is_unicast(struct net *net, int af, 571 union nf_inet_addr *addr) 572 { 573 #ifdef CONFIG_IP_VS_IPV6 574 if (af == AF_INET6) 575 return ipv6_addr_type(&addr->in6) & IPV6_ADDR_UNICAST; 576 #endif 577 return (inet_addr_type(net, addr->ip) == RTN_UNICAST); 578 } 579 580 /* 581 * Pass or drop the packet. 582 * Called by ip_vs_in, when the virtual service is available but 583 * no destination is available for a new connection. 584 */ 585 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb, 586 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph) 587 { 588 __be16 _ports[2], *pptr, dport; 589 struct netns_ipvs *ipvs = svc->ipvs; 590 struct net *net = ipvs->net; 591 592 pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports); 593 if (!pptr) 594 return NF_DROP; 595 dport = likely(!ip_vs_iph_inverse(iph)) ? pptr[1] : pptr[0]; 596 597 /* if it is fwmark-based service, the cache_bypass sysctl is up 598 and the destination is a non-local unicast, then create 599 a cache_bypass connection entry */ 600 if (sysctl_cache_bypass(ipvs) && svc->fwmark && 601 !(iph->hdr_flags & (IP_VS_HDR_INVERSE | IP_VS_HDR_ICMP)) && 602 ip_vs_addr_is_unicast(net, svc->af, &iph->daddr)) { 603 int ret; 604 struct ip_vs_conn *cp; 605 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET && 606 iph->protocol == IPPROTO_UDP) ? 607 IP_VS_CONN_F_ONE_PACKET : 0; 608 union nf_inet_addr daddr = { .all = { 0, 0, 0, 0 } }; 609 610 /* create a new connection entry */ 611 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__); 612 { 613 struct ip_vs_conn_param p; 614 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol, 615 &iph->saddr, pptr[0], 616 &iph->daddr, pptr[1], &p); 617 cp = ip_vs_conn_new(&p, svc->af, &daddr, 0, 618 IP_VS_CONN_F_BYPASS | flags, 619 NULL, skb->mark); 620 if (!cp) 621 return NF_DROP; 622 } 623 624 /* statistics */ 625 ip_vs_in_stats(cp, skb); 626 627 /* set state */ 628 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd); 629 630 /* transmit the first SYN packet */ 631 ret = cp->packet_xmit(skb, cp, pd->pp, iph); 632 /* do not touch skb anymore */ 633 634 if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) && cp->control) 635 atomic_inc(&cp->control->in_pkts); 636 else 637 atomic_inc(&cp->in_pkts); 638 ip_vs_conn_put(cp); 639 return ret; 640 } 641 642 /* 643 * When the virtual ftp service is presented, packets destined 644 * for other services on the VIP may get here (except services 645 * listed in the ipvs table), pass the packets, because it is 646 * not ipvs job to decide to drop the packets. 647 */ 648 if (svc->port == FTPPORT && dport != FTPPORT) 649 return NF_ACCEPT; 650 651 if (unlikely(ip_vs_iph_icmp(iph))) 652 return NF_DROP; 653 654 /* 655 * Notify the client that the destination is unreachable, and 656 * release the socket buffer. 657 * Since it is in IP layer, the TCP socket is not actually 658 * created, the TCP RST packet cannot be sent, instead that 659 * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ 660 */ 661 #ifdef CONFIG_IP_VS_IPV6 662 if (svc->af == AF_INET6) { 663 if (!skb->dev) 664 skb->dev = net->loopback_dev; 665 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 666 } else 667 #endif 668 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 669 670 return NF_DROP; 671 } 672 673 #ifdef CONFIG_SYSCTL 674 675 static int sysctl_snat_reroute(struct netns_ipvs *ipvs) 676 { 677 return ipvs->sysctl_snat_reroute; 678 } 679 680 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs) 681 { 682 return ipvs->sysctl_nat_icmp_send; 683 } 684 685 #else 686 687 static int sysctl_snat_reroute(struct netns_ipvs *ipvs) { return 0; } 688 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs) { return 0; } 689 690 #endif 691 692 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset) 693 { 694 return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0)); 695 } 696 697 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum) 698 { 699 if (NF_INET_LOCAL_IN == hooknum) 700 return IP_DEFRAG_VS_IN; 701 if (NF_INET_FORWARD == hooknum) 702 return IP_DEFRAG_VS_FWD; 703 return IP_DEFRAG_VS_OUT; 704 } 705 706 static inline int ip_vs_gather_frags(struct netns_ipvs *ipvs, 707 struct sk_buff *skb, u_int32_t user) 708 { 709 int err; 710 711 local_bh_disable(); 712 err = ip_defrag(ipvs->net, skb, user); 713 local_bh_enable(); 714 if (!err) 715 ip_send_check(ip_hdr(skb)); 716 717 return err; 718 } 719 720 static int ip_vs_route_me_harder(struct netns_ipvs *ipvs, int af, 721 struct sk_buff *skb, unsigned int hooknum) 722 { 723 if (!sysctl_snat_reroute(ipvs)) 724 return 0; 725 /* Reroute replies only to remote clients (FORWARD and LOCAL_OUT) */ 726 if (NF_INET_LOCAL_IN == hooknum) 727 return 0; 728 #ifdef CONFIG_IP_VS_IPV6 729 if (af == AF_INET6) { 730 struct dst_entry *dst = skb_dst(skb); 731 732 if (dst->dev && !(dst->dev->flags & IFF_LOOPBACK) && 733 ip6_route_me_harder(ipvs->net, skb->sk, skb) != 0) 734 return 1; 735 } else 736 #endif 737 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL) && 738 ip_route_me_harder(ipvs->net, skb->sk, skb, RTN_LOCAL) != 0) 739 return 1; 740 741 return 0; 742 } 743 744 /* 745 * Packet has been made sufficiently writable in caller 746 * - inout: 1=in->out, 0=out->in 747 */ 748 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp, 749 struct ip_vs_conn *cp, int inout) 750 { 751 struct iphdr *iph = ip_hdr(skb); 752 unsigned int icmp_offset = iph->ihl*4; 753 struct icmphdr *icmph = (struct icmphdr *)(skb_network_header(skb) + 754 icmp_offset); 755 struct iphdr *ciph = (struct iphdr *)(icmph + 1); 756 757 if (inout) { 758 iph->saddr = cp->vaddr.ip; 759 ip_send_check(iph); 760 ciph->daddr = cp->vaddr.ip; 761 ip_send_check(ciph); 762 } else { 763 iph->daddr = cp->daddr.ip; 764 ip_send_check(iph); 765 ciph->saddr = cp->daddr.ip; 766 ip_send_check(ciph); 767 } 768 769 /* the TCP/UDP/SCTP port */ 770 if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol || 771 IPPROTO_SCTP == ciph->protocol) { 772 __be16 *ports = (void *)ciph + ciph->ihl*4; 773 774 if (inout) 775 ports[1] = cp->vport; 776 else 777 ports[0] = cp->dport; 778 } 779 780 /* And finally the ICMP checksum */ 781 icmph->checksum = 0; 782 icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset); 783 skb->ip_summed = CHECKSUM_UNNECESSARY; 784 785 if (inout) 786 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph, 787 "Forwarding altered outgoing ICMP"); 788 else 789 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph, 790 "Forwarding altered incoming ICMP"); 791 } 792 793 #ifdef CONFIG_IP_VS_IPV6 794 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp, 795 struct ip_vs_conn *cp, int inout) 796 { 797 struct ipv6hdr *iph = ipv6_hdr(skb); 798 unsigned int icmp_offset = 0; 799 unsigned int offs = 0; /* header offset*/ 800 int protocol; 801 struct icmp6hdr *icmph; 802 struct ipv6hdr *ciph; 803 unsigned short fragoffs; 804 805 ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL); 806 icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset); 807 offs = icmp_offset + sizeof(struct icmp6hdr); 808 ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs); 809 810 protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL); 811 812 if (inout) { 813 iph->saddr = cp->vaddr.in6; 814 ciph->daddr = cp->vaddr.in6; 815 } else { 816 iph->daddr = cp->daddr.in6; 817 ciph->saddr = cp->daddr.in6; 818 } 819 820 /* the TCP/UDP/SCTP port */ 821 if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol || 822 IPPROTO_SCTP == protocol)) { 823 __be16 *ports = (void *)(skb_network_header(skb) + offs); 824 825 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__, 826 ntohs(inout ? ports[1] : ports[0]), 827 ntohs(inout ? cp->vport : cp->dport)); 828 if (inout) 829 ports[1] = cp->vport; 830 else 831 ports[0] = cp->dport; 832 } 833 834 /* And finally the ICMP checksum */ 835 icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 836 skb->len - icmp_offset, 837 IPPROTO_ICMPV6, 0); 838 skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset; 839 skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum); 840 skb->ip_summed = CHECKSUM_PARTIAL; 841 842 if (inout) 843 IP_VS_DBG_PKT(11, AF_INET6, pp, skb, 844 (void *)ciph - (void *)iph, 845 "Forwarding altered outgoing ICMPv6"); 846 else 847 IP_VS_DBG_PKT(11, AF_INET6, pp, skb, 848 (void *)ciph - (void *)iph, 849 "Forwarding altered incoming ICMPv6"); 850 } 851 #endif 852 853 /* Handle relevant response ICMP messages - forward to the right 854 * destination host. 855 */ 856 static int handle_response_icmp(int af, struct sk_buff *skb, 857 union nf_inet_addr *snet, 858 __u8 protocol, struct ip_vs_conn *cp, 859 struct ip_vs_protocol *pp, 860 unsigned int offset, unsigned int ihl, 861 unsigned int hooknum) 862 { 863 unsigned int verdict = NF_DROP; 864 865 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) 866 goto after_nat; 867 868 /* Ensure the checksum is correct */ 869 if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) { 870 /* Failed checksum! */ 871 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n", 872 IP_VS_DBG_ADDR(af, snet)); 873 goto out; 874 } 875 876 if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol || 877 IPPROTO_SCTP == protocol) 878 offset += 2 * sizeof(__u16); 879 if (skb_ensure_writable(skb, offset)) 880 goto out; 881 882 #ifdef CONFIG_IP_VS_IPV6 883 if (af == AF_INET6) 884 ip_vs_nat_icmp_v6(skb, pp, cp, 1); 885 else 886 #endif 887 ip_vs_nat_icmp(skb, pp, cp, 1); 888 889 if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum)) 890 goto out; 891 892 after_nat: 893 /* do the statistics and put it back */ 894 ip_vs_out_stats(cp, skb); 895 896 skb->ipvs_property = 1; 897 if (!(cp->flags & IP_VS_CONN_F_NFCT)) 898 ip_vs_notrack(skb); 899 else 900 ip_vs_update_conntrack(skb, cp, 0); 901 verdict = NF_ACCEPT; 902 903 out: 904 __ip_vs_conn_put(cp); 905 906 return verdict; 907 } 908 909 /* 910 * Handle ICMP messages in the inside-to-outside direction (outgoing). 911 * Find any that might be relevant, check against existing connections. 912 * Currently handles error types - unreachable, quench, ttl exceeded. 913 */ 914 static int ip_vs_out_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb, 915 int *related, unsigned int hooknum) 916 { 917 struct iphdr *iph; 918 struct icmphdr _icmph, *ic; 919 struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */ 920 struct ip_vs_iphdr ciph; 921 struct ip_vs_conn *cp; 922 struct ip_vs_protocol *pp; 923 unsigned int offset, ihl; 924 union nf_inet_addr snet; 925 926 *related = 1; 927 928 /* reassemble IP fragments */ 929 if (ip_is_fragment(ip_hdr(skb))) { 930 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum))) 931 return NF_STOLEN; 932 } 933 934 iph = ip_hdr(skb); 935 offset = ihl = iph->ihl * 4; 936 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph); 937 if (ic == NULL) 938 return NF_DROP; 939 940 IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n", 941 ic->type, ntohs(icmp_id(ic)), 942 &iph->saddr, &iph->daddr); 943 944 /* 945 * Work through seeing if this is for us. 946 * These checks are supposed to be in an order that means easy 947 * things are checked first to speed up processing.... however 948 * this means that some packets will manage to get a long way 949 * down this stack and then be rejected, but that's life. 950 */ 951 if ((ic->type != ICMP_DEST_UNREACH) && 952 (ic->type != ICMP_SOURCE_QUENCH) && 953 (ic->type != ICMP_TIME_EXCEEDED)) { 954 *related = 0; 955 return NF_ACCEPT; 956 } 957 958 /* Now find the contained IP header */ 959 offset += sizeof(_icmph); 960 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph); 961 if (cih == NULL) 962 return NF_ACCEPT; /* The packet looks wrong, ignore */ 963 964 pp = ip_vs_proto_get(cih->protocol); 965 if (!pp) 966 return NF_ACCEPT; 967 968 /* Is the embedded protocol header present? */ 969 if (unlikely(cih->frag_off & htons(IP_OFFSET) && 970 pp->dont_defrag)) 971 return NF_ACCEPT; 972 973 IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset, 974 "Checking outgoing ICMP for"); 975 976 ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, true, &ciph); 977 978 /* The embedded headers contain source and dest in reverse order */ 979 cp = INDIRECT_CALL_1(pp->conn_out_get, ip_vs_conn_out_get_proto, 980 ipvs, AF_INET, skb, &ciph); 981 if (!cp) 982 return NF_ACCEPT; 983 984 snet.ip = iph->saddr; 985 return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp, 986 pp, ciph.len, ihl, hooknum); 987 } 988 989 #ifdef CONFIG_IP_VS_IPV6 990 static int ip_vs_out_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb, 991 int *related, unsigned int hooknum, 992 struct ip_vs_iphdr *ipvsh) 993 { 994 struct icmp6hdr _icmph, *ic; 995 struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */ 996 struct ip_vs_conn *cp; 997 struct ip_vs_protocol *pp; 998 union nf_inet_addr snet; 999 unsigned int offset; 1000 1001 *related = 1; 1002 ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph); 1003 if (ic == NULL) 1004 return NF_DROP; 1005 1006 /* 1007 * Work through seeing if this is for us. 1008 * These checks are supposed to be in an order that means easy 1009 * things are checked first to speed up processing.... however 1010 * this means that some packets will manage to get a long way 1011 * down this stack and then be rejected, but that's life. 1012 */ 1013 if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) { 1014 *related = 0; 1015 return NF_ACCEPT; 1016 } 1017 /* Fragment header that is before ICMP header tells us that: 1018 * it's not an error message since they can't be fragmented. 1019 */ 1020 if (ipvsh->flags & IP6_FH_F_FRAG) 1021 return NF_DROP; 1022 1023 IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n", 1024 ic->icmp6_type, ntohs(icmpv6_id(ic)), 1025 &ipvsh->saddr, &ipvsh->daddr); 1026 1027 if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, ipvsh->len + sizeof(_icmph), 1028 true, &ciph)) 1029 return NF_ACCEPT; /* The packet looks wrong, ignore */ 1030 1031 pp = ip_vs_proto_get(ciph.protocol); 1032 if (!pp) 1033 return NF_ACCEPT; 1034 1035 /* The embedded headers contain source and dest in reverse order */ 1036 cp = INDIRECT_CALL_1(pp->conn_out_get, ip_vs_conn_out_get_proto, 1037 ipvs, AF_INET6, skb, &ciph); 1038 if (!cp) 1039 return NF_ACCEPT; 1040 1041 snet.in6 = ciph.saddr.in6; 1042 offset = ciph.len; 1043 return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp, 1044 pp, offset, sizeof(struct ipv6hdr), 1045 hooknum); 1046 } 1047 #endif 1048 1049 /* 1050 * Check if sctp chunc is ABORT chunk 1051 */ 1052 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len) 1053 { 1054 struct sctp_chunkhdr *sch, schunk; 1055 sch = skb_header_pointer(skb, nh_len + sizeof(struct sctphdr), 1056 sizeof(schunk), &schunk); 1057 if (sch == NULL) 1058 return 0; 1059 if (sch->type == SCTP_CID_ABORT) 1060 return 1; 1061 return 0; 1062 } 1063 1064 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len) 1065 { 1066 struct tcphdr _tcph, *th; 1067 1068 th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph); 1069 if (th == NULL) 1070 return 0; 1071 return th->rst; 1072 } 1073 1074 static inline bool is_new_conn(const struct sk_buff *skb, 1075 struct ip_vs_iphdr *iph) 1076 { 1077 switch (iph->protocol) { 1078 case IPPROTO_TCP: { 1079 struct tcphdr _tcph, *th; 1080 1081 th = skb_header_pointer(skb, iph->len, sizeof(_tcph), &_tcph); 1082 if (th == NULL) 1083 return false; 1084 return th->syn; 1085 } 1086 case IPPROTO_SCTP: { 1087 struct sctp_chunkhdr *sch, schunk; 1088 1089 sch = skb_header_pointer(skb, iph->len + sizeof(struct sctphdr), 1090 sizeof(schunk), &schunk); 1091 if (sch == NULL) 1092 return false; 1093 return sch->type == SCTP_CID_INIT; 1094 } 1095 default: 1096 return false; 1097 } 1098 } 1099 1100 static inline bool is_new_conn_expected(const struct ip_vs_conn *cp, 1101 int conn_reuse_mode) 1102 { 1103 /* Controlled (FTP DATA or persistence)? */ 1104 if (cp->control) 1105 return false; 1106 1107 switch (cp->protocol) { 1108 case IPPROTO_TCP: 1109 return (cp->state == IP_VS_TCP_S_TIME_WAIT) || 1110 (cp->state == IP_VS_TCP_S_CLOSE) || 1111 ((conn_reuse_mode & 2) && 1112 (cp->state == IP_VS_TCP_S_FIN_WAIT) && 1113 (cp->flags & IP_VS_CONN_F_NOOUTPUT)); 1114 case IPPROTO_SCTP: 1115 return cp->state == IP_VS_SCTP_S_CLOSED; 1116 default: 1117 return false; 1118 } 1119 } 1120 1121 /* Generic function to create new connections for outgoing RS packets 1122 * 1123 * Pre-requisites for successful connection creation: 1124 * 1) Virtual Service is NOT fwmark based: 1125 * In fwmark-VS actual vaddr and vport are unknown to IPVS 1126 * 2) Real Server and Virtual Service were NOT configured without port: 1127 * This is to allow match of different VS to the same RS ip-addr 1128 */ 1129 struct ip_vs_conn *ip_vs_new_conn_out(struct ip_vs_service *svc, 1130 struct ip_vs_dest *dest, 1131 struct sk_buff *skb, 1132 const struct ip_vs_iphdr *iph, 1133 __be16 dport, 1134 __be16 cport) 1135 { 1136 struct ip_vs_conn_param param; 1137 struct ip_vs_conn *ct = NULL, *cp = NULL; 1138 const union nf_inet_addr *vaddr, *daddr, *caddr; 1139 union nf_inet_addr snet; 1140 __be16 vport; 1141 unsigned int flags; 1142 1143 vaddr = &svc->addr; 1144 vport = svc->port; 1145 daddr = &iph->saddr; 1146 caddr = &iph->daddr; 1147 1148 /* check pre-requisites are satisfied */ 1149 if (svc->fwmark) 1150 return NULL; 1151 if (!vport || !dport) 1152 return NULL; 1153 1154 /* for persistent service first create connection template */ 1155 if (svc->flags & IP_VS_SVC_F_PERSISTENT) { 1156 /* apply netmask the same way ingress-side does */ 1157 #ifdef CONFIG_IP_VS_IPV6 1158 if (svc->af == AF_INET6) 1159 ipv6_addr_prefix(&snet.in6, &caddr->in6, 1160 (__force __u32)svc->netmask); 1161 else 1162 #endif 1163 snet.ip = caddr->ip & svc->netmask; 1164 /* fill params and create template if not existent */ 1165 if (ip_vs_conn_fill_param_persist(svc, skb, iph->protocol, 1166 &snet, 0, vaddr, 1167 vport, ¶m) < 0) 1168 return NULL; 1169 ct = ip_vs_ct_in_get(¶m); 1170 /* check if template exists and points to the same dest */ 1171 if (!ct || !ip_vs_check_template(ct, dest)) { 1172 ct = ip_vs_conn_new(¶m, dest->af, daddr, dport, 1173 IP_VS_CONN_F_TEMPLATE, dest, 0); 1174 if (!ct) { 1175 kfree(param.pe_data); 1176 return NULL; 1177 } 1178 ct->timeout = svc->timeout; 1179 } else { 1180 kfree(param.pe_data); 1181 } 1182 } 1183 1184 /* connection flags */ 1185 flags = ((svc->flags & IP_VS_SVC_F_ONEPACKET) && 1186 iph->protocol == IPPROTO_UDP) ? IP_VS_CONN_F_ONE_PACKET : 0; 1187 /* create connection */ 1188 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol, 1189 caddr, cport, vaddr, vport, ¶m); 1190 cp = ip_vs_conn_new(¶m, dest->af, daddr, dport, flags, dest, 0); 1191 if (!cp) { 1192 if (ct) 1193 ip_vs_conn_put(ct); 1194 return NULL; 1195 } 1196 if (ct) { 1197 ip_vs_control_add(cp, ct); 1198 ip_vs_conn_put(ct); 1199 } 1200 ip_vs_conn_stats(cp, svc); 1201 1202 /* return connection (will be used to handle outgoing packet) */ 1203 IP_VS_DBG_BUF(6, "New connection RS-initiated:%c c:%s:%u v:%s:%u " 1204 "d:%s:%u conn->flags:%X conn->refcnt:%d\n", 1205 ip_vs_fwd_tag(cp), 1206 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport), 1207 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport), 1208 IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport), 1209 cp->flags, refcount_read(&cp->refcnt)); 1210 return cp; 1211 } 1212 1213 /* Handle outgoing packets which are considered requests initiated by 1214 * real servers, so that subsequent responses from external client can be 1215 * routed to the right real server. 1216 * Used also for outgoing responses in OPS mode. 1217 * 1218 * Connection management is handled by persistent-engine specific callback. 1219 */ 1220 static struct ip_vs_conn *__ip_vs_rs_conn_out(unsigned int hooknum, 1221 struct netns_ipvs *ipvs, 1222 int af, struct sk_buff *skb, 1223 const struct ip_vs_iphdr *iph) 1224 { 1225 struct ip_vs_dest *dest; 1226 struct ip_vs_conn *cp = NULL; 1227 __be16 _ports[2], *pptr; 1228 1229 if (hooknum == NF_INET_LOCAL_IN) 1230 return NULL; 1231 1232 pptr = frag_safe_skb_hp(skb, iph->len, 1233 sizeof(_ports), _ports); 1234 if (!pptr) 1235 return NULL; 1236 1237 dest = ip_vs_find_real_service(ipvs, af, iph->protocol, 1238 &iph->saddr, pptr[0]); 1239 if (dest) { 1240 struct ip_vs_service *svc; 1241 struct ip_vs_pe *pe; 1242 1243 svc = rcu_dereference(dest->svc); 1244 if (svc) { 1245 pe = rcu_dereference(svc->pe); 1246 if (pe && pe->conn_out) 1247 cp = pe->conn_out(svc, dest, skb, iph, 1248 pptr[0], pptr[1]); 1249 } 1250 } 1251 1252 return cp; 1253 } 1254 1255 /* Handle response packets: rewrite addresses and send away... 1256 */ 1257 static unsigned int 1258 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd, 1259 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph, 1260 unsigned int hooknum) 1261 { 1262 struct ip_vs_protocol *pp = pd->pp; 1263 1264 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) 1265 goto after_nat; 1266 1267 IP_VS_DBG_PKT(11, af, pp, skb, iph->off, "Outgoing packet"); 1268 1269 if (skb_ensure_writable(skb, iph->len)) 1270 goto drop; 1271 1272 /* mangle the packet */ 1273 if (pp->snat_handler && 1274 !SNAT_CALL(pp->snat_handler, skb, pp, cp, iph)) 1275 goto drop; 1276 1277 #ifdef CONFIG_IP_VS_IPV6 1278 if (af == AF_INET6) 1279 ipv6_hdr(skb)->saddr = cp->vaddr.in6; 1280 else 1281 #endif 1282 { 1283 ip_hdr(skb)->saddr = cp->vaddr.ip; 1284 ip_send_check(ip_hdr(skb)); 1285 } 1286 1287 /* 1288 * nf_iterate does not expect change in the skb->dst->dev. 1289 * It looks like it is not fatal to enable this code for hooks 1290 * where our handlers are at the end of the chain list and 1291 * when all next handlers use skb->dst->dev and not outdev. 1292 * It will definitely route properly the inout NAT traffic 1293 * when multiple paths are used. 1294 */ 1295 1296 /* For policy routing, packets originating from this 1297 * machine itself may be routed differently to packets 1298 * passing through. We want this packet to be routed as 1299 * if it came from this machine itself. So re-compute 1300 * the routing information. 1301 */ 1302 if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum)) 1303 goto drop; 1304 1305 IP_VS_DBG_PKT(10, af, pp, skb, iph->off, "After SNAT"); 1306 1307 after_nat: 1308 ip_vs_out_stats(cp, skb); 1309 ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd); 1310 skb->ipvs_property = 1; 1311 if (!(cp->flags & IP_VS_CONN_F_NFCT)) 1312 ip_vs_notrack(skb); 1313 else 1314 ip_vs_update_conntrack(skb, cp, 0); 1315 ip_vs_conn_put(cp); 1316 1317 return NF_ACCEPT; 1318 1319 drop: 1320 ip_vs_conn_put(cp); 1321 kfree_skb(skb); 1322 return NF_STOLEN; 1323 } 1324 1325 /* 1326 * Check if outgoing packet belongs to the established ip_vs_conn. 1327 */ 1328 static unsigned int 1329 ip_vs_out_hook(void *priv, struct sk_buff *skb, const struct nf_hook_state *state) 1330 { 1331 struct netns_ipvs *ipvs = net_ipvs(state->net); 1332 unsigned int hooknum = state->hook; 1333 struct ip_vs_iphdr iph; 1334 struct ip_vs_protocol *pp; 1335 struct ip_vs_proto_data *pd; 1336 struct ip_vs_conn *cp; 1337 int af = state->pf; 1338 struct sock *sk; 1339 1340 /* Already marked as IPVS request or reply? */ 1341 if (skb->ipvs_property) 1342 return NF_ACCEPT; 1343 1344 sk = skb_to_full_sk(skb); 1345 /* Bad... Do not break raw sockets */ 1346 if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT && 1347 af == AF_INET)) { 1348 1349 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag) 1350 return NF_ACCEPT; 1351 } 1352 1353 if (unlikely(!skb_dst(skb))) 1354 return NF_ACCEPT; 1355 1356 if (!ipvs->enable) 1357 return NF_ACCEPT; 1358 1359 ip_vs_fill_iph_skb(af, skb, false, &iph); 1360 #ifdef CONFIG_IP_VS_IPV6 1361 if (af == AF_INET6) { 1362 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) { 1363 int related; 1364 int verdict = ip_vs_out_icmp_v6(ipvs, skb, &related, 1365 hooknum, &iph); 1366 1367 if (related) 1368 return verdict; 1369 } 1370 } else 1371 #endif 1372 if (unlikely(iph.protocol == IPPROTO_ICMP)) { 1373 int related; 1374 int verdict = ip_vs_out_icmp(ipvs, skb, &related, hooknum); 1375 1376 if (related) 1377 return verdict; 1378 } 1379 1380 pd = ip_vs_proto_data_get(ipvs, iph.protocol); 1381 if (unlikely(!pd)) 1382 return NF_ACCEPT; 1383 pp = pd->pp; 1384 1385 /* reassemble IP fragments */ 1386 #ifdef CONFIG_IP_VS_IPV6 1387 if (af == AF_INET) 1388 #endif 1389 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) { 1390 if (ip_vs_gather_frags(ipvs, skb, 1391 ip_vs_defrag_user(hooknum))) 1392 return NF_STOLEN; 1393 1394 ip_vs_fill_iph_skb(AF_INET, skb, false, &iph); 1395 } 1396 1397 /* 1398 * Check if the packet belongs to an existing entry 1399 */ 1400 cp = INDIRECT_CALL_1(pp->conn_out_get, ip_vs_conn_out_get_proto, 1401 ipvs, af, skb, &iph); 1402 1403 if (likely(cp)) 1404 return handle_response(af, skb, pd, cp, &iph, hooknum); 1405 1406 /* Check for real-server-started requests */ 1407 if (atomic_read(&ipvs->conn_out_counter)) { 1408 /* Currently only for UDP: 1409 * connection oriented protocols typically use 1410 * ephemeral ports for outgoing connections, so 1411 * related incoming responses would not match any VS 1412 */ 1413 if (pp->protocol == IPPROTO_UDP) { 1414 cp = __ip_vs_rs_conn_out(hooknum, ipvs, af, skb, &iph); 1415 if (likely(cp)) 1416 return handle_response(af, skb, pd, cp, &iph, 1417 hooknum); 1418 } 1419 } 1420 1421 if (sysctl_nat_icmp_send(ipvs) && 1422 (pp->protocol == IPPROTO_TCP || 1423 pp->protocol == IPPROTO_UDP || 1424 pp->protocol == IPPROTO_SCTP)) { 1425 __be16 _ports[2], *pptr; 1426 1427 pptr = frag_safe_skb_hp(skb, iph.len, 1428 sizeof(_ports), _ports); 1429 if (pptr == NULL) 1430 return NF_ACCEPT; /* Not for me */ 1431 if (ip_vs_has_real_service(ipvs, af, iph.protocol, &iph.saddr, 1432 pptr[0])) { 1433 /* 1434 * Notify the real server: there is no 1435 * existing entry if it is not RST 1436 * packet or not TCP packet. 1437 */ 1438 if ((iph.protocol != IPPROTO_TCP && 1439 iph.protocol != IPPROTO_SCTP) 1440 || ((iph.protocol == IPPROTO_TCP 1441 && !is_tcp_reset(skb, iph.len)) 1442 || (iph.protocol == IPPROTO_SCTP 1443 && !is_sctp_abort(skb, 1444 iph.len)))) { 1445 #ifdef CONFIG_IP_VS_IPV6 1446 if (af == AF_INET6) { 1447 if (!skb->dev) 1448 skb->dev = ipvs->net->loopback_dev; 1449 icmpv6_send(skb, 1450 ICMPV6_DEST_UNREACH, 1451 ICMPV6_PORT_UNREACH, 1452 0); 1453 } else 1454 #endif 1455 icmp_send(skb, 1456 ICMP_DEST_UNREACH, 1457 ICMP_PORT_UNREACH, 0); 1458 return NF_DROP; 1459 } 1460 } 1461 } 1462 1463 IP_VS_DBG_PKT(12, af, pp, skb, iph.off, 1464 "ip_vs_out: packet continues traversal as normal"); 1465 return NF_ACCEPT; 1466 } 1467 1468 static unsigned int 1469 ip_vs_try_to_schedule(struct netns_ipvs *ipvs, int af, struct sk_buff *skb, 1470 struct ip_vs_proto_data *pd, 1471 int *verdict, struct ip_vs_conn **cpp, 1472 struct ip_vs_iphdr *iph) 1473 { 1474 struct ip_vs_protocol *pp = pd->pp; 1475 1476 if (!iph->fragoffs) { 1477 /* No (second) fragments need to enter here, as nf_defrag_ipv6 1478 * replayed fragment zero will already have created the cp 1479 */ 1480 1481 /* Schedule and create new connection entry into cpp */ 1482 if (!pp->conn_schedule(ipvs, af, skb, pd, verdict, cpp, iph)) 1483 return 0; 1484 } 1485 1486 if (unlikely(!*cpp)) { 1487 /* sorry, all this trouble for a no-hit :) */ 1488 IP_VS_DBG_PKT(12, af, pp, skb, iph->off, 1489 "ip_vs_in: packet continues traversal as normal"); 1490 1491 /* Fragment couldn't be mapped to a conn entry */ 1492 if (iph->fragoffs) 1493 IP_VS_DBG_PKT(7, af, pp, skb, iph->off, 1494 "unhandled fragment"); 1495 1496 *verdict = NF_ACCEPT; 1497 return 0; 1498 } 1499 1500 return 1; 1501 } 1502 1503 /* Check the UDP tunnel and return its header length */ 1504 static int ipvs_udp_decap(struct netns_ipvs *ipvs, struct sk_buff *skb, 1505 unsigned int offset, __u16 af, 1506 const union nf_inet_addr *daddr, __u8 *proto) 1507 { 1508 struct udphdr _udph, *udph; 1509 struct ip_vs_dest *dest; 1510 1511 udph = skb_header_pointer(skb, offset, sizeof(_udph), &_udph); 1512 if (!udph) 1513 goto unk; 1514 offset += sizeof(struct udphdr); 1515 dest = ip_vs_find_tunnel(ipvs, af, daddr, udph->dest); 1516 if (!dest) 1517 goto unk; 1518 if (dest->tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GUE) { 1519 struct guehdr _gueh, *gueh; 1520 1521 gueh = skb_header_pointer(skb, offset, sizeof(_gueh), &_gueh); 1522 if (!gueh) 1523 goto unk; 1524 if (gueh->control != 0 || gueh->version != 0) 1525 goto unk; 1526 /* Later we can support also IPPROTO_IPV6 */ 1527 if (gueh->proto_ctype != IPPROTO_IPIP) 1528 goto unk; 1529 *proto = gueh->proto_ctype; 1530 return sizeof(struct udphdr) + sizeof(struct guehdr) + 1531 (gueh->hlen << 2); 1532 } 1533 1534 unk: 1535 return 0; 1536 } 1537 1538 /* Check the GRE tunnel and return its header length */ 1539 static int ipvs_gre_decap(struct netns_ipvs *ipvs, struct sk_buff *skb, 1540 unsigned int offset, __u16 af, 1541 const union nf_inet_addr *daddr, __u8 *proto) 1542 { 1543 struct gre_base_hdr _greh, *greh; 1544 struct ip_vs_dest *dest; 1545 1546 greh = skb_header_pointer(skb, offset, sizeof(_greh), &_greh); 1547 if (!greh) 1548 goto unk; 1549 dest = ip_vs_find_tunnel(ipvs, af, daddr, 0); 1550 if (!dest) 1551 goto unk; 1552 if (dest->tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GRE) { 1553 __be16 type; 1554 1555 /* Only support version 0 and C (csum) */ 1556 if ((greh->flags & ~GRE_CSUM) != 0) 1557 goto unk; 1558 type = greh->protocol; 1559 /* Later we can support also IPPROTO_IPV6 */ 1560 if (type != htons(ETH_P_IP)) 1561 goto unk; 1562 *proto = IPPROTO_IPIP; 1563 return gre_calc_hlen(gre_flags_to_tnl_flags(greh->flags)); 1564 } 1565 1566 unk: 1567 return 0; 1568 } 1569 1570 /* 1571 * Handle ICMP messages in the outside-to-inside direction (incoming). 1572 * Find any that might be relevant, check against existing connections, 1573 * forward to the right destination host if relevant. 1574 * Currently handles error types - unreachable, quench, ttl exceeded. 1575 */ 1576 static int 1577 ip_vs_in_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb, int *related, 1578 unsigned int hooknum) 1579 { 1580 struct iphdr *iph; 1581 struct icmphdr _icmph, *ic; 1582 struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */ 1583 struct ip_vs_iphdr ciph; 1584 struct ip_vs_conn *cp; 1585 struct ip_vs_protocol *pp; 1586 struct ip_vs_proto_data *pd; 1587 unsigned int offset, offset2, ihl, verdict; 1588 bool tunnel, new_cp = false; 1589 union nf_inet_addr *raddr; 1590 char *outer_proto = "IPIP"; 1591 1592 *related = 1; 1593 1594 /* reassemble IP fragments */ 1595 if (ip_is_fragment(ip_hdr(skb))) { 1596 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum))) 1597 return NF_STOLEN; 1598 } 1599 1600 iph = ip_hdr(skb); 1601 offset = ihl = iph->ihl * 4; 1602 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph); 1603 if (ic == NULL) 1604 return NF_DROP; 1605 1606 IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n", 1607 ic->type, ntohs(icmp_id(ic)), 1608 &iph->saddr, &iph->daddr); 1609 1610 /* 1611 * Work through seeing if this is for us. 1612 * These checks are supposed to be in an order that means easy 1613 * things are checked first to speed up processing.... however 1614 * this means that some packets will manage to get a long way 1615 * down this stack and then be rejected, but that's life. 1616 */ 1617 if ((ic->type != ICMP_DEST_UNREACH) && 1618 (ic->type != ICMP_SOURCE_QUENCH) && 1619 (ic->type != ICMP_TIME_EXCEEDED)) { 1620 *related = 0; 1621 return NF_ACCEPT; 1622 } 1623 1624 /* Now find the contained IP header */ 1625 offset += sizeof(_icmph); 1626 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph); 1627 if (cih == NULL) 1628 return NF_ACCEPT; /* The packet looks wrong, ignore */ 1629 raddr = (union nf_inet_addr *)&cih->daddr; 1630 1631 /* Special case for errors for IPIP/UDP/GRE tunnel packets */ 1632 tunnel = false; 1633 if (cih->protocol == IPPROTO_IPIP) { 1634 struct ip_vs_dest *dest; 1635 1636 if (unlikely(cih->frag_off & htons(IP_OFFSET))) 1637 return NF_ACCEPT; 1638 /* Error for our IPIP must arrive at LOCAL_IN */ 1639 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL)) 1640 return NF_ACCEPT; 1641 dest = ip_vs_find_tunnel(ipvs, AF_INET, raddr, 0); 1642 /* Only for known tunnel */ 1643 if (!dest || dest->tun_type != IP_VS_CONN_F_TUNNEL_TYPE_IPIP) 1644 return NF_ACCEPT; 1645 offset += cih->ihl * 4; 1646 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph); 1647 if (cih == NULL) 1648 return NF_ACCEPT; /* The packet looks wrong, ignore */ 1649 tunnel = true; 1650 } else if ((cih->protocol == IPPROTO_UDP || /* Can be UDP encap */ 1651 cih->protocol == IPPROTO_GRE) && /* Can be GRE encap */ 1652 /* Error for our tunnel must arrive at LOCAL_IN */ 1653 (skb_rtable(skb)->rt_flags & RTCF_LOCAL)) { 1654 __u8 iproto; 1655 int ulen; 1656 1657 /* Non-first fragment has no UDP/GRE header */ 1658 if (unlikely(cih->frag_off & htons(IP_OFFSET))) 1659 return NF_ACCEPT; 1660 offset2 = offset + cih->ihl * 4; 1661 if (cih->protocol == IPPROTO_UDP) { 1662 ulen = ipvs_udp_decap(ipvs, skb, offset2, AF_INET, 1663 raddr, &iproto); 1664 outer_proto = "UDP"; 1665 } else { 1666 ulen = ipvs_gre_decap(ipvs, skb, offset2, AF_INET, 1667 raddr, &iproto); 1668 outer_proto = "GRE"; 1669 } 1670 if (ulen > 0) { 1671 /* Skip IP and UDP/GRE tunnel headers */ 1672 offset = offset2 + ulen; 1673 /* Now we should be at the original IP header */ 1674 cih = skb_header_pointer(skb, offset, sizeof(_ciph), 1675 &_ciph); 1676 if (cih && cih->version == 4 && cih->ihl >= 5 && 1677 iproto == IPPROTO_IPIP) 1678 tunnel = true; 1679 else 1680 return NF_ACCEPT; 1681 } 1682 } 1683 1684 pd = ip_vs_proto_data_get(ipvs, cih->protocol); 1685 if (!pd) 1686 return NF_ACCEPT; 1687 pp = pd->pp; 1688 1689 /* Is the embedded protocol header present? */ 1690 if (unlikely(cih->frag_off & htons(IP_OFFSET) && 1691 pp->dont_defrag)) 1692 return NF_ACCEPT; 1693 1694 IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset, 1695 "Checking incoming ICMP for"); 1696 1697 offset2 = offset; 1698 ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, !tunnel, &ciph); 1699 offset = ciph.len; 1700 1701 /* The embedded headers contain source and dest in reverse order. 1702 * For IPIP/UDP/GRE tunnel this is error for request, not for reply. 1703 */ 1704 cp = INDIRECT_CALL_1(pp->conn_in_get, ip_vs_conn_in_get_proto, 1705 ipvs, AF_INET, skb, &ciph); 1706 1707 if (!cp) { 1708 int v; 1709 1710 if (tunnel || !sysctl_schedule_icmp(ipvs)) 1711 return NF_ACCEPT; 1712 1713 if (!ip_vs_try_to_schedule(ipvs, AF_INET, skb, pd, &v, &cp, &ciph)) 1714 return v; 1715 new_cp = true; 1716 } 1717 1718 verdict = NF_DROP; 1719 1720 /* Ensure the checksum is correct */ 1721 if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) { 1722 /* Failed checksum! */ 1723 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n", 1724 &iph->saddr); 1725 goto out; 1726 } 1727 1728 if (tunnel) { 1729 __be32 info = ic->un.gateway; 1730 __u8 type = ic->type; 1731 __u8 code = ic->code; 1732 1733 /* Update the MTU */ 1734 if (ic->type == ICMP_DEST_UNREACH && 1735 ic->code == ICMP_FRAG_NEEDED) { 1736 struct ip_vs_dest *dest = cp->dest; 1737 u32 mtu = ntohs(ic->un.frag.mtu); 1738 __be16 frag_off = cih->frag_off; 1739 1740 /* Strip outer IP and ICMP, go to IPIP/UDP/GRE header */ 1741 if (pskb_pull(skb, ihl + sizeof(_icmph)) == NULL) 1742 goto ignore_tunnel; 1743 offset2 -= ihl + sizeof(_icmph); 1744 skb_reset_network_header(skb); 1745 IP_VS_DBG(12, "ICMP for %s %pI4->%pI4: mtu=%u\n", 1746 outer_proto, &ip_hdr(skb)->saddr, 1747 &ip_hdr(skb)->daddr, mtu); 1748 ipv4_update_pmtu(skb, ipvs->net, mtu, 0, 0); 1749 /* Client uses PMTUD? */ 1750 if (!(frag_off & htons(IP_DF))) 1751 goto ignore_tunnel; 1752 /* Prefer the resulting PMTU */ 1753 if (dest) { 1754 struct ip_vs_dest_dst *dest_dst; 1755 1756 dest_dst = rcu_dereference(dest->dest_dst); 1757 if (dest_dst) 1758 mtu = dst_mtu(dest_dst->dst_cache); 1759 } 1760 if (mtu > 68 + sizeof(struct iphdr)) 1761 mtu -= sizeof(struct iphdr); 1762 info = htonl(mtu); 1763 } 1764 /* Strip outer IP, ICMP and IPIP/UDP/GRE, go to IP header of 1765 * original request. 1766 */ 1767 if (pskb_pull(skb, offset2) == NULL) 1768 goto ignore_tunnel; 1769 skb_reset_network_header(skb); 1770 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n", 1771 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, 1772 type, code, ntohl(info)); 1773 icmp_send(skb, type, code, info); 1774 /* ICMP can be shorter but anyways, account it */ 1775 ip_vs_out_stats(cp, skb); 1776 1777 ignore_tunnel: 1778 consume_skb(skb); 1779 verdict = NF_STOLEN; 1780 goto out; 1781 } 1782 1783 /* do the statistics and put it back */ 1784 ip_vs_in_stats(cp, skb); 1785 if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol || 1786 IPPROTO_SCTP == cih->protocol) 1787 offset += 2 * sizeof(__u16); 1788 verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph); 1789 1790 out: 1791 if (likely(!new_cp)) 1792 __ip_vs_conn_put(cp); 1793 else 1794 ip_vs_conn_put(cp); 1795 1796 return verdict; 1797 } 1798 1799 #ifdef CONFIG_IP_VS_IPV6 1800 static int ip_vs_in_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb, 1801 int *related, unsigned int hooknum, 1802 struct ip_vs_iphdr *iph) 1803 { 1804 struct icmp6hdr _icmph, *ic; 1805 struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */ 1806 struct ip_vs_conn *cp; 1807 struct ip_vs_protocol *pp; 1808 struct ip_vs_proto_data *pd; 1809 unsigned int offset, verdict; 1810 bool new_cp = false; 1811 1812 *related = 1; 1813 1814 ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph); 1815 if (ic == NULL) 1816 return NF_DROP; 1817 1818 /* 1819 * Work through seeing if this is for us. 1820 * These checks are supposed to be in an order that means easy 1821 * things are checked first to speed up processing.... however 1822 * this means that some packets will manage to get a long way 1823 * down this stack and then be rejected, but that's life. 1824 */ 1825 if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) { 1826 *related = 0; 1827 return NF_ACCEPT; 1828 } 1829 /* Fragment header that is before ICMP header tells us that: 1830 * it's not an error message since they can't be fragmented. 1831 */ 1832 if (iph->flags & IP6_FH_F_FRAG) 1833 return NF_DROP; 1834 1835 IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n", 1836 ic->icmp6_type, ntohs(icmpv6_id(ic)), 1837 &iph->saddr, &iph->daddr); 1838 1839 offset = iph->len + sizeof(_icmph); 1840 if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, offset, true, &ciph)) 1841 return NF_ACCEPT; 1842 1843 pd = ip_vs_proto_data_get(ipvs, ciph.protocol); 1844 if (!pd) 1845 return NF_ACCEPT; 1846 pp = pd->pp; 1847 1848 /* Cannot handle fragmented embedded protocol */ 1849 if (ciph.fragoffs) 1850 return NF_ACCEPT; 1851 1852 IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset, 1853 "Checking incoming ICMPv6 for"); 1854 1855 /* The embedded headers contain source and dest in reverse order 1856 * if not from localhost 1857 */ 1858 cp = INDIRECT_CALL_1(pp->conn_in_get, ip_vs_conn_in_get_proto, 1859 ipvs, AF_INET6, skb, &ciph); 1860 1861 if (!cp) { 1862 int v; 1863 1864 if (!sysctl_schedule_icmp(ipvs)) 1865 return NF_ACCEPT; 1866 1867 if (!ip_vs_try_to_schedule(ipvs, AF_INET6, skb, pd, &v, &cp, &ciph)) 1868 return v; 1869 1870 new_cp = true; 1871 } 1872 1873 /* VS/TUN, VS/DR and LOCALNODE just let it go */ 1874 if ((hooknum == NF_INET_LOCAL_OUT) && 1875 (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) { 1876 verdict = NF_ACCEPT; 1877 goto out; 1878 } 1879 1880 /* do the statistics and put it back */ 1881 ip_vs_in_stats(cp, skb); 1882 1883 /* Need to mangle contained IPv6 header in ICMPv6 packet */ 1884 offset = ciph.len; 1885 if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol || 1886 IPPROTO_SCTP == ciph.protocol) 1887 offset += 2 * sizeof(__u16); /* Also mangle ports */ 1888 1889 verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset, hooknum, &ciph); 1890 1891 out: 1892 if (likely(!new_cp)) 1893 __ip_vs_conn_put(cp); 1894 else 1895 ip_vs_conn_put(cp); 1896 1897 return verdict; 1898 } 1899 #endif 1900 1901 1902 /* 1903 * Check if it's for virtual services, look it up, 1904 * and send it on its way... 1905 */ 1906 static unsigned int 1907 ip_vs_in_hook(void *priv, struct sk_buff *skb, const struct nf_hook_state *state) 1908 { 1909 struct netns_ipvs *ipvs = net_ipvs(state->net); 1910 unsigned int hooknum = state->hook; 1911 struct ip_vs_iphdr iph; 1912 struct ip_vs_protocol *pp; 1913 struct ip_vs_proto_data *pd; 1914 struct ip_vs_conn *cp; 1915 int ret, pkts; 1916 struct sock *sk; 1917 int af = state->pf; 1918 1919 /* Already marked as IPVS request or reply? */ 1920 if (skb->ipvs_property) 1921 return NF_ACCEPT; 1922 1923 /* 1924 * Big tappo: 1925 * - remote client: only PACKET_HOST 1926 * - route: used for struct net when skb->dev is unset 1927 */ 1928 if (unlikely((skb->pkt_type != PACKET_HOST && 1929 hooknum != NF_INET_LOCAL_OUT) || 1930 !skb_dst(skb))) { 1931 ip_vs_fill_iph_skb(af, skb, false, &iph); 1932 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s" 1933 " ignored in hook %u\n", 1934 skb->pkt_type, iph.protocol, 1935 IP_VS_DBG_ADDR(af, &iph.daddr), hooknum); 1936 return NF_ACCEPT; 1937 } 1938 /* ipvs enabled in this netns ? */ 1939 if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable)) 1940 return NF_ACCEPT; 1941 1942 ip_vs_fill_iph_skb(af, skb, false, &iph); 1943 1944 /* Bad... Do not break raw sockets */ 1945 sk = skb_to_full_sk(skb); 1946 if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT && 1947 af == AF_INET)) { 1948 1949 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag) 1950 return NF_ACCEPT; 1951 } 1952 1953 #ifdef CONFIG_IP_VS_IPV6 1954 if (af == AF_INET6) { 1955 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) { 1956 int related; 1957 int verdict = ip_vs_in_icmp_v6(ipvs, skb, &related, 1958 hooknum, &iph); 1959 1960 if (related) 1961 return verdict; 1962 } 1963 } else 1964 #endif 1965 if (unlikely(iph.protocol == IPPROTO_ICMP)) { 1966 int related; 1967 int verdict = ip_vs_in_icmp(ipvs, skb, &related, 1968 hooknum); 1969 1970 if (related) 1971 return verdict; 1972 } 1973 1974 /* Protocol supported? */ 1975 pd = ip_vs_proto_data_get(ipvs, iph.protocol); 1976 if (unlikely(!pd)) { 1977 /* The only way we'll see this packet again is if it's 1978 * encapsulated, so mark it with ipvs_property=1 so we 1979 * skip it if we're ignoring tunneled packets 1980 */ 1981 if (sysctl_ignore_tunneled(ipvs)) 1982 skb->ipvs_property = 1; 1983 1984 return NF_ACCEPT; 1985 } 1986 pp = pd->pp; 1987 /* 1988 * Check if the packet belongs to an existing connection entry 1989 */ 1990 cp = INDIRECT_CALL_1(pp->conn_in_get, ip_vs_conn_in_get_proto, 1991 ipvs, af, skb, &iph); 1992 1993 if (!iph.fragoffs && is_new_conn(skb, &iph) && cp) { 1994 int conn_reuse_mode = sysctl_conn_reuse_mode(ipvs); 1995 bool old_ct = false, resched = false; 1996 1997 if (unlikely(sysctl_expire_nodest_conn(ipvs)) && cp->dest && 1998 unlikely(!atomic_read(&cp->dest->weight))) { 1999 resched = true; 2000 old_ct = ip_vs_conn_uses_old_conntrack(cp, skb); 2001 } else if (conn_reuse_mode && 2002 is_new_conn_expected(cp, conn_reuse_mode)) { 2003 old_ct = ip_vs_conn_uses_old_conntrack(cp, skb); 2004 if (!atomic_read(&cp->n_control)) { 2005 resched = true; 2006 } else { 2007 /* Do not reschedule controlling connection 2008 * that uses conntrack while it is still 2009 * referenced by controlled connection(s). 2010 */ 2011 resched = !old_ct; 2012 } 2013 } 2014 2015 if (resched) { 2016 if (!old_ct) 2017 cp->flags &= ~IP_VS_CONN_F_NFCT; 2018 if (!atomic_read(&cp->n_control)) 2019 ip_vs_conn_expire_now(cp); 2020 __ip_vs_conn_put(cp); 2021 if (old_ct) 2022 return NF_DROP; 2023 cp = NULL; 2024 } 2025 } 2026 2027 /* Check the server status */ 2028 if (cp && cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) { 2029 /* the destination server is not available */ 2030 if (sysctl_expire_nodest_conn(ipvs)) { 2031 bool old_ct = ip_vs_conn_uses_old_conntrack(cp, skb); 2032 2033 if (!old_ct) 2034 cp->flags &= ~IP_VS_CONN_F_NFCT; 2035 2036 ip_vs_conn_expire_now(cp); 2037 __ip_vs_conn_put(cp); 2038 if (old_ct) 2039 return NF_DROP; 2040 cp = NULL; 2041 } else { 2042 __ip_vs_conn_put(cp); 2043 return NF_DROP; 2044 } 2045 } 2046 2047 if (unlikely(!cp)) { 2048 int v; 2049 2050 if (!ip_vs_try_to_schedule(ipvs, af, skb, pd, &v, &cp, &iph)) 2051 return v; 2052 } 2053 2054 IP_VS_DBG_PKT(11, af, pp, skb, iph.off, "Incoming packet"); 2055 2056 ip_vs_in_stats(cp, skb); 2057 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd); 2058 if (cp->packet_xmit) 2059 ret = cp->packet_xmit(skb, cp, pp, &iph); 2060 /* do not touch skb anymore */ 2061 else { 2062 IP_VS_DBG_RL("warning: packet_xmit is null"); 2063 ret = NF_ACCEPT; 2064 } 2065 2066 /* Increase its packet counter and check if it is needed 2067 * to be synchronized 2068 * 2069 * Sync connection if it is about to close to 2070 * encorage the standby servers to update the connections timeout 2071 * 2072 * For ONE_PKT let ip_vs_sync_conn() do the filter work. 2073 */ 2074 2075 if (cp->flags & IP_VS_CONN_F_ONE_PACKET) 2076 pkts = sysctl_sync_threshold(ipvs); 2077 else 2078 pkts = atomic_inc_return(&cp->in_pkts); 2079 2080 if (ipvs->sync_state & IP_VS_STATE_MASTER) 2081 ip_vs_sync_conn(ipvs, cp, pkts); 2082 else if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) && cp->control) 2083 /* increment is done inside ip_vs_sync_conn too */ 2084 atomic_inc(&cp->control->in_pkts); 2085 2086 ip_vs_conn_put(cp); 2087 return ret; 2088 } 2089 2090 /* 2091 * It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP 2092 * related packets destined for 0.0.0.0/0. 2093 * When fwmark-based virtual service is used, such as transparent 2094 * cache cluster, TCP packets can be marked and routed to ip_vs_in, 2095 * but ICMP destined for 0.0.0.0/0 cannot not be easily marked and 2096 * sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain 2097 * and send them to ip_vs_in_icmp. 2098 */ 2099 static unsigned int 2100 ip_vs_forward_icmp(void *priv, struct sk_buff *skb, 2101 const struct nf_hook_state *state) 2102 { 2103 struct netns_ipvs *ipvs = net_ipvs(state->net); 2104 int r; 2105 2106 /* ipvs enabled in this netns ? */ 2107 if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable)) 2108 return NF_ACCEPT; 2109 2110 if (state->pf == NFPROTO_IPV4) { 2111 if (ip_hdr(skb)->protocol != IPPROTO_ICMP) 2112 return NF_ACCEPT; 2113 #ifdef CONFIG_IP_VS_IPV6 2114 } else { 2115 struct ip_vs_iphdr iphdr; 2116 2117 ip_vs_fill_iph_skb(AF_INET6, skb, false, &iphdr); 2118 2119 if (iphdr.protocol != IPPROTO_ICMPV6) 2120 return NF_ACCEPT; 2121 2122 return ip_vs_in_icmp_v6(ipvs, skb, &r, state->hook, &iphdr); 2123 #endif 2124 } 2125 2126 return ip_vs_in_icmp(ipvs, skb, &r, state->hook); 2127 } 2128 2129 static const struct nf_hook_ops ip_vs_ops4[] = { 2130 /* After packet filtering, change source only for VS/NAT */ 2131 { 2132 .hook = ip_vs_out_hook, 2133 .pf = NFPROTO_IPV4, 2134 .hooknum = NF_INET_LOCAL_IN, 2135 .priority = NF_IP_PRI_NAT_SRC - 2, 2136 }, 2137 /* After packet filtering, forward packet through VS/DR, VS/TUN, 2138 * or VS/NAT(change destination), so that filtering rules can be 2139 * applied to IPVS. */ 2140 { 2141 .hook = ip_vs_in_hook, 2142 .pf = NFPROTO_IPV4, 2143 .hooknum = NF_INET_LOCAL_IN, 2144 .priority = NF_IP_PRI_NAT_SRC - 1, 2145 }, 2146 /* Before ip_vs_in, change source only for VS/NAT */ 2147 { 2148 .hook = ip_vs_out_hook, 2149 .pf = NFPROTO_IPV4, 2150 .hooknum = NF_INET_LOCAL_OUT, 2151 .priority = NF_IP_PRI_NAT_DST + 1, 2152 }, 2153 /* After mangle, schedule and forward local requests */ 2154 { 2155 .hook = ip_vs_in_hook, 2156 .pf = NFPROTO_IPV4, 2157 .hooknum = NF_INET_LOCAL_OUT, 2158 .priority = NF_IP_PRI_NAT_DST + 2, 2159 }, 2160 /* After packet filtering (but before ip_vs_out_icmp), catch icmp 2161 * destined for 0.0.0.0/0, which is for incoming IPVS connections */ 2162 { 2163 .hook = ip_vs_forward_icmp, 2164 .pf = NFPROTO_IPV4, 2165 .hooknum = NF_INET_FORWARD, 2166 .priority = 99, 2167 }, 2168 /* After packet filtering, change source only for VS/NAT */ 2169 { 2170 .hook = ip_vs_out_hook, 2171 .pf = NFPROTO_IPV4, 2172 .hooknum = NF_INET_FORWARD, 2173 .priority = 100, 2174 }, 2175 }; 2176 2177 #ifdef CONFIG_IP_VS_IPV6 2178 static const struct nf_hook_ops ip_vs_ops6[] = { 2179 /* After packet filtering, change source only for VS/NAT */ 2180 { 2181 .hook = ip_vs_out_hook, 2182 .pf = NFPROTO_IPV6, 2183 .hooknum = NF_INET_LOCAL_IN, 2184 .priority = NF_IP6_PRI_NAT_SRC - 2, 2185 }, 2186 /* After packet filtering, forward packet through VS/DR, VS/TUN, 2187 * or VS/NAT(change destination), so that filtering rules can be 2188 * applied to IPVS. */ 2189 { 2190 .hook = ip_vs_in_hook, 2191 .pf = NFPROTO_IPV6, 2192 .hooknum = NF_INET_LOCAL_IN, 2193 .priority = NF_IP6_PRI_NAT_SRC - 1, 2194 }, 2195 /* Before ip_vs_in, change source only for VS/NAT */ 2196 { 2197 .hook = ip_vs_out_hook, 2198 .pf = NFPROTO_IPV6, 2199 .hooknum = NF_INET_LOCAL_OUT, 2200 .priority = NF_IP6_PRI_NAT_DST + 1, 2201 }, 2202 /* After mangle, schedule and forward local requests */ 2203 { 2204 .hook = ip_vs_in_hook, 2205 .pf = NFPROTO_IPV6, 2206 .hooknum = NF_INET_LOCAL_OUT, 2207 .priority = NF_IP6_PRI_NAT_DST + 2, 2208 }, 2209 /* After packet filtering (but before ip_vs_out_icmp), catch icmp 2210 * destined for 0.0.0.0/0, which is for incoming IPVS connections */ 2211 { 2212 .hook = ip_vs_forward_icmp, 2213 .pf = NFPROTO_IPV6, 2214 .hooknum = NF_INET_FORWARD, 2215 .priority = 99, 2216 }, 2217 /* After packet filtering, change source only for VS/NAT */ 2218 { 2219 .hook = ip_vs_out_hook, 2220 .pf = NFPROTO_IPV6, 2221 .hooknum = NF_INET_FORWARD, 2222 .priority = 100, 2223 }, 2224 }; 2225 #endif 2226 2227 int ip_vs_register_hooks(struct netns_ipvs *ipvs, unsigned int af) 2228 { 2229 const struct nf_hook_ops *ops; 2230 unsigned int count; 2231 unsigned int afmask; 2232 int ret = 0; 2233 2234 if (af == AF_INET6) { 2235 #ifdef CONFIG_IP_VS_IPV6 2236 ops = ip_vs_ops6; 2237 count = ARRAY_SIZE(ip_vs_ops6); 2238 afmask = 2; 2239 #else 2240 return -EINVAL; 2241 #endif 2242 } else { 2243 ops = ip_vs_ops4; 2244 count = ARRAY_SIZE(ip_vs_ops4); 2245 afmask = 1; 2246 } 2247 2248 if (!(ipvs->hooks_afmask & afmask)) { 2249 ret = nf_register_net_hooks(ipvs->net, ops, count); 2250 if (ret >= 0) 2251 ipvs->hooks_afmask |= afmask; 2252 } 2253 return ret; 2254 } 2255 2256 void ip_vs_unregister_hooks(struct netns_ipvs *ipvs, unsigned int af) 2257 { 2258 const struct nf_hook_ops *ops; 2259 unsigned int count; 2260 unsigned int afmask; 2261 2262 if (af == AF_INET6) { 2263 #ifdef CONFIG_IP_VS_IPV6 2264 ops = ip_vs_ops6; 2265 count = ARRAY_SIZE(ip_vs_ops6); 2266 afmask = 2; 2267 #else 2268 return; 2269 #endif 2270 } else { 2271 ops = ip_vs_ops4; 2272 count = ARRAY_SIZE(ip_vs_ops4); 2273 afmask = 1; 2274 } 2275 2276 if (ipvs->hooks_afmask & afmask) { 2277 nf_unregister_net_hooks(ipvs->net, ops, count); 2278 ipvs->hooks_afmask &= ~afmask; 2279 } 2280 } 2281 2282 /* 2283 * Initialize IP Virtual Server netns mem. 2284 */ 2285 static int __net_init __ip_vs_init(struct net *net) 2286 { 2287 struct netns_ipvs *ipvs; 2288 2289 ipvs = net_generic(net, ip_vs_net_id); 2290 if (ipvs == NULL) 2291 return -ENOMEM; 2292 2293 /* Hold the beast until a service is registered */ 2294 ipvs->enable = 0; 2295 ipvs->net = net; 2296 /* Counters used for creating unique names */ 2297 ipvs->gen = atomic_read(&ipvs_netns_cnt); 2298 atomic_inc(&ipvs_netns_cnt); 2299 net->ipvs = ipvs; 2300 2301 if (ip_vs_estimator_net_init(ipvs) < 0) 2302 goto estimator_fail; 2303 2304 if (ip_vs_control_net_init(ipvs) < 0) 2305 goto control_fail; 2306 2307 if (ip_vs_protocol_net_init(ipvs) < 0) 2308 goto protocol_fail; 2309 2310 if (ip_vs_app_net_init(ipvs) < 0) 2311 goto app_fail; 2312 2313 if (ip_vs_conn_net_init(ipvs) < 0) 2314 goto conn_fail; 2315 2316 if (ip_vs_sync_net_init(ipvs) < 0) 2317 goto sync_fail; 2318 2319 return 0; 2320 /* 2321 * Error handling 2322 */ 2323 2324 sync_fail: 2325 ip_vs_conn_net_cleanup(ipvs); 2326 conn_fail: 2327 ip_vs_app_net_cleanup(ipvs); 2328 app_fail: 2329 ip_vs_protocol_net_cleanup(ipvs); 2330 protocol_fail: 2331 ip_vs_control_net_cleanup(ipvs); 2332 control_fail: 2333 ip_vs_estimator_net_cleanup(ipvs); 2334 estimator_fail: 2335 net->ipvs = NULL; 2336 return -ENOMEM; 2337 } 2338 2339 static void __net_exit __ip_vs_cleanup_batch(struct list_head *net_list) 2340 { 2341 struct netns_ipvs *ipvs; 2342 struct net *net; 2343 2344 ip_vs_service_nets_cleanup(net_list); /* ip_vs_flush() with locks */ 2345 list_for_each_entry(net, net_list, exit_list) { 2346 ipvs = net_ipvs(net); 2347 ip_vs_conn_net_cleanup(ipvs); 2348 ip_vs_app_net_cleanup(ipvs); 2349 ip_vs_protocol_net_cleanup(ipvs); 2350 ip_vs_control_net_cleanup(ipvs); 2351 ip_vs_estimator_net_cleanup(ipvs); 2352 IP_VS_DBG(2, "ipvs netns %d released\n", ipvs->gen); 2353 net->ipvs = NULL; 2354 } 2355 } 2356 2357 static void __net_exit __ip_vs_dev_cleanup_batch(struct list_head *net_list) 2358 { 2359 struct netns_ipvs *ipvs; 2360 struct net *net; 2361 2362 list_for_each_entry(net, net_list, exit_list) { 2363 ipvs = net_ipvs(net); 2364 ip_vs_unregister_hooks(ipvs, AF_INET); 2365 ip_vs_unregister_hooks(ipvs, AF_INET6); 2366 ipvs->enable = 0; /* Disable packet reception */ 2367 smp_wmb(); 2368 ip_vs_sync_net_cleanup(ipvs); 2369 } 2370 } 2371 2372 static struct pernet_operations ipvs_core_ops = { 2373 .init = __ip_vs_init, 2374 .exit_batch = __ip_vs_cleanup_batch, 2375 .id = &ip_vs_net_id, 2376 .size = sizeof(struct netns_ipvs), 2377 }; 2378 2379 static struct pernet_operations ipvs_core_dev_ops = { 2380 .exit_batch = __ip_vs_dev_cleanup_batch, 2381 }; 2382 2383 /* 2384 * Initialize IP Virtual Server 2385 */ 2386 static int __init ip_vs_init(void) 2387 { 2388 int ret; 2389 2390 ret = ip_vs_control_init(); 2391 if (ret < 0) { 2392 pr_err("can't setup control.\n"); 2393 goto exit; 2394 } 2395 2396 ip_vs_protocol_init(); 2397 2398 ret = ip_vs_conn_init(); 2399 if (ret < 0) { 2400 pr_err("can't setup connection table.\n"); 2401 goto cleanup_protocol; 2402 } 2403 2404 ret = register_pernet_subsys(&ipvs_core_ops); /* Alloc ip_vs struct */ 2405 if (ret < 0) 2406 goto cleanup_conn; 2407 2408 ret = register_pernet_device(&ipvs_core_dev_ops); 2409 if (ret < 0) 2410 goto cleanup_sub; 2411 2412 ret = ip_vs_register_nl_ioctl(); 2413 if (ret < 0) { 2414 pr_err("can't register netlink/ioctl.\n"); 2415 goto cleanup_dev; 2416 } 2417 2418 pr_info("ipvs loaded.\n"); 2419 2420 return ret; 2421 2422 cleanup_dev: 2423 unregister_pernet_device(&ipvs_core_dev_ops); 2424 cleanup_sub: 2425 unregister_pernet_subsys(&ipvs_core_ops); 2426 cleanup_conn: 2427 ip_vs_conn_cleanup(); 2428 cleanup_protocol: 2429 ip_vs_protocol_cleanup(); 2430 ip_vs_control_cleanup(); 2431 exit: 2432 return ret; 2433 } 2434 2435 static void __exit ip_vs_cleanup(void) 2436 { 2437 ip_vs_unregister_nl_ioctl(); 2438 unregister_pernet_device(&ipvs_core_dev_ops); 2439 unregister_pernet_subsys(&ipvs_core_ops); /* free ip_vs struct */ 2440 ip_vs_conn_cleanup(); 2441 ip_vs_protocol_cleanup(); 2442 ip_vs_control_cleanup(); 2443 /* common rcu_barrier() used by: 2444 * - ip_vs_control_cleanup() 2445 */ 2446 rcu_barrier(); 2447 pr_info("ipvs unloaded.\n"); 2448 } 2449 2450 module_init(ip_vs_init); 2451 module_exit(ip_vs_cleanup); 2452 MODULE_LICENSE("GPL"); 2453