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 s->cnt.inpkts++; 136 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 s->cnt.inpkts++; 143 s->cnt.inbytes += skb->len; 144 u64_stats_update_end(&s->syncp); 145 146 s = this_cpu_ptr(ipvs->tot_stats.cpustats); 147 u64_stats_update_begin(&s->syncp); 148 s->cnt.inpkts++; 149 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 s->cnt.outpkts++; 172 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 s->cnt.outpkts++; 179 s->cnt.outbytes += skb->len; 180 u64_stats_update_end(&s->syncp); 181 182 s = this_cpu_ptr(ipvs->tot_stats.cpustats); 183 u64_stats_update_begin(&s->syncp); 184 s->cnt.outpkts++; 185 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 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 s->cnt.conns++; 209 u64_stats_update_end(&s->syncp); 210 211 s = this_cpu_ptr(ipvs->tot_stats.cpustats); 212 u64_stats_update_begin(&s->syncp); 213 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 EnterFunction(12); 1144 vaddr = &svc->addr; 1145 vport = svc->port; 1146 daddr = &iph->saddr; 1147 caddr = &iph->daddr; 1148 1149 /* check pre-requisites are satisfied */ 1150 if (svc->fwmark) 1151 return NULL; 1152 if (!vport || !dport) 1153 return NULL; 1154 1155 /* for persistent service first create connection template */ 1156 if (svc->flags & IP_VS_SVC_F_PERSISTENT) { 1157 /* apply netmask the same way ingress-side does */ 1158 #ifdef CONFIG_IP_VS_IPV6 1159 if (svc->af == AF_INET6) 1160 ipv6_addr_prefix(&snet.in6, &caddr->in6, 1161 (__force __u32)svc->netmask); 1162 else 1163 #endif 1164 snet.ip = caddr->ip & svc->netmask; 1165 /* fill params and create template if not existent */ 1166 if (ip_vs_conn_fill_param_persist(svc, skb, iph->protocol, 1167 &snet, 0, vaddr, 1168 vport, ¶m) < 0) 1169 return NULL; 1170 ct = ip_vs_ct_in_get(¶m); 1171 /* check if template exists and points to the same dest */ 1172 if (!ct || !ip_vs_check_template(ct, dest)) { 1173 ct = ip_vs_conn_new(¶m, dest->af, daddr, dport, 1174 IP_VS_CONN_F_TEMPLATE, dest, 0); 1175 if (!ct) { 1176 kfree(param.pe_data); 1177 return NULL; 1178 } 1179 ct->timeout = svc->timeout; 1180 } else { 1181 kfree(param.pe_data); 1182 } 1183 } 1184 1185 /* connection flags */ 1186 flags = ((svc->flags & IP_VS_SVC_F_ONEPACKET) && 1187 iph->protocol == IPPROTO_UDP) ? IP_VS_CONN_F_ONE_PACKET : 0; 1188 /* create connection */ 1189 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol, 1190 caddr, cport, vaddr, vport, ¶m); 1191 cp = ip_vs_conn_new(¶m, dest->af, daddr, dport, flags, dest, 0); 1192 if (!cp) { 1193 if (ct) 1194 ip_vs_conn_put(ct); 1195 return NULL; 1196 } 1197 if (ct) { 1198 ip_vs_control_add(cp, ct); 1199 ip_vs_conn_put(ct); 1200 } 1201 ip_vs_conn_stats(cp, svc); 1202 1203 /* return connection (will be used to handle outgoing packet) */ 1204 IP_VS_DBG_BUF(6, "New connection RS-initiated:%c c:%s:%u v:%s:%u " 1205 "d:%s:%u conn->flags:%X conn->refcnt:%d\n", 1206 ip_vs_fwd_tag(cp), 1207 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport), 1208 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport), 1209 IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport), 1210 cp->flags, refcount_read(&cp->refcnt)); 1211 LeaveFunction(12); 1212 return cp; 1213 } 1214 1215 /* Handle outgoing packets which are considered requests initiated by 1216 * real servers, so that subsequent responses from external client can be 1217 * routed to the right real server. 1218 * Used also for outgoing responses in OPS mode. 1219 * 1220 * Connection management is handled by persistent-engine specific callback. 1221 */ 1222 static struct ip_vs_conn *__ip_vs_rs_conn_out(unsigned int hooknum, 1223 struct netns_ipvs *ipvs, 1224 int af, struct sk_buff *skb, 1225 const struct ip_vs_iphdr *iph) 1226 { 1227 struct ip_vs_dest *dest; 1228 struct ip_vs_conn *cp = NULL; 1229 __be16 _ports[2], *pptr; 1230 1231 if (hooknum == NF_INET_LOCAL_IN) 1232 return NULL; 1233 1234 pptr = frag_safe_skb_hp(skb, iph->len, 1235 sizeof(_ports), _ports); 1236 if (!pptr) 1237 return NULL; 1238 1239 dest = ip_vs_find_real_service(ipvs, af, iph->protocol, 1240 &iph->saddr, pptr[0]); 1241 if (dest) { 1242 struct ip_vs_service *svc; 1243 struct ip_vs_pe *pe; 1244 1245 svc = rcu_dereference(dest->svc); 1246 if (svc) { 1247 pe = rcu_dereference(svc->pe); 1248 if (pe && pe->conn_out) 1249 cp = pe->conn_out(svc, dest, skb, iph, 1250 pptr[0], pptr[1]); 1251 } 1252 } 1253 1254 return cp; 1255 } 1256 1257 /* Handle response packets: rewrite addresses and send away... 1258 */ 1259 static unsigned int 1260 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd, 1261 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph, 1262 unsigned int hooknum) 1263 { 1264 struct ip_vs_protocol *pp = pd->pp; 1265 1266 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) 1267 goto after_nat; 1268 1269 IP_VS_DBG_PKT(11, af, pp, skb, iph->off, "Outgoing packet"); 1270 1271 if (skb_ensure_writable(skb, iph->len)) 1272 goto drop; 1273 1274 /* mangle the packet */ 1275 if (pp->snat_handler && 1276 !SNAT_CALL(pp->snat_handler, skb, pp, cp, iph)) 1277 goto drop; 1278 1279 #ifdef CONFIG_IP_VS_IPV6 1280 if (af == AF_INET6) 1281 ipv6_hdr(skb)->saddr = cp->vaddr.in6; 1282 else 1283 #endif 1284 { 1285 ip_hdr(skb)->saddr = cp->vaddr.ip; 1286 ip_send_check(ip_hdr(skb)); 1287 } 1288 1289 /* 1290 * nf_iterate does not expect change in the skb->dst->dev. 1291 * It looks like it is not fatal to enable this code for hooks 1292 * where our handlers are at the end of the chain list and 1293 * when all next handlers use skb->dst->dev and not outdev. 1294 * It will definitely route properly the inout NAT traffic 1295 * when multiple paths are used. 1296 */ 1297 1298 /* For policy routing, packets originating from this 1299 * machine itself may be routed differently to packets 1300 * passing through. We want this packet to be routed as 1301 * if it came from this machine itself. So re-compute 1302 * the routing information. 1303 */ 1304 if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum)) 1305 goto drop; 1306 1307 IP_VS_DBG_PKT(10, af, pp, skb, iph->off, "After SNAT"); 1308 1309 after_nat: 1310 ip_vs_out_stats(cp, skb); 1311 ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd); 1312 skb->ipvs_property = 1; 1313 if (!(cp->flags & IP_VS_CONN_F_NFCT)) 1314 ip_vs_notrack(skb); 1315 else 1316 ip_vs_update_conntrack(skb, cp, 0); 1317 ip_vs_conn_put(cp); 1318 1319 LeaveFunction(11); 1320 return NF_ACCEPT; 1321 1322 drop: 1323 ip_vs_conn_put(cp); 1324 kfree_skb(skb); 1325 LeaveFunction(11); 1326 return NF_STOLEN; 1327 } 1328 1329 /* 1330 * Check if outgoing packet belongs to the established ip_vs_conn. 1331 */ 1332 static unsigned int 1333 ip_vs_out_hook(void *priv, struct sk_buff *skb, const struct nf_hook_state *state) 1334 { 1335 struct netns_ipvs *ipvs = net_ipvs(state->net); 1336 unsigned int hooknum = state->hook; 1337 struct ip_vs_iphdr iph; 1338 struct ip_vs_protocol *pp; 1339 struct ip_vs_proto_data *pd; 1340 struct ip_vs_conn *cp; 1341 int af = state->pf; 1342 struct sock *sk; 1343 1344 EnterFunction(11); 1345 1346 /* Already marked as IPVS request or reply? */ 1347 if (skb->ipvs_property) 1348 return NF_ACCEPT; 1349 1350 sk = skb_to_full_sk(skb); 1351 /* Bad... Do not break raw sockets */ 1352 if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT && 1353 af == AF_INET)) { 1354 1355 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag) 1356 return NF_ACCEPT; 1357 } 1358 1359 if (unlikely(!skb_dst(skb))) 1360 return NF_ACCEPT; 1361 1362 if (!ipvs->enable) 1363 return NF_ACCEPT; 1364 1365 ip_vs_fill_iph_skb(af, skb, false, &iph); 1366 #ifdef CONFIG_IP_VS_IPV6 1367 if (af == AF_INET6) { 1368 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) { 1369 int related; 1370 int verdict = ip_vs_out_icmp_v6(ipvs, skb, &related, 1371 hooknum, &iph); 1372 1373 if (related) 1374 return verdict; 1375 } 1376 } else 1377 #endif 1378 if (unlikely(iph.protocol == IPPROTO_ICMP)) { 1379 int related; 1380 int verdict = ip_vs_out_icmp(ipvs, skb, &related, hooknum); 1381 1382 if (related) 1383 return verdict; 1384 } 1385 1386 pd = ip_vs_proto_data_get(ipvs, iph.protocol); 1387 if (unlikely(!pd)) 1388 return NF_ACCEPT; 1389 pp = pd->pp; 1390 1391 /* reassemble IP fragments */ 1392 #ifdef CONFIG_IP_VS_IPV6 1393 if (af == AF_INET) 1394 #endif 1395 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) { 1396 if (ip_vs_gather_frags(ipvs, skb, 1397 ip_vs_defrag_user(hooknum))) 1398 return NF_STOLEN; 1399 1400 ip_vs_fill_iph_skb(AF_INET, skb, false, &iph); 1401 } 1402 1403 /* 1404 * Check if the packet belongs to an existing entry 1405 */ 1406 cp = INDIRECT_CALL_1(pp->conn_out_get, ip_vs_conn_out_get_proto, 1407 ipvs, af, skb, &iph); 1408 1409 if (likely(cp)) 1410 return handle_response(af, skb, pd, cp, &iph, hooknum); 1411 1412 /* Check for real-server-started requests */ 1413 if (atomic_read(&ipvs->conn_out_counter)) { 1414 /* Currently only for UDP: 1415 * connection oriented protocols typically use 1416 * ephemeral ports for outgoing connections, so 1417 * related incoming responses would not match any VS 1418 */ 1419 if (pp->protocol == IPPROTO_UDP) { 1420 cp = __ip_vs_rs_conn_out(hooknum, ipvs, af, skb, &iph); 1421 if (likely(cp)) 1422 return handle_response(af, skb, pd, cp, &iph, 1423 hooknum); 1424 } 1425 } 1426 1427 if (sysctl_nat_icmp_send(ipvs) && 1428 (pp->protocol == IPPROTO_TCP || 1429 pp->protocol == IPPROTO_UDP || 1430 pp->protocol == IPPROTO_SCTP)) { 1431 __be16 _ports[2], *pptr; 1432 1433 pptr = frag_safe_skb_hp(skb, iph.len, 1434 sizeof(_ports), _ports); 1435 if (pptr == NULL) 1436 return NF_ACCEPT; /* Not for me */ 1437 if (ip_vs_has_real_service(ipvs, af, iph.protocol, &iph.saddr, 1438 pptr[0])) { 1439 /* 1440 * Notify the real server: there is no 1441 * existing entry if it is not RST 1442 * packet or not TCP packet. 1443 */ 1444 if ((iph.protocol != IPPROTO_TCP && 1445 iph.protocol != IPPROTO_SCTP) 1446 || ((iph.protocol == IPPROTO_TCP 1447 && !is_tcp_reset(skb, iph.len)) 1448 || (iph.protocol == IPPROTO_SCTP 1449 && !is_sctp_abort(skb, 1450 iph.len)))) { 1451 #ifdef CONFIG_IP_VS_IPV6 1452 if (af == AF_INET6) { 1453 if (!skb->dev) 1454 skb->dev = ipvs->net->loopback_dev; 1455 icmpv6_send(skb, 1456 ICMPV6_DEST_UNREACH, 1457 ICMPV6_PORT_UNREACH, 1458 0); 1459 } else 1460 #endif 1461 icmp_send(skb, 1462 ICMP_DEST_UNREACH, 1463 ICMP_PORT_UNREACH, 0); 1464 return NF_DROP; 1465 } 1466 } 1467 } 1468 1469 IP_VS_DBG_PKT(12, af, pp, skb, iph.off, 1470 "ip_vs_out: packet continues traversal as normal"); 1471 return NF_ACCEPT; 1472 } 1473 1474 static unsigned int 1475 ip_vs_try_to_schedule(struct netns_ipvs *ipvs, int af, struct sk_buff *skb, 1476 struct ip_vs_proto_data *pd, 1477 int *verdict, struct ip_vs_conn **cpp, 1478 struct ip_vs_iphdr *iph) 1479 { 1480 struct ip_vs_protocol *pp = pd->pp; 1481 1482 if (!iph->fragoffs) { 1483 /* No (second) fragments need to enter here, as nf_defrag_ipv6 1484 * replayed fragment zero will already have created the cp 1485 */ 1486 1487 /* Schedule and create new connection entry into cpp */ 1488 if (!pp->conn_schedule(ipvs, af, skb, pd, verdict, cpp, iph)) 1489 return 0; 1490 } 1491 1492 if (unlikely(!*cpp)) { 1493 /* sorry, all this trouble for a no-hit :) */ 1494 IP_VS_DBG_PKT(12, af, pp, skb, iph->off, 1495 "ip_vs_in: packet continues traversal as normal"); 1496 1497 /* Fragment couldn't be mapped to a conn entry */ 1498 if (iph->fragoffs) 1499 IP_VS_DBG_PKT(7, af, pp, skb, iph->off, 1500 "unhandled fragment"); 1501 1502 *verdict = NF_ACCEPT; 1503 return 0; 1504 } 1505 1506 return 1; 1507 } 1508 1509 /* Check the UDP tunnel and return its header length */ 1510 static int ipvs_udp_decap(struct netns_ipvs *ipvs, struct sk_buff *skb, 1511 unsigned int offset, __u16 af, 1512 const union nf_inet_addr *daddr, __u8 *proto) 1513 { 1514 struct udphdr _udph, *udph; 1515 struct ip_vs_dest *dest; 1516 1517 udph = skb_header_pointer(skb, offset, sizeof(_udph), &_udph); 1518 if (!udph) 1519 goto unk; 1520 offset += sizeof(struct udphdr); 1521 dest = ip_vs_find_tunnel(ipvs, af, daddr, udph->dest); 1522 if (!dest) 1523 goto unk; 1524 if (dest->tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GUE) { 1525 struct guehdr _gueh, *gueh; 1526 1527 gueh = skb_header_pointer(skb, offset, sizeof(_gueh), &_gueh); 1528 if (!gueh) 1529 goto unk; 1530 if (gueh->control != 0 || gueh->version != 0) 1531 goto unk; 1532 /* Later we can support also IPPROTO_IPV6 */ 1533 if (gueh->proto_ctype != IPPROTO_IPIP) 1534 goto unk; 1535 *proto = gueh->proto_ctype; 1536 return sizeof(struct udphdr) + sizeof(struct guehdr) + 1537 (gueh->hlen << 2); 1538 } 1539 1540 unk: 1541 return 0; 1542 } 1543 1544 /* Check the GRE tunnel and return its header length */ 1545 static int ipvs_gre_decap(struct netns_ipvs *ipvs, struct sk_buff *skb, 1546 unsigned int offset, __u16 af, 1547 const union nf_inet_addr *daddr, __u8 *proto) 1548 { 1549 struct gre_base_hdr _greh, *greh; 1550 struct ip_vs_dest *dest; 1551 1552 greh = skb_header_pointer(skb, offset, sizeof(_greh), &_greh); 1553 if (!greh) 1554 goto unk; 1555 dest = ip_vs_find_tunnel(ipvs, af, daddr, 0); 1556 if (!dest) 1557 goto unk; 1558 if (dest->tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GRE) { 1559 __be16 type; 1560 1561 /* Only support version 0 and C (csum) */ 1562 if ((greh->flags & ~GRE_CSUM) != 0) 1563 goto unk; 1564 type = greh->protocol; 1565 /* Later we can support also IPPROTO_IPV6 */ 1566 if (type != htons(ETH_P_IP)) 1567 goto unk; 1568 *proto = IPPROTO_IPIP; 1569 return gre_calc_hlen(gre_flags_to_tnl_flags(greh->flags)); 1570 } 1571 1572 unk: 1573 return 0; 1574 } 1575 1576 /* 1577 * Handle ICMP messages in the outside-to-inside direction (incoming). 1578 * Find any that might be relevant, check against existing connections, 1579 * forward to the right destination host if relevant. 1580 * Currently handles error types - unreachable, quench, ttl exceeded. 1581 */ 1582 static int 1583 ip_vs_in_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb, int *related, 1584 unsigned int hooknum) 1585 { 1586 struct iphdr *iph; 1587 struct icmphdr _icmph, *ic; 1588 struct iphdr _ciph, *cih; /* The ip header contained within the ICMP */ 1589 struct ip_vs_iphdr ciph; 1590 struct ip_vs_conn *cp; 1591 struct ip_vs_protocol *pp; 1592 struct ip_vs_proto_data *pd; 1593 unsigned int offset, offset2, ihl, verdict; 1594 bool tunnel, new_cp = false; 1595 union nf_inet_addr *raddr; 1596 char *outer_proto = "IPIP"; 1597 1598 *related = 1; 1599 1600 /* reassemble IP fragments */ 1601 if (ip_is_fragment(ip_hdr(skb))) { 1602 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum))) 1603 return NF_STOLEN; 1604 } 1605 1606 iph = ip_hdr(skb); 1607 offset = ihl = iph->ihl * 4; 1608 ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph); 1609 if (ic == NULL) 1610 return NF_DROP; 1611 1612 IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n", 1613 ic->type, ntohs(icmp_id(ic)), 1614 &iph->saddr, &iph->daddr); 1615 1616 /* 1617 * Work through seeing if this is for us. 1618 * These checks are supposed to be in an order that means easy 1619 * things are checked first to speed up processing.... however 1620 * this means that some packets will manage to get a long way 1621 * down this stack and then be rejected, but that's life. 1622 */ 1623 if ((ic->type != ICMP_DEST_UNREACH) && 1624 (ic->type != ICMP_SOURCE_QUENCH) && 1625 (ic->type != ICMP_TIME_EXCEEDED)) { 1626 *related = 0; 1627 return NF_ACCEPT; 1628 } 1629 1630 /* Now find the contained IP header */ 1631 offset += sizeof(_icmph); 1632 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph); 1633 if (cih == NULL) 1634 return NF_ACCEPT; /* The packet looks wrong, ignore */ 1635 raddr = (union nf_inet_addr *)&cih->daddr; 1636 1637 /* Special case for errors for IPIP/UDP/GRE tunnel packets */ 1638 tunnel = false; 1639 if (cih->protocol == IPPROTO_IPIP) { 1640 struct ip_vs_dest *dest; 1641 1642 if (unlikely(cih->frag_off & htons(IP_OFFSET))) 1643 return NF_ACCEPT; 1644 /* Error for our IPIP must arrive at LOCAL_IN */ 1645 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL)) 1646 return NF_ACCEPT; 1647 dest = ip_vs_find_tunnel(ipvs, AF_INET, raddr, 0); 1648 /* Only for known tunnel */ 1649 if (!dest || dest->tun_type != IP_VS_CONN_F_TUNNEL_TYPE_IPIP) 1650 return NF_ACCEPT; 1651 offset += cih->ihl * 4; 1652 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph); 1653 if (cih == NULL) 1654 return NF_ACCEPT; /* The packet looks wrong, ignore */ 1655 tunnel = true; 1656 } else if ((cih->protocol == IPPROTO_UDP || /* Can be UDP encap */ 1657 cih->protocol == IPPROTO_GRE) && /* Can be GRE encap */ 1658 /* Error for our tunnel must arrive at LOCAL_IN */ 1659 (skb_rtable(skb)->rt_flags & RTCF_LOCAL)) { 1660 __u8 iproto; 1661 int ulen; 1662 1663 /* Non-first fragment has no UDP/GRE header */ 1664 if (unlikely(cih->frag_off & htons(IP_OFFSET))) 1665 return NF_ACCEPT; 1666 offset2 = offset + cih->ihl * 4; 1667 if (cih->protocol == IPPROTO_UDP) { 1668 ulen = ipvs_udp_decap(ipvs, skb, offset2, AF_INET, 1669 raddr, &iproto); 1670 outer_proto = "UDP"; 1671 } else { 1672 ulen = ipvs_gre_decap(ipvs, skb, offset2, AF_INET, 1673 raddr, &iproto); 1674 outer_proto = "GRE"; 1675 } 1676 if (ulen > 0) { 1677 /* Skip IP and UDP/GRE tunnel headers */ 1678 offset = offset2 + ulen; 1679 /* Now we should be at the original IP header */ 1680 cih = skb_header_pointer(skb, offset, sizeof(_ciph), 1681 &_ciph); 1682 if (cih && cih->version == 4 && cih->ihl >= 5 && 1683 iproto == IPPROTO_IPIP) 1684 tunnel = true; 1685 else 1686 return NF_ACCEPT; 1687 } 1688 } 1689 1690 pd = ip_vs_proto_data_get(ipvs, cih->protocol); 1691 if (!pd) 1692 return NF_ACCEPT; 1693 pp = pd->pp; 1694 1695 /* Is the embedded protocol header present? */ 1696 if (unlikely(cih->frag_off & htons(IP_OFFSET) && 1697 pp->dont_defrag)) 1698 return NF_ACCEPT; 1699 1700 IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset, 1701 "Checking incoming ICMP for"); 1702 1703 offset2 = offset; 1704 ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, !tunnel, &ciph); 1705 offset = ciph.len; 1706 1707 /* The embedded headers contain source and dest in reverse order. 1708 * For IPIP/UDP/GRE tunnel this is error for request, not for reply. 1709 */ 1710 cp = INDIRECT_CALL_1(pp->conn_in_get, ip_vs_conn_in_get_proto, 1711 ipvs, AF_INET, skb, &ciph); 1712 1713 if (!cp) { 1714 int v; 1715 1716 if (tunnel || !sysctl_schedule_icmp(ipvs)) 1717 return NF_ACCEPT; 1718 1719 if (!ip_vs_try_to_schedule(ipvs, AF_INET, skb, pd, &v, &cp, &ciph)) 1720 return v; 1721 new_cp = true; 1722 } 1723 1724 verdict = NF_DROP; 1725 1726 /* Ensure the checksum is correct */ 1727 if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) { 1728 /* Failed checksum! */ 1729 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n", 1730 &iph->saddr); 1731 goto out; 1732 } 1733 1734 if (tunnel) { 1735 __be32 info = ic->un.gateway; 1736 __u8 type = ic->type; 1737 __u8 code = ic->code; 1738 1739 /* Update the MTU */ 1740 if (ic->type == ICMP_DEST_UNREACH && 1741 ic->code == ICMP_FRAG_NEEDED) { 1742 struct ip_vs_dest *dest = cp->dest; 1743 u32 mtu = ntohs(ic->un.frag.mtu); 1744 __be16 frag_off = cih->frag_off; 1745 1746 /* Strip outer IP and ICMP, go to IPIP/UDP/GRE header */ 1747 if (pskb_pull(skb, ihl + sizeof(_icmph)) == NULL) 1748 goto ignore_tunnel; 1749 offset2 -= ihl + sizeof(_icmph); 1750 skb_reset_network_header(skb); 1751 IP_VS_DBG(12, "ICMP for %s %pI4->%pI4: mtu=%u\n", 1752 outer_proto, &ip_hdr(skb)->saddr, 1753 &ip_hdr(skb)->daddr, mtu); 1754 ipv4_update_pmtu(skb, ipvs->net, mtu, 0, 0); 1755 /* Client uses PMTUD? */ 1756 if (!(frag_off & htons(IP_DF))) 1757 goto ignore_tunnel; 1758 /* Prefer the resulting PMTU */ 1759 if (dest) { 1760 struct ip_vs_dest_dst *dest_dst; 1761 1762 dest_dst = rcu_dereference(dest->dest_dst); 1763 if (dest_dst) 1764 mtu = dst_mtu(dest_dst->dst_cache); 1765 } 1766 if (mtu > 68 + sizeof(struct iphdr)) 1767 mtu -= sizeof(struct iphdr); 1768 info = htonl(mtu); 1769 } 1770 /* Strip outer IP, ICMP and IPIP/UDP/GRE, go to IP header of 1771 * original request. 1772 */ 1773 if (pskb_pull(skb, offset2) == NULL) 1774 goto ignore_tunnel; 1775 skb_reset_network_header(skb); 1776 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n", 1777 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, 1778 type, code, ntohl(info)); 1779 icmp_send(skb, type, code, info); 1780 /* ICMP can be shorter but anyways, account it */ 1781 ip_vs_out_stats(cp, skb); 1782 1783 ignore_tunnel: 1784 consume_skb(skb); 1785 verdict = NF_STOLEN; 1786 goto out; 1787 } 1788 1789 /* do the statistics and put it back */ 1790 ip_vs_in_stats(cp, skb); 1791 if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol || 1792 IPPROTO_SCTP == cih->protocol) 1793 offset += 2 * sizeof(__u16); 1794 verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph); 1795 1796 out: 1797 if (likely(!new_cp)) 1798 __ip_vs_conn_put(cp); 1799 else 1800 ip_vs_conn_put(cp); 1801 1802 return verdict; 1803 } 1804 1805 #ifdef CONFIG_IP_VS_IPV6 1806 static int ip_vs_in_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb, 1807 int *related, unsigned int hooknum, 1808 struct ip_vs_iphdr *iph) 1809 { 1810 struct icmp6hdr _icmph, *ic; 1811 struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */ 1812 struct ip_vs_conn *cp; 1813 struct ip_vs_protocol *pp; 1814 struct ip_vs_proto_data *pd; 1815 unsigned int offset, verdict; 1816 bool new_cp = false; 1817 1818 *related = 1; 1819 1820 ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph); 1821 if (ic == NULL) 1822 return NF_DROP; 1823 1824 /* 1825 * Work through seeing if this is for us. 1826 * These checks are supposed to be in an order that means easy 1827 * things are checked first to speed up processing.... however 1828 * this means that some packets will manage to get a long way 1829 * down this stack and then be rejected, but that's life. 1830 */ 1831 if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) { 1832 *related = 0; 1833 return NF_ACCEPT; 1834 } 1835 /* Fragment header that is before ICMP header tells us that: 1836 * it's not an error message since they can't be fragmented. 1837 */ 1838 if (iph->flags & IP6_FH_F_FRAG) 1839 return NF_DROP; 1840 1841 IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n", 1842 ic->icmp6_type, ntohs(icmpv6_id(ic)), 1843 &iph->saddr, &iph->daddr); 1844 1845 offset = iph->len + sizeof(_icmph); 1846 if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, offset, true, &ciph)) 1847 return NF_ACCEPT; 1848 1849 pd = ip_vs_proto_data_get(ipvs, ciph.protocol); 1850 if (!pd) 1851 return NF_ACCEPT; 1852 pp = pd->pp; 1853 1854 /* Cannot handle fragmented embedded protocol */ 1855 if (ciph.fragoffs) 1856 return NF_ACCEPT; 1857 1858 IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset, 1859 "Checking incoming ICMPv6 for"); 1860 1861 /* The embedded headers contain source and dest in reverse order 1862 * if not from localhost 1863 */ 1864 cp = INDIRECT_CALL_1(pp->conn_in_get, ip_vs_conn_in_get_proto, 1865 ipvs, AF_INET6, skb, &ciph); 1866 1867 if (!cp) { 1868 int v; 1869 1870 if (!sysctl_schedule_icmp(ipvs)) 1871 return NF_ACCEPT; 1872 1873 if (!ip_vs_try_to_schedule(ipvs, AF_INET6, skb, pd, &v, &cp, &ciph)) 1874 return v; 1875 1876 new_cp = true; 1877 } 1878 1879 /* VS/TUN, VS/DR and LOCALNODE just let it go */ 1880 if ((hooknum == NF_INET_LOCAL_OUT) && 1881 (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) { 1882 verdict = NF_ACCEPT; 1883 goto out; 1884 } 1885 1886 /* do the statistics and put it back */ 1887 ip_vs_in_stats(cp, skb); 1888 1889 /* Need to mangle contained IPv6 header in ICMPv6 packet */ 1890 offset = ciph.len; 1891 if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol || 1892 IPPROTO_SCTP == ciph.protocol) 1893 offset += 2 * sizeof(__u16); /* Also mangle ports */ 1894 1895 verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset, hooknum, &ciph); 1896 1897 out: 1898 if (likely(!new_cp)) 1899 __ip_vs_conn_put(cp); 1900 else 1901 ip_vs_conn_put(cp); 1902 1903 return verdict; 1904 } 1905 #endif 1906 1907 1908 /* 1909 * Check if it's for virtual services, look it up, 1910 * and send it on its way... 1911 */ 1912 static unsigned int 1913 ip_vs_in_hook(void *priv, struct sk_buff *skb, const struct nf_hook_state *state) 1914 { 1915 struct netns_ipvs *ipvs = net_ipvs(state->net); 1916 unsigned int hooknum = state->hook; 1917 struct ip_vs_iphdr iph; 1918 struct ip_vs_protocol *pp; 1919 struct ip_vs_proto_data *pd; 1920 struct ip_vs_conn *cp; 1921 int ret, pkts; 1922 struct sock *sk; 1923 int af = state->pf; 1924 1925 /* Already marked as IPVS request or reply? */ 1926 if (skb->ipvs_property) 1927 return NF_ACCEPT; 1928 1929 /* 1930 * Big tappo: 1931 * - remote client: only PACKET_HOST 1932 * - route: used for struct net when skb->dev is unset 1933 */ 1934 if (unlikely((skb->pkt_type != PACKET_HOST && 1935 hooknum != NF_INET_LOCAL_OUT) || 1936 !skb_dst(skb))) { 1937 ip_vs_fill_iph_skb(af, skb, false, &iph); 1938 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s" 1939 " ignored in hook %u\n", 1940 skb->pkt_type, iph.protocol, 1941 IP_VS_DBG_ADDR(af, &iph.daddr), hooknum); 1942 return NF_ACCEPT; 1943 } 1944 /* ipvs enabled in this netns ? */ 1945 if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable)) 1946 return NF_ACCEPT; 1947 1948 ip_vs_fill_iph_skb(af, skb, false, &iph); 1949 1950 /* Bad... Do not break raw sockets */ 1951 sk = skb_to_full_sk(skb); 1952 if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT && 1953 af == AF_INET)) { 1954 1955 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag) 1956 return NF_ACCEPT; 1957 } 1958 1959 #ifdef CONFIG_IP_VS_IPV6 1960 if (af == AF_INET6) { 1961 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) { 1962 int related; 1963 int verdict = ip_vs_in_icmp_v6(ipvs, skb, &related, 1964 hooknum, &iph); 1965 1966 if (related) 1967 return verdict; 1968 } 1969 } else 1970 #endif 1971 if (unlikely(iph.protocol == IPPROTO_ICMP)) { 1972 int related; 1973 int verdict = ip_vs_in_icmp(ipvs, skb, &related, 1974 hooknum); 1975 1976 if (related) 1977 return verdict; 1978 } 1979 1980 /* Protocol supported? */ 1981 pd = ip_vs_proto_data_get(ipvs, iph.protocol); 1982 if (unlikely(!pd)) { 1983 /* The only way we'll see this packet again is if it's 1984 * encapsulated, so mark it with ipvs_property=1 so we 1985 * skip it if we're ignoring tunneled packets 1986 */ 1987 if (sysctl_ignore_tunneled(ipvs)) 1988 skb->ipvs_property = 1; 1989 1990 return NF_ACCEPT; 1991 } 1992 pp = pd->pp; 1993 /* 1994 * Check if the packet belongs to an existing connection entry 1995 */ 1996 cp = INDIRECT_CALL_1(pp->conn_in_get, ip_vs_conn_in_get_proto, 1997 ipvs, af, skb, &iph); 1998 1999 if (!iph.fragoffs && is_new_conn(skb, &iph) && cp) { 2000 int conn_reuse_mode = sysctl_conn_reuse_mode(ipvs); 2001 bool old_ct = false, resched = false; 2002 2003 if (unlikely(sysctl_expire_nodest_conn(ipvs)) && cp->dest && 2004 unlikely(!atomic_read(&cp->dest->weight))) { 2005 resched = true; 2006 old_ct = ip_vs_conn_uses_old_conntrack(cp, skb); 2007 } else if (conn_reuse_mode && 2008 is_new_conn_expected(cp, conn_reuse_mode)) { 2009 old_ct = ip_vs_conn_uses_old_conntrack(cp, skb); 2010 if (!atomic_read(&cp->n_control)) { 2011 resched = true; 2012 } else { 2013 /* Do not reschedule controlling connection 2014 * that uses conntrack while it is still 2015 * referenced by controlled connection(s). 2016 */ 2017 resched = !old_ct; 2018 } 2019 } 2020 2021 if (resched) { 2022 if (!old_ct) 2023 cp->flags &= ~IP_VS_CONN_F_NFCT; 2024 if (!atomic_read(&cp->n_control)) 2025 ip_vs_conn_expire_now(cp); 2026 __ip_vs_conn_put(cp); 2027 if (old_ct) 2028 return NF_DROP; 2029 cp = NULL; 2030 } 2031 } 2032 2033 /* Check the server status */ 2034 if (cp && cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) { 2035 /* the destination server is not available */ 2036 if (sysctl_expire_nodest_conn(ipvs)) { 2037 bool old_ct = ip_vs_conn_uses_old_conntrack(cp, skb); 2038 2039 if (!old_ct) 2040 cp->flags &= ~IP_VS_CONN_F_NFCT; 2041 2042 ip_vs_conn_expire_now(cp); 2043 __ip_vs_conn_put(cp); 2044 if (old_ct) 2045 return NF_DROP; 2046 cp = NULL; 2047 } else { 2048 __ip_vs_conn_put(cp); 2049 return NF_DROP; 2050 } 2051 } 2052 2053 if (unlikely(!cp)) { 2054 int v; 2055 2056 if (!ip_vs_try_to_schedule(ipvs, af, skb, pd, &v, &cp, &iph)) 2057 return v; 2058 } 2059 2060 IP_VS_DBG_PKT(11, af, pp, skb, iph.off, "Incoming packet"); 2061 2062 ip_vs_in_stats(cp, skb); 2063 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd); 2064 if (cp->packet_xmit) 2065 ret = cp->packet_xmit(skb, cp, pp, &iph); 2066 /* do not touch skb anymore */ 2067 else { 2068 IP_VS_DBG_RL("warning: packet_xmit is null"); 2069 ret = NF_ACCEPT; 2070 } 2071 2072 /* Increase its packet counter and check if it is needed 2073 * to be synchronized 2074 * 2075 * Sync connection if it is about to close to 2076 * encorage the standby servers to update the connections timeout 2077 * 2078 * For ONE_PKT let ip_vs_sync_conn() do the filter work. 2079 */ 2080 2081 if (cp->flags & IP_VS_CONN_F_ONE_PACKET) 2082 pkts = sysctl_sync_threshold(ipvs); 2083 else 2084 pkts = atomic_inc_return(&cp->in_pkts); 2085 2086 if (ipvs->sync_state & IP_VS_STATE_MASTER) 2087 ip_vs_sync_conn(ipvs, cp, pkts); 2088 else if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) && cp->control) 2089 /* increment is done inside ip_vs_sync_conn too */ 2090 atomic_inc(&cp->control->in_pkts); 2091 2092 ip_vs_conn_put(cp); 2093 return ret; 2094 } 2095 2096 /* 2097 * It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP 2098 * related packets destined for 0.0.0.0/0. 2099 * When fwmark-based virtual service is used, such as transparent 2100 * cache cluster, TCP packets can be marked and routed to ip_vs_in, 2101 * but ICMP destined for 0.0.0.0/0 cannot not be easily marked and 2102 * sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain 2103 * and send them to ip_vs_in_icmp. 2104 */ 2105 static unsigned int 2106 ip_vs_forward_icmp(void *priv, struct sk_buff *skb, 2107 const struct nf_hook_state *state) 2108 { 2109 struct netns_ipvs *ipvs = net_ipvs(state->net); 2110 int r; 2111 2112 /* ipvs enabled in this netns ? */ 2113 if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable)) 2114 return NF_ACCEPT; 2115 2116 if (state->pf == NFPROTO_IPV4) { 2117 if (ip_hdr(skb)->protocol != IPPROTO_ICMP) 2118 return NF_ACCEPT; 2119 #ifdef CONFIG_IP_VS_IPV6 2120 } else { 2121 struct ip_vs_iphdr iphdr; 2122 2123 ip_vs_fill_iph_skb(AF_INET6, skb, false, &iphdr); 2124 2125 if (iphdr.protocol != IPPROTO_ICMPV6) 2126 return NF_ACCEPT; 2127 2128 return ip_vs_in_icmp_v6(ipvs, skb, &r, state->hook, &iphdr); 2129 #endif 2130 } 2131 2132 return ip_vs_in_icmp(ipvs, skb, &r, state->hook); 2133 } 2134 2135 static const struct nf_hook_ops ip_vs_ops4[] = { 2136 /* After packet filtering, change source only for VS/NAT */ 2137 { 2138 .hook = ip_vs_out_hook, 2139 .pf = NFPROTO_IPV4, 2140 .hooknum = NF_INET_LOCAL_IN, 2141 .priority = NF_IP_PRI_NAT_SRC - 2, 2142 }, 2143 /* After packet filtering, forward packet through VS/DR, VS/TUN, 2144 * or VS/NAT(change destination), so that filtering rules can be 2145 * applied to IPVS. */ 2146 { 2147 .hook = ip_vs_in_hook, 2148 .pf = NFPROTO_IPV4, 2149 .hooknum = NF_INET_LOCAL_IN, 2150 .priority = NF_IP_PRI_NAT_SRC - 1, 2151 }, 2152 /* Before ip_vs_in, change source only for VS/NAT */ 2153 { 2154 .hook = ip_vs_out_hook, 2155 .pf = NFPROTO_IPV4, 2156 .hooknum = NF_INET_LOCAL_OUT, 2157 .priority = NF_IP_PRI_NAT_DST + 1, 2158 }, 2159 /* After mangle, schedule and forward local requests */ 2160 { 2161 .hook = ip_vs_in_hook, 2162 .pf = NFPROTO_IPV4, 2163 .hooknum = NF_INET_LOCAL_OUT, 2164 .priority = NF_IP_PRI_NAT_DST + 2, 2165 }, 2166 /* After packet filtering (but before ip_vs_out_icmp), catch icmp 2167 * destined for 0.0.0.0/0, which is for incoming IPVS connections */ 2168 { 2169 .hook = ip_vs_forward_icmp, 2170 .pf = NFPROTO_IPV4, 2171 .hooknum = NF_INET_FORWARD, 2172 .priority = 99, 2173 }, 2174 /* After packet filtering, change source only for VS/NAT */ 2175 { 2176 .hook = ip_vs_out_hook, 2177 .pf = NFPROTO_IPV4, 2178 .hooknum = NF_INET_FORWARD, 2179 .priority = 100, 2180 }, 2181 }; 2182 2183 #ifdef CONFIG_IP_VS_IPV6 2184 static const struct nf_hook_ops ip_vs_ops6[] = { 2185 /* After packet filtering, change source only for VS/NAT */ 2186 { 2187 .hook = ip_vs_out_hook, 2188 .pf = NFPROTO_IPV6, 2189 .hooknum = NF_INET_LOCAL_IN, 2190 .priority = NF_IP6_PRI_NAT_SRC - 2, 2191 }, 2192 /* After packet filtering, forward packet through VS/DR, VS/TUN, 2193 * or VS/NAT(change destination), so that filtering rules can be 2194 * applied to IPVS. */ 2195 { 2196 .hook = ip_vs_in_hook, 2197 .pf = NFPROTO_IPV6, 2198 .hooknum = NF_INET_LOCAL_IN, 2199 .priority = NF_IP6_PRI_NAT_SRC - 1, 2200 }, 2201 /* Before ip_vs_in, change source only for VS/NAT */ 2202 { 2203 .hook = ip_vs_out_hook, 2204 .pf = NFPROTO_IPV6, 2205 .hooknum = NF_INET_LOCAL_OUT, 2206 .priority = NF_IP6_PRI_NAT_DST + 1, 2207 }, 2208 /* After mangle, schedule and forward local requests */ 2209 { 2210 .hook = ip_vs_in_hook, 2211 .pf = NFPROTO_IPV6, 2212 .hooknum = NF_INET_LOCAL_OUT, 2213 .priority = NF_IP6_PRI_NAT_DST + 2, 2214 }, 2215 /* After packet filtering (but before ip_vs_out_icmp), catch icmp 2216 * destined for 0.0.0.0/0, which is for incoming IPVS connections */ 2217 { 2218 .hook = ip_vs_forward_icmp, 2219 .pf = NFPROTO_IPV6, 2220 .hooknum = NF_INET_FORWARD, 2221 .priority = 99, 2222 }, 2223 /* After packet filtering, change source only for VS/NAT */ 2224 { 2225 .hook = ip_vs_out_hook, 2226 .pf = NFPROTO_IPV6, 2227 .hooknum = NF_INET_FORWARD, 2228 .priority = 100, 2229 }, 2230 }; 2231 #endif 2232 2233 int ip_vs_register_hooks(struct netns_ipvs *ipvs, unsigned int af) 2234 { 2235 const struct nf_hook_ops *ops; 2236 unsigned int count; 2237 unsigned int afmask; 2238 int ret = 0; 2239 2240 if (af == AF_INET6) { 2241 #ifdef CONFIG_IP_VS_IPV6 2242 ops = ip_vs_ops6; 2243 count = ARRAY_SIZE(ip_vs_ops6); 2244 afmask = 2; 2245 #else 2246 return -EINVAL; 2247 #endif 2248 } else { 2249 ops = ip_vs_ops4; 2250 count = ARRAY_SIZE(ip_vs_ops4); 2251 afmask = 1; 2252 } 2253 2254 if (!(ipvs->hooks_afmask & afmask)) { 2255 ret = nf_register_net_hooks(ipvs->net, ops, count); 2256 if (ret >= 0) 2257 ipvs->hooks_afmask |= afmask; 2258 } 2259 return ret; 2260 } 2261 2262 void ip_vs_unregister_hooks(struct netns_ipvs *ipvs, unsigned int af) 2263 { 2264 const struct nf_hook_ops *ops; 2265 unsigned int count; 2266 unsigned int afmask; 2267 2268 if (af == AF_INET6) { 2269 #ifdef CONFIG_IP_VS_IPV6 2270 ops = ip_vs_ops6; 2271 count = ARRAY_SIZE(ip_vs_ops6); 2272 afmask = 2; 2273 #else 2274 return; 2275 #endif 2276 } else { 2277 ops = ip_vs_ops4; 2278 count = ARRAY_SIZE(ip_vs_ops4); 2279 afmask = 1; 2280 } 2281 2282 if (ipvs->hooks_afmask & afmask) { 2283 nf_unregister_net_hooks(ipvs->net, ops, count); 2284 ipvs->hooks_afmask &= ~afmask; 2285 } 2286 } 2287 2288 /* 2289 * Initialize IP Virtual Server netns mem. 2290 */ 2291 static int __net_init __ip_vs_init(struct net *net) 2292 { 2293 struct netns_ipvs *ipvs; 2294 2295 ipvs = net_generic(net, ip_vs_net_id); 2296 if (ipvs == NULL) 2297 return -ENOMEM; 2298 2299 /* Hold the beast until a service is registered */ 2300 ipvs->enable = 0; 2301 ipvs->net = net; 2302 /* Counters used for creating unique names */ 2303 ipvs->gen = atomic_read(&ipvs_netns_cnt); 2304 atomic_inc(&ipvs_netns_cnt); 2305 net->ipvs = ipvs; 2306 2307 if (ip_vs_estimator_net_init(ipvs) < 0) 2308 goto estimator_fail; 2309 2310 if (ip_vs_control_net_init(ipvs) < 0) 2311 goto control_fail; 2312 2313 if (ip_vs_protocol_net_init(ipvs) < 0) 2314 goto protocol_fail; 2315 2316 if (ip_vs_app_net_init(ipvs) < 0) 2317 goto app_fail; 2318 2319 if (ip_vs_conn_net_init(ipvs) < 0) 2320 goto conn_fail; 2321 2322 if (ip_vs_sync_net_init(ipvs) < 0) 2323 goto sync_fail; 2324 2325 return 0; 2326 /* 2327 * Error handling 2328 */ 2329 2330 sync_fail: 2331 ip_vs_conn_net_cleanup(ipvs); 2332 conn_fail: 2333 ip_vs_app_net_cleanup(ipvs); 2334 app_fail: 2335 ip_vs_protocol_net_cleanup(ipvs); 2336 protocol_fail: 2337 ip_vs_control_net_cleanup(ipvs); 2338 control_fail: 2339 ip_vs_estimator_net_cleanup(ipvs); 2340 estimator_fail: 2341 net->ipvs = NULL; 2342 return -ENOMEM; 2343 } 2344 2345 static void __net_exit __ip_vs_cleanup_batch(struct list_head *net_list) 2346 { 2347 struct netns_ipvs *ipvs; 2348 struct net *net; 2349 2350 ip_vs_service_nets_cleanup(net_list); /* ip_vs_flush() with locks */ 2351 list_for_each_entry(net, net_list, exit_list) { 2352 ipvs = net_ipvs(net); 2353 ip_vs_conn_net_cleanup(ipvs); 2354 ip_vs_app_net_cleanup(ipvs); 2355 ip_vs_protocol_net_cleanup(ipvs); 2356 ip_vs_control_net_cleanup(ipvs); 2357 ip_vs_estimator_net_cleanup(ipvs); 2358 IP_VS_DBG(2, "ipvs netns %d released\n", ipvs->gen); 2359 net->ipvs = NULL; 2360 } 2361 } 2362 2363 static void __net_exit __ip_vs_dev_cleanup_batch(struct list_head *net_list) 2364 { 2365 struct netns_ipvs *ipvs; 2366 struct net *net; 2367 2368 EnterFunction(2); 2369 list_for_each_entry(net, net_list, exit_list) { 2370 ipvs = net_ipvs(net); 2371 ip_vs_unregister_hooks(ipvs, AF_INET); 2372 ip_vs_unregister_hooks(ipvs, AF_INET6); 2373 ipvs->enable = 0; /* Disable packet reception */ 2374 smp_wmb(); 2375 ip_vs_sync_net_cleanup(ipvs); 2376 } 2377 LeaveFunction(2); 2378 } 2379 2380 static struct pernet_operations ipvs_core_ops = { 2381 .init = __ip_vs_init, 2382 .exit_batch = __ip_vs_cleanup_batch, 2383 .id = &ip_vs_net_id, 2384 .size = sizeof(struct netns_ipvs), 2385 }; 2386 2387 static struct pernet_operations ipvs_core_dev_ops = { 2388 .exit_batch = __ip_vs_dev_cleanup_batch, 2389 }; 2390 2391 /* 2392 * Initialize IP Virtual Server 2393 */ 2394 static int __init ip_vs_init(void) 2395 { 2396 int ret; 2397 2398 ret = ip_vs_control_init(); 2399 if (ret < 0) { 2400 pr_err("can't setup control.\n"); 2401 goto exit; 2402 } 2403 2404 ip_vs_protocol_init(); 2405 2406 ret = ip_vs_conn_init(); 2407 if (ret < 0) { 2408 pr_err("can't setup connection table.\n"); 2409 goto cleanup_protocol; 2410 } 2411 2412 ret = register_pernet_subsys(&ipvs_core_ops); /* Alloc ip_vs struct */ 2413 if (ret < 0) 2414 goto cleanup_conn; 2415 2416 ret = register_pernet_device(&ipvs_core_dev_ops); 2417 if (ret < 0) 2418 goto cleanup_sub; 2419 2420 ret = ip_vs_register_nl_ioctl(); 2421 if (ret < 0) { 2422 pr_err("can't register netlink/ioctl.\n"); 2423 goto cleanup_dev; 2424 } 2425 2426 pr_info("ipvs loaded.\n"); 2427 2428 return ret; 2429 2430 cleanup_dev: 2431 unregister_pernet_device(&ipvs_core_dev_ops); 2432 cleanup_sub: 2433 unregister_pernet_subsys(&ipvs_core_ops); 2434 cleanup_conn: 2435 ip_vs_conn_cleanup(); 2436 cleanup_protocol: 2437 ip_vs_protocol_cleanup(); 2438 ip_vs_control_cleanup(); 2439 exit: 2440 return ret; 2441 } 2442 2443 static void __exit ip_vs_cleanup(void) 2444 { 2445 ip_vs_unregister_nl_ioctl(); 2446 unregister_pernet_device(&ipvs_core_dev_ops); 2447 unregister_pernet_subsys(&ipvs_core_ops); /* free ip_vs struct */ 2448 ip_vs_conn_cleanup(); 2449 ip_vs_protocol_cleanup(); 2450 ip_vs_control_cleanup(); 2451 pr_info("ipvs unloaded.\n"); 2452 } 2453 2454 module_init(ip_vs_init); 2455 module_exit(ip_vs_cleanup); 2456 MODULE_LICENSE("GPL"); 2457