net/netfilter/ipvs/ip_vs_core.c

   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/route.h>
  38 #include <net/ip6_checksum.h>
  39 #include <net/netns/generic.h>          /* net_generic() */
  40
  41 #include <linux/netfilter.h>
  42 #include <linux/netfilter_ipv4.h>
  43
  44 #ifdef CONFIG_IP_VS_IPV6
  45 #include <net/ipv6.h>
  46 #include <linux/netfilter_ipv6.h>
  47 #include <net/ip6_route.h>
  48 #endif
  49
  50 #include <net/ip_vs.h>
  51 #include <linux/indirect_call_wrapper.h>
  52
  53
  54 EXPORT_SYMBOL(register_ip_vs_scheduler);
  55 EXPORT_SYMBOL(unregister_ip_vs_scheduler);
  56 EXPORT_SYMBOL(ip_vs_proto_name);
  57 EXPORT_SYMBOL(ip_vs_conn_new);
  58 EXPORT_SYMBOL(ip_vs_conn_in_get);
  59 EXPORT_SYMBOL(ip_vs_conn_out_get);
  60 #ifdef CONFIG_IP_VS_PROTO_TCP
  61 EXPORT_SYMBOL(ip_vs_tcp_conn_listen);
  62 #endif
  63 EXPORT_SYMBOL(ip_vs_conn_put);
  64 #ifdef CONFIG_IP_VS_DEBUG
  65 EXPORT_SYMBOL(ip_vs_get_debug_level);
  66 #endif
  67 EXPORT_SYMBOL(ip_vs_new_conn_out);
  68
  69 #ifdef CONFIG_IP_VS_PROTO_TCP
  70 INDIRECT_CALLABLE_DECLARE(int
  71         tcp_snat_handler(struct sk_buff *skb, struct ip_vs_protocol *pp,
  72                          struct ip_vs_conn *cp, struct ip_vs_iphdr *iph));
  73 #endif
  74
  75 #ifdef CONFIG_IP_VS_PROTO_UDP
  76 INDIRECT_CALLABLE_DECLARE(int
  77         udp_snat_handler(struct sk_buff *skb, struct ip_vs_protocol *pp,
  78                          struct ip_vs_conn *cp, struct ip_vs_iphdr *iph));
  79 #endif
  80
  81 #if defined(CONFIG_IP_VS_PROTO_TCP) && defined(CONFIG_IP_VS_PROTO_UDP)
  82 #define SNAT_CALL(f, ...) \
  83         INDIRECT_CALL_2(f, tcp_snat_handler, udp_snat_handler, __VA_ARGS__)
  84 #elif defined(CONFIG_IP_VS_PROTO_TCP)
  85 #define SNAT_CALL(f, ...) INDIRECT_CALL_1(f, tcp_snat_handler, __VA_ARGS__)
  86 #elif defined(CONFIG_IP_VS_PROTO_UDP)
  87 #define SNAT_CALL(f, ...) INDIRECT_CALL_1(f, udp_snat_handler, __VA_ARGS__)
  88 #else
  89 #define SNAT_CALL(f, ...) f(__VA_ARGS__)
  90 #endif
  91
  92 static unsigned int ip_vs_net_id __read_mostly;
  93 /* netns cnt used for uniqueness */
  94 static atomic_t ipvs_netns_cnt = ATOMIC_INIT(0);
  95
  96 /* ID used in ICMP lookups */
  97 #define icmp_id(icmph)          (((icmph)->un).echo.id)
  98 #define icmpv6_id(icmph)        (icmph->icmp6_dataun.u_echo.identifier)
  99
 100 const char *ip_vs_proto_name(unsigned int proto)
 101 {
 102         static char buf[20];
 103
 104         switch (proto) {
 105         case IPPROTO_IP:
 106                 return "IP";
 107         case IPPROTO_UDP:
 108                 return "UDP";
 109         case IPPROTO_TCP:
 110                 return "TCP";
 111         case IPPROTO_SCTP:
 112                 return "SCTP";
 113         case IPPROTO_ICMP:
 114                 return "ICMP";
 115 #ifdef CONFIG_IP_VS_IPV6
 116         case IPPROTO_ICMPV6:
 117                 return "ICMPv6";
 118 #endif
 119         default:
 120                 sprintf(buf, "IP_%u", proto);
 121                 return buf;
 122         }
 123 }
 124
 125 void ip_vs_init_hash_table(struct list_head *table, int rows)
 126 {
 127         while (--rows >= 0)
 128                 INIT_LIST_HEAD(&table[rows]);
 129 }
 130
 131 static inline void
 132 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 133 {
 134         struct ip_vs_dest *dest = cp->dest;
 135         struct netns_ipvs *ipvs = cp->ipvs;
 136
 137         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 138                 struct ip_vs_cpu_stats *s;
 139                 struct ip_vs_service *svc;
 140
 141                 local_bh_disable();
 142
 143                 s = this_cpu_ptr(dest->stats.cpustats);
 144                 u64_stats_update_begin(&s->syncp);
 145                 s->cnt.inpkts++;
 146                 s->cnt.inbytes += skb->len;
 147                 u64_stats_update_end(&s->syncp);
 148
 149                 svc = rcu_dereference(dest->svc);
 150                 s = this_cpu_ptr(svc->stats.cpustats);
 151                 u64_stats_update_begin(&s->syncp);
 152                 s->cnt.inpkts++;
 153                 s->cnt.inbytes += skb->len;
 154                 u64_stats_update_end(&s->syncp);
 155
 156                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 157                 u64_stats_update_begin(&s->syncp);
 158                 s->cnt.inpkts++;
 159                 s->cnt.inbytes += skb->len;
 160                 u64_stats_update_end(&s->syncp);
 161
 162                 local_bh_enable();
 163         }
 164 }
 165
 166
 167 static inline void
 168 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 169 {
 170         struct ip_vs_dest *dest = cp->dest;
 171         struct netns_ipvs *ipvs = cp->ipvs;
 172
 173         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 174                 struct ip_vs_cpu_stats *s;
 175                 struct ip_vs_service *svc;
 176
 177                 local_bh_disable();
 178
 179                 s = this_cpu_ptr(dest->stats.cpustats);
 180                 u64_stats_update_begin(&s->syncp);
 181                 s->cnt.outpkts++;
 182                 s->cnt.outbytes += skb->len;
 183                 u64_stats_update_end(&s->syncp);
 184
 185                 svc = rcu_dereference(dest->svc);
 186                 s = this_cpu_ptr(svc->stats.cpustats);
 187                 u64_stats_update_begin(&s->syncp);
 188                 s->cnt.outpkts++;
 189                 s->cnt.outbytes += skb->len;
 190                 u64_stats_update_end(&s->syncp);
 191
 192                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 193                 u64_stats_update_begin(&s->syncp);
 194                 s->cnt.outpkts++;
 195                 s->cnt.outbytes += skb->len;
 196                 u64_stats_update_end(&s->syncp);
 197
 198                 local_bh_enable();
 199         }
 200 }
 201
 202
 203 static inline void
 204 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
 205 {
 206         struct netns_ipvs *ipvs = svc->ipvs;
 207         struct ip_vs_cpu_stats *s;
 208
 209         local_bh_disable();
 210
 211         s = this_cpu_ptr(cp->dest->stats.cpustats);
 212         u64_stats_update_begin(&s->syncp);
 213         s->cnt.conns++;
 214         u64_stats_update_end(&s->syncp);
 215
 216         s = this_cpu_ptr(svc->stats.cpustats);
 217         u64_stats_update_begin(&s->syncp);
 218         s->cnt.conns++;
 219         u64_stats_update_end(&s->syncp);
 220
 221         s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 222         u64_stats_update_begin(&s->syncp);
 223         s->cnt.conns++;
 224         u64_stats_update_end(&s->syncp);
 225
 226         local_bh_enable();
 227 }
 228
 229
 230 static inline void
 231 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
 232                 const struct sk_buff *skb,
 233                 struct ip_vs_proto_data *pd)
 234 {
 235         if (likely(pd->pp->state_transition))
 236                 pd->pp->state_transition(cp, direction, skb, pd);
 237 }
 238
 239 static inline int
 240 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
 241                               struct sk_buff *skb, int protocol,
 242                               const union nf_inet_addr *caddr, __be16 cport,
 243                               const union nf_inet_addr *vaddr, __be16 vport,
 244                               struct ip_vs_conn_param *p)
 245 {
 246         ip_vs_conn_fill_param(svc->ipvs, svc->af, protocol, caddr, cport, vaddr,
 247                               vport, p);
 248         p->pe = rcu_dereference(svc->pe);
 249         if (p->pe && p->pe->fill_param)
 250                 return p->pe->fill_param(p, skb);
 251
 252         return 0;
 253 }
 254
 255 /*
 256  *  IPVS persistent scheduling function
 257  *  It creates a connection entry according to its template if exists,
 258  *  or selects a server and creates a connection entry plus a template.
 259  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
 260  *  Protocols supported: TCP, UDP
 261  */
 262 static struct ip_vs_conn *
 263 ip_vs_sched_persist(struct ip_vs_service *svc,
 264                     struct sk_buff *skb, __be16 src_port, __be16 dst_port,
 265                     int *ignored, struct ip_vs_iphdr *iph)
 266 {
 267         struct ip_vs_conn *cp = NULL;
 268         struct ip_vs_dest *dest;
 269         struct ip_vs_conn *ct;
 270         __be16 dport = 0;               /* destination port to forward */
 271         unsigned int flags;
 272         struct ip_vs_conn_param param;
 273         const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
 274         union nf_inet_addr snet;        /* source network of the client,
 275                                            after masking */
 276         const union nf_inet_addr *src_addr, *dst_addr;
 277
 278         if (likely(!ip_vs_iph_inverse(iph))) {
 279                 src_addr = &iph->saddr;
 280                 dst_addr = &iph->daddr;
 281         } else {
 282                 src_addr = &iph->daddr;
 283                 dst_addr = &iph->saddr;
 284         }
 285
 286
 287         /* Mask saddr with the netmask to adjust template granularity */
 288 #ifdef CONFIG_IP_VS_IPV6
 289         if (svc->af == AF_INET6)
 290                 ipv6_addr_prefix(&snet.in6, &src_addr->in6,
 291                                  (__force __u32) svc->netmask);
 292         else
 293 #endif
 294                 snet.ip = src_addr->ip & svc->netmask;
 295
 296         IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
 297                       "mnet %s\n",
 298                       IP_VS_DBG_ADDR(svc->af, src_addr), ntohs(src_port),
 299                       IP_VS_DBG_ADDR(svc->af, dst_addr), ntohs(dst_port),
 300                       IP_VS_DBG_ADDR(svc->af, &snet));
 301
 302         /*
 303          * As far as we know, FTP is a very complicated network protocol, and
 304          * it uses control connection and data connections. For active FTP,
 305          * FTP server initialize data connection to the client, its source port
 306          * is often 20. For passive FTP, FTP server tells the clients the port
 307          * that it passively listens to,  and the client issues the data
 308          * connection. In the tunneling or direct routing mode, the load
 309          * balancer is on the client-to-server half of connection, the port
 310          * number is unknown to the load balancer. So, a conn template like
 311          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
 312          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
 313          * is created for other persistent services.
 314          */
 315         {
 316                 int protocol = iph->protocol;
 317                 const union nf_inet_addr *vaddr = dst_addr;
 318                 __be16 vport = 0;
 319
 320                 if (dst_port == svc->port) {
 321                         /* non-FTP template:
 322                          * <protocol, caddr, 0, vaddr, vport, daddr, dport>
 323                          * FTP template:
 324                          * <protocol, caddr, 0, vaddr, 0, daddr, 0>
 325                          */
 326                         if (svc->port != FTPPORT)
 327                                 vport = dst_port;
 328                 } else {
 329                         /* Note: persistent fwmark-based services and
 330                          * persistent port zero service are handled here.
 331                          * fwmark template:
 332                          * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
 333                          * port zero template:
 334                          * <protocol,caddr,0,vaddr,0,daddr,0>
 335                          */
 336                         if (svc->fwmark) {
 337                                 protocol = IPPROTO_IP;
 338                                 vaddr = &fwmark;
 339                         }
 340                 }
 341                 /* return *ignored = -1 so NF_DROP can be used */
 342                 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
 343                                                   vaddr, vport, &param) < 0) {
 344                         *ignored = -1;
 345                         return NULL;
 346                 }
 347         }
 348
 349         /* Check if a template already exists */
 350         ct = ip_vs_ct_in_get(&param);
 351         if (!ct || !ip_vs_check_template(ct, NULL)) {
 352                 struct ip_vs_scheduler *sched;
 353
 354                 /*
 355                  * No template found or the dest of the connection
 356                  * template is not available.
 357                  * return *ignored=0 i.e. ICMP and NF_DROP
 358                  */
 359                 sched = rcu_dereference(svc->scheduler);
 360                 if (sched) {
 361                         /* read svc->sched_data after svc->scheduler */
 362                         smp_rmb();
 363                         dest = sched->schedule(svc, skb, iph);
 364                 } else {
 365                         dest = NULL;
 366                 }
 367                 if (!dest) {
 368                         IP_VS_DBG(1, "p-schedule: no dest found.\n");
 369                         kfree(param.pe_data);
 370                         *ignored = 0;
 371                         return NULL;
 372                 }
 373
 374                 if (dst_port == svc->port && svc->port != FTPPORT)
 375                         dport = dest->port;
 376
 377                 /* Create a template
 378                  * This adds param.pe_data to the template,
 379                  * and thus param.pe_data will be destroyed
 380                  * when the template expires */
 381                 ct = ip_vs_conn_new(&param, dest->af, &dest->addr, dport,
 382                                     IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
 383                 if (ct == NULL) {
 384                         kfree(param.pe_data);
 385                         *ignored = -1;
 386                         return NULL;
 387                 }
 388
 389                 ct->timeout = svc->timeout;
 390         } else {
 391                 /* set destination with the found template */
 392                 dest = ct->dest;
 393                 kfree(param.pe_data);
 394         }
 395
 396         dport = dst_port;
 397         if (dport == svc->port && dest->port)
 398                 dport = dest->port;
 399
 400         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 401                  && iph->protocol == IPPROTO_UDP) ?
 402                 IP_VS_CONN_F_ONE_PACKET : 0;
 403
 404         /*
 405          *    Create a new connection according to the template
 406          */
 407         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol, src_addr,
 408                               src_port, dst_addr, dst_port, &param);
 409
 410         cp = ip_vs_conn_new(&param, dest->af, &dest->addr, dport, flags, dest,
 411                             skb->mark);
 412         if (cp == NULL) {
 413                 ip_vs_conn_put(ct);
 414                 *ignored = -1;
 415                 return NULL;
 416         }
 417
 418         /*
 419          *    Add its control
 420          */
 421         ip_vs_control_add(cp, ct);
 422         ip_vs_conn_put(ct);
 423
 424         ip_vs_conn_stats(cp, svc);
 425         return cp;
 426 }
 427
 428
 429 /*
 430  *  IPVS main scheduling function
 431  *  It selects a server according to the virtual service, and
 432  *  creates a connection entry.
 433  *  Protocols supported: TCP, UDP
 434  *
 435  *  Usage of *ignored
 436  *
 437  * 1 :   protocol tried to schedule (eg. on SYN), found svc but the
 438  *       svc/scheduler decides that this packet should be accepted with
 439  *       NF_ACCEPT because it must not be scheduled.
 440  *
 441  * 0 :   scheduler can not find destination, so try bypass or
 442  *       return ICMP and then NF_DROP (ip_vs_leave).
 443  *
 444  * -1 :  scheduler tried to schedule but fatal error occurred, eg.
 445  *       ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
 446  *       failure such as missing Call-ID, ENOMEM on skb_linearize
 447  *       or pe_data. In this case we should return NF_DROP without
 448  *       any attempts to send ICMP with ip_vs_leave.
 449  */
 450 struct ip_vs_conn *
 451 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
 452                struct ip_vs_proto_data *pd, int *ignored,
 453                struct ip_vs_iphdr *iph)
 454 {
 455         struct ip_vs_protocol *pp = pd->pp;
 456         struct ip_vs_conn *cp = NULL;
 457         struct ip_vs_scheduler *sched;
 458         struct ip_vs_dest *dest;
 459         __be16 _ports[2], *pptr, cport, vport;
 460         const void *caddr, *vaddr;
 461         unsigned int flags;
 462
 463         *ignored = 1;
 464         /*
 465          * IPv6 frags, only the first hit here.
 466          */
 467         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports);
 468         if (pptr == NULL)
 469                 return NULL;
 470
 471         if (likely(!ip_vs_iph_inverse(iph))) {
 472                 cport = pptr[0];
 473                 caddr = &iph->saddr;
 474                 vport = pptr[1];
 475                 vaddr = &iph->daddr;
 476         } else {
 477                 cport = pptr[1];
 478                 caddr = &iph->daddr;
 479                 vport = pptr[0];
 480                 vaddr = &iph->saddr;
 481         }
 482
 483         /*
 484          * FTPDATA needs this check when using local real server.
 485          * Never schedule Active FTPDATA connections from real server.
 486          * For LVS-NAT they must be already created. For other methods
 487          * with persistence the connection is created on SYN+ACK.
 488          */
 489         if (cport == FTPDATA) {
 490                 IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
 491                               "Not scheduling FTPDATA");
 492                 return NULL;
 493         }
 494
 495         /*
 496          *    Do not schedule replies from local real server.
 497          */
 498         if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK)) {
 499                 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
 500                 cp = INDIRECT_CALL_1(pp->conn_in_get,
 501                                      ip_vs_conn_in_get_proto, svc->ipvs,
 502                                      svc->af, skb, iph);
 503                 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
 504
 505                 if (cp) {
 506                         IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
 507                                       "Not scheduling reply for existing"
 508                                       " connection");
 509                         __ip_vs_conn_put(cp);
 510                         return NULL;
 511                 }
 512         }
 513
 514         /*
 515          *    Persistent service
 516          */
 517         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 518                 return ip_vs_sched_persist(svc, skb, cport, vport, ignored,
 519                                            iph);
 520
 521         *ignored = 0;
 522
 523         /*
 524          *    Non-persistent service
 525          */
 526         if (!svc->fwmark && vport != svc->port) {
 527                 if (!svc->port)
 528                         pr_err("Schedule: port zero only supported "
 529                                "in persistent services, "
 530                                "check your ipvs configuration\n");
 531                 return NULL;
 532         }
 533
 534         sched = rcu_dereference(svc->scheduler);
 535         if (sched) {
 536                 /* read svc->sched_data after svc->scheduler */
 537                 smp_rmb();
 538                 dest = sched->schedule(svc, skb, iph);
 539         } else {
 540                 dest = NULL;
 541         }
 542         if (dest == NULL) {
 543                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 544                 return NULL;
 545         }
 546
 547         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 548                  && iph->protocol == IPPROTO_UDP) ?
 549                 IP_VS_CONN_F_ONE_PACKET : 0;
 550
 551         /*
 552          *    Create a connection entry.
 553          */
 554         {
 555                 struct ip_vs_conn_param p;
 556
 557                 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
 558                                       caddr, cport, vaddr, vport, &p);
 559                 cp = ip_vs_conn_new(&p, dest->af, &dest->addr,
 560                                     dest->port ? dest->port : vport,
 561                                     flags, dest, skb->mark);
 562                 if (!cp) {
 563                         *ignored = -1;
 564                         return NULL;
 565                 }
 566         }
 567
 568         IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
 569                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
 570                       ip_vs_fwd_tag(cp),
 571                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 572                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 573                       IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
 574                       cp->flags, refcount_read(&cp->refcnt));
 575
 576         ip_vs_conn_stats(cp, svc);
 577         return cp;
 578 }
 579
 580 static inline int ip_vs_addr_is_unicast(struct net *net, int af,
 581                                         union nf_inet_addr *addr)
 582 {
 583 #ifdef CONFIG_IP_VS_IPV6
 584         if (af == AF_INET6)
 585                 return ipv6_addr_type(&addr->in6) & IPV6_ADDR_UNICAST;
 586 #endif
 587         return (inet_addr_type(net, addr->ip) == RTN_UNICAST);
 588 }
 589
 590 /*
 591  *  Pass or drop the packet.
 592  *  Called by ip_vs_in, when the virtual service is available but
 593  *  no destination is available for a new connection.
 594  */
 595 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 596                 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
 597 {
 598         __be16 _ports[2], *pptr, dport;
 599         struct netns_ipvs *ipvs = svc->ipvs;
 600         struct net *net = ipvs->net;
 601
 602         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports);
 603         if (!pptr)
 604                 return NF_DROP;
 605         dport = likely(!ip_vs_iph_inverse(iph)) ? pptr[1] : pptr[0];
 606
 607         /* if it is fwmark-based service, the cache_bypass sysctl is up
 608            and the destination is a non-local unicast, then create
 609            a cache_bypass connection entry */
 610         if (sysctl_cache_bypass(ipvs) && svc->fwmark &&
 611             !(iph->hdr_flags & (IP_VS_HDR_INVERSE | IP_VS_HDR_ICMP)) &&
 612             ip_vs_addr_is_unicast(net, svc->af, &iph->daddr)) {
 613                 int ret;
 614                 struct ip_vs_conn *cp;
 615                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
 616                                       iph->protocol == IPPROTO_UDP) ?
 617                                       IP_VS_CONN_F_ONE_PACKET : 0;
 618                 union nf_inet_addr daddr =  { .all = { 0, 0, 0, 0 } };
 619
 620                 /* create a new connection entry */
 621                 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
 622                 {
 623                         struct ip_vs_conn_param p;
 624                         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
 625                                               &iph->saddr, pptr[0],
 626                                               &iph->daddr, pptr[1], &p);
 627                         cp = ip_vs_conn_new(&p, svc->af, &daddr, 0,
 628                                             IP_VS_CONN_F_BYPASS | flags,
 629                                             NULL, skb->mark);
 630                         if (!cp)
 631                                 return NF_DROP;
 632                 }
 633
 634                 /* statistics */
 635                 ip_vs_in_stats(cp, skb);
 636
 637                 /* set state */
 638                 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
 639
 640                 /* transmit the first SYN packet */
 641                 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
 642                 /* do not touch skb anymore */
 643
 644                 if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) && cp->control)
 645                         atomic_inc(&cp->control->in_pkts);
 646                 else
 647                         atomic_inc(&cp->in_pkts);
 648                 ip_vs_conn_put(cp);
 649                 return ret;
 650         }
 651
 652         /*
 653          * When the virtual ftp service is presented, packets destined
 654          * for other services on the VIP may get here (except services
 655          * listed in the ipvs table), pass the packets, because it is
 656          * not ipvs job to decide to drop the packets.
 657          */
 658         if (svc->port == FTPPORT && dport != FTPPORT)
 659                 return NF_ACCEPT;
 660
 661         if (unlikely(ip_vs_iph_icmp(iph)))
 662                 return NF_DROP;
 663
 664         /*
 665          * Notify the client that the destination is unreachable, and
 666          * release the socket buffer.
 667          * Since it is in IP layer, the TCP socket is not actually
 668          * created, the TCP RST packet cannot be sent, instead that
 669          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 670          */
 671 #ifdef CONFIG_IP_VS_IPV6
 672         if (svc->af == AF_INET6) {
 673                 if (!skb->dev)
 674                         skb->dev = net->loopback_dev;
 675                 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 676         } else
 677 #endif
 678                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 679
 680         return NF_DROP;
 681 }
 682
 683 #ifdef CONFIG_SYSCTL
 684
 685 static int sysctl_snat_reroute(struct netns_ipvs *ipvs)
 686 {
 687         return ipvs->sysctl_snat_reroute;
 688 }
 689
 690 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs)
 691 {
 692         return ipvs->sysctl_nat_icmp_send;
 693 }
 694
 695 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
 696 {
 697         return ipvs->sysctl_expire_nodest_conn;
 698 }
 699
 700 #else
 701
 702 static int sysctl_snat_reroute(struct netns_ipvs *ipvs) { return 0; }
 703 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs) { return 0; }
 704 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
 705
 706 #endif
 707
 708 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 709 {
 710         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 711 }
 712
 713 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
 714 {
 715         if (NF_INET_LOCAL_IN == hooknum)
 716                 return IP_DEFRAG_VS_IN;
 717         if (NF_INET_FORWARD == hooknum)
 718                 return IP_DEFRAG_VS_FWD;
 719         return IP_DEFRAG_VS_OUT;
 720 }
 721
 722 static inline int ip_vs_gather_frags(struct netns_ipvs *ipvs,
 723                                      struct sk_buff *skb, u_int32_t user)
 724 {
 725         int err;
 726
 727         local_bh_disable();
 728         err = ip_defrag(ipvs->net, skb, user);
 729         local_bh_enable();
 730         if (!err)
 731                 ip_send_check(ip_hdr(skb));
 732
 733         return err;
 734 }
 735
 736 static int ip_vs_route_me_harder(struct netns_ipvs *ipvs, int af,
 737                                  struct sk_buff *skb, unsigned int hooknum)
 738 {
 739         if (!sysctl_snat_reroute(ipvs))
 740                 return 0;
 741         /* Reroute replies only to remote clients (FORWARD and LOCAL_OUT) */
 742         if (NF_INET_LOCAL_IN == hooknum)
 743                 return 0;
 744 #ifdef CONFIG_IP_VS_IPV6
 745         if (af == AF_INET6) {
 746                 struct dst_entry *dst = skb_dst(skb);
 747
 748                 if (dst->dev && !(dst->dev->flags & IFF_LOOPBACK) &&
 749                     ip6_route_me_harder(ipvs->net, skb) != 0)
 750                         return 1;
 751         } else
 752 #endif
 753                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
 754                     ip_route_me_harder(ipvs->net, skb, RTN_LOCAL) != 0)
 755                         return 1;
 756
 757         return 0;
 758 }
 759
 760 /*
 761  * Packet has been made sufficiently writable in caller
 762  * - inout: 1=in->out, 0=out->in
 763  */
 764 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 765                     struct ip_vs_conn *cp, int inout)
 766 {
 767         struct iphdr *iph        = ip_hdr(skb);
 768         unsigned int icmp_offset = iph->ihl*4;
 769         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
 770                                                       icmp_offset);
 771         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 772
 773         if (inout) {
 774                 iph->saddr = cp->vaddr.ip;
 775                 ip_send_check(iph);
 776                 ciph->daddr = cp->vaddr.ip;
 777                 ip_send_check(ciph);
 778         } else {
 779                 iph->daddr = cp->daddr.ip;
 780                 ip_send_check(iph);
 781                 ciph->saddr = cp->daddr.ip;
 782                 ip_send_check(ciph);
 783         }
 784
 785         /* the TCP/UDP/SCTP port */
 786         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
 787             IPPROTO_SCTP == ciph->protocol) {
 788                 __be16 *ports = (void *)ciph + ciph->ihl*4;
 789
 790                 if (inout)
 791                         ports[1] = cp->vport;
 792                 else
 793                         ports[0] = cp->dport;
 794         }
 795
 796         /* And finally the ICMP checksum */
 797         icmph->checksum = 0;
 798         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 799         skb->ip_summed = CHECKSUM_UNNECESSARY;
 800
 801         if (inout)
 802                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 803                         "Forwarding altered outgoing ICMP");
 804         else
 805                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 806                         "Forwarding altered incoming ICMP");
 807 }
 808
 809 #ifdef CONFIG_IP_VS_IPV6
 810 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
 811                     struct ip_vs_conn *cp, int inout)
 812 {
 813         struct ipv6hdr *iph      = ipv6_hdr(skb);
 814         unsigned int icmp_offset = 0;
 815         unsigned int offs        = 0; /* header offset*/
 816         int protocol;
 817         struct icmp6hdr *icmph;
 818         struct ipv6hdr *ciph;
 819         unsigned short fragoffs;
 820
 821         ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
 822         icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
 823         offs = icmp_offset + sizeof(struct icmp6hdr);
 824         ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
 825
 826         protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
 827
 828         if (inout) {
 829                 iph->saddr = cp->vaddr.in6;
 830                 ciph->daddr = cp->vaddr.in6;
 831         } else {
 832                 iph->daddr = cp->daddr.in6;
 833                 ciph->saddr = cp->daddr.in6;
 834         }
 835
 836         /* the TCP/UDP/SCTP port */
 837         if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 838                           IPPROTO_SCTP == protocol)) {
 839                 __be16 *ports = (void *)(skb_network_header(skb) + offs);
 840
 841                 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
 842                               ntohs(inout ? ports[1] : ports[0]),
 843                               ntohs(inout ? cp->vport : cp->dport));
 844                 if (inout)
 845                         ports[1] = cp->vport;
 846                 else
 847                         ports[0] = cp->dport;
 848         }
 849
 850         /* And finally the ICMP checksum */
 851         icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
 852                                               skb->len - icmp_offset,
 853                                               IPPROTO_ICMPV6, 0);
 854         skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
 855         skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
 856         skb->ip_summed = CHECKSUM_PARTIAL;
 857
 858         if (inout)
 859                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 860                               (void *)ciph - (void *)iph,
 861                               "Forwarding altered outgoing ICMPv6");
 862         else
 863                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 864                               (void *)ciph - (void *)iph,
 865                               "Forwarding altered incoming ICMPv6");
 866 }
 867 #endif
 868
 869 /* Handle relevant response ICMP messages - forward to the right
 870  * destination host.
 871  */
 872 static int handle_response_icmp(int af, struct sk_buff *skb,
 873                                 union nf_inet_addr *snet,
 874                                 __u8 protocol, struct ip_vs_conn *cp,
 875                                 struct ip_vs_protocol *pp,
 876                                 unsigned int offset, unsigned int ihl,
 877                                 unsigned int hooknum)
 878 {
 879         unsigned int verdict = NF_DROP;
 880
 881         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
 882                 goto ignore_cp;
 883
 884         /* Ensure the checksum is correct */
 885         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 886                 /* Failed checksum! */
 887                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
 888                               IP_VS_DBG_ADDR(af, snet));
 889                 goto out;
 890         }
 891
 892         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 893             IPPROTO_SCTP == protocol)
 894                 offset += 2 * sizeof(__u16);
 895         if (!skb_make_writable(skb, offset))
 896                 goto out;
 897
 898 #ifdef CONFIG_IP_VS_IPV6
 899         if (af == AF_INET6)
 900                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
 901         else
 902 #endif
 903                 ip_vs_nat_icmp(skb, pp, cp, 1);
 904
 905         if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
 906                 goto out;
 907
 908         /* do the statistics and put it back */
 909         ip_vs_out_stats(cp, skb);
 910
 911         skb->ipvs_property = 1;
 912         if (!(cp->flags & IP_VS_CONN_F_NFCT))
 913                 ip_vs_notrack(skb);
 914         else
 915                 ip_vs_update_conntrack(skb, cp, 0);
 916
 917 ignore_cp:
 918         verdict = NF_ACCEPT;
 919
 920 out:
 921         __ip_vs_conn_put(cp);
 922
 923         return verdict;
 924 }
 925
 926 /*
 927  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 928  *      Find any that might be relevant, check against existing connections.
 929  *      Currently handles error types - unreachable, quench, ttl exceeded.
 930  */
 931 static int ip_vs_out_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb,
 932                           int *related, unsigned int hooknum)
 933 {
 934         struct iphdr *iph;
 935         struct icmphdr  _icmph, *ic;
 936         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 937         struct ip_vs_iphdr ciph;
 938         struct ip_vs_conn *cp;
 939         struct ip_vs_protocol *pp;
 940         unsigned int offset, ihl;
 941         union nf_inet_addr snet;
 942
 943         *related = 1;
 944
 945         /* reassemble IP fragments */
 946         if (ip_is_fragment(ip_hdr(skb))) {
 947                 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
 948                         return NF_STOLEN;
 949         }
 950
 951         iph = ip_hdr(skb);
 952         offset = ihl = iph->ihl * 4;
 953         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 954         if (ic == NULL)
 955                 return NF_DROP;
 956
 957         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
 958                   ic->type, ntohs(icmp_id(ic)),
 959                   &iph->saddr, &iph->daddr);
 960
 961         /*
 962          * Work through seeing if this is for us.
 963          * These checks are supposed to be in an order that means easy
 964          * things are checked first to speed up processing.... however
 965          * this means that some packets will manage to get a long way
 966          * down this stack and then be rejected, but that's life.
 967          */
 968         if ((ic->type != ICMP_DEST_UNREACH) &&
 969             (ic->type != ICMP_SOURCE_QUENCH) &&
 970             (ic->type != ICMP_TIME_EXCEEDED)) {
 971                 *related = 0;
 972                 return NF_ACCEPT;
 973         }
 974
 975         /* Now find the contained IP header */
 976         offset += sizeof(_icmph);
 977         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 978         if (cih == NULL)
 979                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 980
 981         pp = ip_vs_proto_get(cih->protocol);
 982         if (!pp)
 983                 return NF_ACCEPT;
 984
 985         /* Is the embedded protocol header present? */
 986         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 987                      pp->dont_defrag))
 988                 return NF_ACCEPT;
 989
 990         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
 991                       "Checking outgoing ICMP for");
 992
 993         ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, true, &ciph);
 994
 995         /* The embedded headers contain source and dest in reverse order */
 996         cp = INDIRECT_CALL_1(pp->conn_out_get, ip_vs_conn_out_get_proto,
 997                              ipvs, AF_INET, skb, &ciph);
 998         if (!cp)
 999                 return NF_ACCEPT;
1000
1001         snet.ip = iph->saddr;
1002         return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
1003                                     pp, ciph.len, ihl, hooknum);
1004 }
1005
1006 #ifdef CONFIG_IP_VS_IPV6
1007 static int ip_vs_out_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
1008                              int *related,  unsigned int hooknum,
1009                              struct ip_vs_iphdr *ipvsh)
1010 {
1011         struct icmp6hdr _icmph, *ic;
1012         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1013         struct ip_vs_conn *cp;
1014         struct ip_vs_protocol *pp;
1015         union nf_inet_addr snet;
1016         unsigned int offset;
1017
1018         *related = 1;
1019         ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph);
1020         if (ic == NULL)
1021                 return NF_DROP;
1022
1023         /*
1024          * Work through seeing if this is for us.
1025          * These checks are supposed to be in an order that means easy
1026          * things are checked first to speed up processing.... however
1027          * this means that some packets will manage to get a long way
1028          * down this stack and then be rejected, but that's life.
1029          */
1030         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1031                 *related = 0;
1032                 return NF_ACCEPT;
1033         }
1034         /* Fragment header that is before ICMP header tells us that:
1035          * it's not an error message since they can't be fragmented.
1036          */
1037         if (ipvsh->flags & IP6_FH_F_FRAG)
1038                 return NF_DROP;
1039
1040         IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1041                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1042                   &ipvsh->saddr, &ipvsh->daddr);
1043
1044         if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, ipvsh->len + sizeof(_icmph),
1045                                      true, &ciph))
1046                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1047
1048         pp = ip_vs_proto_get(ciph.protocol);
1049         if (!pp)
1050                 return NF_ACCEPT;
1051
1052         /* The embedded headers contain source and dest in reverse order */
1053         cp = INDIRECT_CALL_1(pp->conn_out_get, ip_vs_conn_out_get_proto,
1054                              ipvs, AF_INET6, skb, &ciph);
1055         if (!cp)
1056                 return NF_ACCEPT;
1057
1058         snet.in6 = ciph.saddr.in6;
1059         offset = ciph.len;
1060         return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
1061                                     pp, offset, sizeof(struct ipv6hdr),
1062                                     hooknum);
1063 }
1064 #endif
1065
1066 /*
1067  * Check if sctp chunc is ABORT chunk
1068  */
1069 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
1070 {
1071         struct sctp_chunkhdr *sch, schunk;
1072         sch = skb_header_pointer(skb, nh_len + sizeof(struct sctphdr),
1073                                  sizeof(schunk), &schunk);
1074         if (sch == NULL)
1075                 return 0;
1076         if (sch->type == SCTP_CID_ABORT)
1077                 return 1;
1078         return 0;
1079 }
1080
1081 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
1082 {
1083         struct tcphdr _tcph, *th;
1084
1085         th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
1086         if (th == NULL)
1087                 return 0;
1088         return th->rst;
1089 }
1090
1091 static inline bool is_new_conn(const struct sk_buff *skb,
1092                                struct ip_vs_iphdr *iph)
1093 {
1094         switch (iph->protocol) {
1095         case IPPROTO_TCP: {
1096                 struct tcphdr _tcph, *th;
1097
1098                 th = skb_header_pointer(skb, iph->len, sizeof(_tcph), &_tcph);
1099                 if (th == NULL)
1100                         return false;
1101                 return th->syn;
1102         }
1103         case IPPROTO_SCTP: {
1104                 struct sctp_chunkhdr *sch, schunk;
1105
1106                 sch = skb_header_pointer(skb, iph->len + sizeof(struct sctphdr),
1107                                          sizeof(schunk), &schunk);
1108                 if (sch == NULL)
1109                         return false;
1110                 return sch->type == SCTP_CID_INIT;
1111         }
1112         default:
1113                 return false;
1114         }
1115 }
1116
1117 static inline bool is_new_conn_expected(const struct ip_vs_conn *cp,
1118                                         int conn_reuse_mode)
1119 {
1120         /* Controlled (FTP DATA or persistence)? */
1121         if (cp->control)
1122                 return false;
1123
1124         switch (cp->protocol) {
1125         case IPPROTO_TCP:
1126                 return (cp->state == IP_VS_TCP_S_TIME_WAIT) ||
1127                        (cp->state == IP_VS_TCP_S_CLOSE) ||
1128                         ((conn_reuse_mode & 2) &&
1129                          (cp->state == IP_VS_TCP_S_FIN_WAIT) &&
1130                          (cp->flags & IP_VS_CONN_F_NOOUTPUT));
1131         case IPPROTO_SCTP:
1132                 return cp->state == IP_VS_SCTP_S_CLOSED;
1133         default:
1134                 return false;
1135         }
1136 }
1137
1138 /* Generic function to create new connections for outgoing RS packets
1139  *
1140  * Pre-requisites for successful connection creation:
1141  * 1) Virtual Service is NOT fwmark based:
1142  *    In fwmark-VS actual vaddr and vport are unknown to IPVS
1143  * 2) Real Server and Virtual Service were NOT configured without port:
1144  *    This is to allow match of different VS to the same RS ip-addr
1145  */
1146 struct ip_vs_conn *ip_vs_new_conn_out(struct ip_vs_service *svc,
1147                                       struct ip_vs_dest *dest,
1148                                       struct sk_buff *skb,
1149                                       const struct ip_vs_iphdr *iph,
1150                                       __be16 dport,
1151                                       __be16 cport)
1152 {
1153         struct ip_vs_conn_param param;
1154         struct ip_vs_conn *ct = NULL, *cp = NULL;
1155         const union nf_inet_addr *vaddr, *daddr, *caddr;
1156         union nf_inet_addr snet;
1157         __be16 vport;
1158         unsigned int flags;
1159
1160         EnterFunction(12);
1161         vaddr = &svc->addr;
1162         vport = svc->port;
1163         daddr = &iph->saddr;
1164         caddr = &iph->daddr;
1165
1166         /* check pre-requisites are satisfied */
1167         if (svc->fwmark)
1168                 return NULL;
1169         if (!vport || !dport)
1170                 return NULL;
1171
1172         /* for persistent service first create connection template */
1173         if (svc->flags & IP_VS_SVC_F_PERSISTENT) {
1174                 /* apply netmask the same way ingress-side does */
1175 #ifdef CONFIG_IP_VS_IPV6
1176                 if (svc->af == AF_INET6)
1177                         ipv6_addr_prefix(&snet.in6, &caddr->in6,
1178                                          (__force __u32)svc->netmask);
1179                 else
1180 #endif
1181                         snet.ip = caddr->ip & svc->netmask;
1182                 /* fill params and create template if not existent */
1183                 if (ip_vs_conn_fill_param_persist(svc, skb, iph->protocol,
1184                                                   &snet, 0, vaddr,
1185                                                   vport, &param) < 0)
1186                         return NULL;
1187                 ct = ip_vs_ct_in_get(&param);
1188                 /* check if template exists and points to the same dest */
1189                 if (!ct || !ip_vs_check_template(ct, dest)) {
1190                         ct = ip_vs_conn_new(&param, dest->af, daddr, dport,
1191                                             IP_VS_CONN_F_TEMPLATE, dest, 0);
1192                         if (!ct) {
1193                                 kfree(param.pe_data);
1194                                 return NULL;
1195                         }
1196                         ct->timeout = svc->timeout;
1197                 } else {
1198                         kfree(param.pe_data);
1199                 }
1200         }
1201
1202         /* connection flags */
1203         flags = ((svc->flags & IP_VS_SVC_F_ONEPACKET) &&
1204                  iph->protocol == IPPROTO_UDP) ? IP_VS_CONN_F_ONE_PACKET : 0;
1205         /* create connection */
1206         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
1207                               caddr, cport, vaddr, vport, &param);
1208         cp = ip_vs_conn_new(&param, dest->af, daddr, dport, flags, dest, 0);
1209         if (!cp) {
1210                 if (ct)
1211                         ip_vs_conn_put(ct);
1212                 return NULL;
1213         }
1214         if (ct) {
1215                 ip_vs_control_add(cp, ct);
1216                 ip_vs_conn_put(ct);
1217         }
1218         ip_vs_conn_stats(cp, svc);
1219
1220         /* return connection (will be used to handle outgoing packet) */
1221         IP_VS_DBG_BUF(6, "New connection RS-initiated:%c c:%s:%u v:%s:%u "
1222                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
1223                       ip_vs_fwd_tag(cp),
1224                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
1225                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
1226                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
1227                       cp->flags, refcount_read(&cp->refcnt));
1228         LeaveFunction(12);
1229         return cp;
1230 }
1231
1232 /* Handle outgoing packets which are considered requests initiated by
1233  * real servers, so that subsequent responses from external client can be
1234  * routed to the right real server.
1235  * Used also for outgoing responses in OPS mode.
1236  *
1237  * Connection management is handled by persistent-engine specific callback.
1238  */
1239 static struct ip_vs_conn *__ip_vs_rs_conn_out(unsigned int hooknum,
1240                                               struct netns_ipvs *ipvs,
1241                                               int af, struct sk_buff *skb,
1242                                               const struct ip_vs_iphdr *iph)
1243 {
1244         struct ip_vs_dest *dest;
1245         struct ip_vs_conn *cp = NULL;
1246         __be16 _ports[2], *pptr;
1247
1248         if (hooknum == NF_INET_LOCAL_IN)
1249                 return NULL;
1250
1251         pptr = frag_safe_skb_hp(skb, iph->len,
1252                                 sizeof(_ports), _ports);
1253         if (!pptr)
1254                 return NULL;
1255
1256         dest = ip_vs_find_real_service(ipvs, af, iph->protocol,
1257                                        &iph->saddr, pptr[0]);
1258         if (dest) {
1259                 struct ip_vs_service *svc;
1260                 struct ip_vs_pe *pe;
1261
1262                 svc = rcu_dereference(dest->svc);
1263                 if (svc) {
1264                         pe = rcu_dereference(svc->pe);
1265                         if (pe && pe->conn_out)
1266                                 cp = pe->conn_out(svc, dest, skb, iph,
1267                                                   pptr[0], pptr[1]);
1268                 }
1269         }
1270
1271         return cp;
1272 }
1273
1274 /* Handle response packets: rewrite addresses and send away...
1275  */
1276 static unsigned int
1277 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1278                 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph,
1279                 unsigned int hooknum)
1280 {
1281         struct ip_vs_protocol *pp = pd->pp;
1282
1283         IP_VS_DBG_PKT(11, af, pp, skb, iph->off, "Outgoing packet");
1284
1285         if (!skb_make_writable(skb, iph->len))
1286                 goto drop;
1287
1288         /* mangle the packet */
1289         if (pp->snat_handler &&
1290             !SNAT_CALL(pp->snat_handler, skb, pp, cp, iph))
1291                 goto drop;
1292
1293 #ifdef CONFIG_IP_VS_IPV6
1294         if (af == AF_INET6)
1295                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1296         else
1297 #endif
1298         {
1299                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1300                 ip_send_check(ip_hdr(skb));
1301         }
1302
1303         /*
1304          * nf_iterate does not expect change in the skb->dst->dev.
1305          * It looks like it is not fatal to enable this code for hooks
1306          * where our handlers are at the end of the chain list and
1307          * when all next handlers use skb->dst->dev and not outdev.
1308          * It will definitely route properly the inout NAT traffic
1309          * when multiple paths are used.
1310          */
1311
1312         /* For policy routing, packets originating from this
1313          * machine itself may be routed differently to packets
1314          * passing through.  We want this packet to be routed as
1315          * if it came from this machine itself.  So re-compute
1316          * the routing information.
1317          */
1318         if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
1319                 goto drop;
1320
1321         IP_VS_DBG_PKT(10, af, pp, skb, iph->off, "After SNAT");
1322
1323         ip_vs_out_stats(cp, skb);
1324         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1325         skb->ipvs_property = 1;
1326         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1327                 ip_vs_notrack(skb);
1328         else
1329                 ip_vs_update_conntrack(skb, cp, 0);
1330         ip_vs_conn_put(cp);
1331
1332         LeaveFunction(11);
1333         return NF_ACCEPT;
1334
1335 drop:
1336         ip_vs_conn_put(cp);
1337         kfree_skb(skb);
1338         LeaveFunction(11);
1339         return NF_STOLEN;
1340 }
1341
1342 /*
1343  *      Check if outgoing packet belongs to the established ip_vs_conn.
1344  */
1345 static unsigned int
1346 ip_vs_out(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1347 {
1348         struct ip_vs_iphdr iph;
1349         struct ip_vs_protocol *pp;
1350         struct ip_vs_proto_data *pd;
1351         struct ip_vs_conn *cp;
1352         struct sock *sk;
1353
1354         EnterFunction(11);
1355
1356         /* Already marked as IPVS request or reply? */
1357         if (skb->ipvs_property)
1358                 return NF_ACCEPT;
1359
1360         sk = skb_to_full_sk(skb);
1361         /* Bad... Do not break raw sockets */
1362         if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
1363                      af == AF_INET)) {
1364
1365                 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
1366                         return NF_ACCEPT;
1367         }
1368
1369         if (unlikely(!skb_dst(skb)))
1370                 return NF_ACCEPT;
1371
1372         if (!ipvs->enable)
1373                 return NF_ACCEPT;
1374
1375         ip_vs_fill_iph_skb(af, skb, false, &iph);
1376 #ifdef CONFIG_IP_VS_IPV6
1377         if (af == AF_INET6) {
1378                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1379                         int related;
1380                         int verdict = ip_vs_out_icmp_v6(ipvs, skb, &related,
1381                                                         hooknum, &iph);
1382
1383                         if (related)
1384                                 return verdict;
1385                 }
1386         } else
1387 #endif
1388                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1389                         int related;
1390                         int verdict = ip_vs_out_icmp(ipvs, skb, &related, hooknum);
1391
1392                         if (related)
1393                                 return verdict;
1394                 }
1395
1396         pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1397         if (unlikely(!pd))
1398                 return NF_ACCEPT;
1399         pp = pd->pp;
1400
1401         /* reassemble IP fragments */
1402 #ifdef CONFIG_IP_VS_IPV6
1403         if (af == AF_INET)
1404 #endif
1405                 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1406                         if (ip_vs_gather_frags(ipvs, skb,
1407                                                ip_vs_defrag_user(hooknum)))
1408                                 return NF_STOLEN;
1409
1410                         ip_vs_fill_iph_skb(AF_INET, skb, false, &iph);
1411                 }
1412
1413         /*
1414          * Check if the packet belongs to an existing entry
1415          */
1416         cp = INDIRECT_CALL_1(pp->conn_out_get, ip_vs_conn_out_get_proto,
1417                              ipvs, af, skb, &iph);
1418
1419         if (likely(cp)) {
1420                 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
1421                         goto ignore_cp;
1422                 return handle_response(af, skb, pd, cp, &iph, hooknum);
1423         }
1424
1425         /* Check for real-server-started requests */
1426         if (atomic_read(&ipvs->conn_out_counter)) {
1427                 /* Currently only for UDP:
1428                  * connection oriented protocols typically use
1429                  * ephemeral ports for outgoing connections, so
1430                  * related incoming responses would not match any VS
1431                  */
1432                 if (pp->protocol == IPPROTO_UDP) {
1433                         cp = __ip_vs_rs_conn_out(hooknum, ipvs, af, skb, &iph);
1434                         if (likely(cp))
1435                                 return handle_response(af, skb, pd, cp, &iph,
1436                                                        hooknum);
1437                 }
1438         }
1439
1440         if (sysctl_nat_icmp_send(ipvs) &&
1441             (pp->protocol == IPPROTO_TCP ||
1442              pp->protocol == IPPROTO_UDP ||
1443              pp->protocol == IPPROTO_SCTP)) {
1444                 __be16 _ports[2], *pptr;
1445
1446                 pptr = frag_safe_skb_hp(skb, iph.len,
1447                                          sizeof(_ports), _ports);
1448                 if (pptr == NULL)
1449                         return NF_ACCEPT;       /* Not for me */
1450                 if (ip_vs_has_real_service(ipvs, af, iph.protocol, &iph.saddr,
1451                                            pptr[0])) {
1452                         /*
1453                          * Notify the real server: there is no
1454                          * existing entry if it is not RST
1455                          * packet or not TCP packet.
1456                          */
1457                         if ((iph.protocol != IPPROTO_TCP &&
1458                              iph.protocol != IPPROTO_SCTP)
1459                              || ((iph.protocol == IPPROTO_TCP
1460                                   && !is_tcp_reset(skb, iph.len))
1461                                  || (iph.protocol == IPPROTO_SCTP
1462                                         && !is_sctp_abort(skb,
1463                                                 iph.len)))) {
1464 #ifdef CONFIG_IP_VS_IPV6
1465                                 if (af == AF_INET6) {
1466                                         if (!skb->dev)
1467                                                 skb->dev = ipvs->net->loopback_dev;
1468                                         icmpv6_send(skb,
1469                                                     ICMPV6_DEST_UNREACH,
1470                                                     ICMPV6_PORT_UNREACH,
1471                                                     0);
1472                                 } else
1473 #endif
1474                                         icmp_send(skb,
1475                                                   ICMP_DEST_UNREACH,
1476                                                   ICMP_PORT_UNREACH, 0);
1477                                 return NF_DROP;
1478                         }
1479                 }
1480         }
1481
1482 out:
1483         IP_VS_DBG_PKT(12, af, pp, skb, iph.off,
1484                       "ip_vs_out: packet continues traversal as normal");
1485         return NF_ACCEPT;
1486
1487 ignore_cp:
1488         __ip_vs_conn_put(cp);
1489         goto out;
1490 }
1491
1492 /*
1493  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1494  *      used only for VS/NAT.
1495  *      Check if packet is reply for established ip_vs_conn.
1496  */
1497 static unsigned int
1498 ip_vs_reply4(void *priv, struct sk_buff *skb,
1499              const struct nf_hook_state *state)
1500 {
1501         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1502 }
1503
1504 /*
1505  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1506  *      Check if packet is reply for established ip_vs_conn.
1507  */
1508 static unsigned int
1509 ip_vs_local_reply4(void *priv, struct sk_buff *skb,
1510                    const struct nf_hook_state *state)
1511 {
1512         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1513 }
1514
1515 #ifdef CONFIG_IP_VS_IPV6
1516
1517 /*
1518  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1519  *      used only for VS/NAT.
1520  *      Check if packet is reply for established ip_vs_conn.
1521  */
1522 static unsigned int
1523 ip_vs_reply6(void *priv, struct sk_buff *skb,
1524              const struct nf_hook_state *state)
1525 {
1526         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1527 }
1528
1529 /*
1530  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1531  *      Check if packet is reply for established ip_vs_conn.
1532  */
1533 static unsigned int
1534 ip_vs_local_reply6(void *priv, struct sk_buff *skb,
1535                    const struct nf_hook_state *state)
1536 {
1537         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1538 }
1539
1540 #endif
1541
1542 static unsigned int
1543 ip_vs_try_to_schedule(struct netns_ipvs *ipvs, int af, struct sk_buff *skb,
1544                       struct ip_vs_proto_data *pd,
1545                       int *verdict, struct ip_vs_conn **cpp,
1546                       struct ip_vs_iphdr *iph)
1547 {
1548         struct ip_vs_protocol *pp = pd->pp;
1549
1550         if (!iph->fragoffs) {
1551                 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1552                  * replayed fragment zero will already have created the cp
1553                  */
1554
1555                 /* Schedule and create new connection entry into cpp */
1556                 if (!pp->conn_schedule(ipvs, af, skb, pd, verdict, cpp, iph))
1557                         return 0;
1558         }
1559
1560         if (unlikely(!*cpp)) {
1561                 /* sorry, all this trouble for a no-hit :) */
1562                 IP_VS_DBG_PKT(12, af, pp, skb, iph->off,
1563                               "ip_vs_in: packet continues traversal as normal");
1564
1565                 /* Fragment couldn't be mapped to a conn entry */
1566                 if (iph->fragoffs)
1567                         IP_VS_DBG_PKT(7, af, pp, skb, iph->off,
1568                                       "unhandled fragment");
1569
1570                 *verdict = NF_ACCEPT;
1571                 return 0;
1572         }
1573
1574         return 1;
1575 }
1576
1577 /*
1578  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1579  *      Find any that might be relevant, check against existing connections,
1580  *      forward to the right destination host if relevant.
1581  *      Currently handles error types - unreachable, quench, ttl exceeded.
1582  */
1583 static int
1584 ip_vs_in_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb, int *related,
1585               unsigned int hooknum)
1586 {
1587         struct iphdr *iph;
1588         struct icmphdr  _icmph, *ic;
1589         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1590         struct ip_vs_iphdr ciph;
1591         struct ip_vs_conn *cp;
1592         struct ip_vs_protocol *pp;
1593         struct ip_vs_proto_data *pd;
1594         unsigned int offset, offset2, ihl, verdict;
1595         bool ipip, new_cp = false;
1596
1597         *related = 1;
1598
1599         /* reassemble IP fragments */
1600         if (ip_is_fragment(ip_hdr(skb))) {
1601                 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
1602                         return NF_STOLEN;
1603         }
1604
1605         iph = ip_hdr(skb);
1606         offset = ihl = iph->ihl * 4;
1607         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1608         if (ic == NULL)
1609                 return NF_DROP;
1610
1611         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1612                   ic->type, ntohs(icmp_id(ic)),
1613                   &iph->saddr, &iph->daddr);
1614
1615         /*
1616          * Work through seeing if this is for us.
1617          * These checks are supposed to be in an order that means easy
1618          * things are checked first to speed up processing.... however
1619          * this means that some packets will manage to get a long way
1620          * down this stack and then be rejected, but that's life.
1621          */
1622         if ((ic->type != ICMP_DEST_UNREACH) &&
1623             (ic->type != ICMP_SOURCE_QUENCH) &&
1624             (ic->type != ICMP_TIME_EXCEEDED)) {
1625                 *related = 0;
1626                 return NF_ACCEPT;
1627         }
1628
1629         /* Now find the contained IP header */
1630         offset += sizeof(_icmph);
1631         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1632         if (cih == NULL)
1633                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1634
1635         /* Special case for errors for IPIP packets */
1636         ipip = false;
1637         if (cih->protocol == IPPROTO_IPIP) {
1638                 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1639                         return NF_ACCEPT;
1640                 /* Error for our IPIP must arrive at LOCAL_IN */
1641                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1642                         return NF_ACCEPT;
1643                 offset += cih->ihl * 4;
1644                 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1645                 if (cih == NULL)
1646                         return NF_ACCEPT; /* The packet looks wrong, ignore */
1647                 ipip = true;
1648         }
1649
1650         pd = ip_vs_proto_data_get(ipvs, cih->protocol);
1651         if (!pd)
1652                 return NF_ACCEPT;
1653         pp = pd->pp;
1654
1655         /* Is the embedded protocol header present? */
1656         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1657                      pp->dont_defrag))
1658                 return NF_ACCEPT;
1659
1660         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1661                       "Checking incoming ICMP for");
1662
1663         offset2 = offset;
1664         ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, !ipip, &ciph);
1665         offset = ciph.len;
1666
1667         /* The embedded headers contain source and dest in reverse order.
1668          * For IPIP this is error for request, not for reply.
1669          */
1670         cp = INDIRECT_CALL_1(pp->conn_in_get, ip_vs_conn_in_get_proto,
1671                              ipvs, AF_INET, skb, &ciph);
1672
1673         if (!cp) {
1674                 int v;
1675
1676                 if (ipip || !sysctl_schedule_icmp(ipvs))
1677                         return NF_ACCEPT;
1678
1679                 if (!ip_vs_try_to_schedule(ipvs, AF_INET, skb, pd, &v, &cp, &ciph))
1680                         return v;
1681                 new_cp = true;
1682         }
1683
1684         verdict = NF_DROP;
1685
1686         /* Ensure the checksum is correct */
1687         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1688                 /* Failed checksum! */
1689                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1690                           &iph->saddr);
1691                 goto out;
1692         }
1693
1694         if (ipip) {
1695                 __be32 info = ic->un.gateway;
1696                 __u8 type = ic->type;
1697                 __u8 code = ic->code;
1698
1699                 /* Update the MTU */
1700                 if (ic->type == ICMP_DEST_UNREACH &&
1701                     ic->code == ICMP_FRAG_NEEDED) {
1702                         struct ip_vs_dest *dest = cp->dest;
1703                         u32 mtu = ntohs(ic->un.frag.mtu);
1704                         __be16 frag_off = cih->frag_off;
1705
1706                         /* Strip outer IP and ICMP, go to IPIP header */
1707                         if (pskb_pull(skb, ihl + sizeof(_icmph)) == NULL)
1708                                 goto ignore_ipip;
1709                         offset2 -= ihl + sizeof(_icmph);
1710                         skb_reset_network_header(skb);
1711                         IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1712                                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1713                         ipv4_update_pmtu(skb, ipvs->net, mtu, 0, 0);
1714                         /* Client uses PMTUD? */
1715                         if (!(frag_off & htons(IP_DF)))
1716                                 goto ignore_ipip;
1717                         /* Prefer the resulting PMTU */
1718                         if (dest) {
1719                                 struct ip_vs_dest_dst *dest_dst;
1720
1721                                 dest_dst = rcu_dereference(dest->dest_dst);
1722                                 if (dest_dst)
1723                                         mtu = dst_mtu(dest_dst->dst_cache);
1724                         }
1725                         if (mtu > 68 + sizeof(struct iphdr))
1726                                 mtu -= sizeof(struct iphdr);
1727                         info = htonl(mtu);
1728                 }
1729                 /* Strip outer IP, ICMP and IPIP, go to IP header of
1730                  * original request.
1731                  */
1732                 if (pskb_pull(skb, offset2) == NULL)
1733                         goto ignore_ipip;
1734                 skb_reset_network_header(skb);
1735                 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1736                         &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1737                         type, code, ntohl(info));
1738                 icmp_send(skb, type, code, info);
1739                 /* ICMP can be shorter but anyways, account it */
1740                 ip_vs_out_stats(cp, skb);
1741
1742 ignore_ipip:
1743                 consume_skb(skb);
1744                 verdict = NF_STOLEN;
1745                 goto out;
1746         }
1747
1748         /* do the statistics and put it back */
1749         ip_vs_in_stats(cp, skb);
1750         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol ||
1751             IPPROTO_SCTP == cih->protocol)
1752                 offset += 2 * sizeof(__u16);
1753         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1754
1755 out:
1756         if (likely(!new_cp))
1757                 __ip_vs_conn_put(cp);
1758         else
1759                 ip_vs_conn_put(cp);
1760
1761         return verdict;
1762 }
1763
1764 #ifdef CONFIG_IP_VS_IPV6
1765 static int ip_vs_in_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
1766                             int *related, unsigned int hooknum,
1767                             struct ip_vs_iphdr *iph)
1768 {
1769         struct icmp6hdr _icmph, *ic;
1770         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1771         struct ip_vs_conn *cp;
1772         struct ip_vs_protocol *pp;
1773         struct ip_vs_proto_data *pd;
1774         unsigned int offset, verdict;
1775         bool new_cp = false;
1776
1777         *related = 1;
1778
1779         ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph);
1780         if (ic == NULL)
1781                 return NF_DROP;
1782
1783         /*
1784          * Work through seeing if this is for us.
1785          * These checks are supposed to be in an order that means easy
1786          * things are checked first to speed up processing.... however
1787          * this means that some packets will manage to get a long way
1788          * down this stack and then be rejected, but that's life.
1789          */
1790         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1791                 *related = 0;
1792                 return NF_ACCEPT;
1793         }
1794         /* Fragment header that is before ICMP header tells us that:
1795          * it's not an error message since they can't be fragmented.
1796          */
1797         if (iph->flags & IP6_FH_F_FRAG)
1798                 return NF_DROP;
1799
1800         IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1801                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1802                   &iph->saddr, &iph->daddr);
1803
1804         offset = iph->len + sizeof(_icmph);
1805         if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, offset, true, &ciph))
1806                 return NF_ACCEPT;
1807
1808         pd = ip_vs_proto_data_get(ipvs, ciph.protocol);
1809         if (!pd)
1810                 return NF_ACCEPT;
1811         pp = pd->pp;
1812
1813         /* Cannot handle fragmented embedded protocol */
1814         if (ciph.fragoffs)
1815                 return NF_ACCEPT;
1816
1817         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
1818                       "Checking incoming ICMPv6 for");
1819
1820         /* The embedded headers contain source and dest in reverse order
1821          * if not from localhost
1822          */
1823         cp = INDIRECT_CALL_1(pp->conn_in_get, ip_vs_conn_in_get_proto,
1824                              ipvs, AF_INET6, skb, &ciph);
1825
1826         if (!cp) {
1827                 int v;
1828
1829                 if (!sysctl_schedule_icmp(ipvs))
1830                         return NF_ACCEPT;
1831
1832                 if (!ip_vs_try_to_schedule(ipvs, AF_INET6, skb, pd, &v, &cp, &ciph))
1833                         return v;
1834
1835                 new_cp = true;
1836         }
1837
1838         /* VS/TUN, VS/DR and LOCALNODE just let it go */
1839         if ((hooknum == NF_INET_LOCAL_OUT) &&
1840             (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1841                 verdict = NF_ACCEPT;
1842                 goto out;
1843         }
1844
1845         /* do the statistics and put it back */
1846         ip_vs_in_stats(cp, skb);
1847
1848         /* Need to mangle contained IPv6 header in ICMPv6 packet */
1849         offset = ciph.len;
1850         if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1851             IPPROTO_SCTP == ciph.protocol)
1852                 offset += 2 * sizeof(__u16); /* Also mangle ports */
1853
1854         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset, hooknum, &ciph);
1855
1856 out:
1857         if (likely(!new_cp))
1858                 __ip_vs_conn_put(cp);
1859         else
1860                 ip_vs_conn_put(cp);
1861
1862         return verdict;
1863 }
1864 #endif
1865
1866
1867 /*
1868  *      Check if it's for virtual services, look it up,
1869  *      and send it on its way...
1870  */
1871 static unsigned int
1872 ip_vs_in(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1873 {
1874         struct ip_vs_iphdr iph;
1875         struct ip_vs_protocol *pp;
1876         struct ip_vs_proto_data *pd;
1877         struct ip_vs_conn *cp;
1878         int ret, pkts;
1879         int conn_reuse_mode;
1880         struct sock *sk;
1881
1882         /* Already marked as IPVS request or reply? */
1883         if (skb->ipvs_property)
1884                 return NF_ACCEPT;
1885
1886         /*
1887          *      Big tappo:
1888          *      - remote client: only PACKET_HOST
1889          *      - route: used for struct net when skb->dev is unset
1890          */
1891         if (unlikely((skb->pkt_type != PACKET_HOST &&
1892                       hooknum != NF_INET_LOCAL_OUT) ||
1893                      !skb_dst(skb))) {
1894                 ip_vs_fill_iph_skb(af, skb, false, &iph);
1895                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1896                               " ignored in hook %u\n",
1897                               skb->pkt_type, iph.protocol,
1898                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1899                 return NF_ACCEPT;
1900         }
1901         /* ipvs enabled in this netns ? */
1902         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1903                 return NF_ACCEPT;
1904
1905         ip_vs_fill_iph_skb(af, skb, false, &iph);
1906
1907         /* Bad... Do not break raw sockets */
1908         sk = skb_to_full_sk(skb);
1909         if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
1910                      af == AF_INET)) {
1911
1912                 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
1913                         return NF_ACCEPT;
1914         }
1915
1916 #ifdef CONFIG_IP_VS_IPV6
1917         if (af == AF_INET6) {
1918                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1919                         int related;
1920                         int verdict = ip_vs_in_icmp_v6(ipvs, skb, &related,
1921                                                        hooknum, &iph);
1922
1923                         if (related)
1924                                 return verdict;
1925                 }
1926         } else
1927 #endif
1928                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1929                         int related;
1930                         int verdict = ip_vs_in_icmp(ipvs, skb, &related,
1931                                                     hooknum);
1932
1933                         if (related)
1934                                 return verdict;
1935                 }
1936
1937         /* Protocol supported? */
1938         pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1939         if (unlikely(!pd)) {
1940                 /* The only way we'll see this packet again is if it's
1941                  * encapsulated, so mark it with ipvs_property=1 so we
1942                  * skip it if we're ignoring tunneled packets
1943                  */
1944                 if (sysctl_ignore_tunneled(ipvs))
1945                         skb->ipvs_property = 1;
1946
1947                 return NF_ACCEPT;
1948         }
1949         pp = pd->pp;
1950         /*
1951          * Check if the packet belongs to an existing connection entry
1952          */
1953         cp = INDIRECT_CALL_1(pp->conn_in_get, ip_vs_conn_in_get_proto,
1954                              ipvs, af, skb, &iph);
1955
1956         conn_reuse_mode = sysctl_conn_reuse_mode(ipvs);
1957         if (conn_reuse_mode && !iph.fragoffs && is_new_conn(skb, &iph) && cp) {
1958                 bool uses_ct = false, resched = false;
1959
1960                 if (unlikely(sysctl_expire_nodest_conn(ipvs)) && cp->dest &&
1961                     unlikely(!atomic_read(&cp->dest->weight))) {
1962                         resched = true;
1963                         uses_ct = ip_vs_conn_uses_conntrack(cp, skb);
1964                 } else if (is_new_conn_expected(cp, conn_reuse_mode)) {
1965                         uses_ct = ip_vs_conn_uses_conntrack(cp, skb);
1966                         if (!atomic_read(&cp->n_control)) {
1967                                 resched = true;
1968                         } else {
1969                                 /* Do not reschedule controlling connection
1970                                  * that uses conntrack while it is still
1971                                  * referenced by controlled connection(s).
1972                                  */
1973                                 resched = !uses_ct;
1974                         }
1975                 }
1976
1977                 if (resched) {
1978                         if (!atomic_read(&cp->n_control))
1979                                 ip_vs_conn_expire_now(cp);
1980                         __ip_vs_conn_put(cp);
1981                         if (uses_ct)
1982                                 return NF_DROP;
1983                         cp = NULL;
1984                 }
1985         }
1986
1987         if (unlikely(!cp)) {
1988                 int v;
1989
1990                 if (!ip_vs_try_to_schedule(ipvs, af, skb, pd, &v, &cp, &iph))
1991                         return v;
1992         }
1993
1994         IP_VS_DBG_PKT(11, af, pp, skb, iph.off, "Incoming packet");
1995
1996         /* Check the server status */
1997         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1998                 /* the destination server is not available */
1999
2000                 __u32 flags = cp->flags;
2001
2002                 /* when timer already started, silently drop the packet.*/
2003                 if (timer_pending(&cp->timer))
2004                         __ip_vs_conn_put(cp);
2005                 else
2006                         ip_vs_conn_put(cp);
2007
2008                 if (sysctl_expire_nodest_conn(ipvs) &&
2009                     !(flags & IP_VS_CONN_F_ONE_PACKET)) {
2010                         /* try to expire the connection immediately */
2011                         ip_vs_conn_expire_now(cp);
2012                 }
2013
2014                 return NF_DROP;
2015         }
2016
2017         ip_vs_in_stats(cp, skb);
2018         ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
2019         if (cp->packet_xmit)
2020                 ret = cp->packet_xmit(skb, cp, pp, &iph);
2021                 /* do not touch skb anymore */
2022         else {
2023                 IP_VS_DBG_RL("warning: packet_xmit is null");
2024                 ret = NF_ACCEPT;
2025         }
2026
2027         /* Increase its packet counter and check if it is needed
2028          * to be synchronized
2029          *
2030          * Sync connection if it is about to close to
2031          * encorage the standby servers to update the connections timeout
2032          *
2033          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
2034          */
2035
2036         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
2037                 pkts = sysctl_sync_threshold(ipvs);
2038         else
2039                 pkts = atomic_add_return(1, &cp->in_pkts);
2040
2041         if (ipvs->sync_state & IP_VS_STATE_MASTER)
2042                 ip_vs_sync_conn(ipvs, cp, pkts);
2043         else if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) && cp->control)
2044                 /* increment is done inside ip_vs_sync_conn too */
2045                 atomic_inc(&cp->control->in_pkts);
2046
2047         ip_vs_conn_put(cp);
2048         return ret;
2049 }
2050
2051 /*
2052  *      AF_INET handler in NF_INET_LOCAL_IN chain
2053  *      Schedule and forward packets from remote clients
2054  */
2055 static unsigned int
2056 ip_vs_remote_request4(void *priv, struct sk_buff *skb,
2057                       const struct nf_hook_state *state)
2058 {
2059         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
2060 }
2061
2062 /*
2063  *      AF_INET handler in NF_INET_LOCAL_OUT chain
2064  *      Schedule and forward packets from local clients
2065  */
2066 static unsigned int
2067 ip_vs_local_request4(void *priv, struct sk_buff *skb,
2068                      const struct nf_hook_state *state)
2069 {
2070         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
2071 }
2072
2073 #ifdef CONFIG_IP_VS_IPV6
2074
2075 /*
2076  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
2077  *      Schedule and forward packets from remote clients
2078  */
2079 static unsigned int
2080 ip_vs_remote_request6(void *priv, struct sk_buff *skb,
2081                       const struct nf_hook_state *state)
2082 {
2083         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
2084 }
2085
2086 /*
2087  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
2088  *      Schedule and forward packets from local clients
2089  */
2090 static unsigned int
2091 ip_vs_local_request6(void *priv, struct sk_buff *skb,
2092                      const struct nf_hook_state *state)
2093 {
2094         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
2095 }
2096
2097 #endif
2098
2099
2100 /*
2101  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
2102  *      related packets destined for 0.0.0.0/0.
2103  *      When fwmark-based virtual service is used, such as transparent
2104  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
2105  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
2106  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
2107  *      and send them to ip_vs_in_icmp.
2108  */
2109 static unsigned int
2110 ip_vs_forward_icmp(void *priv, struct sk_buff *skb,
2111                    const struct nf_hook_state *state)
2112 {
2113         int r;
2114         struct netns_ipvs *ipvs = net_ipvs(state->net);
2115
2116         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
2117                 return NF_ACCEPT;
2118
2119         /* ipvs enabled in this netns ? */
2120         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
2121                 return NF_ACCEPT;
2122
2123         return ip_vs_in_icmp(ipvs, skb, &r, state->hook);
2124 }
2125
2126 #ifdef CONFIG_IP_VS_IPV6
2127 static unsigned int
2128 ip_vs_forward_icmp_v6(void *priv, struct sk_buff *skb,
2129                       const struct nf_hook_state *state)
2130 {
2131         int r;
2132         struct netns_ipvs *ipvs = net_ipvs(state->net);
2133         struct ip_vs_iphdr iphdr;
2134
2135         ip_vs_fill_iph_skb(AF_INET6, skb, false, &iphdr);
2136         if (iphdr.protocol != IPPROTO_ICMPV6)
2137                 return NF_ACCEPT;
2138
2139         /* ipvs enabled in this netns ? */
2140         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
2141                 return NF_ACCEPT;
2142
2143         return ip_vs_in_icmp_v6(ipvs, skb, &r, state->hook, &iphdr);
2144 }
2145 #endif
2146
2147
2148 static const struct nf_hook_ops ip_vs_ops[] = {
2149         /* After packet filtering, change source only for VS/NAT */
2150         {
2151                 .hook           = ip_vs_reply4,
2152                 .pf             = NFPROTO_IPV4,
2153                 .hooknum        = NF_INET_LOCAL_IN,
2154                 .priority       = NF_IP_PRI_NAT_SRC - 2,
2155         },
2156         /* After packet filtering, forward packet through VS/DR, VS/TUN,
2157          * or VS/NAT(change destination), so that filtering rules can be
2158          * applied to IPVS. */
2159         {
2160                 .hook           = ip_vs_remote_request4,
2161                 .pf             = NFPROTO_IPV4,
2162                 .hooknum        = NF_INET_LOCAL_IN,
2163                 .priority       = NF_IP_PRI_NAT_SRC - 1,
2164         },
2165         /* Before ip_vs_in, change source only for VS/NAT */
2166         {
2167                 .hook           = ip_vs_local_reply4,
2168                 .pf             = NFPROTO_IPV4,
2169                 .hooknum        = NF_INET_LOCAL_OUT,
2170                 .priority       = NF_IP_PRI_NAT_DST + 1,
2171         },
2172         /* After mangle, schedule and forward local requests */
2173         {
2174                 .hook           = ip_vs_local_request4,
2175                 .pf             = NFPROTO_IPV4,
2176                 .hooknum        = NF_INET_LOCAL_OUT,
2177                 .priority       = NF_IP_PRI_NAT_DST + 2,
2178         },
2179         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2180          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2181         {
2182                 .hook           = ip_vs_forward_icmp,
2183                 .pf             = NFPROTO_IPV4,
2184                 .hooknum        = NF_INET_FORWARD,
2185                 .priority       = 99,
2186         },
2187         /* After packet filtering, change source only for VS/NAT */
2188         {
2189                 .hook           = ip_vs_reply4,
2190                 .pf             = NFPROTO_IPV4,
2191                 .hooknum        = NF_INET_FORWARD,
2192                 .priority       = 100,
2193         },
2194 #ifdef CONFIG_IP_VS_IPV6
2195         /* After packet filtering, change source only for VS/NAT */
2196         {
2197                 .hook           = ip_vs_reply6,
2198                 .pf             = NFPROTO_IPV6,
2199                 .hooknum        = NF_INET_LOCAL_IN,
2200                 .priority       = NF_IP6_PRI_NAT_SRC - 2,
2201         },
2202         /* After packet filtering, forward packet through VS/DR, VS/TUN,
2203          * or VS/NAT(change destination), so that filtering rules can be
2204          * applied to IPVS. */
2205         {
2206                 .hook           = ip_vs_remote_request6,
2207                 .pf             = NFPROTO_IPV6,
2208                 .hooknum        = NF_INET_LOCAL_IN,
2209                 .priority       = NF_IP6_PRI_NAT_SRC - 1,
2210         },
2211         /* Before ip_vs_in, change source only for VS/NAT */
2212         {
2213                 .hook           = ip_vs_local_reply6,
2214                 .pf             = NFPROTO_IPV6,
2215                 .hooknum        = NF_INET_LOCAL_OUT,
2216                 .priority       = NF_IP6_PRI_NAT_DST + 1,
2217         },
2218         /* After mangle, schedule and forward local requests */
2219         {
2220                 .hook           = ip_vs_local_request6,
2221                 .pf             = NFPROTO_IPV6,
2222                 .hooknum        = NF_INET_LOCAL_OUT,
2223                 .priority       = NF_IP6_PRI_NAT_DST + 2,
2224         },
2225         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2226          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2227         {
2228                 .hook           = ip_vs_forward_icmp_v6,
2229                 .pf             = NFPROTO_IPV6,
2230                 .hooknum        = NF_INET_FORWARD,
2231                 .priority       = 99,
2232         },
2233         /* After packet filtering, change source only for VS/NAT */
2234         {
2235                 .hook           = ip_vs_reply6,
2236                 .pf             = NFPROTO_IPV6,
2237                 .hooknum        = NF_INET_FORWARD,
2238                 .priority       = 100,
2239         },
2240 #endif
2241 };
2242 /*
2243  *      Initialize IP Virtual Server netns mem.
2244  */
2245 static int __net_init __ip_vs_init(struct net *net)
2246 {
2247         struct netns_ipvs *ipvs;
2248         int ret;
2249
2250         ipvs = net_generic(net, ip_vs_net_id);
2251         if (ipvs == NULL)
2252                 return -ENOMEM;
2253
2254         /* Hold the beast until a service is registerd */
2255         ipvs->enable = 0;
2256         ipvs->net = net;
2257         /* Counters used for creating unique names */
2258         ipvs->gen = atomic_read(&ipvs_netns_cnt);
2259         atomic_inc(&ipvs_netns_cnt);
2260         net->ipvs = ipvs;
2261
2262         if (ip_vs_estimator_net_init(ipvs) < 0)
2263                 goto estimator_fail;
2264
2265         if (ip_vs_control_net_init(ipvs) < 0)
2266                 goto control_fail;
2267
2268         if (ip_vs_protocol_net_init(ipvs) < 0)
2269                 goto protocol_fail;
2270
2271         if (ip_vs_app_net_init(ipvs) < 0)
2272                 goto app_fail;
2273
2274         if (ip_vs_conn_net_init(ipvs) < 0)
2275                 goto conn_fail;
2276
2277         if (ip_vs_sync_net_init(ipvs) < 0)
2278                 goto sync_fail;
2279
2280         ret = nf_register_net_hooks(net, ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2281         if (ret < 0)
2282                 goto hook_fail;
2283
2284         return 0;
2285 /*
2286  * Error handling
2287  */
2288
2289 hook_fail:
2290         ip_vs_sync_net_cleanup(ipvs);
2291 sync_fail:
2292         ip_vs_conn_net_cleanup(ipvs);
2293 conn_fail:
2294         ip_vs_app_net_cleanup(ipvs);
2295 app_fail:
2296         ip_vs_protocol_net_cleanup(ipvs);
2297 protocol_fail:
2298         ip_vs_control_net_cleanup(ipvs);
2299 control_fail:
2300         ip_vs_estimator_net_cleanup(ipvs);
2301 estimator_fail:
2302         net->ipvs = NULL;
2303         return -ENOMEM;
2304 }
2305
2306 static void __net_exit __ip_vs_cleanup(struct net *net)
2307 {
2308         struct netns_ipvs *ipvs = net_ipvs(net);
2309
2310         ip_vs_service_net_cleanup(ipvs);        /* ip_vs_flush() with locks */
2311         ip_vs_conn_net_cleanup(ipvs);
2312         ip_vs_app_net_cleanup(ipvs);
2313         ip_vs_protocol_net_cleanup(ipvs);
2314         ip_vs_control_net_cleanup(ipvs);
2315         ip_vs_estimator_net_cleanup(ipvs);
2316         IP_VS_DBG(2, "ipvs netns %d released\n", ipvs->gen);
2317         net->ipvs = NULL;
2318 }
2319
2320 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
2321 {
2322         struct netns_ipvs *ipvs = net_ipvs(net);
2323         EnterFunction(2);
2324         nf_unregister_net_hooks(net, ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2325         ipvs->enable = 0;       /* Disable packet reception */
2326         smp_wmb();
2327         ip_vs_sync_net_cleanup(ipvs);
2328         LeaveFunction(2);
2329 }
2330
2331 static struct pernet_operations ipvs_core_ops = {
2332         .init = __ip_vs_init,
2333         .exit = __ip_vs_cleanup,
2334         .id   = &ip_vs_net_id,
2335         .size = sizeof(struct netns_ipvs),
2336 };
2337
2338 static struct pernet_operations ipvs_core_dev_ops = {
2339         .exit = __ip_vs_dev_cleanup,
2340 };
2341
2342 /*
2343  *      Initialize IP Virtual Server
2344  */
2345 static int __init ip_vs_init(void)
2346 {
2347         int ret;
2348
2349         ret = ip_vs_control_init();
2350         if (ret < 0) {
2351                 pr_err("can't setup control.\n");
2352                 goto exit;
2353         }
2354
2355         ip_vs_protocol_init();
2356
2357         ret = ip_vs_conn_init();
2358         if (ret < 0) {
2359                 pr_err("can't setup connection table.\n");
2360                 goto cleanup_protocol;
2361         }
2362
2363         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
2364         if (ret < 0)
2365                 goto cleanup_conn;
2366
2367         ret = register_pernet_device(&ipvs_core_dev_ops);
2368         if (ret < 0)
2369                 goto cleanup_sub;
2370
2371         ret = ip_vs_register_nl_ioctl();
2372         if (ret < 0) {
2373                 pr_err("can't register netlink/ioctl.\n");
2374                 goto cleanup_dev;
2375         }
2376
2377         pr_info("ipvs loaded.\n");
2378
2379         return ret;
2380
2381 cleanup_dev:
2382         unregister_pernet_device(&ipvs_core_dev_ops);
2383 cleanup_sub:
2384         unregister_pernet_subsys(&ipvs_core_ops);
2385 cleanup_conn:
2386         ip_vs_conn_cleanup();
2387 cleanup_protocol:
2388         ip_vs_protocol_cleanup();
2389         ip_vs_control_cleanup();
2390 exit:
2391         return ret;
2392 }
2393
2394 static void __exit ip_vs_cleanup(void)
2395 {
2396         ip_vs_unregister_nl_ioctl();
2397         unregister_pernet_device(&ipvs_core_dev_ops);
2398         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
2399         ip_vs_conn_cleanup();
2400         ip_vs_protocol_cleanup();
2401         ip_vs_control_cleanup();
2402         pr_info("ipvs unloaded.\n");
2403 }
2404
2405 module_init(ip_vs_init);
2406 module_exit(ip_vs_cleanup);
2407 MODULE_LICENSE("GPL");