net/netfilter/ipvs/ip_vs_core.c

   1 /*
   2  * IPVS         An implementation of the IP virtual server support for the
   3  *              LINUX operating system.  IPVS is now implemented as a module
   4  *              over the Netfilter framework. IPVS can be used to build a
   5  *              high-performance and highly available server based on a
   6  *              cluster of servers.
   7  *
   8  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
   9  *              Peter Kese <peter.kese@ijs.si>
  10  *              Julian Anastasov <ja@ssi.bg>
  11  *
  12  *              This program is free software; you can redistribute it and/or
  13  *              modify it under the terms of the GNU General Public License
  14  *              as published by the Free Software Foundation; either version
  15  *              2 of the License, or (at your option) any later version.
  16  *
  17  * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
  18  * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
  19  * and others.
  20  *
  21  * Changes:
  22  *      Paul `Rusty' Russell            properly handle non-linear skbs
  23  *      Harald Welte                    don't use nfcache
  24  *
  25  */
  26
  27 #define KMSG_COMPONENT "IPVS"
  28 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  29
  30 #include <linux/module.h>
  31 #include <linux/kernel.h>
  32 #include <linux/ip.h>
  33 #include <linux/tcp.h>
  34 #include <linux/sctp.h>
  35 #include <linux/icmp.h>
  36 #include <linux/slab.h>
  37
  38 #include <net/ip.h>
  39 #include <net/tcp.h>
  40 #include <net/udp.h>
  41 #include <net/icmp.h>                   /* for icmp_send */
  42 #include <net/route.h>
  43 #include <net/ip6_checksum.h>
  44 #include <net/netns/generic.h>          /* net_generic() */
  45
  46 #include <linux/netfilter.h>
  47 #include <linux/netfilter_ipv4.h>
  48
  49 #ifdef CONFIG_IP_VS_IPV6
  50 #include <net/ipv6.h>
  51 #include <linux/netfilter_ipv6.h>
  52 #include <net/ip6_route.h>
  53 #endif
  54
  55 #include <net/ip_vs.h>
  56 #include <linux/indirect_call_wrapper.h>
  57
  58
  59 EXPORT_SYMBOL(register_ip_vs_scheduler);
  60 EXPORT_SYMBOL(unregister_ip_vs_scheduler);
  61 EXPORT_SYMBOL(ip_vs_proto_name);
  62 EXPORT_SYMBOL(ip_vs_conn_new);
  63 EXPORT_SYMBOL(ip_vs_conn_in_get);
  64 EXPORT_SYMBOL(ip_vs_conn_out_get);
  65 #ifdef CONFIG_IP_VS_PROTO_TCP
  66 EXPORT_SYMBOL(ip_vs_tcp_conn_listen);
  67 #endif
  68 EXPORT_SYMBOL(ip_vs_conn_put);
  69 #ifdef CONFIG_IP_VS_DEBUG
  70 EXPORT_SYMBOL(ip_vs_get_debug_level);
  71 #endif
  72 EXPORT_SYMBOL(ip_vs_new_conn_out);
  73
  74 #ifdef CONFIG_IP_VS_PROTO_TCP
  75 INDIRECT_CALLABLE_DECLARE(int
  76         tcp_snat_handler(struct sk_buff *skb, struct ip_vs_protocol *pp,
  77                          struct ip_vs_conn *cp, struct ip_vs_iphdr *iph));
  78 #endif
  79
  80 #ifdef CONFIG_IP_VS_PROTO_UDP
  81 INDIRECT_CALLABLE_DECLARE(int
  82         udp_snat_handler(struct sk_buff *skb, struct ip_vs_protocol *pp,
  83                          struct ip_vs_conn *cp, struct ip_vs_iphdr *iph));
  84 #endif
  85
  86 #if defined(CONFIG_IP_VS_PROTO_TCP) && defined(CONFIG_IP_VS_PROTO_UDP)
  87 #define SNAT_CALL(f, ...) \
  88         INDIRECT_CALL_2(f, tcp_snat_handler, udp_snat_handler, __VA_ARGS__)
  89 #elif defined(CONFIG_IP_VS_PROTO_TCP)
  90 #define SNAT_CALL(f, ...) INDIRECT_CALL_1(f, tcp_snat_handler, __VA_ARGS__)
  91 #elif defined(CONFIG_IP_VS_PROTO_UDP)
  92 #define SNAT_CALL(f, ...) INDIRECT_CALL_1(f, udp_snat_handler, __VA_ARGS__)
  93 #else
  94 #define SNAT_CALL(f, ...) f(__VA_ARGS__)
  95 #endif
  96
  97 static unsigned int ip_vs_net_id __read_mostly;
  98 /* netns cnt used for uniqueness */
  99 static atomic_t ipvs_netns_cnt = ATOMIC_INIT(0);
 100
 101 /* ID used in ICMP lookups */
 102 #define icmp_id(icmph)          (((icmph)->un).echo.id)
 103 #define icmpv6_id(icmph)        (icmph->icmp6_dataun.u_echo.identifier)
 104
 105 const char *ip_vs_proto_name(unsigned int proto)
 106 {
 107         static char buf[20];
 108
 109         switch (proto) {
 110         case IPPROTO_IP:
 111                 return "IP";
 112         case IPPROTO_UDP:
 113                 return "UDP";
 114         case IPPROTO_TCP:
 115                 return "TCP";
 116         case IPPROTO_SCTP:
 117                 return "SCTP";
 118         case IPPROTO_ICMP:
 119                 return "ICMP";
 120 #ifdef CONFIG_IP_VS_IPV6
 121         case IPPROTO_ICMPV6:
 122                 return "ICMPv6";
 123 #endif
 124         default:
 125                 sprintf(buf, "IP_%u", proto);
 126                 return buf;
 127         }
 128 }
 129
 130 void ip_vs_init_hash_table(struct list_head *table, int rows)
 131 {
 132         while (--rows >= 0)
 133                 INIT_LIST_HEAD(&table[rows]);
 134 }
 135
 136 static inline void
 137 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 138 {
 139         struct ip_vs_dest *dest = cp->dest;
 140         struct netns_ipvs *ipvs = cp->ipvs;
 141
 142         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 143                 struct ip_vs_cpu_stats *s;
 144                 struct ip_vs_service *svc;
 145
 146                 local_bh_disable();
 147
 148                 s = this_cpu_ptr(dest->stats.cpustats);
 149                 u64_stats_update_begin(&s->syncp);
 150                 s->cnt.inpkts++;
 151                 s->cnt.inbytes += skb->len;
 152                 u64_stats_update_end(&s->syncp);
 153
 154                 svc = rcu_dereference(dest->svc);
 155                 s = this_cpu_ptr(svc->stats.cpustats);
 156                 u64_stats_update_begin(&s->syncp);
 157                 s->cnt.inpkts++;
 158                 s->cnt.inbytes += skb->len;
 159                 u64_stats_update_end(&s->syncp);
 160
 161                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 162                 u64_stats_update_begin(&s->syncp);
 163                 s->cnt.inpkts++;
 164                 s->cnt.inbytes += skb->len;
 165                 u64_stats_update_end(&s->syncp);
 166
 167                 local_bh_enable();
 168         }
 169 }
 170
 171
 172 static inline void
 173 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 174 {
 175         struct ip_vs_dest *dest = cp->dest;
 176         struct netns_ipvs *ipvs = cp->ipvs;
 177
 178         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 179                 struct ip_vs_cpu_stats *s;
 180                 struct ip_vs_service *svc;
 181
 182                 local_bh_disable();
 183
 184                 s = this_cpu_ptr(dest->stats.cpustats);
 185                 u64_stats_update_begin(&s->syncp);
 186                 s->cnt.outpkts++;
 187                 s->cnt.outbytes += skb->len;
 188                 u64_stats_update_end(&s->syncp);
 189
 190                 svc = rcu_dereference(dest->svc);
 191                 s = this_cpu_ptr(svc->stats.cpustats);
 192                 u64_stats_update_begin(&s->syncp);
 193                 s->cnt.outpkts++;
 194                 s->cnt.outbytes += skb->len;
 195                 u64_stats_update_end(&s->syncp);
 196
 197                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 198                 u64_stats_update_begin(&s->syncp);
 199                 s->cnt.outpkts++;
 200                 s->cnt.outbytes += skb->len;
 201                 u64_stats_update_end(&s->syncp);
 202
 203                 local_bh_enable();
 204         }
 205 }
 206
 207
 208 static inline void
 209 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
 210 {
 211         struct netns_ipvs *ipvs = svc->ipvs;
 212         struct ip_vs_cpu_stats *s;
 213
 214         local_bh_disable();
 215
 216         s = this_cpu_ptr(cp->dest->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(svc->stats.cpustats);
 222         u64_stats_update_begin(&s->syncp);
 223         s->cnt.conns++;
 224         u64_stats_update_end(&s->syncp);
 225
 226         s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 227         u64_stats_update_begin(&s->syncp);
 228         s->cnt.conns++;
 229         u64_stats_update_end(&s->syncp);
 230
 231         local_bh_enable();
 232 }
 233
 234
 235 static inline void
 236 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
 237                 const struct sk_buff *skb,
 238                 struct ip_vs_proto_data *pd)
 239 {
 240         if (likely(pd->pp->state_transition))
 241                 pd->pp->state_transition(cp, direction, skb, pd);
 242 }
 243
 244 static inline int
 245 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
 246                               struct sk_buff *skb, int protocol,
 247                               const union nf_inet_addr *caddr, __be16 cport,
 248                               const union nf_inet_addr *vaddr, __be16 vport,
 249                               struct ip_vs_conn_param *p)
 250 {
 251         ip_vs_conn_fill_param(svc->ipvs, svc->af, protocol, caddr, cport, vaddr,
 252                               vport, p);
 253         p->pe = rcu_dereference(svc->pe);
 254         if (p->pe && p->pe->fill_param)
 255                 return p->pe->fill_param(p, skb);
 256
 257         return 0;
 258 }
 259
 260 /*
 261  *  IPVS persistent scheduling function
 262  *  It creates a connection entry according to its template if exists,
 263  *  or selects a server and creates a connection entry plus a template.
 264  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
 265  *  Protocols supported: TCP, UDP
 266  */
 267 static struct ip_vs_conn *
 268 ip_vs_sched_persist(struct ip_vs_service *svc,
 269                     struct sk_buff *skb, __be16 src_port, __be16 dst_port,
 270                     int *ignored, struct ip_vs_iphdr *iph)
 271 {
 272         struct ip_vs_conn *cp = NULL;
 273         struct ip_vs_dest *dest;
 274         struct ip_vs_conn *ct;
 275         __be16 dport = 0;               /* destination port to forward */
 276         unsigned int flags;
 277         struct ip_vs_conn_param param;
 278         const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
 279         union nf_inet_addr snet;        /* source network of the client,
 280                                            after masking */
 281         const union nf_inet_addr *src_addr, *dst_addr;
 282
 283         if (likely(!ip_vs_iph_inverse(iph))) {
 284                 src_addr = &iph->saddr;
 285                 dst_addr = &iph->daddr;
 286         } else {
 287                 src_addr = &iph->daddr;
 288                 dst_addr = &iph->saddr;
 289         }
 290
 291
 292         /* Mask saddr with the netmask to adjust template granularity */
 293 #ifdef CONFIG_IP_VS_IPV6
 294         if (svc->af == AF_INET6)
 295                 ipv6_addr_prefix(&snet.in6, &src_addr->in6,
 296                                  (__force __u32) svc->netmask);
 297         else
 298 #endif
 299                 snet.ip = src_addr->ip & svc->netmask;
 300
 301         IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
 302                       "mnet %s\n",
 303                       IP_VS_DBG_ADDR(svc->af, src_addr), ntohs(src_port),
 304                       IP_VS_DBG_ADDR(svc->af, dst_addr), ntohs(dst_port),
 305                       IP_VS_DBG_ADDR(svc->af, &snet));
 306
 307         /*
 308          * As far as we know, FTP is a very complicated network protocol, and
 309          * it uses control connection and data connections. For active FTP,
 310          * FTP server initialize data connection to the client, its source port
 311          * is often 20. For passive FTP, FTP server tells the clients the port
 312          * that it passively listens to,  and the client issues the data
 313          * connection. In the tunneling or direct routing mode, the load
 314          * balancer is on the client-to-server half of connection, the port
 315          * number is unknown to the load balancer. So, a conn template like
 316          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
 317          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
 318          * is created for other persistent services.
 319          */
 320         {
 321                 int protocol = iph->protocol;
 322                 const union nf_inet_addr *vaddr = dst_addr;
 323                 __be16 vport = 0;
 324
 325                 if (dst_port == svc->port) {
 326                         /* non-FTP template:
 327                          * <protocol, caddr, 0, vaddr, vport, daddr, dport>
 328                          * FTP template:
 329                          * <protocol, caddr, 0, vaddr, 0, daddr, 0>
 330                          */
 331                         if (svc->port != FTPPORT)
 332                                 vport = dst_port;
 333                 } else {
 334                         /* Note: persistent fwmark-based services and
 335                          * persistent port zero service are handled here.
 336                          * fwmark template:
 337                          * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
 338                          * port zero template:
 339                          * <protocol,caddr,0,vaddr,0,daddr,0>
 340                          */
 341                         if (svc->fwmark) {
 342                                 protocol = IPPROTO_IP;
 343                                 vaddr = &fwmark;
 344                         }
 345                 }
 346                 /* return *ignored = -1 so NF_DROP can be used */
 347                 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
 348                                                   vaddr, vport, &param) < 0) {
 349                         *ignored = -1;
 350                         return NULL;
 351                 }
 352         }
 353
 354         /* Check if a template already exists */
 355         ct = ip_vs_ct_in_get(&param);
 356         if (!ct || !ip_vs_check_template(ct, NULL)) {
 357                 struct ip_vs_scheduler *sched;
 358
 359                 /*
 360                  * No template found or the dest of the connection
 361                  * template is not available.
 362                  * return *ignored=0 i.e. ICMP and NF_DROP
 363                  */
 364                 sched = rcu_dereference(svc->scheduler);
 365                 if (sched) {
 366                         /* read svc->sched_data after svc->scheduler */
 367                         smp_rmb();
 368                         dest = sched->schedule(svc, skb, iph);
 369                 } else {
 370                         dest = NULL;
 371                 }
 372                 if (!dest) {
 373                         IP_VS_DBG(1, "p-schedule: no dest found.\n");
 374                         kfree(param.pe_data);
 375                         *ignored = 0;
 376                         return NULL;
 377                 }
 378
 379                 if (dst_port == svc->port && svc->port != FTPPORT)
 380                         dport = dest->port;
 381
 382                 /* Create a template
 383                  * This adds param.pe_data to the template,
 384                  * and thus param.pe_data will be destroyed
 385                  * when the template expires */
 386                 ct = ip_vs_conn_new(&param, dest->af, &dest->addr, dport,
 387                                     IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
 388                 if (ct == NULL) {
 389                         kfree(param.pe_data);
 390                         *ignored = -1;
 391                         return NULL;
 392                 }
 393
 394                 ct->timeout = svc->timeout;
 395         } else {
 396                 /* set destination with the found template */
 397                 dest = ct->dest;
 398                 kfree(param.pe_data);
 399         }
 400
 401         dport = dst_port;
 402         if (dport == svc->port && dest->port)
 403                 dport = dest->port;
 404
 405         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 406                  && iph->protocol == IPPROTO_UDP) ?
 407                 IP_VS_CONN_F_ONE_PACKET : 0;
 408
 409         /*
 410          *    Create a new connection according to the template
 411          */
 412         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol, src_addr,
 413                               src_port, dst_addr, dst_port, &param);
 414
 415         cp = ip_vs_conn_new(&param, dest->af, &dest->addr, dport, flags, dest,
 416                             skb->mark);
 417         if (cp == NULL) {
 418                 ip_vs_conn_put(ct);
 419                 *ignored = -1;
 420                 return NULL;
 421         }
 422
 423         /*
 424          *    Add its control
 425          */
 426         ip_vs_control_add(cp, ct);
 427         ip_vs_conn_put(ct);
 428
 429         ip_vs_conn_stats(cp, svc);
 430         return cp;
 431 }
 432
 433
 434 /*
 435  *  IPVS main scheduling function
 436  *  It selects a server according to the virtual service, and
 437  *  creates a connection entry.
 438  *  Protocols supported: TCP, UDP
 439  *
 440  *  Usage of *ignored
 441  *
 442  * 1 :   protocol tried to schedule (eg. on SYN), found svc but the
 443  *       svc/scheduler decides that this packet should be accepted with
 444  *       NF_ACCEPT because it must not be scheduled.
 445  *
 446  * 0 :   scheduler can not find destination, so try bypass or
 447  *       return ICMP and then NF_DROP (ip_vs_leave).
 448  *
 449  * -1 :  scheduler tried to schedule but fatal error occurred, eg.
 450  *       ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
 451  *       failure such as missing Call-ID, ENOMEM on skb_linearize
 452  *       or pe_data. In this case we should return NF_DROP without
 453  *       any attempts to send ICMP with ip_vs_leave.
 454  */
 455 struct ip_vs_conn *
 456 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
 457                struct ip_vs_proto_data *pd, int *ignored,
 458                struct ip_vs_iphdr *iph)
 459 {
 460         struct ip_vs_protocol *pp = pd->pp;
 461         struct ip_vs_conn *cp = NULL;
 462         struct ip_vs_scheduler *sched;
 463         struct ip_vs_dest *dest;
 464         __be16 _ports[2], *pptr, cport, vport;
 465         const void *caddr, *vaddr;
 466         unsigned int flags;
 467
 468         *ignored = 1;
 469         /*
 470          * IPv6 frags, only the first hit here.
 471          */
 472         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports);
 473         if (pptr == NULL)
 474                 return NULL;
 475
 476         if (likely(!ip_vs_iph_inverse(iph))) {
 477                 cport = pptr[0];
 478                 caddr = &iph->saddr;
 479                 vport = pptr[1];
 480                 vaddr = &iph->daddr;
 481         } else {
 482                 cport = pptr[1];
 483                 caddr = &iph->daddr;
 484                 vport = pptr[0];
 485                 vaddr = &iph->saddr;
 486         }
 487
 488         /*
 489          * FTPDATA needs this check when using local real server.
 490          * Never schedule Active FTPDATA connections from real server.
 491          * For LVS-NAT they must be already created. For other methods
 492          * with persistence the connection is created on SYN+ACK.
 493          */
 494         if (cport == FTPDATA) {
 495                 IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
 496                               "Not scheduling FTPDATA");
 497                 return NULL;
 498         }
 499
 500         /*
 501          *    Do not schedule replies from local real server.
 502          */
 503         if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK)) {
 504                 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
 505                 cp = INDIRECT_CALL_1(pp->conn_in_get,
 506                                      ip_vs_conn_in_get_proto, svc->ipvs,
 507                                      svc->af, skb, iph);
 508                 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
 509
 510                 if (cp) {
 511                         IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
 512                                       "Not scheduling reply for existing"
 513                                       " connection");
 514                         __ip_vs_conn_put(cp);
 515                         return NULL;
 516                 }
 517         }
 518
 519         /*
 520          *    Persistent service
 521          */
 522         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 523                 return ip_vs_sched_persist(svc, skb, cport, vport, ignored,
 524                                            iph);
 525
 526         *ignored = 0;
 527
 528         /*
 529          *    Non-persistent service
 530          */
 531         if (!svc->fwmark && vport != svc->port) {
 532                 if (!svc->port)
 533                         pr_err("Schedule: port zero only supported "
 534                                "in persistent services, "
 535                                "check your ipvs configuration\n");
 536                 return NULL;
 537         }
 538
 539         sched = rcu_dereference(svc->scheduler);
 540         if (sched) {
 541                 /* read svc->sched_data after svc->scheduler */
 542                 smp_rmb();
 543                 dest = sched->schedule(svc, skb, iph);
 544         } else {
 545                 dest = NULL;
 546         }
 547         if (dest == NULL) {
 548                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 549                 return NULL;
 550         }
 551
 552         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 553                  && iph->protocol == IPPROTO_UDP) ?
 554                 IP_VS_CONN_F_ONE_PACKET : 0;
 555
 556         /*
 557          *    Create a connection entry.
 558          */
 559         {
 560                 struct ip_vs_conn_param p;
 561
 562                 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
 563                                       caddr, cport, vaddr, vport, &p);
 564                 cp = ip_vs_conn_new(&p, dest->af, &dest->addr,
 565                                     dest->port ? dest->port : vport,
 566                                     flags, dest, skb->mark);
 567                 if (!cp) {
 568                         *ignored = -1;
 569                         return NULL;
 570                 }
 571         }
 572
 573         IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
 574                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
 575                       ip_vs_fwd_tag(cp),
 576                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 577                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 578                       IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
 579                       cp->flags, refcount_read(&cp->refcnt));
 580
 581         ip_vs_conn_stats(cp, svc);
 582         return cp;
 583 }
 584
 585 static inline int ip_vs_addr_is_unicast(struct net *net, int af,
 586                                         union nf_inet_addr *addr)
 587 {
 588 #ifdef CONFIG_IP_VS_IPV6
 589         if (af == AF_INET6)
 590                 return ipv6_addr_type(&addr->in6) & IPV6_ADDR_UNICAST;
 591 #endif
 592         return (inet_addr_type(net, addr->ip) == RTN_UNICAST);
 593 }
 594
 595 /*
 596  *  Pass or drop the packet.
 597  *  Called by ip_vs_in, when the virtual service is available but
 598  *  no destination is available for a new connection.
 599  */
 600 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 601                 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
 602 {
 603         __be16 _ports[2], *pptr, dport;
 604         struct netns_ipvs *ipvs = svc->ipvs;
 605         struct net *net = ipvs->net;
 606
 607         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports);
 608         if (!pptr)
 609                 return NF_DROP;
 610         dport = likely(!ip_vs_iph_inverse(iph)) ? pptr[1] : pptr[0];
 611
 612         /* if it is fwmark-based service, the cache_bypass sysctl is up
 613            and the destination is a non-local unicast, then create
 614            a cache_bypass connection entry */
 615         if (sysctl_cache_bypass(ipvs) && svc->fwmark &&
 616             !(iph->hdr_flags & (IP_VS_HDR_INVERSE | IP_VS_HDR_ICMP)) &&
 617             ip_vs_addr_is_unicast(net, svc->af, &iph->daddr)) {
 618                 int ret;
 619                 struct ip_vs_conn *cp;
 620                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
 621                                       iph->protocol == IPPROTO_UDP) ?
 622                                       IP_VS_CONN_F_ONE_PACKET : 0;
 623                 union nf_inet_addr daddr =  { .all = { 0, 0, 0, 0 } };
 624
 625                 /* create a new connection entry */
 626                 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
 627                 {
 628                         struct ip_vs_conn_param p;
 629                         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
 630                                               &iph->saddr, pptr[0],
 631                                               &iph->daddr, pptr[1], &p);
 632                         cp = ip_vs_conn_new(&p, svc->af, &daddr, 0,
 633                                             IP_VS_CONN_F_BYPASS | flags,
 634                                             NULL, skb->mark);
 635                         if (!cp)
 636                                 return NF_DROP;
 637                 }
 638
 639                 /* statistics */
 640                 ip_vs_in_stats(cp, skb);
 641
 642                 /* set state */
 643                 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
 644
 645                 /* transmit the first SYN packet */
 646                 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
 647                 /* do not touch skb anymore */
 648
 649                 if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) && cp->control)
 650                         atomic_inc(&cp->control->in_pkts);
 651                 else
 652                         atomic_inc(&cp->in_pkts);
 653                 ip_vs_conn_put(cp);
 654                 return ret;
 655         }
 656
 657         /*
 658          * When the virtual ftp service is presented, packets destined
 659          * for other services on the VIP may get here (except services
 660          * listed in the ipvs table), pass the packets, because it is
 661          * not ipvs job to decide to drop the packets.
 662          */
 663         if (svc->port == FTPPORT && dport != FTPPORT)
 664                 return NF_ACCEPT;
 665
 666         if (unlikely(ip_vs_iph_icmp(iph)))
 667                 return NF_DROP;
 668
 669         /*
 670          * Notify the client that the destination is unreachable, and
 671          * release the socket buffer.
 672          * Since it is in IP layer, the TCP socket is not actually
 673          * created, the TCP RST packet cannot be sent, instead that
 674          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 675          */
 676 #ifdef CONFIG_IP_VS_IPV6
 677         if (svc->af == AF_INET6) {
 678                 if (!skb->dev)
 679                         skb->dev = net->loopback_dev;
 680                 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 681         } else
 682 #endif
 683                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 684
 685         return NF_DROP;
 686 }
 687
 688 #ifdef CONFIG_SYSCTL
 689
 690 static int sysctl_snat_reroute(struct netns_ipvs *ipvs)
 691 {
 692         return ipvs->sysctl_snat_reroute;
 693 }
 694
 695 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs)
 696 {
 697         return ipvs->sysctl_nat_icmp_send;
 698 }
 699
 700 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
 701 {
 702         return ipvs->sysctl_expire_nodest_conn;
 703 }
 704
 705 #else
 706
 707 static int sysctl_snat_reroute(struct netns_ipvs *ipvs) { return 0; }
 708 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs) { return 0; }
 709 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
 710
 711 #endif
 712
 713 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 714 {
 715         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 716 }
 717
 718 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
 719 {
 720         if (NF_INET_LOCAL_IN == hooknum)
 721                 return IP_DEFRAG_VS_IN;
 722         if (NF_INET_FORWARD == hooknum)
 723                 return IP_DEFRAG_VS_FWD;
 724         return IP_DEFRAG_VS_OUT;
 725 }
 726
 727 static inline int ip_vs_gather_frags(struct netns_ipvs *ipvs,
 728                                      struct sk_buff *skb, u_int32_t user)
 729 {
 730         int err;
 731
 732         local_bh_disable();
 733         err = ip_defrag(ipvs->net, skb, user);
 734         local_bh_enable();
 735         if (!err)
 736                 ip_send_check(ip_hdr(skb));
 737
 738         return err;
 739 }
 740
 741 static int ip_vs_route_me_harder(struct netns_ipvs *ipvs, int af,
 742                                  struct sk_buff *skb, unsigned int hooknum)
 743 {
 744         if (!sysctl_snat_reroute(ipvs))
 745                 return 0;
 746         /* Reroute replies only to remote clients (FORWARD and LOCAL_OUT) */
 747         if (NF_INET_LOCAL_IN == hooknum)
 748                 return 0;
 749 #ifdef CONFIG_IP_VS_IPV6
 750         if (af == AF_INET6) {
 751                 struct dst_entry *dst = skb_dst(skb);
 752
 753                 if (dst->dev && !(dst->dev->flags & IFF_LOOPBACK) &&
 754                     ip6_route_me_harder(ipvs->net, skb) != 0)
 755                         return 1;
 756         } else
 757 #endif
 758                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
 759                     ip_route_me_harder(ipvs->net, skb, RTN_LOCAL) != 0)
 760                         return 1;
 761
 762         return 0;
 763 }
 764
 765 /*
 766  * Packet has been made sufficiently writable in caller
 767  * - inout: 1=in->out, 0=out->in
 768  */
 769 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 770                     struct ip_vs_conn *cp, int inout)
 771 {
 772         struct iphdr *iph        = ip_hdr(skb);
 773         unsigned int icmp_offset = iph->ihl*4;
 774         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
 775                                                       icmp_offset);
 776         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 777
 778         if (inout) {
 779                 iph->saddr = cp->vaddr.ip;
 780                 ip_send_check(iph);
 781                 ciph->daddr = cp->vaddr.ip;
 782                 ip_send_check(ciph);
 783         } else {
 784                 iph->daddr = cp->daddr.ip;
 785                 ip_send_check(iph);
 786                 ciph->saddr = cp->daddr.ip;
 787                 ip_send_check(ciph);
 788         }
 789
 790         /* the TCP/UDP/SCTP port */
 791         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
 792             IPPROTO_SCTP == ciph->protocol) {
 793                 __be16 *ports = (void *)ciph + ciph->ihl*4;
 794
 795                 if (inout)
 796                         ports[1] = cp->vport;
 797                 else
 798                         ports[0] = cp->dport;
 799         }
 800
 801         /* And finally the ICMP checksum */
 802         icmph->checksum = 0;
 803         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 804         skb->ip_summed = CHECKSUM_UNNECESSARY;
 805
 806         if (inout)
 807                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 808                         "Forwarding altered outgoing ICMP");
 809         else
 810                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 811                         "Forwarding altered incoming ICMP");
 812 }
 813
 814 #ifdef CONFIG_IP_VS_IPV6
 815 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
 816                     struct ip_vs_conn *cp, int inout)
 817 {
 818         struct ipv6hdr *iph      = ipv6_hdr(skb);
 819         unsigned int icmp_offset = 0;
 820         unsigned int offs        = 0; /* header offset*/
 821         int protocol;
 822         struct icmp6hdr *icmph;
 823         struct ipv6hdr *ciph;
 824         unsigned short fragoffs;
 825
 826         ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
 827         icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
 828         offs = icmp_offset + sizeof(struct icmp6hdr);
 829         ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
 830
 831         protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
 832
 833         if (inout) {
 834                 iph->saddr = cp->vaddr.in6;
 835                 ciph->daddr = cp->vaddr.in6;
 836         } else {
 837                 iph->daddr = cp->daddr.in6;
 838                 ciph->saddr = cp->daddr.in6;
 839         }
 840
 841         /* the TCP/UDP/SCTP port */
 842         if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 843                           IPPROTO_SCTP == protocol)) {
 844                 __be16 *ports = (void *)(skb_network_header(skb) + offs);
 845
 846                 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
 847                               ntohs(inout ? ports[1] : ports[0]),
 848                               ntohs(inout ? cp->vport : cp->dport));
 849                 if (inout)
 850                         ports[1] = cp->vport;
 851                 else
 852                         ports[0] = cp->dport;
 853         }
 854
 855         /* And finally the ICMP checksum */
 856         icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
 857                                               skb->len - icmp_offset,
 858                                               IPPROTO_ICMPV6, 0);
 859         skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
 860         skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
 861         skb->ip_summed = CHECKSUM_PARTIAL;
 862
 863         if (inout)
 864                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 865                               (void *)ciph - (void *)iph,
 866                               "Forwarding altered outgoing ICMPv6");
 867         else
 868                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 869                               (void *)ciph - (void *)iph,
 870                               "Forwarding altered incoming ICMPv6");
 871 }
 872 #endif
 873
 874 /* Handle relevant response ICMP messages - forward to the right
 875  * destination host.
 876  */
 877 static int handle_response_icmp(int af, struct sk_buff *skb,
 878                                 union nf_inet_addr *snet,
 879                                 __u8 protocol, struct ip_vs_conn *cp,
 880                                 struct ip_vs_protocol *pp,
 881                                 unsigned int offset, unsigned int ihl,
 882                                 unsigned int hooknum)
 883 {
 884         unsigned int verdict = NF_DROP;
 885
 886         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
 887                 goto ignore_cp;
 888
 889         /* Ensure the checksum is correct */
 890         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 891                 /* Failed checksum! */
 892                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
 893                               IP_VS_DBG_ADDR(af, snet));
 894                 goto out;
 895         }
 896
 897         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 898             IPPROTO_SCTP == protocol)
 899                 offset += 2 * sizeof(__u16);
 900         if (!skb_make_writable(skb, offset))
 901                 goto out;
 902
 903 #ifdef CONFIG_IP_VS_IPV6
 904         if (af == AF_INET6)
 905                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
 906         else
 907 #endif
 908                 ip_vs_nat_icmp(skb, pp, cp, 1);
 909
 910         if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
 911                 goto out;
 912
 913         /* do the statistics and put it back */
 914         ip_vs_out_stats(cp, skb);
 915
 916         skb->ipvs_property = 1;
 917         if (!(cp->flags & IP_VS_CONN_F_NFCT))
 918                 ip_vs_notrack(skb);
 919         else
 920                 ip_vs_update_conntrack(skb, cp, 0);
 921
 922 ignore_cp:
 923         verdict = NF_ACCEPT;
 924
 925 out:
 926         __ip_vs_conn_put(cp);
 927
 928         return verdict;
 929 }
 930
 931 /*
 932  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 933  *      Find any that might be relevant, check against existing connections.
 934  *      Currently handles error types - unreachable, quench, ttl exceeded.
 935  */
 936 static int ip_vs_out_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb,
 937                           int *related, unsigned int hooknum)
 938 {
 939         struct iphdr *iph;
 940         struct icmphdr  _icmph, *ic;
 941         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 942         struct ip_vs_iphdr ciph;
 943         struct ip_vs_conn *cp;
 944         struct ip_vs_protocol *pp;
 945         unsigned int offset, ihl;
 946         union nf_inet_addr snet;
 947
 948         *related = 1;
 949
 950         /* reassemble IP fragments */
 951         if (ip_is_fragment(ip_hdr(skb))) {
 952                 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
 953                         return NF_STOLEN;
 954         }
 955
 956         iph = ip_hdr(skb);
 957         offset = ihl = iph->ihl * 4;
 958         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 959         if (ic == NULL)
 960                 return NF_DROP;
 961
 962         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
 963                   ic->type, ntohs(icmp_id(ic)),
 964                   &iph->saddr, &iph->daddr);
 965
 966         /*
 967          * Work through seeing if this is for us.
 968          * These checks are supposed to be in an order that means easy
 969          * things are checked first to speed up processing.... however
 970          * this means that some packets will manage to get a long way
 971          * down this stack and then be rejected, but that's life.
 972          */
 973         if ((ic->type != ICMP_DEST_UNREACH) &&
 974             (ic->type != ICMP_SOURCE_QUENCH) &&
 975             (ic->type != ICMP_TIME_EXCEEDED)) {
 976                 *related = 0;
 977                 return NF_ACCEPT;
 978         }
 979
 980         /* Now find the contained IP header */
 981         offset += sizeof(_icmph);
 982         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 983         if (cih == NULL)
 984                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 985
 986         pp = ip_vs_proto_get(cih->protocol);
 987         if (!pp)
 988                 return NF_ACCEPT;
 989
 990         /* Is the embedded protocol header present? */
 991         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 992                      pp->dont_defrag))
 993                 return NF_ACCEPT;
 994
 995         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
 996                       "Checking outgoing ICMP for");
 997
 998         ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, true, &ciph);
 999
1000         /* The embedded headers contain source and dest in reverse order */
1001         cp = INDIRECT_CALL_1(pp->conn_out_get, ip_vs_conn_out_get_proto,
1002                              ipvs, AF_INET, skb, &ciph);
1003         if (!cp)
1004                 return NF_ACCEPT;
1005
1006         snet.ip = iph->saddr;
1007         return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
1008                                     pp, ciph.len, ihl, hooknum);
1009 }
1010
1011 #ifdef CONFIG_IP_VS_IPV6
1012 static int ip_vs_out_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
1013                              int *related,  unsigned int hooknum,
1014                              struct ip_vs_iphdr *ipvsh)
1015 {
1016         struct icmp6hdr _icmph, *ic;
1017         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1018         struct ip_vs_conn *cp;
1019         struct ip_vs_protocol *pp;
1020         union nf_inet_addr snet;
1021         unsigned int offset;
1022
1023         *related = 1;
1024         ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph);
1025         if (ic == NULL)
1026                 return NF_DROP;
1027
1028         /*
1029          * Work through seeing if this is for us.
1030          * These checks are supposed to be in an order that means easy
1031          * things are checked first to speed up processing.... however
1032          * this means that some packets will manage to get a long way
1033          * down this stack and then be rejected, but that's life.
1034          */
1035         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1036                 *related = 0;
1037                 return NF_ACCEPT;
1038         }
1039         /* Fragment header that is before ICMP header tells us that:
1040          * it's not an error message since they can't be fragmented.
1041          */
1042         if (ipvsh->flags & IP6_FH_F_FRAG)
1043                 return NF_DROP;
1044
1045         IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1046                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1047                   &ipvsh->saddr, &ipvsh->daddr);
1048
1049         if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, ipvsh->len + sizeof(_icmph),
1050                                      true, &ciph))
1051                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1052
1053         pp = ip_vs_proto_get(ciph.protocol);
1054         if (!pp)
1055                 return NF_ACCEPT;
1056
1057         /* The embedded headers contain source and dest in reverse order */
1058         cp = INDIRECT_CALL_1(pp->conn_out_get, ip_vs_conn_out_get_proto,
1059                              ipvs, AF_INET6, skb, &ciph);
1060         if (!cp)
1061                 return NF_ACCEPT;
1062
1063         snet.in6 = ciph.saddr.in6;
1064         offset = ciph.len;
1065         return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
1066                                     pp, offset, sizeof(struct ipv6hdr),
1067                                     hooknum);
1068 }
1069 #endif
1070
1071 /*
1072  * Check if sctp chunc is ABORT chunk
1073  */
1074 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
1075 {
1076         struct sctp_chunkhdr *sch, schunk;
1077         sch = skb_header_pointer(skb, nh_len + sizeof(struct sctphdr),
1078                                  sizeof(schunk), &schunk);
1079         if (sch == NULL)
1080                 return 0;
1081         if (sch->type == SCTP_CID_ABORT)
1082                 return 1;
1083         return 0;
1084 }
1085
1086 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
1087 {
1088         struct tcphdr _tcph, *th;
1089
1090         th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
1091         if (th == NULL)
1092                 return 0;
1093         return th->rst;
1094 }
1095
1096 static inline bool is_new_conn(const struct sk_buff *skb,
1097                                struct ip_vs_iphdr *iph)
1098 {
1099         switch (iph->protocol) {
1100         case IPPROTO_TCP: {
1101                 struct tcphdr _tcph, *th;
1102
1103                 th = skb_header_pointer(skb, iph->len, sizeof(_tcph), &_tcph);
1104                 if (th == NULL)
1105                         return false;
1106                 return th->syn;
1107         }
1108         case IPPROTO_SCTP: {
1109                 struct sctp_chunkhdr *sch, schunk;
1110
1111                 sch = skb_header_pointer(skb, iph->len + sizeof(struct sctphdr),
1112                                          sizeof(schunk), &schunk);
1113                 if (sch == NULL)
1114                         return false;
1115                 return sch->type == SCTP_CID_INIT;
1116         }
1117         default:
1118                 return false;
1119         }
1120 }
1121
1122 static inline bool is_new_conn_expected(const struct ip_vs_conn *cp,
1123                                         int conn_reuse_mode)
1124 {
1125         /* Controlled (FTP DATA or persistence)? */
1126         if (cp->control)
1127                 return false;
1128
1129         switch (cp->protocol) {
1130         case IPPROTO_TCP:
1131                 return (cp->state == IP_VS_TCP_S_TIME_WAIT) ||
1132                        (cp->state == IP_VS_TCP_S_CLOSE) ||
1133                         ((conn_reuse_mode & 2) &&
1134                          (cp->state == IP_VS_TCP_S_FIN_WAIT) &&
1135                          (cp->flags & IP_VS_CONN_F_NOOUTPUT));
1136         case IPPROTO_SCTP:
1137                 return cp->state == IP_VS_SCTP_S_CLOSED;
1138         default:
1139                 return false;
1140         }
1141 }
1142
1143 /* Generic function to create new connections for outgoing RS packets
1144  *
1145  * Pre-requisites for successful connection creation:
1146  * 1) Virtual Service is NOT fwmark based:
1147  *    In fwmark-VS actual vaddr and vport are unknown to IPVS
1148  * 2) Real Server and Virtual Service were NOT configured without port:
1149  *    This is to allow match of different VS to the same RS ip-addr
1150  */
1151 struct ip_vs_conn *ip_vs_new_conn_out(struct ip_vs_service *svc,
1152                                       struct ip_vs_dest *dest,
1153                                       struct sk_buff *skb,
1154                                       const struct ip_vs_iphdr *iph,
1155                                       __be16 dport,
1156                                       __be16 cport)
1157 {
1158         struct ip_vs_conn_param param;
1159         struct ip_vs_conn *ct = NULL, *cp = NULL;
1160         const union nf_inet_addr *vaddr, *daddr, *caddr;
1161         union nf_inet_addr snet;
1162         __be16 vport;
1163         unsigned int flags;
1164
1165         EnterFunction(12);
1166         vaddr = &svc->addr;
1167         vport = svc->port;
1168         daddr = &iph->saddr;
1169         caddr = &iph->daddr;
1170
1171         /* check pre-requisites are satisfied */
1172         if (svc->fwmark)
1173                 return NULL;
1174         if (!vport || !dport)
1175                 return NULL;
1176
1177         /* for persistent service first create connection template */
1178         if (svc->flags & IP_VS_SVC_F_PERSISTENT) {
1179                 /* apply netmask the same way ingress-side does */
1180 #ifdef CONFIG_IP_VS_IPV6
1181                 if (svc->af == AF_INET6)
1182                         ipv6_addr_prefix(&snet.in6, &caddr->in6,
1183                                          (__force __u32)svc->netmask);
1184                 else
1185 #endif
1186                         snet.ip = caddr->ip & svc->netmask;
1187                 /* fill params and create template if not existent */
1188                 if (ip_vs_conn_fill_param_persist(svc, skb, iph->protocol,
1189                                                   &snet, 0, vaddr,
1190                                                   vport, &param) < 0)
1191                         return NULL;
1192                 ct = ip_vs_ct_in_get(&param);
1193                 /* check if template exists and points to the same dest */
1194                 if (!ct || !ip_vs_check_template(ct, dest)) {
1195                         ct = ip_vs_conn_new(&param, dest->af, daddr, dport,
1196                                             IP_VS_CONN_F_TEMPLATE, dest, 0);
1197                         if (!ct) {
1198                                 kfree(param.pe_data);
1199                                 return NULL;
1200                         }
1201                         ct->timeout = svc->timeout;
1202                 } else {
1203                         kfree(param.pe_data);
1204                 }
1205         }
1206
1207         /* connection flags */
1208         flags = ((svc->flags & IP_VS_SVC_F_ONEPACKET) &&
1209                  iph->protocol == IPPROTO_UDP) ? IP_VS_CONN_F_ONE_PACKET : 0;
1210         /* create connection */
1211         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
1212                               caddr, cport, vaddr, vport, &param);
1213         cp = ip_vs_conn_new(&param, dest->af, daddr, dport, flags, dest, 0);
1214         if (!cp) {
1215                 if (ct)
1216                         ip_vs_conn_put(ct);
1217                 return NULL;
1218         }
1219         if (ct) {
1220                 ip_vs_control_add(cp, ct);
1221                 ip_vs_conn_put(ct);
1222         }
1223         ip_vs_conn_stats(cp, svc);
1224
1225         /* return connection (will be used to handle outgoing packet) */
1226         IP_VS_DBG_BUF(6, "New connection RS-initiated:%c c:%s:%u v:%s:%u "
1227                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
1228                       ip_vs_fwd_tag(cp),
1229                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
1230                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
1231                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
1232                       cp->flags, refcount_read(&cp->refcnt));
1233         LeaveFunction(12);
1234         return cp;
1235 }
1236
1237 /* Handle outgoing packets which are considered requests initiated by
1238  * real servers, so that subsequent responses from external client can be
1239  * routed to the right real server.
1240  * Used also for outgoing responses in OPS mode.
1241  *
1242  * Connection management is handled by persistent-engine specific callback.
1243  */
1244 static struct ip_vs_conn *__ip_vs_rs_conn_out(unsigned int hooknum,
1245                                               struct netns_ipvs *ipvs,
1246                                               int af, struct sk_buff *skb,
1247                                               const struct ip_vs_iphdr *iph)
1248 {
1249         struct ip_vs_dest *dest;
1250         struct ip_vs_conn *cp = NULL;
1251         __be16 _ports[2], *pptr;
1252
1253         if (hooknum == NF_INET_LOCAL_IN)
1254                 return NULL;
1255
1256         pptr = frag_safe_skb_hp(skb, iph->len,
1257                                 sizeof(_ports), _ports);
1258         if (!pptr)
1259                 return NULL;
1260
1261         dest = ip_vs_find_real_service(ipvs, af, iph->protocol,
1262                                        &iph->saddr, pptr[0]);
1263         if (dest) {
1264                 struct ip_vs_service *svc;
1265                 struct ip_vs_pe *pe;
1266
1267                 svc = rcu_dereference(dest->svc);
1268                 if (svc) {
1269                         pe = rcu_dereference(svc->pe);
1270                         if (pe && pe->conn_out)
1271                                 cp = pe->conn_out(svc, dest, skb, iph,
1272                                                   pptr[0], pptr[1]);
1273                 }
1274         }
1275
1276         return cp;
1277 }
1278
1279 /* Handle response packets: rewrite addresses and send away...
1280  */
1281 static unsigned int
1282 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1283                 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph,
1284                 unsigned int hooknum)
1285 {
1286         struct ip_vs_protocol *pp = pd->pp;
1287
1288         IP_VS_DBG_PKT(11, af, pp, skb, iph->off, "Outgoing packet");
1289
1290         if (!skb_make_writable(skb, iph->len))
1291                 goto drop;
1292
1293         /* mangle the packet */
1294         if (pp->snat_handler &&
1295             !SNAT_CALL(pp->snat_handler, skb, pp, cp, iph))
1296                 goto drop;
1297
1298 #ifdef CONFIG_IP_VS_IPV6
1299         if (af == AF_INET6)
1300                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1301         else
1302 #endif
1303         {
1304                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1305                 ip_send_check(ip_hdr(skb));
1306         }
1307
1308         /*
1309          * nf_iterate does not expect change in the skb->dst->dev.
1310          * It looks like it is not fatal to enable this code for hooks
1311          * where our handlers are at the end of the chain list and
1312          * when all next handlers use skb->dst->dev and not outdev.
1313          * It will definitely route properly the inout NAT traffic
1314          * when multiple paths are used.
1315          */
1316
1317         /* For policy routing, packets originating from this
1318          * machine itself may be routed differently to packets
1319          * passing through.  We want this packet to be routed as
1320          * if it came from this machine itself.  So re-compute
1321          * the routing information.
1322          */
1323         if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
1324                 goto drop;
1325
1326         IP_VS_DBG_PKT(10, af, pp, skb, iph->off, "After SNAT");
1327
1328         ip_vs_out_stats(cp, skb);
1329         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1330         skb->ipvs_property = 1;
1331         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1332                 ip_vs_notrack(skb);
1333         else
1334                 ip_vs_update_conntrack(skb, cp, 0);
1335         ip_vs_conn_put(cp);
1336
1337         LeaveFunction(11);
1338         return NF_ACCEPT;
1339
1340 drop:
1341         ip_vs_conn_put(cp);
1342         kfree_skb(skb);
1343         LeaveFunction(11);
1344         return NF_STOLEN;
1345 }
1346
1347 /*
1348  *      Check if outgoing packet belongs to the established ip_vs_conn.
1349  */
1350 static unsigned int
1351 ip_vs_out(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1352 {
1353         struct ip_vs_iphdr iph;
1354         struct ip_vs_protocol *pp;
1355         struct ip_vs_proto_data *pd;
1356         struct ip_vs_conn *cp;
1357         struct sock *sk;
1358
1359         EnterFunction(11);
1360
1361         /* Already marked as IPVS request or reply? */
1362         if (skb->ipvs_property)
1363                 return NF_ACCEPT;
1364
1365         sk = skb_to_full_sk(skb);
1366         /* Bad... Do not break raw sockets */
1367         if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
1368                      af == AF_INET)) {
1369
1370                 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
1371                         return NF_ACCEPT;
1372         }
1373
1374         if (unlikely(!skb_dst(skb)))
1375                 return NF_ACCEPT;
1376
1377         if (!ipvs->enable)
1378                 return NF_ACCEPT;
1379
1380         ip_vs_fill_iph_skb(af, skb, false, &iph);
1381 #ifdef CONFIG_IP_VS_IPV6
1382         if (af == AF_INET6) {
1383                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1384                         int related;
1385                         int verdict = ip_vs_out_icmp_v6(ipvs, skb, &related,
1386                                                         hooknum, &iph);
1387
1388                         if (related)
1389                                 return verdict;
1390                 }
1391         } else
1392 #endif
1393                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1394                         int related;
1395                         int verdict = ip_vs_out_icmp(ipvs, skb, &related, hooknum);
1396
1397                         if (related)
1398                                 return verdict;
1399                 }
1400
1401         pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1402         if (unlikely(!pd))
1403                 return NF_ACCEPT;
1404         pp = pd->pp;
1405
1406         /* reassemble IP fragments */
1407 #ifdef CONFIG_IP_VS_IPV6
1408         if (af == AF_INET)
1409 #endif
1410                 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1411                         if (ip_vs_gather_frags(ipvs, skb,
1412                                                ip_vs_defrag_user(hooknum)))
1413                                 return NF_STOLEN;
1414
1415                         ip_vs_fill_iph_skb(AF_INET, skb, false, &iph);
1416                 }
1417
1418         /*
1419          * Check if the packet belongs to an existing entry
1420          */
1421         cp = INDIRECT_CALL_1(pp->conn_out_get, ip_vs_conn_out_get_proto,
1422                              ipvs, af, skb, &iph);
1423
1424         if (likely(cp)) {
1425                 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
1426                         goto ignore_cp;
1427                 return handle_response(af, skb, pd, cp, &iph, hooknum);
1428         }
1429
1430         /* Check for real-server-started requests */
1431         if (atomic_read(&ipvs->conn_out_counter)) {
1432                 /* Currently only for UDP:
1433                  * connection oriented protocols typically use
1434                  * ephemeral ports for outgoing connections, so
1435                  * related incoming responses would not match any VS
1436                  */
1437                 if (pp->protocol == IPPROTO_UDP) {
1438                         cp = __ip_vs_rs_conn_out(hooknum, ipvs, af, skb, &iph);
1439                         if (likely(cp))
1440                                 return handle_response(af, skb, pd, cp, &iph,
1441                                                        hooknum);
1442                 }
1443         }
1444
1445         if (sysctl_nat_icmp_send(ipvs) &&
1446             (pp->protocol == IPPROTO_TCP ||
1447              pp->protocol == IPPROTO_UDP ||
1448              pp->protocol == IPPROTO_SCTP)) {
1449                 __be16 _ports[2], *pptr;
1450
1451                 pptr = frag_safe_skb_hp(skb, iph.len,
1452                                          sizeof(_ports), _ports);
1453                 if (pptr == NULL)
1454                         return NF_ACCEPT;       /* Not for me */
1455                 if (ip_vs_has_real_service(ipvs, af, iph.protocol, &iph.saddr,
1456                                            pptr[0])) {
1457                         /*
1458                          * Notify the real server: there is no
1459                          * existing entry if it is not RST
1460                          * packet or not TCP packet.
1461                          */
1462                         if ((iph.protocol != IPPROTO_TCP &&
1463                              iph.protocol != IPPROTO_SCTP)
1464                              || ((iph.protocol == IPPROTO_TCP
1465                                   && !is_tcp_reset(skb, iph.len))
1466                                  || (iph.protocol == IPPROTO_SCTP
1467                                         && !is_sctp_abort(skb,
1468                                                 iph.len)))) {
1469 #ifdef CONFIG_IP_VS_IPV6
1470                                 if (af == AF_INET6) {
1471                                         if (!skb->dev)
1472                                                 skb->dev = ipvs->net->loopback_dev;
1473                                         icmpv6_send(skb,
1474                                                     ICMPV6_DEST_UNREACH,
1475                                                     ICMPV6_PORT_UNREACH,
1476                                                     0);
1477                                 } else
1478 #endif
1479                                         icmp_send(skb,
1480                                                   ICMP_DEST_UNREACH,
1481                                                   ICMP_PORT_UNREACH, 0);
1482                                 return NF_DROP;
1483                         }
1484                 }
1485         }
1486
1487 out:
1488         IP_VS_DBG_PKT(12, af, pp, skb, iph.off,
1489                       "ip_vs_out: packet continues traversal as normal");
1490         return NF_ACCEPT;
1491
1492 ignore_cp:
1493         __ip_vs_conn_put(cp);
1494         goto out;
1495 }
1496
1497 /*
1498  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1499  *      used only for VS/NAT.
1500  *      Check if packet is reply for established ip_vs_conn.
1501  */
1502 static unsigned int
1503 ip_vs_reply4(void *priv, struct sk_buff *skb,
1504              const struct nf_hook_state *state)
1505 {
1506         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1507 }
1508
1509 /*
1510  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1511  *      Check if packet is reply for established ip_vs_conn.
1512  */
1513 static unsigned int
1514 ip_vs_local_reply4(void *priv, struct sk_buff *skb,
1515                    const struct nf_hook_state *state)
1516 {
1517         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1518 }
1519
1520 #ifdef CONFIG_IP_VS_IPV6
1521
1522 /*
1523  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1524  *      used only for VS/NAT.
1525  *      Check if packet is reply for established ip_vs_conn.
1526  */
1527 static unsigned int
1528 ip_vs_reply6(void *priv, struct sk_buff *skb,
1529              const struct nf_hook_state *state)
1530 {
1531         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1532 }
1533
1534 /*
1535  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1536  *      Check if packet is reply for established ip_vs_conn.
1537  */
1538 static unsigned int
1539 ip_vs_local_reply6(void *priv, struct sk_buff *skb,
1540                    const struct nf_hook_state *state)
1541 {
1542         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1543 }
1544
1545 #endif
1546
1547 static unsigned int
1548 ip_vs_try_to_schedule(struct netns_ipvs *ipvs, int af, struct sk_buff *skb,
1549                       struct ip_vs_proto_data *pd,
1550                       int *verdict, struct ip_vs_conn **cpp,
1551                       struct ip_vs_iphdr *iph)
1552 {
1553         struct ip_vs_protocol *pp = pd->pp;
1554
1555         if (!iph->fragoffs) {
1556                 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1557                  * replayed fragment zero will already have created the cp
1558                  */
1559
1560                 /* Schedule and create new connection entry into cpp */
1561                 if (!pp->conn_schedule(ipvs, af, skb, pd, verdict, cpp, iph))
1562                         return 0;
1563         }
1564
1565         if (unlikely(!*cpp)) {
1566                 /* sorry, all this trouble for a no-hit :) */
1567                 IP_VS_DBG_PKT(12, af, pp, skb, iph->off,
1568                               "ip_vs_in: packet continues traversal as normal");
1569
1570                 /* Fragment couldn't be mapped to a conn entry */
1571                 if (iph->fragoffs)
1572                         IP_VS_DBG_PKT(7, af, pp, skb, iph->off,
1573                                       "unhandled fragment");
1574
1575                 *verdict = NF_ACCEPT;
1576                 return 0;
1577         }
1578
1579         return 1;
1580 }
1581
1582 /*
1583  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1584  *      Find any that might be relevant, check against existing connections,
1585  *      forward to the right destination host if relevant.
1586  *      Currently handles error types - unreachable, quench, ttl exceeded.
1587  */
1588 static int
1589 ip_vs_in_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb, int *related,
1590               unsigned int hooknum)
1591 {
1592         struct iphdr *iph;
1593         struct icmphdr  _icmph, *ic;
1594         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1595         struct ip_vs_iphdr ciph;
1596         struct ip_vs_conn *cp;
1597         struct ip_vs_protocol *pp;
1598         struct ip_vs_proto_data *pd;
1599         unsigned int offset, offset2, ihl, verdict;
1600         bool ipip, new_cp = false;
1601
1602         *related = 1;
1603
1604         /* reassemble IP fragments */
1605         if (ip_is_fragment(ip_hdr(skb))) {
1606                 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
1607                         return NF_STOLEN;
1608         }
1609
1610         iph = ip_hdr(skb);
1611         offset = ihl = iph->ihl * 4;
1612         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1613         if (ic == NULL)
1614                 return NF_DROP;
1615
1616         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1617                   ic->type, ntohs(icmp_id(ic)),
1618                   &iph->saddr, &iph->daddr);
1619
1620         /*
1621          * Work through seeing if this is for us.
1622          * These checks are supposed to be in an order that means easy
1623          * things are checked first to speed up processing.... however
1624          * this means that some packets will manage to get a long way
1625          * down this stack and then be rejected, but that's life.
1626          */
1627         if ((ic->type != ICMP_DEST_UNREACH) &&
1628             (ic->type != ICMP_SOURCE_QUENCH) &&
1629             (ic->type != ICMP_TIME_EXCEEDED)) {
1630                 *related = 0;
1631                 return NF_ACCEPT;
1632         }
1633
1634         /* Now find the contained IP header */
1635         offset += sizeof(_icmph);
1636         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1637         if (cih == NULL)
1638                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1639
1640         /* Special case for errors for IPIP packets */
1641         ipip = false;
1642         if (cih->protocol == IPPROTO_IPIP) {
1643                 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1644                         return NF_ACCEPT;
1645                 /* Error for our IPIP must arrive at LOCAL_IN */
1646                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1647                         return NF_ACCEPT;
1648                 offset += cih->ihl * 4;
1649                 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1650                 if (cih == NULL)
1651                         return NF_ACCEPT; /* The packet looks wrong, ignore */
1652                 ipip = true;
1653         }
1654
1655         pd = ip_vs_proto_data_get(ipvs, cih->protocol);
1656         if (!pd)
1657                 return NF_ACCEPT;
1658         pp = pd->pp;
1659
1660         /* Is the embedded protocol header present? */
1661         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1662                      pp->dont_defrag))
1663                 return NF_ACCEPT;
1664
1665         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1666                       "Checking incoming ICMP for");
1667
1668         offset2 = offset;
1669         ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, !ipip, &ciph);
1670         offset = ciph.len;
1671
1672         /* The embedded headers contain source and dest in reverse order.
1673          * For IPIP this is error for request, not for reply.
1674          */
1675         cp = INDIRECT_CALL_1(pp->conn_in_get, ip_vs_conn_in_get_proto,
1676                              ipvs, AF_INET, skb, &ciph);
1677
1678         if (!cp) {
1679                 int v;
1680
1681                 if (!sysctl_schedule_icmp(ipvs))
1682                         return NF_ACCEPT;
1683
1684                 if (!ip_vs_try_to_schedule(ipvs, AF_INET, skb, pd, &v, &cp, &ciph))
1685                         return v;
1686                 new_cp = true;
1687         }
1688
1689         verdict = NF_DROP;
1690
1691         /* Ensure the checksum is correct */
1692         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1693                 /* Failed checksum! */
1694                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1695                           &iph->saddr);
1696                 goto out;
1697         }
1698
1699         if (ipip) {
1700                 __be32 info = ic->un.gateway;
1701                 __u8 type = ic->type;
1702                 __u8 code = ic->code;
1703
1704                 /* Update the MTU */
1705                 if (ic->type == ICMP_DEST_UNREACH &&
1706                     ic->code == ICMP_FRAG_NEEDED) {
1707                         struct ip_vs_dest *dest = cp->dest;
1708                         u32 mtu = ntohs(ic->un.frag.mtu);
1709                         __be16 frag_off = cih->frag_off;
1710
1711                         /* Strip outer IP and ICMP, go to IPIP header */
1712                         if (pskb_pull(skb, ihl + sizeof(_icmph)) == NULL)
1713                                 goto ignore_ipip;
1714                         offset2 -= ihl + sizeof(_icmph);
1715                         skb_reset_network_header(skb);
1716                         IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1717                                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1718                         ipv4_update_pmtu(skb, ipvs->net, mtu, 0, 0);
1719                         /* Client uses PMTUD? */
1720                         if (!(frag_off & htons(IP_DF)))
1721                                 goto ignore_ipip;
1722                         /* Prefer the resulting PMTU */
1723                         if (dest) {
1724                                 struct ip_vs_dest_dst *dest_dst;
1725
1726                                 dest_dst = rcu_dereference(dest->dest_dst);
1727                                 if (dest_dst)
1728                                         mtu = dst_mtu(dest_dst->dst_cache);
1729                         }
1730                         if (mtu > 68 + sizeof(struct iphdr))
1731                                 mtu -= sizeof(struct iphdr);
1732                         info = htonl(mtu);
1733                 }
1734                 /* Strip outer IP, ICMP and IPIP, go to IP header of
1735                  * original request.
1736                  */
1737                 if (pskb_pull(skb, offset2) == NULL)
1738                         goto ignore_ipip;
1739                 skb_reset_network_header(skb);
1740                 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1741                         &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1742                         type, code, ntohl(info));
1743                 icmp_send(skb, type, code, info);
1744                 /* ICMP can be shorter but anyways, account it */
1745                 ip_vs_out_stats(cp, skb);
1746
1747 ignore_ipip:
1748                 consume_skb(skb);
1749                 verdict = NF_STOLEN;
1750                 goto out;
1751         }
1752
1753         /* do the statistics and put it back */
1754         ip_vs_in_stats(cp, skb);
1755         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol ||
1756             IPPROTO_SCTP == cih->protocol)
1757                 offset += 2 * sizeof(__u16);
1758         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1759
1760 out:
1761         if (likely(!new_cp))
1762                 __ip_vs_conn_put(cp);
1763         else
1764                 ip_vs_conn_put(cp);
1765
1766         return verdict;
1767 }
1768
1769 #ifdef CONFIG_IP_VS_IPV6
1770 static int ip_vs_in_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
1771                             int *related, unsigned int hooknum,
1772                             struct ip_vs_iphdr *iph)
1773 {
1774         struct icmp6hdr _icmph, *ic;
1775         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1776         struct ip_vs_conn *cp;
1777         struct ip_vs_protocol *pp;
1778         struct ip_vs_proto_data *pd;
1779         unsigned int offset, verdict;
1780         bool new_cp = false;
1781
1782         *related = 1;
1783
1784         ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph);
1785         if (ic == NULL)
1786                 return NF_DROP;
1787
1788         /*
1789          * Work through seeing if this is for us.
1790          * These checks are supposed to be in an order that means easy
1791          * things are checked first to speed up processing.... however
1792          * this means that some packets will manage to get a long way
1793          * down this stack and then be rejected, but that's life.
1794          */
1795         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1796                 *related = 0;
1797                 return NF_ACCEPT;
1798         }
1799         /* Fragment header that is before ICMP header tells us that:
1800          * it's not an error message since they can't be fragmented.
1801          */
1802         if (iph->flags & IP6_FH_F_FRAG)
1803                 return NF_DROP;
1804
1805         IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1806                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1807                   &iph->saddr, &iph->daddr);
1808
1809         offset = iph->len + sizeof(_icmph);
1810         if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, offset, true, &ciph))
1811                 return NF_ACCEPT;
1812
1813         pd = ip_vs_proto_data_get(ipvs, ciph.protocol);
1814         if (!pd)
1815                 return NF_ACCEPT;
1816         pp = pd->pp;
1817
1818         /* Cannot handle fragmented embedded protocol */
1819         if (ciph.fragoffs)
1820                 return NF_ACCEPT;
1821
1822         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
1823                       "Checking incoming ICMPv6 for");
1824
1825         /* The embedded headers contain source and dest in reverse order
1826          * if not from localhost
1827          */
1828         cp = INDIRECT_CALL_1(pp->conn_in_get, ip_vs_conn_in_get_proto,
1829                              ipvs, AF_INET6, skb, &ciph);
1830
1831         if (!cp) {
1832                 int v;
1833
1834                 if (!sysctl_schedule_icmp(ipvs))
1835                         return NF_ACCEPT;
1836
1837                 if (!ip_vs_try_to_schedule(ipvs, AF_INET6, skb, pd, &v, &cp, &ciph))
1838                         return v;
1839
1840                 new_cp = true;
1841         }
1842
1843         /* VS/TUN, VS/DR and LOCALNODE just let it go */
1844         if ((hooknum == NF_INET_LOCAL_OUT) &&
1845             (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1846                 verdict = NF_ACCEPT;
1847                 goto out;
1848         }
1849
1850         /* do the statistics and put it back */
1851         ip_vs_in_stats(cp, skb);
1852
1853         /* Need to mangle contained IPv6 header in ICMPv6 packet */
1854         offset = ciph.len;
1855         if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1856             IPPROTO_SCTP == ciph.protocol)
1857                 offset += 2 * sizeof(__u16); /* Also mangle ports */
1858
1859         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset, hooknum, &ciph);
1860
1861 out:
1862         if (likely(!new_cp))
1863                 __ip_vs_conn_put(cp);
1864         else
1865                 ip_vs_conn_put(cp);
1866
1867         return verdict;
1868 }
1869 #endif
1870
1871
1872 /*
1873  *      Check if it's for virtual services, look it up,
1874  *      and send it on its way...
1875  */
1876 static unsigned int
1877 ip_vs_in(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1878 {
1879         struct ip_vs_iphdr iph;
1880         struct ip_vs_protocol *pp;
1881         struct ip_vs_proto_data *pd;
1882         struct ip_vs_conn *cp;
1883         int ret, pkts;
1884         int conn_reuse_mode;
1885         struct sock *sk;
1886
1887         /* Already marked as IPVS request or reply? */
1888         if (skb->ipvs_property)
1889                 return NF_ACCEPT;
1890
1891         /*
1892          *      Big tappo:
1893          *      - remote client: only PACKET_HOST
1894          *      - route: used for struct net when skb->dev is unset
1895          */
1896         if (unlikely((skb->pkt_type != PACKET_HOST &&
1897                       hooknum != NF_INET_LOCAL_OUT) ||
1898                      !skb_dst(skb))) {
1899                 ip_vs_fill_iph_skb(af, skb, false, &iph);
1900                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1901                               " ignored in hook %u\n",
1902                               skb->pkt_type, iph.protocol,
1903                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1904                 return NF_ACCEPT;
1905         }
1906         /* ipvs enabled in this netns ? */
1907         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1908                 return NF_ACCEPT;
1909
1910         ip_vs_fill_iph_skb(af, skb, false, &iph);
1911
1912         /* Bad... Do not break raw sockets */
1913         sk = skb_to_full_sk(skb);
1914         if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
1915                      af == AF_INET)) {
1916
1917                 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
1918                         return NF_ACCEPT;
1919         }
1920
1921 #ifdef CONFIG_IP_VS_IPV6
1922         if (af == AF_INET6) {
1923                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1924                         int related;
1925                         int verdict = ip_vs_in_icmp_v6(ipvs, skb, &related,
1926                                                        hooknum, &iph);
1927
1928                         if (related)
1929                                 return verdict;
1930                 }
1931         } else
1932 #endif
1933                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1934                         int related;
1935                         int verdict = ip_vs_in_icmp(ipvs, skb, &related,
1936                                                     hooknum);
1937
1938                         if (related)
1939                                 return verdict;
1940                 }
1941
1942         /* Protocol supported? */
1943         pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1944         if (unlikely(!pd)) {
1945                 /* The only way we'll see this packet again is if it's
1946                  * encapsulated, so mark it with ipvs_property=1 so we
1947                  * skip it if we're ignoring tunneled packets
1948                  */
1949                 if (sysctl_ignore_tunneled(ipvs))
1950                         skb->ipvs_property = 1;
1951
1952                 return NF_ACCEPT;
1953         }
1954         pp = pd->pp;
1955         /*
1956          * Check if the packet belongs to an existing connection entry
1957          */
1958         cp = INDIRECT_CALL_1(pp->conn_in_get, ip_vs_conn_in_get_proto,
1959                              ipvs, af, skb, &iph);
1960
1961         conn_reuse_mode = sysctl_conn_reuse_mode(ipvs);
1962         if (conn_reuse_mode && !iph.fragoffs && is_new_conn(skb, &iph) && cp) {
1963                 bool uses_ct = false, resched = false;
1964
1965                 if (unlikely(sysctl_expire_nodest_conn(ipvs)) && cp->dest &&
1966                     unlikely(!atomic_read(&cp->dest->weight))) {
1967                         resched = true;
1968                         uses_ct = ip_vs_conn_uses_conntrack(cp, skb);
1969                 } else if (is_new_conn_expected(cp, conn_reuse_mode)) {
1970                         uses_ct = ip_vs_conn_uses_conntrack(cp, skb);
1971                         if (!atomic_read(&cp->n_control)) {
1972                                 resched = true;
1973                         } else {
1974                                 /* Do not reschedule controlling connection
1975                                  * that uses conntrack while it is still
1976                                  * referenced by controlled connection(s).
1977                                  */
1978                                 resched = !uses_ct;
1979                         }
1980                 }
1981
1982                 if (resched) {
1983                         if (!atomic_read(&cp->n_control))
1984                                 ip_vs_conn_expire_now(cp);
1985                         __ip_vs_conn_put(cp);
1986                         if (uses_ct)
1987                                 return NF_DROP;
1988                         cp = NULL;
1989                 }
1990         }
1991
1992         if (unlikely(!cp)) {
1993                 int v;
1994
1995                 if (!ip_vs_try_to_schedule(ipvs, af, skb, pd, &v, &cp, &iph))
1996                         return v;
1997         }
1998
1999         IP_VS_DBG_PKT(11, af, pp, skb, iph.off, "Incoming packet");
2000
2001         /* Check the server status */
2002         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
2003                 /* the destination server is not available */
2004
2005                 __u32 flags = cp->flags;
2006
2007                 /* when timer already started, silently drop the packet.*/
2008                 if (timer_pending(&cp->timer))
2009                         __ip_vs_conn_put(cp);
2010                 else
2011                         ip_vs_conn_put(cp);
2012
2013                 if (sysctl_expire_nodest_conn(ipvs) &&
2014                     !(flags & IP_VS_CONN_F_ONE_PACKET)) {
2015                         /* try to expire the connection immediately */
2016                         ip_vs_conn_expire_now(cp);
2017                 }
2018
2019                 return NF_DROP;
2020         }
2021
2022         ip_vs_in_stats(cp, skb);
2023         ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
2024         if (cp->packet_xmit)
2025                 ret = cp->packet_xmit(skb, cp, pp, &iph);
2026                 /* do not touch skb anymore */
2027         else {
2028                 IP_VS_DBG_RL("warning: packet_xmit is null");
2029                 ret = NF_ACCEPT;
2030         }
2031
2032         /* Increase its packet counter and check if it is needed
2033          * to be synchronized
2034          *
2035          * Sync connection if it is about to close to
2036          * encorage the standby servers to update the connections timeout
2037          *
2038          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
2039          */
2040
2041         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
2042                 pkts = sysctl_sync_threshold(ipvs);
2043         else
2044                 pkts = atomic_add_return(1, &cp->in_pkts);
2045
2046         if (ipvs->sync_state & IP_VS_STATE_MASTER)
2047                 ip_vs_sync_conn(ipvs, cp, pkts);
2048         else if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) && cp->control)
2049                 /* increment is done inside ip_vs_sync_conn too */
2050                 atomic_inc(&cp->control->in_pkts);
2051
2052         ip_vs_conn_put(cp);
2053         return ret;
2054 }
2055
2056 /*
2057  *      AF_INET handler in NF_INET_LOCAL_IN chain
2058  *      Schedule and forward packets from remote clients
2059  */
2060 static unsigned int
2061 ip_vs_remote_request4(void *priv, struct sk_buff *skb,
2062                       const struct nf_hook_state *state)
2063 {
2064         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
2065 }
2066
2067 /*
2068  *      AF_INET handler in NF_INET_LOCAL_OUT chain
2069  *      Schedule and forward packets from local clients
2070  */
2071 static unsigned int
2072 ip_vs_local_request4(void *priv, struct sk_buff *skb,
2073                      const struct nf_hook_state *state)
2074 {
2075         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
2076 }
2077
2078 #ifdef CONFIG_IP_VS_IPV6
2079
2080 /*
2081  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
2082  *      Schedule and forward packets from remote clients
2083  */
2084 static unsigned int
2085 ip_vs_remote_request6(void *priv, struct sk_buff *skb,
2086                       const struct nf_hook_state *state)
2087 {
2088         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
2089 }
2090
2091 /*
2092  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
2093  *      Schedule and forward packets from local clients
2094  */
2095 static unsigned int
2096 ip_vs_local_request6(void *priv, struct sk_buff *skb,
2097                      const struct nf_hook_state *state)
2098 {
2099         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
2100 }
2101
2102 #endif
2103
2104
2105 /*
2106  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
2107  *      related packets destined for 0.0.0.0/0.
2108  *      When fwmark-based virtual service is used, such as transparent
2109  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
2110  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
2111  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
2112  *      and send them to ip_vs_in_icmp.
2113  */
2114 static unsigned int
2115 ip_vs_forward_icmp(void *priv, struct sk_buff *skb,
2116                    const struct nf_hook_state *state)
2117 {
2118         int r;
2119         struct netns_ipvs *ipvs = net_ipvs(state->net);
2120
2121         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
2122                 return NF_ACCEPT;
2123
2124         /* ipvs enabled in this netns ? */
2125         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
2126                 return NF_ACCEPT;
2127
2128         return ip_vs_in_icmp(ipvs, skb, &r, state->hook);
2129 }
2130
2131 #ifdef CONFIG_IP_VS_IPV6
2132 static unsigned int
2133 ip_vs_forward_icmp_v6(void *priv, struct sk_buff *skb,
2134                       const struct nf_hook_state *state)
2135 {
2136         int r;
2137         struct netns_ipvs *ipvs = net_ipvs(state->net);
2138         struct ip_vs_iphdr iphdr;
2139
2140         ip_vs_fill_iph_skb(AF_INET6, skb, false, &iphdr);
2141         if (iphdr.protocol != IPPROTO_ICMPV6)
2142                 return NF_ACCEPT;
2143
2144         /* ipvs enabled in this netns ? */
2145         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
2146                 return NF_ACCEPT;
2147
2148         return ip_vs_in_icmp_v6(ipvs, skb, &r, state->hook, &iphdr);
2149 }
2150 #endif
2151
2152
2153 static const struct nf_hook_ops ip_vs_ops[] = {
2154         /* After packet filtering, change source only for VS/NAT */
2155         {
2156                 .hook           = ip_vs_reply4,
2157                 .pf             = NFPROTO_IPV4,
2158                 .hooknum        = NF_INET_LOCAL_IN,
2159                 .priority       = NF_IP_PRI_NAT_SRC - 2,
2160         },
2161         /* After packet filtering, forward packet through VS/DR, VS/TUN,
2162          * or VS/NAT(change destination), so that filtering rules can be
2163          * applied to IPVS. */
2164         {
2165                 .hook           = ip_vs_remote_request4,
2166                 .pf             = NFPROTO_IPV4,
2167                 .hooknum        = NF_INET_LOCAL_IN,
2168                 .priority       = NF_IP_PRI_NAT_SRC - 1,
2169         },
2170         /* Before ip_vs_in, change source only for VS/NAT */
2171         {
2172                 .hook           = ip_vs_local_reply4,
2173                 .pf             = NFPROTO_IPV4,
2174                 .hooknum        = NF_INET_LOCAL_OUT,
2175                 .priority       = NF_IP_PRI_NAT_DST + 1,
2176         },
2177         /* After mangle, schedule and forward local requests */
2178         {
2179                 .hook           = ip_vs_local_request4,
2180                 .pf             = NFPROTO_IPV4,
2181                 .hooknum        = NF_INET_LOCAL_OUT,
2182                 .priority       = NF_IP_PRI_NAT_DST + 2,
2183         },
2184         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2185          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2186         {
2187                 .hook           = ip_vs_forward_icmp,
2188                 .pf             = NFPROTO_IPV4,
2189                 .hooknum        = NF_INET_FORWARD,
2190                 .priority       = 99,
2191         },
2192         /* After packet filtering, change source only for VS/NAT */
2193         {
2194                 .hook           = ip_vs_reply4,
2195                 .pf             = NFPROTO_IPV4,
2196                 .hooknum        = NF_INET_FORWARD,
2197                 .priority       = 100,
2198         },
2199 #ifdef CONFIG_IP_VS_IPV6
2200         /* After packet filtering, change source only for VS/NAT */
2201         {
2202                 .hook           = ip_vs_reply6,
2203                 .pf             = NFPROTO_IPV6,
2204                 .hooknum        = NF_INET_LOCAL_IN,
2205                 .priority       = NF_IP6_PRI_NAT_SRC - 2,
2206         },
2207         /* After packet filtering, forward packet through VS/DR, VS/TUN,
2208          * or VS/NAT(change destination), so that filtering rules can be
2209          * applied to IPVS. */
2210         {
2211                 .hook           = ip_vs_remote_request6,
2212                 .pf             = NFPROTO_IPV6,
2213                 .hooknum        = NF_INET_LOCAL_IN,
2214                 .priority       = NF_IP6_PRI_NAT_SRC - 1,
2215         },
2216         /* Before ip_vs_in, change source only for VS/NAT */
2217         {
2218                 .hook           = ip_vs_local_reply6,
2219                 .pf             = NFPROTO_IPV6,
2220                 .hooknum        = NF_INET_LOCAL_OUT,
2221                 .priority       = NF_IP6_PRI_NAT_DST + 1,
2222         },
2223         /* After mangle, schedule and forward local requests */
2224         {
2225                 .hook           = ip_vs_local_request6,
2226                 .pf             = NFPROTO_IPV6,
2227                 .hooknum        = NF_INET_LOCAL_OUT,
2228                 .priority       = NF_IP6_PRI_NAT_DST + 2,
2229         },
2230         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2231          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2232         {
2233                 .hook           = ip_vs_forward_icmp_v6,
2234                 .pf             = NFPROTO_IPV6,
2235                 .hooknum        = NF_INET_FORWARD,
2236                 .priority       = 99,
2237         },
2238         /* After packet filtering, change source only for VS/NAT */
2239         {
2240                 .hook           = ip_vs_reply6,
2241                 .pf             = NFPROTO_IPV6,
2242                 .hooknum        = NF_INET_FORWARD,
2243                 .priority       = 100,
2244         },
2245 #endif
2246 };
2247 /*
2248  *      Initialize IP Virtual Server netns mem.
2249  */
2250 static int __net_init __ip_vs_init(struct net *net)
2251 {
2252         struct netns_ipvs *ipvs;
2253         int ret;
2254
2255         ipvs = net_generic(net, ip_vs_net_id);
2256         if (ipvs == NULL)
2257                 return -ENOMEM;
2258
2259         /* Hold the beast until a service is registerd */
2260         ipvs->enable = 0;
2261         ipvs->net = net;
2262         /* Counters used for creating unique names */
2263         ipvs->gen = atomic_read(&ipvs_netns_cnt);
2264         atomic_inc(&ipvs_netns_cnt);
2265         net->ipvs = ipvs;
2266
2267         if (ip_vs_estimator_net_init(ipvs) < 0)
2268                 goto estimator_fail;
2269
2270         if (ip_vs_control_net_init(ipvs) < 0)
2271                 goto control_fail;
2272
2273         if (ip_vs_protocol_net_init(ipvs) < 0)
2274                 goto protocol_fail;
2275
2276         if (ip_vs_app_net_init(ipvs) < 0)
2277                 goto app_fail;
2278
2279         if (ip_vs_conn_net_init(ipvs) < 0)
2280                 goto conn_fail;
2281
2282         if (ip_vs_sync_net_init(ipvs) < 0)
2283                 goto sync_fail;
2284
2285         ret = nf_register_net_hooks(net, ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2286         if (ret < 0)
2287                 goto hook_fail;
2288
2289         return 0;
2290 /*
2291  * Error handling
2292  */
2293
2294 hook_fail:
2295         ip_vs_sync_net_cleanup(ipvs);
2296 sync_fail:
2297         ip_vs_conn_net_cleanup(ipvs);
2298 conn_fail:
2299         ip_vs_app_net_cleanup(ipvs);
2300 app_fail:
2301         ip_vs_protocol_net_cleanup(ipvs);
2302 protocol_fail:
2303         ip_vs_control_net_cleanup(ipvs);
2304 control_fail:
2305         ip_vs_estimator_net_cleanup(ipvs);
2306 estimator_fail:
2307         net->ipvs = NULL;
2308         return -ENOMEM;
2309 }
2310
2311 static void __net_exit __ip_vs_cleanup(struct net *net)
2312 {
2313         struct netns_ipvs *ipvs = net_ipvs(net);
2314
2315         nf_unregister_net_hooks(net, ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2316         ip_vs_service_net_cleanup(ipvs);        /* ip_vs_flush() with locks */
2317         ip_vs_conn_net_cleanup(ipvs);
2318         ip_vs_app_net_cleanup(ipvs);
2319         ip_vs_protocol_net_cleanup(ipvs);
2320         ip_vs_control_net_cleanup(ipvs);
2321         ip_vs_estimator_net_cleanup(ipvs);
2322         IP_VS_DBG(2, "ipvs netns %d released\n", ipvs->gen);
2323         net->ipvs = NULL;
2324 }
2325
2326 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
2327 {
2328         struct netns_ipvs *ipvs = net_ipvs(net);
2329         EnterFunction(2);
2330         ipvs->enable = 0;       /* Disable packet reception */
2331         smp_wmb();
2332         ip_vs_sync_net_cleanup(ipvs);
2333         LeaveFunction(2);
2334 }
2335
2336 static struct pernet_operations ipvs_core_ops = {
2337         .init = __ip_vs_init,
2338         .exit = __ip_vs_cleanup,
2339         .id   = &ip_vs_net_id,
2340         .size = sizeof(struct netns_ipvs),
2341 };
2342
2343 static struct pernet_operations ipvs_core_dev_ops = {
2344         .exit = __ip_vs_dev_cleanup,
2345 };
2346
2347 /*
2348  *      Initialize IP Virtual Server
2349  */
2350 static int __init ip_vs_init(void)
2351 {
2352         int ret;
2353
2354         ret = ip_vs_control_init();
2355         if (ret < 0) {
2356                 pr_err("can't setup control.\n");
2357                 goto exit;
2358         }
2359
2360         ip_vs_protocol_init();
2361
2362         ret = ip_vs_conn_init();
2363         if (ret < 0) {
2364                 pr_err("can't setup connection table.\n");
2365                 goto cleanup_protocol;
2366         }
2367
2368         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
2369         if (ret < 0)
2370                 goto cleanup_conn;
2371
2372         ret = register_pernet_device(&ipvs_core_dev_ops);
2373         if (ret < 0)
2374                 goto cleanup_sub;
2375
2376         ret = ip_vs_register_nl_ioctl();
2377         if (ret < 0) {
2378                 pr_err("can't register netlink/ioctl.\n");
2379                 goto cleanup_dev;
2380         }
2381
2382         pr_info("ipvs loaded.\n");
2383
2384         return ret;
2385
2386 cleanup_dev:
2387         unregister_pernet_device(&ipvs_core_dev_ops);
2388 cleanup_sub:
2389         unregister_pernet_subsys(&ipvs_core_ops);
2390 cleanup_conn:
2391         ip_vs_conn_cleanup();
2392 cleanup_protocol:
2393         ip_vs_protocol_cleanup();
2394         ip_vs_control_cleanup();
2395 exit:
2396         return ret;
2397 }
2398
2399 static void __exit ip_vs_cleanup(void)
2400 {
2401         ip_vs_unregister_nl_ioctl();
2402         unregister_pernet_device(&ipvs_core_dev_ops);
2403         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
2404         ip_vs_conn_cleanup();
2405         ip_vs_protocol_cleanup();
2406         ip_vs_control_cleanup();
2407         pr_info("ipvs unloaded.\n");
2408 }
2409
2410 module_init(ip_vs_init);
2411 module_exit(ip_vs_cleanup);
2412 MODULE_LICENSE("GPL");