Pileus Git - ~andy/linux/blob - 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
  57
  58 EXPORT_SYMBOL(register_ip_vs_scheduler);
  59 EXPORT_SYMBOL(unregister_ip_vs_scheduler);
  60 EXPORT_SYMBOL(ip_vs_proto_name);
  61 EXPORT_SYMBOL(ip_vs_conn_new);
  62 EXPORT_SYMBOL(ip_vs_conn_in_get);
  63 EXPORT_SYMBOL(ip_vs_conn_out_get);
  64 #ifdef CONFIG_IP_VS_PROTO_TCP
  65 EXPORT_SYMBOL(ip_vs_tcp_conn_listen);
  66 #endif
  67 EXPORT_SYMBOL(ip_vs_conn_put);
  68 #ifdef CONFIG_IP_VS_DEBUG
  69 EXPORT_SYMBOL(ip_vs_get_debug_level);
  70 #endif
  71
  72 int ip_vs_net_id __read_mostly;
  73 #ifdef IP_VS_GENERIC_NETNS
  74 EXPORT_SYMBOL(ip_vs_net_id);
  75 #endif
  76 /* netns cnt used for uniqueness */
  77 static atomic_t ipvs_netns_cnt = ATOMIC_INIT(0);
  78
  79 /* ID used in ICMP lookups */
  80 #define icmp_id(icmph)          (((icmph)->un).echo.id)
  81 #define icmpv6_id(icmph)        (icmph->icmp6_dataun.u_echo.identifier)
  82
  83 const char *ip_vs_proto_name(unsigned proto)
  84 {
  85         static char buf[20];
  86
  87         switch (proto) {
  88         case IPPROTO_IP:
  89                 return "IP";
  90         case IPPROTO_UDP:
  91                 return "UDP";
  92         case IPPROTO_TCP:
  93                 return "TCP";
  94         case IPPROTO_SCTP:
  95                 return "SCTP";
  96         case IPPROTO_ICMP:
  97                 return "ICMP";
  98 #ifdef CONFIG_IP_VS_IPV6
  99         case IPPROTO_ICMPV6:
 100                 return "ICMPv6";
 101 #endif
 102         default:
 103                 sprintf(buf, "IP_%d", proto);
 104                 return buf;
 105         }
 106 }
 107
 108 void ip_vs_init_hash_table(struct list_head *table, int rows)
 109 {
 110         while (--rows >= 0)
 111                 INIT_LIST_HEAD(&table[rows]);
 112 }
 113
 114 static inline void
 115 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 116 {
 117         struct ip_vs_dest *dest = cp->dest;
 118         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 119
 120         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 121                 struct ip_vs_cpu_stats *s;
 122
 123                 s = this_cpu_ptr(dest->stats.cpustats);
 124                 s->ustats.inpkts++;
 125                 u64_stats_update_begin(&s->syncp);
 126                 s->ustats.inbytes += skb->len;
 127                 u64_stats_update_end(&s->syncp);
 128
 129                 s = this_cpu_ptr(dest->svc->stats.cpustats);
 130                 s->ustats.inpkts++;
 131                 u64_stats_update_begin(&s->syncp);
 132                 s->ustats.inbytes += skb->len;
 133                 u64_stats_update_end(&s->syncp);
 134
 135                 s = this_cpu_ptr(ipvs->cpustats);
 136                 s->ustats.inpkts++;
 137                 u64_stats_update_begin(&s->syncp);
 138                 s->ustats.inbytes += skb->len;
 139                 u64_stats_update_end(&s->syncp);
 140         }
 141 }
 142
 143
 144 static inline void
 145 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 146 {
 147         struct ip_vs_dest *dest = cp->dest;
 148         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 149
 150         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 151                 struct ip_vs_cpu_stats *s;
 152
 153                 s = this_cpu_ptr(dest->stats.cpustats);
 154                 s->ustats.outpkts++;
 155                 u64_stats_update_begin(&s->syncp);
 156                 s->ustats.outbytes += skb->len;
 157                 u64_stats_update_end(&s->syncp);
 158
 159                 s = this_cpu_ptr(dest->svc->stats.cpustats);
 160                 s->ustats.outpkts++;
 161                 u64_stats_update_begin(&s->syncp);
 162                 s->ustats.outbytes += skb->len;
 163                 u64_stats_update_end(&s->syncp);
 164
 165                 s = this_cpu_ptr(ipvs->cpustats);
 166                 s->ustats.outpkts++;
 167                 u64_stats_update_begin(&s->syncp);
 168                 s->ustats.outbytes += skb->len;
 169                 u64_stats_update_end(&s->syncp);
 170         }
 171 }
 172
 173
 174 static inline void
 175 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
 176 {
 177         struct netns_ipvs *ipvs = net_ipvs(svc->net);
 178         struct ip_vs_cpu_stats *s;
 179
 180         s = this_cpu_ptr(cp->dest->stats.cpustats);
 181         s->ustats.conns++;
 182
 183         s = this_cpu_ptr(svc->stats.cpustats);
 184         s->ustats.conns++;
 185
 186         s = this_cpu_ptr(ipvs->cpustats);
 187         s->ustats.conns++;
 188 }
 189
 190
 191 static inline int
 192 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
 193                 const struct sk_buff *skb,
 194                 struct ip_vs_proto_data *pd)
 195 {
 196         if (unlikely(!pd->pp->state_transition))
 197                 return 0;
 198         return pd->pp->state_transition(cp, direction, skb, pd);
 199 }
 200
 201 static inline int
 202 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
 203                               struct sk_buff *skb, int protocol,
 204                               const union nf_inet_addr *caddr, __be16 cport,
 205                               const union nf_inet_addr *vaddr, __be16 vport,
 206                               struct ip_vs_conn_param *p)
 207 {
 208         ip_vs_conn_fill_param(svc->net, svc->af, protocol, caddr, cport, vaddr,
 209                               vport, p);
 210         p->pe = svc->pe;
 211         if (p->pe && p->pe->fill_param)
 212                 return p->pe->fill_param(p, skb);
 213
 214         return 0;
 215 }
 216
 217 /*
 218  *  IPVS persistent scheduling function
 219  *  It creates a connection entry according to its template if exists,
 220  *  or selects a server and creates a connection entry plus a template.
 221  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
 222  *  Protocols supported: TCP, UDP
 223  */
 224 static struct ip_vs_conn *
 225 ip_vs_sched_persist(struct ip_vs_service *svc,
 226                     struct sk_buff *skb,
 227                     __be16 src_port, __be16 dst_port, int *ignored)
 228 {
 229         struct ip_vs_conn *cp = NULL;
 230         struct ip_vs_iphdr iph;
 231         struct ip_vs_dest *dest;
 232         struct ip_vs_conn *ct;
 233         __be16 dport = 0;               /* destination port to forward */
 234         unsigned int flags;
 235         struct ip_vs_conn_param param;
 236         union nf_inet_addr snet;        /* source network of the client,
 237                                            after masking */
 238
 239         ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph);
 240
 241         /* Mask saddr with the netmask to adjust template granularity */
 242 #ifdef CONFIG_IP_VS_IPV6
 243         if (svc->af == AF_INET6)
 244                 ipv6_addr_prefix(&snet.in6, &iph.saddr.in6, svc->netmask);
 245         else
 246 #endif
 247                 snet.ip = iph.saddr.ip & svc->netmask;
 248
 249         IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
 250                       "mnet %s\n",
 251                       IP_VS_DBG_ADDR(svc->af, &iph.saddr), ntohs(src_port),
 252                       IP_VS_DBG_ADDR(svc->af, &iph.daddr), ntohs(dst_port),
 253                       IP_VS_DBG_ADDR(svc->af, &snet));
 254
 255         /*
 256          * As far as we know, FTP is a very complicated network protocol, and
 257          * it uses control connection and data connections. For active FTP,
 258          * FTP server initialize data connection to the client, its source port
 259          * is often 20. For passive FTP, FTP server tells the clients the port
 260          * that it passively listens to,  and the client issues the data
 261          * connection. In the tunneling or direct routing mode, the load
 262          * balancer is on the client-to-server half of connection, the port
 263          * number is unknown to the load balancer. So, a conn template like
 264          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
 265          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
 266          * is created for other persistent services.
 267          */
 268         {
 269                 int protocol = iph.protocol;
 270                 const union nf_inet_addr *vaddr = &iph.daddr;
 271                 const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
 272                 __be16 vport = 0;
 273
 274                 if (dst_port == svc->port) {
 275                         /* non-FTP template:
 276                          * <protocol, caddr, 0, vaddr, vport, daddr, dport>
 277                          * FTP template:
 278                          * <protocol, caddr, 0, vaddr, 0, daddr, 0>
 279                          */
 280                         if (svc->port != FTPPORT)
 281                                 vport = dst_port;
 282                 } else {
 283                         /* Note: persistent fwmark-based services and
 284                          * persistent port zero service are handled here.
 285                          * fwmark template:
 286                          * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
 287                          * port zero template:
 288                          * <protocol,caddr,0,vaddr,0,daddr,0>
 289                          */
 290                         if (svc->fwmark) {
 291                                 protocol = IPPROTO_IP;
 292                                 vaddr = &fwmark;
 293                         }
 294                 }
 295                 /* return *ignored = -1 so NF_DROP can be used */
 296                 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
 297                                                   vaddr, vport, &param) < 0) {
 298                         *ignored = -1;
 299                         return NULL;
 300                 }
 301         }
 302
 303         /* Check if a template already exists */
 304         ct = ip_vs_ct_in_get(&param);
 305         if (!ct || !ip_vs_check_template(ct)) {
 306                 /*
 307                  * No template found or the dest of the connection
 308                  * template is not available.
 309                  * return *ignored=0 i.e. ICMP and NF_DROP
 310                  */
 311                 dest = svc->scheduler->schedule(svc, skb);
 312                 if (!dest) {
 313                         IP_VS_DBG(1, "p-schedule: no dest found.\n");
 314                         kfree(param.pe_data);
 315                         *ignored = 0;
 316                         return NULL;
 317                 }
 318
 319                 if (dst_port == svc->port && svc->port != FTPPORT)
 320                         dport = dest->port;
 321
 322                 /* Create a template
 323                  * This adds param.pe_data to the template,
 324                  * and thus param.pe_data will be destroyed
 325                  * when the template expires */
 326                 ct = ip_vs_conn_new(&param, &dest->addr, dport,
 327                                     IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
 328                 if (ct == NULL) {
 329                         kfree(param.pe_data);
 330                         *ignored = -1;
 331                         return NULL;
 332                 }
 333
 334                 ct->timeout = svc->timeout;
 335         } else {
 336                 /* set destination with the found template */
 337                 dest = ct->dest;
 338                 kfree(param.pe_data);
 339         }
 340
 341         dport = dst_port;
 342         if (dport == svc->port && dest->port)
 343                 dport = dest->port;
 344
 345         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 346                  && iph.protocol == IPPROTO_UDP)?
 347                 IP_VS_CONN_F_ONE_PACKET : 0;
 348
 349         /*
 350          *    Create a new connection according to the template
 351          */
 352         ip_vs_conn_fill_param(svc->net, svc->af, iph.protocol, &iph.saddr,
 353                               src_port, &iph.daddr, dst_port, &param);
 354
 355         cp = ip_vs_conn_new(&param, &dest->addr, dport, flags, dest, skb->mark);
 356         if (cp == NULL) {
 357                 ip_vs_conn_put(ct);
 358                 *ignored = -1;
 359                 return NULL;
 360         }
 361
 362         /*
 363          *    Add its control
 364          */
 365         ip_vs_control_add(cp, ct);
 366         ip_vs_conn_put(ct);
 367
 368         ip_vs_conn_stats(cp, svc);
 369         return cp;
 370 }
 371
 372
 373 /*
 374  *  IPVS main scheduling function
 375  *  It selects a server according to the virtual service, and
 376  *  creates a connection entry.
 377  *  Protocols supported: TCP, UDP
 378  *
 379  *  Usage of *ignored
 380  *
 381  * 1 :   protocol tried to schedule (eg. on SYN), found svc but the
 382  *       svc/scheduler decides that this packet should be accepted with
 383  *       NF_ACCEPT because it must not be scheduled.
 384  *
 385  * 0 :   scheduler can not find destination, so try bypass or
 386  *       return ICMP and then NF_DROP (ip_vs_leave).
 387  *
 388  * -1 :  scheduler tried to schedule but fatal error occurred, eg.
 389  *       ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
 390  *       failure such as missing Call-ID, ENOMEM on skb_linearize
 391  *       or pe_data. In this case we should return NF_DROP without
 392  *       any attempts to send ICMP with ip_vs_leave.
 393  */
 394 struct ip_vs_conn *
 395 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
 396                struct ip_vs_proto_data *pd, int *ignored)
 397 {
 398         struct ip_vs_protocol *pp = pd->pp;
 399         struct ip_vs_conn *cp = NULL;
 400         struct ip_vs_iphdr iph;
 401         struct ip_vs_dest *dest;
 402         __be16 _ports[2], *pptr;
 403         unsigned int flags;
 404
 405         *ignored = 1;
 406         ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph);
 407         pptr = skb_header_pointer(skb, iph.len, sizeof(_ports), _ports);
 408         if (pptr == NULL)
 409                 return NULL;
 410
 411         /*
 412          * FTPDATA needs this check when using local real server.
 413          * Never schedule Active FTPDATA connections from real server.
 414          * For LVS-NAT they must be already created. For other methods
 415          * with persistence the connection is created on SYN+ACK.
 416          */
 417         if (pptr[0] == FTPDATA) {
 418                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 419                               "Not scheduling FTPDATA");
 420                 return NULL;
 421         }
 422
 423         /*
 424          *    Do not schedule replies from local real server.
 425          */
 426         if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
 427             (cp = pp->conn_in_get(svc->af, skb, &iph, iph.len, 1))) {
 428                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 429                               "Not scheduling reply for existing connection");
 430                 __ip_vs_conn_put(cp);
 431                 return NULL;
 432         }
 433
 434         /*
 435          *    Persistent service
 436          */
 437         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 438                 return ip_vs_sched_persist(svc, skb, pptr[0], pptr[1], ignored);
 439
 440         *ignored = 0;
 441
 442         /*
 443          *    Non-persistent service
 444          */
 445         if (!svc->fwmark && pptr[1] != svc->port) {
 446                 if (!svc->port)
 447                         pr_err("Schedule: port zero only supported "
 448                                "in persistent services, "
 449                                "check your ipvs configuration\n");
 450                 return NULL;
 451         }
 452
 453         dest = svc->scheduler->schedule(svc, skb);
 454         if (dest == NULL) {
 455                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 456                 return NULL;
 457         }
 458
 459         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 460                  && iph.protocol == IPPROTO_UDP)?
 461                 IP_VS_CONN_F_ONE_PACKET : 0;
 462
 463         /*
 464          *    Create a connection entry.
 465          */
 466         {
 467                 struct ip_vs_conn_param p;
 468
 469                 ip_vs_conn_fill_param(svc->net, svc->af, iph.protocol,
 470                                       &iph.saddr, pptr[0], &iph.daddr, pptr[1],
 471                                       &p);
 472                 cp = ip_vs_conn_new(&p, &dest->addr,
 473                                     dest->port ? dest->port : pptr[1],
 474                                     flags, dest, skb->mark);
 475                 if (!cp) {
 476                         *ignored = -1;
 477                         return NULL;
 478                 }
 479         }
 480
 481         IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
 482                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
 483                       ip_vs_fwd_tag(cp),
 484                       IP_VS_DBG_ADDR(svc->af, &cp->caddr), ntohs(cp->cport),
 485                       IP_VS_DBG_ADDR(svc->af, &cp->vaddr), ntohs(cp->vport),
 486                       IP_VS_DBG_ADDR(svc->af, &cp->daddr), ntohs(cp->dport),
 487                       cp->flags, atomic_read(&cp->refcnt));
 488
 489         ip_vs_conn_stats(cp, svc);
 490         return cp;
 491 }
 492
 493
 494 /*
 495  *  Pass or drop the packet.
 496  *  Called by ip_vs_in, when the virtual service is available but
 497  *  no destination is available for a new connection.
 498  */
 499 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 500                 struct ip_vs_proto_data *pd)
 501 {
 502         struct net *net;
 503         struct netns_ipvs *ipvs;
 504         __be16 _ports[2], *pptr;
 505         struct ip_vs_iphdr iph;
 506         int unicast;
 507
 508         ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph);
 509
 510         pptr = skb_header_pointer(skb, iph.len, sizeof(_ports), _ports);
 511         if (pptr == NULL) {
 512                 ip_vs_service_put(svc);
 513                 return NF_DROP;
 514         }
 515         net = skb_net(skb);
 516
 517 #ifdef CONFIG_IP_VS_IPV6
 518         if (svc->af == AF_INET6)
 519                 unicast = ipv6_addr_type(&iph.daddr.in6) & IPV6_ADDR_UNICAST;
 520         else
 521 #endif
 522                 unicast = (inet_addr_type(net, iph.daddr.ip) == RTN_UNICAST);
 523
 524         /* if it is fwmark-based service, the cache_bypass sysctl is up
 525            and the destination is a non-local unicast, then create
 526            a cache_bypass connection entry */
 527         ipvs = net_ipvs(net);
 528         if (ipvs->sysctl_cache_bypass && svc->fwmark && unicast) {
 529                 int ret, cs;
 530                 struct ip_vs_conn *cp;
 531                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
 532                                       iph.protocol == IPPROTO_UDP)?
 533                                       IP_VS_CONN_F_ONE_PACKET : 0;
 534                 union nf_inet_addr daddr =  { .all = { 0, 0, 0, 0 } };
 535
 536                 ip_vs_service_put(svc);
 537
 538                 /* create a new connection entry */
 539                 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
 540                 {
 541                         struct ip_vs_conn_param p;
 542                         ip_vs_conn_fill_param(svc->net, svc->af, iph.protocol,
 543                                               &iph.saddr, pptr[0],
 544                                               &iph.daddr, pptr[1], &p);
 545                         cp = ip_vs_conn_new(&p, &daddr, 0,
 546                                             IP_VS_CONN_F_BYPASS | flags,
 547                                             NULL, skb->mark);
 548                         if (!cp)
 549                                 return NF_DROP;
 550                 }
 551
 552                 /* statistics */
 553                 ip_vs_in_stats(cp, skb);
 554
 555                 /* set state */
 556                 cs = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
 557
 558                 /* transmit the first SYN packet */
 559                 ret = cp->packet_xmit(skb, cp, pd->pp);
 560                 /* do not touch skb anymore */
 561
 562                 atomic_inc(&cp->in_pkts);
 563                 ip_vs_conn_put(cp);
 564                 return ret;
 565         }
 566
 567         /*
 568          * When the virtual ftp service is presented, packets destined
 569          * for other services on the VIP may get here (except services
 570          * listed in the ipvs table), pass the packets, because it is
 571          * not ipvs job to decide to drop the packets.
 572          */
 573         if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT)) {
 574                 ip_vs_service_put(svc);
 575                 return NF_ACCEPT;
 576         }
 577
 578         ip_vs_service_put(svc);
 579
 580         /*
 581          * Notify the client that the destination is unreachable, and
 582          * release the socket buffer.
 583          * Since it is in IP layer, the TCP socket is not actually
 584          * created, the TCP RST packet cannot be sent, instead that
 585          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 586          */
 587 #ifdef CONFIG_IP_VS_IPV6
 588         if (svc->af == AF_INET6) {
 589                 if (!skb->dev) {
 590                         struct net *net = dev_net(skb_dst(skb)->dev);
 591
 592                         skb->dev = net->loopback_dev;
 593                 }
 594                 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 595         } else
 596 #endif
 597                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 598
 599         return NF_DROP;
 600 }
 601
 602 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 603 {
 604         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 605 }
 606
 607 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
 608 {
 609         if (NF_INET_LOCAL_IN == hooknum)
 610                 return IP_DEFRAG_VS_IN;
 611         if (NF_INET_FORWARD == hooknum)
 612                 return IP_DEFRAG_VS_FWD;
 613         return IP_DEFRAG_VS_OUT;
 614 }
 615
 616 static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user)
 617 {
 618         int err = ip_defrag(skb, user);
 619
 620         if (!err)
 621                 ip_send_check(ip_hdr(skb));
 622
 623         return err;
 624 }
 625
 626 #ifdef CONFIG_IP_VS_IPV6
 627 static inline int ip_vs_gather_frags_v6(struct sk_buff *skb, u_int32_t user)
 628 {
 629         /* TODO IPv6: Find out what to do here for IPv6 */
 630         return 0;
 631 }
 632 #endif
 633
 634 /*
 635  * Packet has been made sufficiently writable in caller
 636  * - inout: 1=in->out, 0=out->in
 637  */
 638 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 639                     struct ip_vs_conn *cp, int inout)
 640 {
 641         struct iphdr *iph        = ip_hdr(skb);
 642         unsigned int icmp_offset = iph->ihl*4;
 643         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
 644                                                       icmp_offset);
 645         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 646
 647         if (inout) {
 648                 iph->saddr = cp->vaddr.ip;
 649                 ip_send_check(iph);
 650                 ciph->daddr = cp->vaddr.ip;
 651                 ip_send_check(ciph);
 652         } else {
 653                 iph->daddr = cp->daddr.ip;
 654                 ip_send_check(iph);
 655                 ciph->saddr = cp->daddr.ip;
 656                 ip_send_check(ciph);
 657         }
 658
 659         /* the TCP/UDP/SCTP port */
 660         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
 661             IPPROTO_SCTP == ciph->protocol) {
 662                 __be16 *ports = (void *)ciph + ciph->ihl*4;
 663
 664                 if (inout)
 665                         ports[1] = cp->vport;
 666                 else
 667                         ports[0] = cp->dport;
 668         }
 669
 670         /* And finally the ICMP checksum */
 671         icmph->checksum = 0;
 672         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 673         skb->ip_summed = CHECKSUM_UNNECESSARY;
 674
 675         if (inout)
 676                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 677                         "Forwarding altered outgoing ICMP");
 678         else
 679                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 680                         "Forwarding altered incoming ICMP");
 681 }
 682
 683 #ifdef CONFIG_IP_VS_IPV6
 684 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
 685                     struct ip_vs_conn *cp, int inout)
 686 {
 687         struct ipv6hdr *iph      = ipv6_hdr(skb);
 688         unsigned int icmp_offset = sizeof(struct ipv6hdr);
 689         struct icmp6hdr *icmph   = (struct icmp6hdr *)(skb_network_header(skb) +
 690                                                       icmp_offset);
 691         struct ipv6hdr *ciph     = (struct ipv6hdr *)(icmph + 1);
 692
 693         if (inout) {
 694                 iph->saddr = cp->vaddr.in6;
 695                 ciph->daddr = cp->vaddr.in6;
 696         } else {
 697                 iph->daddr = cp->daddr.in6;
 698                 ciph->saddr = cp->daddr.in6;
 699         }
 700
 701         /* the TCP/UDP/SCTP port */
 702         if (IPPROTO_TCP == ciph->nexthdr || IPPROTO_UDP == ciph->nexthdr ||
 703             IPPROTO_SCTP == ciph->nexthdr) {
 704                 __be16 *ports = (void *)ciph + sizeof(struct ipv6hdr);
 705
 706                 if (inout)
 707                         ports[1] = cp->vport;
 708                 else
 709                         ports[0] = cp->dport;
 710         }
 711
 712         /* And finally the ICMP checksum */
 713         icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
 714                                               skb->len - icmp_offset,
 715                                               IPPROTO_ICMPV6, 0);
 716         skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
 717         skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
 718         skb->ip_summed = CHECKSUM_PARTIAL;
 719
 720         if (inout)
 721                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 722                               (void *)ciph - (void *)iph,
 723                               "Forwarding altered outgoing ICMPv6");
 724         else
 725                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 726                               (void *)ciph - (void *)iph,
 727                               "Forwarding altered incoming ICMPv6");
 728 }
 729 #endif
 730
 731 /* Handle relevant response ICMP messages - forward to the right
 732  * destination host.
 733  */
 734 static int handle_response_icmp(int af, struct sk_buff *skb,
 735                                 union nf_inet_addr *snet,
 736                                 __u8 protocol, struct ip_vs_conn *cp,
 737                                 struct ip_vs_protocol *pp,
 738                                 unsigned int offset, unsigned int ihl)
 739 {
 740         struct netns_ipvs *ipvs;
 741         unsigned int verdict = NF_DROP;
 742
 743         if (IP_VS_FWD_METHOD(cp) != 0) {
 744                 pr_err("shouldn't reach here, because the box is on the "
 745                        "half connection in the tun/dr module.\n");
 746         }
 747
 748         /* Ensure the checksum is correct */
 749         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 750                 /* Failed checksum! */
 751                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
 752                               IP_VS_DBG_ADDR(af, snet));
 753                 goto out;
 754         }
 755
 756         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 757             IPPROTO_SCTP == protocol)
 758                 offset += 2 * sizeof(__u16);
 759         if (!skb_make_writable(skb, offset))
 760                 goto out;
 761
 762         ipvs = net_ipvs(skb_net(skb));
 763
 764 #ifdef CONFIG_IP_VS_IPV6
 765         if (af == AF_INET6)
 766                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
 767         else
 768 #endif
 769                 ip_vs_nat_icmp(skb, pp, cp, 1);
 770
 771 #ifdef CONFIG_IP_VS_IPV6
 772         if (af == AF_INET6) {
 773                 if (ipvs->sysctl_snat_reroute && ip6_route_me_harder(skb) != 0)
 774                         goto out;
 775         } else
 776 #endif
 777                 if ((ipvs->sysctl_snat_reroute ||
 778                      skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
 779                     ip_route_me_harder(skb, RTN_LOCAL) != 0)
 780                         goto out;
 781
 782         /* do the statistics and put it back */
 783         ip_vs_out_stats(cp, skb);
 784
 785         skb->ipvs_property = 1;
 786         if (!(cp->flags & IP_VS_CONN_F_NFCT))
 787                 ip_vs_notrack(skb);
 788         else
 789                 ip_vs_update_conntrack(skb, cp, 0);
 790         verdict = NF_ACCEPT;
 791
 792 out:
 793         __ip_vs_conn_put(cp);
 794
 795         return verdict;
 796 }
 797
 798 /*
 799  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 800  *      Find any that might be relevant, check against existing connections.
 801  *      Currently handles error types - unreachable, quench, ttl exceeded.
 802  */
 803 static int ip_vs_out_icmp(struct sk_buff *skb, int *related,
 804                           unsigned int hooknum)
 805 {
 806         struct iphdr *iph;
 807         struct icmphdr  _icmph, *ic;
 808         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 809         struct ip_vs_iphdr ciph;
 810         struct ip_vs_conn *cp;
 811         struct ip_vs_protocol *pp;
 812         unsigned int offset, ihl;
 813         union nf_inet_addr snet;
 814
 815         *related = 1;
 816
 817         /* reassemble IP fragments */
 818         if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
 819                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
 820                         return NF_STOLEN;
 821         }
 822
 823         iph = ip_hdr(skb);
 824         offset = ihl = iph->ihl * 4;
 825         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 826         if (ic == NULL)
 827                 return NF_DROP;
 828
 829         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
 830                   ic->type, ntohs(icmp_id(ic)),
 831                   &iph->saddr, &iph->daddr);
 832
 833         /*
 834          * Work through seeing if this is for us.
 835          * These checks are supposed to be in an order that means easy
 836          * things are checked first to speed up processing.... however
 837          * this means that some packets will manage to get a long way
 838          * down this stack and then be rejected, but that's life.
 839          */
 840         if ((ic->type != ICMP_DEST_UNREACH) &&
 841             (ic->type != ICMP_SOURCE_QUENCH) &&
 842             (ic->type != ICMP_TIME_EXCEEDED)) {
 843                 *related = 0;
 844                 return NF_ACCEPT;
 845         }
 846
 847         /* Now find the contained IP header */
 848         offset += sizeof(_icmph);
 849         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 850         if (cih == NULL)
 851                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 852
 853         pp = ip_vs_proto_get(cih->protocol);
 854         if (!pp)
 855                 return NF_ACCEPT;
 856
 857         /* Is the embedded protocol header present? */
 858         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 859                      pp->dont_defrag))
 860                 return NF_ACCEPT;
 861
 862         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
 863                       "Checking outgoing ICMP for");
 864
 865         offset += cih->ihl * 4;
 866
 867         ip_vs_fill_iphdr(AF_INET, cih, &ciph);
 868         /* The embedded headers contain source and dest in reverse order */
 869         cp = pp->conn_out_get(AF_INET, skb, &ciph, offset, 1);
 870         if (!cp)
 871                 return NF_ACCEPT;
 872
 873         snet.ip = iph->saddr;
 874         return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
 875                                     pp, offset, ihl);
 876 }
 877
 878 #ifdef CONFIG_IP_VS_IPV6
 879 static int ip_vs_out_icmp_v6(struct sk_buff *skb, int *related,
 880                              unsigned int hooknum)
 881 {
 882         struct ipv6hdr *iph;
 883         struct icmp6hdr _icmph, *ic;
 884         struct ipv6hdr  _ciph, *cih;    /* The ip header contained
 885                                            within the ICMP */
 886         struct ip_vs_iphdr ciph;
 887         struct ip_vs_conn *cp;
 888         struct ip_vs_protocol *pp;
 889         unsigned int offset;
 890         union nf_inet_addr snet;
 891
 892         *related = 1;
 893
 894         /* reassemble IP fragments */
 895         if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) {
 896                 if (ip_vs_gather_frags_v6(skb, ip_vs_defrag_user(hooknum)))
 897                         return NF_STOLEN;
 898         }
 899
 900         iph = ipv6_hdr(skb);
 901         offset = sizeof(struct ipv6hdr);
 902         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 903         if (ic == NULL)
 904                 return NF_DROP;
 905
 906         IP_VS_DBG(12, "Outgoing ICMPv6 (%d,%d) %pI6->%pI6\n",
 907                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
 908                   &iph->saddr, &iph->daddr);
 909
 910         /*
 911          * Work through seeing if this is for us.
 912          * These checks are supposed to be in an order that means easy
 913          * things are checked first to speed up processing.... however
 914          * this means that some packets will manage to get a long way
 915          * down this stack and then be rejected, but that's life.
 916          */
 917         if ((ic->icmp6_type != ICMPV6_DEST_UNREACH) &&
 918             (ic->icmp6_type != ICMPV6_PKT_TOOBIG) &&
 919             (ic->icmp6_type != ICMPV6_TIME_EXCEED)) {
 920                 *related = 0;
 921                 return NF_ACCEPT;
 922         }
 923
 924         /* Now find the contained IP header */
 925         offset += sizeof(_icmph);
 926         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 927         if (cih == NULL)
 928                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 929
 930         pp = ip_vs_proto_get(cih->nexthdr);
 931         if (!pp)
 932                 return NF_ACCEPT;
 933
 934         /* Is the embedded protocol header present? */
 935         /* TODO: we don't support fragmentation at the moment anyways */
 936         if (unlikely(cih->nexthdr == IPPROTO_FRAGMENT && pp->dont_defrag))
 937                 return NF_ACCEPT;
 938
 939         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
 940                       "Checking outgoing ICMPv6 for");
 941
 942         offset += sizeof(struct ipv6hdr);
 943
 944         ip_vs_fill_iphdr(AF_INET6, cih, &ciph);
 945         /* The embedded headers contain source and dest in reverse order */
 946         cp = pp->conn_out_get(AF_INET6, skb, &ciph, offset, 1);
 947         if (!cp)
 948                 return NF_ACCEPT;
 949
 950         ipv6_addr_copy(&snet.in6, &iph->saddr);
 951         return handle_response_icmp(AF_INET6, skb, &snet, cih->nexthdr, cp,
 952                                     pp, offset, sizeof(struct ipv6hdr));
 953 }
 954 #endif
 955
 956 /*
 957  * Check if sctp chunc is ABORT chunk
 958  */
 959 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
 960 {
 961         sctp_chunkhdr_t *sch, schunk;
 962         sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
 963                         sizeof(schunk), &schunk);
 964         if (sch == NULL)
 965                 return 0;
 966         if (sch->type == SCTP_CID_ABORT)
 967                 return 1;
 968         return 0;
 969 }
 970
 971 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
 972 {
 973         struct tcphdr _tcph, *th;
 974
 975         th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
 976         if (th == NULL)
 977                 return 0;
 978         return th->rst;
 979 }
 980
 981 /* Handle response packets: rewrite addresses and send away...
 982  */
 983 static unsigned int
 984 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
 985                 struct ip_vs_conn *cp, int ihl)
 986 {
 987         struct ip_vs_protocol *pp = pd->pp;
 988         struct netns_ipvs *ipvs;
 989
 990         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Outgoing packet");
 991
 992         if (!skb_make_writable(skb, ihl))
 993                 goto drop;
 994
 995         /* mangle the packet */
 996         if (pp->snat_handler && !pp->snat_handler(skb, pp, cp))
 997                 goto drop;
 998
 999 #ifdef CONFIG_IP_VS_IPV6
1000         if (af == AF_INET6)
1001                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1002         else
1003 #endif
1004         {
1005                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1006                 ip_send_check(ip_hdr(skb));
1007         }
1008
1009         /*
1010          * nf_iterate does not expect change in the skb->dst->dev.
1011          * It looks like it is not fatal to enable this code for hooks
1012          * where our handlers are at the end of the chain list and
1013          * when all next handlers use skb->dst->dev and not outdev.
1014          * It will definitely route properly the inout NAT traffic
1015          * when multiple paths are used.
1016          */
1017
1018         /* For policy routing, packets originating from this
1019          * machine itself may be routed differently to packets
1020          * passing through.  We want this packet to be routed as
1021          * if it came from this machine itself.  So re-compute
1022          * the routing information.
1023          */
1024         ipvs = net_ipvs(skb_net(skb));
1025
1026 #ifdef CONFIG_IP_VS_IPV6
1027         if (af == AF_INET6) {
1028                 if (ipvs->sysctl_snat_reroute && ip6_route_me_harder(skb) != 0)
1029                         goto drop;
1030         } else
1031 #endif
1032                 if ((ipvs->sysctl_snat_reroute ||
1033                      skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
1034                     ip_route_me_harder(skb, RTN_LOCAL) != 0)
1035                         goto drop;
1036
1037         IP_VS_DBG_PKT(10, af, pp, skb, 0, "After SNAT");
1038
1039         ip_vs_out_stats(cp, skb);
1040         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1041         skb->ipvs_property = 1;
1042         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1043                 ip_vs_notrack(skb);
1044         else
1045                 ip_vs_update_conntrack(skb, cp, 0);
1046         ip_vs_conn_put(cp);
1047
1048         LeaveFunction(11);
1049         return NF_ACCEPT;
1050
1051 drop:
1052         ip_vs_conn_put(cp);
1053         kfree_skb(skb);
1054         LeaveFunction(11);
1055         return NF_STOLEN;
1056 }
1057
1058 /*
1059  *      Check if outgoing packet belongs to the established ip_vs_conn.
1060  */
1061 static unsigned int
1062 ip_vs_out(unsigned int hooknum, struct sk_buff *skb, int af)
1063 {
1064         struct net *net = NULL;
1065         struct ip_vs_iphdr iph;
1066         struct ip_vs_protocol *pp;
1067         struct ip_vs_proto_data *pd;
1068         struct ip_vs_conn *cp;
1069         struct netns_ipvs *ipvs;
1070
1071         EnterFunction(11);
1072
1073         /* Already marked as IPVS request or reply? */
1074         if (skb->ipvs_property)
1075                 return NF_ACCEPT;
1076
1077         /* Bad... Do not break raw sockets */
1078         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1079                      af == AF_INET)) {
1080                 struct sock *sk = skb->sk;
1081                 struct inet_sock *inet = inet_sk(skb->sk);
1082
1083                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1084                         return NF_ACCEPT;
1085         }
1086
1087         if (unlikely(!skb_dst(skb)))
1088                 return NF_ACCEPT;
1089
1090         net = skb_net(skb);
1091         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1092 #ifdef CONFIG_IP_VS_IPV6
1093         if (af == AF_INET6) {
1094                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1095                         int related;
1096                         int verdict = ip_vs_out_icmp_v6(skb, &related,
1097                                                         hooknum);
1098
1099                         if (related)
1100                                 return verdict;
1101                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1102                 }
1103         } else
1104 #endif
1105                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1106                         int related;
1107                         int verdict = ip_vs_out_icmp(skb, &related, hooknum);
1108
1109                         if (related)
1110                                 return verdict;
1111                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1112                 }
1113
1114         pd = ip_vs_proto_data_get(net, iph.protocol);
1115         if (unlikely(!pd))
1116                 return NF_ACCEPT;
1117         pp = pd->pp;
1118
1119         /* reassemble IP fragments */
1120 #ifdef CONFIG_IP_VS_IPV6
1121         if (af == AF_INET6) {
1122                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) {
1123                         if (ip_vs_gather_frags_v6(skb,
1124                                                   ip_vs_defrag_user(hooknum)))
1125                                 return NF_STOLEN;
1126                 }
1127
1128                 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1129         } else
1130 #endif
1131                 if (unlikely(ip_hdr(skb)->frag_off & htons(IP_MF|IP_OFFSET) &&
1132                              !pp->dont_defrag)) {
1133                         if (ip_vs_gather_frags(skb,
1134                                                ip_vs_defrag_user(hooknum)))
1135                                 return NF_STOLEN;
1136
1137                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1138                 }
1139
1140         /*
1141          * Check if the packet belongs to an existing entry
1142          */
1143         cp = pp->conn_out_get(af, skb, &iph, iph.len, 0);
1144         ipvs = net_ipvs(net);
1145
1146         if (likely(cp))
1147                 return handle_response(af, skb, pd, cp, iph.len);
1148         if (ipvs->sysctl_nat_icmp_send &&
1149             (pp->protocol == IPPROTO_TCP ||
1150              pp->protocol == IPPROTO_UDP ||
1151              pp->protocol == IPPROTO_SCTP)) {
1152                 __be16 _ports[2], *pptr;
1153
1154                 pptr = skb_header_pointer(skb, iph.len,
1155                                           sizeof(_ports), _ports);
1156                 if (pptr == NULL)
1157                         return NF_ACCEPT;       /* Not for me */
1158                 if (ip_vs_lookup_real_service(net, af, iph.protocol,
1159                                               &iph.saddr,
1160                                               pptr[0])) {
1161                         /*
1162                          * Notify the real server: there is no
1163                          * existing entry if it is not RST
1164                          * packet or not TCP packet.
1165                          */
1166                         if ((iph.protocol != IPPROTO_TCP &&
1167                              iph.protocol != IPPROTO_SCTP)
1168                              || ((iph.protocol == IPPROTO_TCP
1169                                   && !is_tcp_reset(skb, iph.len))
1170                                  || (iph.protocol == IPPROTO_SCTP
1171                                         && !is_sctp_abort(skb,
1172                                                 iph.len)))) {
1173 #ifdef CONFIG_IP_VS_IPV6
1174                                 if (af == AF_INET6) {
1175                                         struct net *net =
1176                                                 dev_net(skb_dst(skb)->dev);
1177
1178                                         if (!skb->dev)
1179                                                 skb->dev = net->loopback_dev;
1180                                         icmpv6_send(skb,
1181                                                     ICMPV6_DEST_UNREACH,
1182                                                     ICMPV6_PORT_UNREACH,
1183                                                     0);
1184                                 } else
1185 #endif
1186                                         icmp_send(skb,
1187                                                   ICMP_DEST_UNREACH,
1188                                                   ICMP_PORT_UNREACH, 0);
1189                                 return NF_DROP;
1190                         }
1191                 }
1192         }
1193         IP_VS_DBG_PKT(12, af, pp, skb, 0,
1194                       "ip_vs_out: packet continues traversal as normal");
1195         return NF_ACCEPT;
1196 }
1197
1198 /*
1199  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1200  *      used only for VS/NAT.
1201  *      Check if packet is reply for established ip_vs_conn.
1202  */
1203 static unsigned int
1204 ip_vs_reply4(unsigned int hooknum, struct sk_buff *skb,
1205              const struct net_device *in, const struct net_device *out,
1206              int (*okfn)(struct sk_buff *))
1207 {
1208         return ip_vs_out(hooknum, skb, AF_INET);
1209 }
1210
1211 /*
1212  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1213  *      Check if packet is reply for established ip_vs_conn.
1214  */
1215 static unsigned int
1216 ip_vs_local_reply4(unsigned int hooknum, struct sk_buff *skb,
1217                    const struct net_device *in, const struct net_device *out,
1218                    int (*okfn)(struct sk_buff *))
1219 {
1220         unsigned int verdict;
1221
1222         /* Disable BH in LOCAL_OUT until all places are fixed */
1223         local_bh_disable();
1224         verdict = ip_vs_out(hooknum, skb, AF_INET);
1225         local_bh_enable();
1226         return verdict;
1227 }
1228
1229 #ifdef CONFIG_IP_VS_IPV6
1230
1231 /*
1232  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1233  *      used only for VS/NAT.
1234  *      Check if packet is reply for established ip_vs_conn.
1235  */
1236 static unsigned int
1237 ip_vs_reply6(unsigned int hooknum, struct sk_buff *skb,
1238              const struct net_device *in, const struct net_device *out,
1239              int (*okfn)(struct sk_buff *))
1240 {
1241         return ip_vs_out(hooknum, skb, AF_INET6);
1242 }
1243
1244 /*
1245  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1246  *      Check if packet is reply for established ip_vs_conn.
1247  */
1248 static unsigned int
1249 ip_vs_local_reply6(unsigned int hooknum, struct sk_buff *skb,
1250                    const struct net_device *in, const struct net_device *out,
1251                    int (*okfn)(struct sk_buff *))
1252 {
1253         unsigned int verdict;
1254
1255         /* Disable BH in LOCAL_OUT until all places are fixed */
1256         local_bh_disable();
1257         verdict = ip_vs_out(hooknum, skb, AF_INET6);
1258         local_bh_enable();
1259         return verdict;
1260 }
1261
1262 #endif
1263
1264 /*
1265  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1266  *      Find any that might be relevant, check against existing connections,
1267  *      forward to the right destination host if relevant.
1268  *      Currently handles error types - unreachable, quench, ttl exceeded.
1269  */
1270 static int
1271 ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum)
1272 {
1273         struct net *net = NULL;
1274         struct iphdr *iph;
1275         struct icmphdr  _icmph, *ic;
1276         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1277         struct ip_vs_iphdr ciph;
1278         struct ip_vs_conn *cp;
1279         struct ip_vs_protocol *pp;
1280         struct ip_vs_proto_data *pd;
1281         unsigned int offset, ihl, verdict;
1282
1283         *related = 1;
1284
1285         /* reassemble IP fragments */
1286         if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
1287                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
1288                         return NF_STOLEN;
1289         }
1290
1291         iph = ip_hdr(skb);
1292         offset = ihl = iph->ihl * 4;
1293         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1294         if (ic == NULL)
1295                 return NF_DROP;
1296
1297         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1298                   ic->type, ntohs(icmp_id(ic)),
1299                   &iph->saddr, &iph->daddr);
1300
1301         /*
1302          * Work through seeing if this is for us.
1303          * These checks are supposed to be in an order that means easy
1304          * things are checked first to speed up processing.... however
1305          * this means that some packets will manage to get a long way
1306          * down this stack and then be rejected, but that's life.
1307          */
1308         if ((ic->type != ICMP_DEST_UNREACH) &&
1309             (ic->type != ICMP_SOURCE_QUENCH) &&
1310             (ic->type != ICMP_TIME_EXCEEDED)) {
1311                 *related = 0;
1312                 return NF_ACCEPT;
1313         }
1314
1315         /* Now find the contained IP header */
1316         offset += sizeof(_icmph);
1317         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1318         if (cih == NULL)
1319                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1320
1321         net = skb_net(skb);
1322         pd = ip_vs_proto_data_get(net, cih->protocol);
1323         if (!pd)
1324                 return NF_ACCEPT;
1325         pp = pd->pp;
1326
1327         /* Is the embedded protocol header present? */
1328         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1329                      pp->dont_defrag))
1330                 return NF_ACCEPT;
1331
1332         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1333                       "Checking incoming ICMP for");
1334
1335         offset += cih->ihl * 4;
1336
1337         ip_vs_fill_iphdr(AF_INET, cih, &ciph);
1338         /* The embedded headers contain source and dest in reverse order */
1339         cp = pp->conn_in_get(AF_INET, skb, &ciph, offset, 1);
1340         if (!cp)
1341                 return NF_ACCEPT;
1342
1343         verdict = NF_DROP;
1344
1345         /* Ensure the checksum is correct */
1346         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1347                 /* Failed checksum! */
1348                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1349                           &iph->saddr);
1350                 goto out;
1351         }
1352
1353         /* do the statistics and put it back */
1354         ip_vs_in_stats(cp, skb);
1355         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol)
1356                 offset += 2 * sizeof(__u16);
1357         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset);
1358         /* LOCALNODE from FORWARD hook is not supported */
1359         if (verdict == NF_ACCEPT && hooknum == NF_INET_FORWARD &&
1360             skb_rtable(skb)->rt_flags & RTCF_LOCAL) {
1361                 IP_VS_DBG(1, "%s(): "
1362                           "local delivery to %pI4 but in FORWARD\n",
1363                           __func__, &skb_rtable(skb)->rt_dst);
1364                 verdict = NF_DROP;
1365         }
1366
1367   out:
1368         __ip_vs_conn_put(cp);
1369
1370         return verdict;
1371 }
1372
1373 #ifdef CONFIG_IP_VS_IPV6
1374 static int
1375 ip_vs_in_icmp_v6(struct sk_buff *skb, int *related, unsigned int hooknum)
1376 {
1377         struct net *net = NULL;
1378         struct ipv6hdr *iph;
1379         struct icmp6hdr _icmph, *ic;
1380         struct ipv6hdr  _ciph, *cih;    /* The ip header contained
1381                                            within the ICMP */
1382         struct ip_vs_iphdr ciph;
1383         struct ip_vs_conn *cp;
1384         struct ip_vs_protocol *pp;
1385         struct ip_vs_proto_data *pd;
1386         unsigned int offset, verdict;
1387         struct rt6_info *rt;
1388
1389         *related = 1;
1390
1391         /* reassemble IP fragments */
1392         if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) {
1393                 if (ip_vs_gather_frags_v6(skb, ip_vs_defrag_user(hooknum)))
1394                         return NF_STOLEN;
1395         }
1396
1397         iph = ipv6_hdr(skb);
1398         offset = sizeof(struct ipv6hdr);
1399         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1400         if (ic == NULL)
1401                 return NF_DROP;
1402
1403         IP_VS_DBG(12, "Incoming ICMPv6 (%d,%d) %pI6->%pI6\n",
1404                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1405                   &iph->saddr, &iph->daddr);
1406
1407         /*
1408          * Work through seeing if this is for us.
1409          * These checks are supposed to be in an order that means easy
1410          * things are checked first to speed up processing.... however
1411          * this means that some packets will manage to get a long way
1412          * down this stack and then be rejected, but that's life.
1413          */
1414         if ((ic->icmp6_type != ICMPV6_DEST_UNREACH) &&
1415             (ic->icmp6_type != ICMPV6_PKT_TOOBIG) &&
1416             (ic->icmp6_type != ICMPV6_TIME_EXCEED)) {
1417                 *related = 0;
1418                 return NF_ACCEPT;
1419         }
1420
1421         /* Now find the contained IP header */
1422         offset += sizeof(_icmph);
1423         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1424         if (cih == NULL)
1425                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1426
1427         net = skb_net(skb);
1428         pd = ip_vs_proto_data_get(net, cih->nexthdr);
1429         if (!pd)
1430                 return NF_ACCEPT;
1431         pp = pd->pp;
1432
1433         /* Is the embedded protocol header present? */
1434         /* TODO: we don't support fragmentation at the moment anyways */
1435         if (unlikely(cih->nexthdr == IPPROTO_FRAGMENT && pp->dont_defrag))
1436                 return NF_ACCEPT;
1437
1438         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
1439                       "Checking incoming ICMPv6 for");
1440
1441         offset += sizeof(struct ipv6hdr);
1442
1443         ip_vs_fill_iphdr(AF_INET6, cih, &ciph);
1444         /* The embedded headers contain source and dest in reverse order */
1445         cp = pp->conn_in_get(AF_INET6, skb, &ciph, offset, 1);
1446         if (!cp)
1447                 return NF_ACCEPT;
1448
1449         verdict = NF_DROP;
1450
1451         /* do the statistics and put it back */
1452         ip_vs_in_stats(cp, skb);
1453         if (IPPROTO_TCP == cih->nexthdr || IPPROTO_UDP == cih->nexthdr ||
1454             IPPROTO_SCTP == cih->nexthdr)
1455                 offset += 2 * sizeof(__u16);
1456         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset);
1457         /* LOCALNODE from FORWARD hook is not supported */
1458         if (verdict == NF_ACCEPT && hooknum == NF_INET_FORWARD &&
1459             (rt = (struct rt6_info *) skb_dst(skb)) &&
1460             rt->rt6i_dev && rt->rt6i_dev->flags & IFF_LOOPBACK) {
1461                 IP_VS_DBG(1, "%s(): "
1462                           "local delivery to %pI6 but in FORWARD\n",
1463                           __func__, &rt->rt6i_dst);
1464                 verdict = NF_DROP;
1465         }
1466
1467         __ip_vs_conn_put(cp);
1468
1469         return verdict;
1470 }
1471 #endif
1472
1473
1474 /*
1475  *      Check if it's for virtual services, look it up,
1476  *      and send it on its way...
1477  */
1478 static unsigned int
1479 ip_vs_in(unsigned int hooknum, struct sk_buff *skb, int af)
1480 {
1481         struct net *net;
1482         struct ip_vs_iphdr iph;
1483         struct ip_vs_protocol *pp;
1484         struct ip_vs_proto_data *pd;
1485         struct ip_vs_conn *cp;
1486         int ret, restart, pkts;
1487         struct netns_ipvs *ipvs;
1488
1489         /* Already marked as IPVS request or reply? */
1490         if (skb->ipvs_property)
1491                 return NF_ACCEPT;
1492
1493         /*
1494          *      Big tappo:
1495          *      - remote client: only PACKET_HOST
1496          *      - route: used for struct net when skb->dev is unset
1497          */
1498         if (unlikely((skb->pkt_type != PACKET_HOST &&
1499                       hooknum != NF_INET_LOCAL_OUT) ||
1500                      !skb_dst(skb))) {
1501                 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1502                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1503                               " ignored in hook %u\n",
1504                               skb->pkt_type, iph.protocol,
1505                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1506                 return NF_ACCEPT;
1507         }
1508         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1509
1510         /* Bad... Do not break raw sockets */
1511         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1512                      af == AF_INET)) {
1513                 struct sock *sk = skb->sk;
1514                 struct inet_sock *inet = inet_sk(skb->sk);
1515
1516                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1517                         return NF_ACCEPT;
1518         }
1519
1520 #ifdef CONFIG_IP_VS_IPV6
1521         if (af == AF_INET6) {
1522                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1523                         int related;
1524                         int verdict = ip_vs_in_icmp_v6(skb, &related, hooknum);
1525
1526                         if (related)
1527                                 return verdict;
1528                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1529                 }
1530         } else
1531 #endif
1532                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1533                         int related;
1534                         int verdict = ip_vs_in_icmp(skb, &related, hooknum);
1535
1536                         if (related)
1537                                 return verdict;
1538                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1539                 }
1540
1541         net = skb_net(skb);
1542         /* Protocol supported? */
1543         pd = ip_vs_proto_data_get(net, iph.protocol);
1544         if (unlikely(!pd))
1545                 return NF_ACCEPT;
1546         pp = pd->pp;
1547         /*
1548          * Check if the packet belongs to an existing connection entry
1549          */
1550         cp = pp->conn_in_get(af, skb, &iph, iph.len, 0);
1551
1552         if (unlikely(!cp)) {
1553                 int v;
1554
1555                 if (!pp->conn_schedule(af, skb, pd, &v, &cp))
1556                         return v;
1557         }
1558
1559         if (unlikely(!cp)) {
1560                 /* sorry, all this trouble for a no-hit :) */
1561                 IP_VS_DBG_PKT(12, af, pp, skb, 0,
1562                               "ip_vs_in: packet continues traversal as normal");
1563                 return NF_ACCEPT;
1564         }
1565
1566         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Incoming packet");
1567         net = skb_net(skb);
1568         ipvs = net_ipvs(net);
1569         /* Check the server status */
1570         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1571                 /* the destination server is not available */
1572
1573                 if (ipvs->sysctl_expire_nodest_conn) {
1574                         /* try to expire the connection immediately */
1575                         ip_vs_conn_expire_now(cp);
1576                 }
1577                 /* don't restart its timer, and silently
1578                    drop the packet. */
1579                 __ip_vs_conn_put(cp);
1580                 return NF_DROP;
1581         }
1582
1583         ip_vs_in_stats(cp, skb);
1584         restart = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1585         if (cp->packet_xmit)
1586                 ret = cp->packet_xmit(skb, cp, pp);
1587                 /* do not touch skb anymore */
1588         else {
1589                 IP_VS_DBG_RL("warning: packet_xmit is null");
1590                 ret = NF_ACCEPT;
1591         }
1592
1593         /* Increase its packet counter and check if it is needed
1594          * to be synchronized
1595          *
1596          * Sync connection if it is about to close to
1597          * encorage the standby servers to update the connections timeout
1598          *
1599          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1600          */
1601
1602         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
1603                 pkts = ipvs->sysctl_sync_threshold[0];
1604         else
1605                 pkts = atomic_add_return(1, &cp->in_pkts);
1606
1607         if ((ipvs->sync_state & IP_VS_STATE_MASTER) &&
1608             cp->protocol == IPPROTO_SCTP) {
1609                 if ((cp->state == IP_VS_SCTP_S_ESTABLISHED &&
1610                         (pkts % ipvs->sysctl_sync_threshold[1]
1611                          == ipvs->sysctl_sync_threshold[0])) ||
1612                                 (cp->old_state != cp->state &&
1613                                  ((cp->state == IP_VS_SCTP_S_CLOSED) ||
1614                                   (cp->state == IP_VS_SCTP_S_SHUT_ACK_CLI) ||
1615                                   (cp->state == IP_VS_SCTP_S_SHUT_ACK_SER)))) {
1616                         ip_vs_sync_conn(net, cp);
1617                         goto out;
1618                 }
1619         }
1620
1621         /* Keep this block last: TCP and others with pp->num_states <= 1 */
1622         else if ((ipvs->sync_state & IP_VS_STATE_MASTER) &&
1623             (((cp->protocol != IPPROTO_TCP ||
1624                cp->state == IP_VS_TCP_S_ESTABLISHED) &&
1625               (pkts % ipvs->sysctl_sync_threshold[1]
1626                == ipvs->sysctl_sync_threshold[0])) ||
1627              ((cp->protocol == IPPROTO_TCP) && (cp->old_state != cp->state) &&
1628               ((cp->state == IP_VS_TCP_S_FIN_WAIT) ||
1629                (cp->state == IP_VS_TCP_S_CLOSE) ||
1630                (cp->state == IP_VS_TCP_S_CLOSE_WAIT) ||
1631                (cp->state == IP_VS_TCP_S_TIME_WAIT)))))
1632                 ip_vs_sync_conn(net, cp);
1633 out:
1634         cp->old_state = cp->state;
1635
1636         ip_vs_conn_put(cp);
1637         return ret;
1638 }
1639
1640 /*
1641  *      AF_INET handler in NF_INET_LOCAL_IN chain
1642  *      Schedule and forward packets from remote clients
1643  */
1644 static unsigned int
1645 ip_vs_remote_request4(unsigned int hooknum, struct sk_buff *skb,
1646                       const struct net_device *in,
1647                       const struct net_device *out,
1648                       int (*okfn)(struct sk_buff *))
1649 {
1650         return ip_vs_in(hooknum, skb, AF_INET);
1651 }
1652
1653 /*
1654  *      AF_INET handler in NF_INET_LOCAL_OUT chain
1655  *      Schedule and forward packets from local clients
1656  */
1657 static unsigned int
1658 ip_vs_local_request4(unsigned int hooknum, struct sk_buff *skb,
1659                      const struct net_device *in, const struct net_device *out,
1660                      int (*okfn)(struct sk_buff *))
1661 {
1662         unsigned int verdict;
1663
1664         /* Disable BH in LOCAL_OUT until all places are fixed */
1665         local_bh_disable();
1666         verdict = ip_vs_in(hooknum, skb, AF_INET);
1667         local_bh_enable();
1668         return verdict;
1669 }
1670
1671 #ifdef CONFIG_IP_VS_IPV6
1672
1673 /*
1674  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
1675  *      Schedule and forward packets from remote clients
1676  */
1677 static unsigned int
1678 ip_vs_remote_request6(unsigned int hooknum, struct sk_buff *skb,
1679                       const struct net_device *in,
1680                       const struct net_device *out,
1681                       int (*okfn)(struct sk_buff *))
1682 {
1683         return ip_vs_in(hooknum, skb, AF_INET6);
1684 }
1685
1686 /*
1687  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
1688  *      Schedule and forward packets from local clients
1689  */
1690 static unsigned int
1691 ip_vs_local_request6(unsigned int hooknum, struct sk_buff *skb,
1692                      const struct net_device *in, const struct net_device *out,
1693                      int (*okfn)(struct sk_buff *))
1694 {
1695         unsigned int verdict;
1696
1697         /* Disable BH in LOCAL_OUT until all places are fixed */
1698         local_bh_disable();
1699         verdict = ip_vs_in(hooknum, skb, AF_INET6);
1700         local_bh_enable();
1701         return verdict;
1702 }
1703
1704 #endif
1705
1706
1707 /*
1708  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1709  *      related packets destined for 0.0.0.0/0.
1710  *      When fwmark-based virtual service is used, such as transparent
1711  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
1712  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1713  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1714  *      and send them to ip_vs_in_icmp.
1715  */
1716 static unsigned int
1717 ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff *skb,
1718                    const struct net_device *in, const struct net_device *out,
1719                    int (*okfn)(struct sk_buff *))
1720 {
1721         int r;
1722
1723         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1724                 return NF_ACCEPT;
1725
1726         return ip_vs_in_icmp(skb, &r, hooknum);
1727 }
1728
1729 #ifdef CONFIG_IP_VS_IPV6
1730 static unsigned int
1731 ip_vs_forward_icmp_v6(unsigned int hooknum, struct sk_buff *skb,
1732                       const struct net_device *in, const struct net_device *out,
1733                       int (*okfn)(struct sk_buff *))
1734 {
1735         int r;
1736
1737         if (ipv6_hdr(skb)->nexthdr != IPPROTO_ICMPV6)
1738                 return NF_ACCEPT;
1739
1740         return ip_vs_in_icmp_v6(skb, &r, hooknum);
1741 }
1742 #endif
1743
1744
1745 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
1746         /* After packet filtering, change source only for VS/NAT */
1747         {
1748                 .hook           = ip_vs_reply4,
1749                 .owner          = THIS_MODULE,
1750                 .pf             = PF_INET,
1751                 .hooknum        = NF_INET_LOCAL_IN,
1752                 .priority       = 99,
1753         },
1754         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1755          * or VS/NAT(change destination), so that filtering rules can be
1756          * applied to IPVS. */
1757         {
1758                 .hook           = ip_vs_remote_request4,
1759                 .owner          = THIS_MODULE,
1760                 .pf             = PF_INET,
1761                 .hooknum        = NF_INET_LOCAL_IN,
1762                 .priority       = 101,
1763         },
1764         /* Before ip_vs_in, change source only for VS/NAT */
1765         {
1766                 .hook           = ip_vs_local_reply4,
1767                 .owner          = THIS_MODULE,
1768                 .pf             = PF_INET,
1769                 .hooknum        = NF_INET_LOCAL_OUT,
1770                 .priority       = -99,
1771         },
1772         /* After mangle, schedule and forward local requests */
1773         {
1774                 .hook           = ip_vs_local_request4,
1775                 .owner          = THIS_MODULE,
1776                 .pf             = PF_INET,
1777                 .hooknum        = NF_INET_LOCAL_OUT,
1778                 .priority       = -98,
1779         },
1780         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1781          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1782         {
1783                 .hook           = ip_vs_forward_icmp,
1784                 .owner          = THIS_MODULE,
1785                 .pf             = PF_INET,
1786                 .hooknum        = NF_INET_FORWARD,
1787                 .priority       = 99,
1788         },
1789         /* After packet filtering, change source only for VS/NAT */
1790         {
1791                 .hook           = ip_vs_reply4,
1792                 .owner          = THIS_MODULE,
1793                 .pf             = PF_INET,
1794                 .hooknum        = NF_INET_FORWARD,
1795                 .priority       = 100,
1796         },
1797 #ifdef CONFIG_IP_VS_IPV6
1798         /* After packet filtering, change source only for VS/NAT */
1799         {
1800                 .hook           = ip_vs_reply6,
1801                 .owner          = THIS_MODULE,
1802                 .pf             = PF_INET6,
1803                 .hooknum        = NF_INET_LOCAL_IN,
1804                 .priority       = 99,
1805         },
1806         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1807          * or VS/NAT(change destination), so that filtering rules can be
1808          * applied to IPVS. */
1809         {
1810                 .hook           = ip_vs_remote_request6,
1811                 .owner          = THIS_MODULE,
1812                 .pf             = PF_INET6,
1813                 .hooknum        = NF_INET_LOCAL_IN,
1814                 .priority       = 101,
1815         },
1816         /* Before ip_vs_in, change source only for VS/NAT */
1817         {
1818                 .hook           = ip_vs_local_reply6,
1819                 .owner          = THIS_MODULE,
1820                 .pf             = PF_INET,
1821                 .hooknum        = NF_INET_LOCAL_OUT,
1822                 .priority       = -99,
1823         },
1824         /* After mangle, schedule and forward local requests */
1825         {
1826                 .hook           = ip_vs_local_request6,
1827                 .owner          = THIS_MODULE,
1828                 .pf             = PF_INET6,
1829                 .hooknum        = NF_INET_LOCAL_OUT,
1830                 .priority       = -98,
1831         },
1832         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1833          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1834         {
1835                 .hook           = ip_vs_forward_icmp_v6,
1836                 .owner          = THIS_MODULE,
1837                 .pf             = PF_INET6,
1838                 .hooknum        = NF_INET_FORWARD,
1839                 .priority       = 99,
1840         },
1841         /* After packet filtering, change source only for VS/NAT */
1842         {
1843                 .hook           = ip_vs_reply6,
1844                 .owner          = THIS_MODULE,
1845                 .pf             = PF_INET6,
1846                 .hooknum        = NF_INET_FORWARD,
1847                 .priority       = 100,
1848         },
1849 #endif
1850 };
1851 /*
1852  *      Initialize IP Virtual Server netns mem.
1853  */
1854 static int __net_init __ip_vs_init(struct net *net)
1855 {
1856         struct netns_ipvs *ipvs;
1857
1858         ipvs = net_generic(net, ip_vs_net_id);
1859         if (ipvs == NULL) {
1860                 pr_err("%s(): no memory.\n", __func__);
1861                 return -ENOMEM;
1862         }
1863         ipvs->net = net;
1864         /* Counters used for creating unique names */
1865         ipvs->gen = atomic_read(&ipvs_netns_cnt);
1866         atomic_inc(&ipvs_netns_cnt);
1867         net->ipvs = ipvs;
1868         printk(KERN_INFO "IPVS: Creating netns size=%zu id=%d\n",
1869                          sizeof(struct netns_ipvs), ipvs->gen);
1870         return 0;
1871 }
1872
1873 static void __net_exit __ip_vs_cleanup(struct net *net)
1874 {
1875         IP_VS_DBG(10, "ipvs netns %d released\n", net_ipvs(net)->gen);
1876 }
1877
1878 static struct pernet_operations ipvs_core_ops = {
1879         .init = __ip_vs_init,
1880         .exit = __ip_vs_cleanup,
1881         .id   = &ip_vs_net_id,
1882         .size = sizeof(struct netns_ipvs),
1883 };
1884
1885 /*
1886  *      Initialize IP Virtual Server
1887  */
1888 static int __init ip_vs_init(void)
1889 {
1890         int ret;
1891
1892         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
1893         if (ret < 0)
1894                 return ret;
1895
1896         ip_vs_estimator_init();
1897         ret = ip_vs_control_init();
1898         if (ret < 0) {
1899                 pr_err("can't setup control.\n");
1900                 goto cleanup_estimator;
1901         }
1902
1903         ip_vs_protocol_init();
1904
1905         ret = ip_vs_app_init();
1906         if (ret < 0) {
1907                 pr_err("can't setup application helper.\n");
1908                 goto cleanup_protocol;
1909         }
1910
1911         ret = ip_vs_conn_init();
1912         if (ret < 0) {
1913                 pr_err("can't setup connection table.\n");
1914                 goto cleanup_app;
1915         }
1916
1917         ret = ip_vs_sync_init();
1918         if (ret < 0) {
1919                 pr_err("can't setup sync data.\n");
1920                 goto cleanup_conn;
1921         }
1922
1923         ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
1924         if (ret < 0) {
1925                 pr_err("can't register hooks.\n");
1926                 goto cleanup_sync;
1927         }
1928
1929         pr_info("ipvs loaded.\n");
1930         return ret;
1931
1932 cleanup_sync:
1933         ip_vs_sync_cleanup();
1934   cleanup_conn:
1935         ip_vs_conn_cleanup();
1936   cleanup_app:
1937         ip_vs_app_cleanup();
1938   cleanup_protocol:
1939         ip_vs_protocol_cleanup();
1940         ip_vs_control_cleanup();
1941   cleanup_estimator:
1942         ip_vs_estimator_cleanup();
1943         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
1944         return ret;
1945 }
1946
1947 static void __exit ip_vs_cleanup(void)
1948 {
1949         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
1950         ip_vs_sync_cleanup();
1951         ip_vs_conn_cleanup();
1952         ip_vs_app_cleanup();
1953         ip_vs_protocol_cleanup();
1954         ip_vs_control_cleanup();
1955         ip_vs_estimator_cleanup();
1956         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
1957         pr_info("ipvs unloaded.\n");
1958 }
1959
1960 module_init(ip_vs_init);
1961 module_exit(ip_vs_cleanup);
1962 MODULE_LICENSE("GPL");