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. Used for NAT and local client.
 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  * Used for NAT and local client.
 983  */
 984 static unsigned int
 985 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
 986                 struct ip_vs_conn *cp, int ihl)
 987 {
 988         struct ip_vs_protocol *pp = pd->pp;
 989         struct netns_ipvs *ipvs;
 990
 991         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Outgoing packet");
 992
 993         if (!skb_make_writable(skb, ihl))
 994                 goto drop;
 995
 996         /* mangle the packet */
 997         if (pp->snat_handler && !pp->snat_handler(skb, pp, cp))
 998                 goto drop;
 999
1000 #ifdef CONFIG_IP_VS_IPV6
1001         if (af == AF_INET6)
1002                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1003         else
1004 #endif
1005         {
1006                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1007                 ip_send_check(ip_hdr(skb));
1008         }
1009
1010         /*
1011          * nf_iterate does not expect change in the skb->dst->dev.
1012          * It looks like it is not fatal to enable this code for hooks
1013          * where our handlers are at the end of the chain list and
1014          * when all next handlers use skb->dst->dev and not outdev.
1015          * It will definitely route properly the inout NAT traffic
1016          * when multiple paths are used.
1017          */
1018
1019         /* For policy routing, packets originating from this
1020          * machine itself may be routed differently to packets
1021          * passing through.  We want this packet to be routed as
1022          * if it came from this machine itself.  So re-compute
1023          * the routing information.
1024          */
1025         ipvs = net_ipvs(skb_net(skb));
1026
1027 #ifdef CONFIG_IP_VS_IPV6
1028         if (af == AF_INET6) {
1029                 if (ipvs->sysctl_snat_reroute && ip6_route_me_harder(skb) != 0)
1030                         goto drop;
1031         } else
1032 #endif
1033                 if ((ipvs->sysctl_snat_reroute ||
1034                      skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
1035                     ip_route_me_harder(skb, RTN_LOCAL) != 0)
1036                         goto drop;
1037
1038         IP_VS_DBG_PKT(10, af, pp, skb, 0, "After SNAT");
1039
1040         ip_vs_out_stats(cp, skb);
1041         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1042         skb->ipvs_property = 1;
1043         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1044                 ip_vs_notrack(skb);
1045         else
1046                 ip_vs_update_conntrack(skb, cp, 0);
1047         ip_vs_conn_put(cp);
1048
1049         LeaveFunction(11);
1050         return NF_ACCEPT;
1051
1052 drop:
1053         ip_vs_conn_put(cp);
1054         kfree_skb(skb);
1055         LeaveFunction(11);
1056         return NF_STOLEN;
1057 }
1058
1059 /*
1060  *      Check if outgoing packet belongs to the established ip_vs_conn.
1061  */
1062 static unsigned int
1063 ip_vs_out(unsigned int hooknum, struct sk_buff *skb, int af)
1064 {
1065         struct net *net = NULL;
1066         struct ip_vs_iphdr iph;
1067         struct ip_vs_protocol *pp;
1068         struct ip_vs_proto_data *pd;
1069         struct ip_vs_conn *cp;
1070         struct netns_ipvs *ipvs;
1071
1072         EnterFunction(11);
1073
1074         /* Already marked as IPVS request or reply? */
1075         if (skb->ipvs_property)
1076                 return NF_ACCEPT;
1077
1078         /* Bad... Do not break raw sockets */
1079         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1080                      af == AF_INET)) {
1081                 struct sock *sk = skb->sk;
1082                 struct inet_sock *inet = inet_sk(skb->sk);
1083
1084                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1085                         return NF_ACCEPT;
1086         }
1087
1088         if (unlikely(!skb_dst(skb)))
1089                 return NF_ACCEPT;
1090
1091         net = skb_net(skb);
1092         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1093 #ifdef CONFIG_IP_VS_IPV6
1094         if (af == AF_INET6) {
1095                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1096                         int related;
1097                         int verdict = ip_vs_out_icmp_v6(skb, &related,
1098                                                         hooknum);
1099
1100                         if (related)
1101                                 return verdict;
1102                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1103                 }
1104         } else
1105 #endif
1106                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1107                         int related;
1108                         int verdict = ip_vs_out_icmp(skb, &related, hooknum);
1109
1110                         if (related)
1111                                 return verdict;
1112                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1113                 }
1114
1115         pd = ip_vs_proto_data_get(net, iph.protocol);
1116         if (unlikely(!pd))
1117                 return NF_ACCEPT;
1118         pp = pd->pp;
1119
1120         /* reassemble IP fragments */
1121 #ifdef CONFIG_IP_VS_IPV6
1122         if (af == AF_INET6) {
1123                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) {
1124                         if (ip_vs_gather_frags_v6(skb,
1125                                                   ip_vs_defrag_user(hooknum)))
1126                                 return NF_STOLEN;
1127                 }
1128
1129                 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1130         } else
1131 #endif
1132                 if (unlikely(ip_hdr(skb)->frag_off & htons(IP_MF|IP_OFFSET) &&
1133                              !pp->dont_defrag)) {
1134                         if (ip_vs_gather_frags(skb,
1135                                                ip_vs_defrag_user(hooknum)))
1136                                 return NF_STOLEN;
1137
1138                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1139                 }
1140
1141         /*
1142          * Check if the packet belongs to an existing entry
1143          */
1144         cp = pp->conn_out_get(af, skb, &iph, iph.len, 0);
1145         ipvs = net_ipvs(net);
1146
1147         if (likely(cp))
1148                 return handle_response(af, skb, pd, cp, iph.len);
1149         if (ipvs->sysctl_nat_icmp_send &&
1150             (pp->protocol == IPPROTO_TCP ||
1151              pp->protocol == IPPROTO_UDP ||
1152              pp->protocol == IPPROTO_SCTP)) {
1153                 __be16 _ports[2], *pptr;
1154
1155                 pptr = skb_header_pointer(skb, iph.len,
1156                                           sizeof(_ports), _ports);
1157                 if (pptr == NULL)
1158                         return NF_ACCEPT;       /* Not for me */
1159                 if (ip_vs_lookup_real_service(net, af, iph.protocol,
1160                                               &iph.saddr,
1161                                               pptr[0])) {
1162                         /*
1163                          * Notify the real server: there is no
1164                          * existing entry if it is not RST
1165                          * packet or not TCP packet.
1166                          */
1167                         if ((iph.protocol != IPPROTO_TCP &&
1168                              iph.protocol != IPPROTO_SCTP)
1169                              || ((iph.protocol == IPPROTO_TCP
1170                                   && !is_tcp_reset(skb, iph.len))
1171                                  || (iph.protocol == IPPROTO_SCTP
1172                                         && !is_sctp_abort(skb,
1173                                                 iph.len)))) {
1174 #ifdef CONFIG_IP_VS_IPV6
1175                                 if (af == AF_INET6) {
1176                                         struct net *net =
1177                                                 dev_net(skb_dst(skb)->dev);
1178
1179                                         if (!skb->dev)
1180                                                 skb->dev = net->loopback_dev;
1181                                         icmpv6_send(skb,
1182                                                     ICMPV6_DEST_UNREACH,
1183                                                     ICMPV6_PORT_UNREACH,
1184                                                     0);
1185                                 } else
1186 #endif
1187                                         icmp_send(skb,
1188                                                   ICMP_DEST_UNREACH,
1189                                                   ICMP_PORT_UNREACH, 0);
1190                                 return NF_DROP;
1191                         }
1192                 }
1193         }
1194         IP_VS_DBG_PKT(12, af, pp, skb, 0,
1195                       "ip_vs_out: packet continues traversal as normal");
1196         return NF_ACCEPT;
1197 }
1198
1199 /*
1200  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1201  *      used only for VS/NAT.
1202  *      Check if packet is reply for established ip_vs_conn.
1203  */
1204 static unsigned int
1205 ip_vs_reply4(unsigned int hooknum, struct sk_buff *skb,
1206              const struct net_device *in, const struct net_device *out,
1207              int (*okfn)(struct sk_buff *))
1208 {
1209         return ip_vs_out(hooknum, skb, AF_INET);
1210 }
1211
1212 /*
1213  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1214  *      Check if packet is reply for established ip_vs_conn.
1215  */
1216 static unsigned int
1217 ip_vs_local_reply4(unsigned int hooknum, struct sk_buff *skb,
1218                    const struct net_device *in, const struct net_device *out,
1219                    int (*okfn)(struct sk_buff *))
1220 {
1221         unsigned int verdict;
1222
1223         /* Disable BH in LOCAL_OUT until all places are fixed */
1224         local_bh_disable();
1225         verdict = ip_vs_out(hooknum, skb, AF_INET);
1226         local_bh_enable();
1227         return verdict;
1228 }
1229
1230 #ifdef CONFIG_IP_VS_IPV6
1231
1232 /*
1233  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1234  *      used only for VS/NAT.
1235  *      Check if packet is reply for established ip_vs_conn.
1236  */
1237 static unsigned int
1238 ip_vs_reply6(unsigned int hooknum, struct sk_buff *skb,
1239              const struct net_device *in, const struct net_device *out,
1240              int (*okfn)(struct sk_buff *))
1241 {
1242         return ip_vs_out(hooknum, skb, AF_INET6);
1243 }
1244
1245 /*
1246  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1247  *      Check if packet is reply for established ip_vs_conn.
1248  */
1249 static unsigned int
1250 ip_vs_local_reply6(unsigned int hooknum, struct sk_buff *skb,
1251                    const struct net_device *in, const struct net_device *out,
1252                    int (*okfn)(struct sk_buff *))
1253 {
1254         unsigned int verdict;
1255
1256         /* Disable BH in LOCAL_OUT until all places are fixed */
1257         local_bh_disable();
1258         verdict = ip_vs_out(hooknum, skb, AF_INET6);
1259         local_bh_enable();
1260         return verdict;
1261 }
1262
1263 #endif
1264
1265 /*
1266  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1267  *      Find any that might be relevant, check against existing connections,
1268  *      forward to the right destination host if relevant.
1269  *      Currently handles error types - unreachable, quench, ttl exceeded.
1270  */
1271 static int
1272 ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum)
1273 {
1274         struct net *net = NULL;
1275         struct iphdr *iph;
1276         struct icmphdr  _icmph, *ic;
1277         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1278         struct ip_vs_iphdr ciph;
1279         struct ip_vs_conn *cp;
1280         struct ip_vs_protocol *pp;
1281         struct ip_vs_proto_data *pd;
1282         unsigned int offset, ihl, verdict;
1283         union nf_inet_addr snet;
1284
1285         *related = 1;
1286
1287         /* reassemble IP fragments */
1288         if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
1289                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
1290                         return NF_STOLEN;
1291         }
1292
1293         iph = ip_hdr(skb);
1294         offset = ihl = iph->ihl * 4;
1295         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1296         if (ic == NULL)
1297                 return NF_DROP;
1298
1299         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1300                   ic->type, ntohs(icmp_id(ic)),
1301                   &iph->saddr, &iph->daddr);
1302
1303         /*
1304          * Work through seeing if this is for us.
1305          * These checks are supposed to be in an order that means easy
1306          * things are checked first to speed up processing.... however
1307          * this means that some packets will manage to get a long way
1308          * down this stack and then be rejected, but that's life.
1309          */
1310         if ((ic->type != ICMP_DEST_UNREACH) &&
1311             (ic->type != ICMP_SOURCE_QUENCH) &&
1312             (ic->type != ICMP_TIME_EXCEEDED)) {
1313                 *related = 0;
1314                 return NF_ACCEPT;
1315         }
1316
1317         /* Now find the contained IP header */
1318         offset += sizeof(_icmph);
1319         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1320         if (cih == NULL)
1321                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1322
1323         net = skb_net(skb);
1324         pd = ip_vs_proto_data_get(net, cih->protocol);
1325         if (!pd)
1326                 return NF_ACCEPT;
1327         pp = pd->pp;
1328
1329         /* Is the embedded protocol header present? */
1330         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1331                      pp->dont_defrag))
1332                 return NF_ACCEPT;
1333
1334         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1335                       "Checking incoming ICMP for");
1336
1337         offset += cih->ihl * 4;
1338
1339         ip_vs_fill_iphdr(AF_INET, cih, &ciph);
1340         /* The embedded headers contain source and dest in reverse order */
1341         cp = pp->conn_in_get(AF_INET, skb, &ciph, offset, 1);
1342         if (!cp) {
1343                 /* The packet could also belong to a local client */
1344                 cp = pp->conn_out_get(AF_INET, skb, &ciph, offset, 1);
1345                 if (cp) {
1346                         snet.ip = iph->saddr;
1347                         return handle_response_icmp(AF_INET, skb, &snet,
1348                                                     cih->protocol, cp, pp,
1349                                                     offset, ihl);
1350                 }
1351                 return NF_ACCEPT;
1352         }
1353
1354         verdict = NF_DROP;
1355
1356         /* Ensure the checksum is correct */
1357         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1358                 /* Failed checksum! */
1359                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1360                           &iph->saddr);
1361                 goto out;
1362         }
1363
1364         /* do the statistics and put it back */
1365         ip_vs_in_stats(cp, skb);
1366         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol)
1367                 offset += 2 * sizeof(__u16);
1368         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset);
1369         /* LOCALNODE from FORWARD hook is not supported */
1370         if (verdict == NF_ACCEPT && hooknum == NF_INET_FORWARD &&
1371             skb_rtable(skb)->rt_flags & RTCF_LOCAL) {
1372                 IP_VS_DBG(1, "%s(): "
1373                           "local delivery to %pI4 but in FORWARD\n",
1374                           __func__, &skb_rtable(skb)->rt_dst);
1375                 verdict = NF_DROP;
1376         }
1377
1378   out:
1379         __ip_vs_conn_put(cp);
1380
1381         return verdict;
1382 }
1383
1384 #ifdef CONFIG_IP_VS_IPV6
1385 static int
1386 ip_vs_in_icmp_v6(struct sk_buff *skb, int *related, unsigned int hooknum)
1387 {
1388         struct net *net = NULL;
1389         struct ipv6hdr *iph;
1390         struct icmp6hdr _icmph, *ic;
1391         struct ipv6hdr  _ciph, *cih;    /* The ip header contained
1392                                            within the ICMP */
1393         struct ip_vs_iphdr ciph;
1394         struct ip_vs_conn *cp;
1395         struct ip_vs_protocol *pp;
1396         struct ip_vs_proto_data *pd;
1397         unsigned int offset, verdict;
1398         union nf_inet_addr snet;
1399         struct rt6_info *rt;
1400
1401         *related = 1;
1402
1403         /* reassemble IP fragments */
1404         if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) {
1405                 if (ip_vs_gather_frags_v6(skb, ip_vs_defrag_user(hooknum)))
1406                         return NF_STOLEN;
1407         }
1408
1409         iph = ipv6_hdr(skb);
1410         offset = sizeof(struct ipv6hdr);
1411         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1412         if (ic == NULL)
1413                 return NF_DROP;
1414
1415         IP_VS_DBG(12, "Incoming ICMPv6 (%d,%d) %pI6->%pI6\n",
1416                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1417                   &iph->saddr, &iph->daddr);
1418
1419         /*
1420          * Work through seeing if this is for us.
1421          * These checks are supposed to be in an order that means easy
1422          * things are checked first to speed up processing.... however
1423          * this means that some packets will manage to get a long way
1424          * down this stack and then be rejected, but that's life.
1425          */
1426         if ((ic->icmp6_type != ICMPV6_DEST_UNREACH) &&
1427             (ic->icmp6_type != ICMPV6_PKT_TOOBIG) &&
1428             (ic->icmp6_type != ICMPV6_TIME_EXCEED)) {
1429                 *related = 0;
1430                 return NF_ACCEPT;
1431         }
1432
1433         /* Now find the contained IP header */
1434         offset += sizeof(_icmph);
1435         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1436         if (cih == NULL)
1437                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1438
1439         net = skb_net(skb);
1440         pd = ip_vs_proto_data_get(net, cih->nexthdr);
1441         if (!pd)
1442                 return NF_ACCEPT;
1443         pp = pd->pp;
1444
1445         /* Is the embedded protocol header present? */
1446         /* TODO: we don't support fragmentation at the moment anyways */
1447         if (unlikely(cih->nexthdr == IPPROTO_FRAGMENT && pp->dont_defrag))
1448                 return NF_ACCEPT;
1449
1450         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
1451                       "Checking incoming ICMPv6 for");
1452
1453         offset += sizeof(struct ipv6hdr);
1454
1455         ip_vs_fill_iphdr(AF_INET6, cih, &ciph);
1456         /* The embedded headers contain source and dest in reverse order */
1457         cp = pp->conn_in_get(AF_INET6, skb, &ciph, offset, 1);
1458         if (!cp) {
1459                 /* The packet could also belong to a local client */
1460                 cp = pp->conn_out_get(AF_INET6, skb, &ciph, offset, 1);
1461                 if (cp) {
1462                         ipv6_addr_copy(&snet.in6, &iph->saddr);
1463                         return handle_response_icmp(AF_INET6, skb, &snet,
1464                                                     cih->nexthdr,
1465                                                     cp, pp, offset,
1466                                                     sizeof(struct ipv6hdr));
1467                 }
1468                 return NF_ACCEPT;
1469         }
1470
1471         verdict = NF_DROP;
1472
1473         /* do the statistics and put it back */
1474         ip_vs_in_stats(cp, skb);
1475         if (IPPROTO_TCP == cih->nexthdr || IPPROTO_UDP == cih->nexthdr ||
1476             IPPROTO_SCTP == cih->nexthdr)
1477                 offset += 2 * sizeof(__u16);
1478         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset);
1479         /* LOCALNODE from FORWARD hook is not supported */
1480         if (verdict == NF_ACCEPT && hooknum == NF_INET_FORWARD &&
1481             (rt = (struct rt6_info *) skb_dst(skb)) &&
1482             rt->rt6i_dev && rt->rt6i_dev->flags & IFF_LOOPBACK) {
1483                 IP_VS_DBG(1, "%s(): "
1484                           "local delivery to %pI6 but in FORWARD\n",
1485                           __func__, &rt->rt6i_dst);
1486                 verdict = NF_DROP;
1487         }
1488
1489         __ip_vs_conn_put(cp);
1490
1491         return verdict;
1492 }
1493 #endif
1494
1495
1496 /*
1497  *      Check if it's for virtual services, look it up,
1498  *      and send it on its way...
1499  */
1500 static unsigned int
1501 ip_vs_in(unsigned int hooknum, struct sk_buff *skb, int af)
1502 {
1503         struct net *net;
1504         struct ip_vs_iphdr iph;
1505         struct ip_vs_protocol *pp;
1506         struct ip_vs_proto_data *pd;
1507         struct ip_vs_conn *cp;
1508         int ret, restart, pkts;
1509         struct netns_ipvs *ipvs;
1510
1511         /* Already marked as IPVS request or reply? */
1512         if (skb->ipvs_property)
1513                 return NF_ACCEPT;
1514
1515         /*
1516          *      Big tappo:
1517          *      - remote client: only PACKET_HOST
1518          *      - route: used for struct net when skb->dev is unset
1519          */
1520         if (unlikely((skb->pkt_type != PACKET_HOST &&
1521                       hooknum != NF_INET_LOCAL_OUT) ||
1522                      !skb_dst(skb))) {
1523                 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1524                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1525                               " ignored in hook %u\n",
1526                               skb->pkt_type, iph.protocol,
1527                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1528                 return NF_ACCEPT;
1529         }
1530         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1531
1532         /* Bad... Do not break raw sockets */
1533         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1534                      af == AF_INET)) {
1535                 struct sock *sk = skb->sk;
1536                 struct inet_sock *inet = inet_sk(skb->sk);
1537
1538                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1539                         return NF_ACCEPT;
1540         }
1541
1542 #ifdef CONFIG_IP_VS_IPV6
1543         if (af == AF_INET6) {
1544                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1545                         int related;
1546                         int verdict = ip_vs_in_icmp_v6(skb, &related, hooknum);
1547
1548                         if (related)
1549                                 return verdict;
1550                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1551                 }
1552         } else
1553 #endif
1554                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1555                         int related;
1556                         int verdict = ip_vs_in_icmp(skb, &related, hooknum);
1557
1558                         if (related)
1559                                 return verdict;
1560                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1561                 }
1562
1563         net = skb_net(skb);
1564         /* Protocol supported? */
1565         pd = ip_vs_proto_data_get(net, iph.protocol);
1566         if (unlikely(!pd))
1567                 return NF_ACCEPT;
1568         pp = pd->pp;
1569         /*
1570          * Check if the packet belongs to an existing connection entry
1571          */
1572         cp = pp->conn_in_get(af, skb, &iph, iph.len, 0);
1573
1574         if (unlikely(!cp)) {
1575                 int v;
1576
1577                 if (!pp->conn_schedule(af, skb, pd, &v, &cp))
1578                         return v;
1579         }
1580
1581         if (unlikely(!cp)) {
1582                 /* sorry, all this trouble for a no-hit :) */
1583                 IP_VS_DBG_PKT(12, af, pp, skb, 0,
1584                               "ip_vs_in: packet continues traversal as normal");
1585                 return NF_ACCEPT;
1586         }
1587
1588         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Incoming packet");
1589         net = skb_net(skb);
1590         ipvs = net_ipvs(net);
1591         /* Check the server status */
1592         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1593                 /* the destination server is not available */
1594
1595                 if (ipvs->sysctl_expire_nodest_conn) {
1596                         /* try to expire the connection immediately */
1597                         ip_vs_conn_expire_now(cp);
1598                 }
1599                 /* don't restart its timer, and silently
1600                    drop the packet. */
1601                 __ip_vs_conn_put(cp);
1602                 return NF_DROP;
1603         }
1604
1605         ip_vs_in_stats(cp, skb);
1606         restart = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1607         if (cp->packet_xmit)
1608                 ret = cp->packet_xmit(skb, cp, pp);
1609                 /* do not touch skb anymore */
1610         else {
1611                 IP_VS_DBG_RL("warning: packet_xmit is null");
1612                 ret = NF_ACCEPT;
1613         }
1614
1615         /* Increase its packet counter and check if it is needed
1616          * to be synchronized
1617          *
1618          * Sync connection if it is about to close to
1619          * encorage the standby servers to update the connections timeout
1620          *
1621          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1622          */
1623
1624         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
1625                 pkts = ipvs->sysctl_sync_threshold[0];
1626         else
1627                 pkts = atomic_add_return(1, &cp->in_pkts);
1628
1629         if ((ipvs->sync_state & IP_VS_STATE_MASTER) &&
1630             cp->protocol == IPPROTO_SCTP) {
1631                 if ((cp->state == IP_VS_SCTP_S_ESTABLISHED &&
1632                         (pkts % ipvs->sysctl_sync_threshold[1]
1633                          == ipvs->sysctl_sync_threshold[0])) ||
1634                                 (cp->old_state != cp->state &&
1635                                  ((cp->state == IP_VS_SCTP_S_CLOSED) ||
1636                                   (cp->state == IP_VS_SCTP_S_SHUT_ACK_CLI) ||
1637                                   (cp->state == IP_VS_SCTP_S_SHUT_ACK_SER)))) {
1638                         ip_vs_sync_conn(net, cp);
1639                         goto out;
1640                 }
1641         }
1642
1643         /* Keep this block last: TCP and others with pp->num_states <= 1 */
1644         else if ((ipvs->sync_state & IP_VS_STATE_MASTER) &&
1645             (((cp->protocol != IPPROTO_TCP ||
1646                cp->state == IP_VS_TCP_S_ESTABLISHED) &&
1647               (pkts % ipvs->sysctl_sync_threshold[1]
1648                == ipvs->sysctl_sync_threshold[0])) ||
1649              ((cp->protocol == IPPROTO_TCP) && (cp->old_state != cp->state) &&
1650               ((cp->state == IP_VS_TCP_S_FIN_WAIT) ||
1651                (cp->state == IP_VS_TCP_S_CLOSE) ||
1652                (cp->state == IP_VS_TCP_S_CLOSE_WAIT) ||
1653                (cp->state == IP_VS_TCP_S_TIME_WAIT)))))
1654                 ip_vs_sync_conn(net, cp);
1655 out:
1656         cp->old_state = cp->state;
1657
1658         ip_vs_conn_put(cp);
1659         return ret;
1660 }
1661
1662 /*
1663  *      AF_INET handler in NF_INET_LOCAL_IN chain
1664  *      Schedule and forward packets from remote clients
1665  */
1666 static unsigned int
1667 ip_vs_remote_request4(unsigned int hooknum, struct sk_buff *skb,
1668                       const struct net_device *in,
1669                       const struct net_device *out,
1670                       int (*okfn)(struct sk_buff *))
1671 {
1672         return ip_vs_in(hooknum, skb, AF_INET);
1673 }
1674
1675 /*
1676  *      AF_INET handler in NF_INET_LOCAL_OUT chain
1677  *      Schedule and forward packets from local clients
1678  */
1679 static unsigned int
1680 ip_vs_local_request4(unsigned int hooknum, struct sk_buff *skb,
1681                      const struct net_device *in, const struct net_device *out,
1682                      int (*okfn)(struct sk_buff *))
1683 {
1684         unsigned int verdict;
1685
1686         /* Disable BH in LOCAL_OUT until all places are fixed */
1687         local_bh_disable();
1688         verdict = ip_vs_in(hooknum, skb, AF_INET);
1689         local_bh_enable();
1690         return verdict;
1691 }
1692
1693 #ifdef CONFIG_IP_VS_IPV6
1694
1695 /*
1696  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
1697  *      Schedule and forward packets from remote clients
1698  */
1699 static unsigned int
1700 ip_vs_remote_request6(unsigned int hooknum, struct sk_buff *skb,
1701                       const struct net_device *in,
1702                       const struct net_device *out,
1703                       int (*okfn)(struct sk_buff *))
1704 {
1705         return ip_vs_in(hooknum, skb, AF_INET6);
1706 }
1707
1708 /*
1709  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
1710  *      Schedule and forward packets from local clients
1711  */
1712 static unsigned int
1713 ip_vs_local_request6(unsigned int hooknum, struct sk_buff *skb,
1714                      const struct net_device *in, const struct net_device *out,
1715                      int (*okfn)(struct sk_buff *))
1716 {
1717         unsigned int verdict;
1718
1719         /* Disable BH in LOCAL_OUT until all places are fixed */
1720         local_bh_disable();
1721         verdict = ip_vs_in(hooknum, skb, AF_INET6);
1722         local_bh_enable();
1723         return verdict;
1724 }
1725
1726 #endif
1727
1728
1729 /*
1730  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1731  *      related packets destined for 0.0.0.0/0.
1732  *      When fwmark-based virtual service is used, such as transparent
1733  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
1734  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1735  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1736  *      and send them to ip_vs_in_icmp.
1737  */
1738 static unsigned int
1739 ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff *skb,
1740                    const struct net_device *in, const struct net_device *out,
1741                    int (*okfn)(struct sk_buff *))
1742 {
1743         int r;
1744
1745         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1746                 return NF_ACCEPT;
1747
1748         return ip_vs_in_icmp(skb, &r, hooknum);
1749 }
1750
1751 #ifdef CONFIG_IP_VS_IPV6
1752 static unsigned int
1753 ip_vs_forward_icmp_v6(unsigned int hooknum, struct sk_buff *skb,
1754                       const struct net_device *in, const struct net_device *out,
1755                       int (*okfn)(struct sk_buff *))
1756 {
1757         int r;
1758
1759         if (ipv6_hdr(skb)->nexthdr != IPPROTO_ICMPV6)
1760                 return NF_ACCEPT;
1761
1762         return ip_vs_in_icmp_v6(skb, &r, hooknum);
1763 }
1764 #endif
1765
1766
1767 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
1768         /* After packet filtering, change source only for VS/NAT */
1769         {
1770                 .hook           = ip_vs_reply4,
1771                 .owner          = THIS_MODULE,
1772                 .pf             = PF_INET,
1773                 .hooknum        = NF_INET_LOCAL_IN,
1774                 .priority       = 99,
1775         },
1776         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1777          * or VS/NAT(change destination), so that filtering rules can be
1778          * applied to IPVS. */
1779         {
1780                 .hook           = ip_vs_remote_request4,
1781                 .owner          = THIS_MODULE,
1782                 .pf             = PF_INET,
1783                 .hooknum        = NF_INET_LOCAL_IN,
1784                 .priority       = 101,
1785         },
1786         /* Before ip_vs_in, change source only for VS/NAT */
1787         {
1788                 .hook           = ip_vs_local_reply4,
1789                 .owner          = THIS_MODULE,
1790                 .pf             = PF_INET,
1791                 .hooknum        = NF_INET_LOCAL_OUT,
1792                 .priority       = -99,
1793         },
1794         /* After mangle, schedule and forward local requests */
1795         {
1796                 .hook           = ip_vs_local_request4,
1797                 .owner          = THIS_MODULE,
1798                 .pf             = PF_INET,
1799                 .hooknum        = NF_INET_LOCAL_OUT,
1800                 .priority       = -98,
1801         },
1802         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1803          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1804         {
1805                 .hook           = ip_vs_forward_icmp,
1806                 .owner          = THIS_MODULE,
1807                 .pf             = PF_INET,
1808                 .hooknum        = NF_INET_FORWARD,
1809                 .priority       = 99,
1810         },
1811         /* After packet filtering, change source only for VS/NAT */
1812         {
1813                 .hook           = ip_vs_reply4,
1814                 .owner          = THIS_MODULE,
1815                 .pf             = PF_INET,
1816                 .hooknum        = NF_INET_FORWARD,
1817                 .priority       = 100,
1818         },
1819 #ifdef CONFIG_IP_VS_IPV6
1820         /* After packet filtering, change source only for VS/NAT */
1821         {
1822                 .hook           = ip_vs_reply6,
1823                 .owner          = THIS_MODULE,
1824                 .pf             = PF_INET6,
1825                 .hooknum        = NF_INET_LOCAL_IN,
1826                 .priority       = 99,
1827         },
1828         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1829          * or VS/NAT(change destination), so that filtering rules can be
1830          * applied to IPVS. */
1831         {
1832                 .hook           = ip_vs_remote_request6,
1833                 .owner          = THIS_MODULE,
1834                 .pf             = PF_INET6,
1835                 .hooknum        = NF_INET_LOCAL_IN,
1836                 .priority       = 101,
1837         },
1838         /* Before ip_vs_in, change source only for VS/NAT */
1839         {
1840                 .hook           = ip_vs_local_reply6,
1841                 .owner          = THIS_MODULE,
1842                 .pf             = PF_INET,
1843                 .hooknum        = NF_INET_LOCAL_OUT,
1844                 .priority       = -99,
1845         },
1846         /* After mangle, schedule and forward local requests */
1847         {
1848                 .hook           = ip_vs_local_request6,
1849                 .owner          = THIS_MODULE,
1850                 .pf             = PF_INET6,
1851                 .hooknum        = NF_INET_LOCAL_OUT,
1852                 .priority       = -98,
1853         },
1854         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1855          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1856         {
1857                 .hook           = ip_vs_forward_icmp_v6,
1858                 .owner          = THIS_MODULE,
1859                 .pf             = PF_INET6,
1860                 .hooknum        = NF_INET_FORWARD,
1861                 .priority       = 99,
1862         },
1863         /* After packet filtering, change source only for VS/NAT */
1864         {
1865                 .hook           = ip_vs_reply6,
1866                 .owner          = THIS_MODULE,
1867                 .pf             = PF_INET6,
1868                 .hooknum        = NF_INET_FORWARD,
1869                 .priority       = 100,
1870         },
1871 #endif
1872 };
1873 /*
1874  *      Initialize IP Virtual Server netns mem.
1875  */
1876 static int __net_init __ip_vs_init(struct net *net)
1877 {
1878         struct netns_ipvs *ipvs;
1879
1880         ipvs = net_generic(net, ip_vs_net_id);
1881         if (ipvs == NULL) {
1882                 pr_err("%s(): no memory.\n", __func__);
1883                 return -ENOMEM;
1884         }
1885         ipvs->net = net;
1886         /* Counters used for creating unique names */
1887         ipvs->gen = atomic_read(&ipvs_netns_cnt);
1888         atomic_inc(&ipvs_netns_cnt);
1889         net->ipvs = ipvs;
1890         printk(KERN_INFO "IPVS: Creating netns size=%zu id=%d\n",
1891                          sizeof(struct netns_ipvs), ipvs->gen);
1892         return 0;
1893 }
1894
1895 static void __net_exit __ip_vs_cleanup(struct net *net)
1896 {
1897         IP_VS_DBG(10, "ipvs netns %d released\n", net_ipvs(net)->gen);
1898 }
1899
1900 static struct pernet_operations ipvs_core_ops = {
1901         .init = __ip_vs_init,
1902         .exit = __ip_vs_cleanup,
1903         .id   = &ip_vs_net_id,
1904         .size = sizeof(struct netns_ipvs),
1905 };
1906
1907 /*
1908  *      Initialize IP Virtual Server
1909  */
1910 static int __init ip_vs_init(void)
1911 {
1912         int ret;
1913
1914         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
1915         if (ret < 0)
1916                 return ret;
1917
1918         ip_vs_estimator_init();
1919         ret = ip_vs_control_init();
1920         if (ret < 0) {
1921                 pr_err("can't setup control.\n");
1922                 goto cleanup_estimator;
1923         }
1924
1925         ip_vs_protocol_init();
1926
1927         ret = ip_vs_app_init();
1928         if (ret < 0) {
1929                 pr_err("can't setup application helper.\n");
1930                 goto cleanup_protocol;
1931         }
1932
1933         ret = ip_vs_conn_init();
1934         if (ret < 0) {
1935                 pr_err("can't setup connection table.\n");
1936                 goto cleanup_app;
1937         }
1938
1939         ret = ip_vs_sync_init();
1940         if (ret < 0) {
1941                 pr_err("can't setup sync data.\n");
1942                 goto cleanup_conn;
1943         }
1944
1945         ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
1946         if (ret < 0) {
1947                 pr_err("can't register hooks.\n");
1948                 goto cleanup_sync;
1949         }
1950
1951         pr_info("ipvs loaded.\n");
1952         return ret;
1953
1954 cleanup_sync:
1955         ip_vs_sync_cleanup();
1956   cleanup_conn:
1957         ip_vs_conn_cleanup();
1958   cleanup_app:
1959         ip_vs_app_cleanup();
1960   cleanup_protocol:
1961         ip_vs_protocol_cleanup();
1962         ip_vs_control_cleanup();
1963   cleanup_estimator:
1964         ip_vs_estimator_cleanup();
1965         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
1966         return ret;
1967 }
1968
1969 static void __exit ip_vs_cleanup(void)
1970 {
1971         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
1972         ip_vs_sync_cleanup();
1973         ip_vs_conn_cleanup();
1974         ip_vs_app_cleanup();
1975         ip_vs_protocol_cleanup();
1976         ip_vs_control_cleanup();
1977         ip_vs_estimator_cleanup();
1978         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
1979         pr_info("ipvs unloaded.\n");
1980 }
1981
1982 module_init(ip_vs_init);
1983 module_exit(ip_vs_cleanup);
1984 MODULE_LICENSE("GPL");