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->tot_stats.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->tot_stats.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->tot_stats.cpustats);
 187         s->ustats.conns++;
 188 }
 189
 190
 191 static inline void
 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 (likely(pd->pp->state_transition))
 197                 pd->pp->state_transition(cp, direction, skb, pd);
 198 }
 199
 200 static inline int
 201 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
 202                               struct sk_buff *skb, int protocol,
 203                               const union nf_inet_addr *caddr, __be16 cport,
 204                               const union nf_inet_addr *vaddr, __be16 vport,
 205                               struct ip_vs_conn_param *p)
 206 {
 207         ip_vs_conn_fill_param(svc->net, svc->af, protocol, caddr, cport, vaddr,
 208                               vport, p);
 209         p->pe = svc->pe;
 210         if (p->pe && p->pe->fill_param)
 211                 return p->pe->fill_param(p, skb);
 212
 213         return 0;
 214 }
 215
 216 /*
 217  *  IPVS persistent scheduling function
 218  *  It creates a connection entry according to its template if exists,
 219  *  or selects a server and creates a connection entry plus a template.
 220  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
 221  *  Protocols supported: TCP, UDP
 222  */
 223 static struct ip_vs_conn *
 224 ip_vs_sched_persist(struct ip_vs_service *svc,
 225                     struct sk_buff *skb,
 226                     __be16 src_port, __be16 dst_port, int *ignored)
 227 {
 228         struct ip_vs_conn *cp = NULL;
 229         struct ip_vs_iphdr iph;
 230         struct ip_vs_dest *dest;
 231         struct ip_vs_conn *ct;
 232         __be16 dport = 0;               /* destination port to forward */
 233         unsigned int flags;
 234         struct ip_vs_conn_param param;
 235         union nf_inet_addr snet;        /* source network of the client,
 236                                            after masking */
 237
 238         ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph);
 239
 240         /* Mask saddr with the netmask to adjust template granularity */
 241 #ifdef CONFIG_IP_VS_IPV6
 242         if (svc->af == AF_INET6)
 243                 ipv6_addr_prefix(&snet.in6, &iph.saddr.in6, svc->netmask);
 244         else
 245 #endif
 246                 snet.ip = iph.saddr.ip & svc->netmask;
 247
 248         IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
 249                       "mnet %s\n",
 250                       IP_VS_DBG_ADDR(svc->af, &iph.saddr), ntohs(src_port),
 251                       IP_VS_DBG_ADDR(svc->af, &iph.daddr), ntohs(dst_port),
 252                       IP_VS_DBG_ADDR(svc->af, &snet));
 253
 254         /*
 255          * As far as we know, FTP is a very complicated network protocol, and
 256          * it uses control connection and data connections. For active FTP,
 257          * FTP server initialize data connection to the client, its source port
 258          * is often 20. For passive FTP, FTP server tells the clients the port
 259          * that it passively listens to,  and the client issues the data
 260          * connection. In the tunneling or direct routing mode, the load
 261          * balancer is on the client-to-server half of connection, the port
 262          * number is unknown to the load balancer. So, a conn template like
 263          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
 264          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
 265          * is created for other persistent services.
 266          */
 267         {
 268                 int protocol = iph.protocol;
 269                 const union nf_inet_addr *vaddr = &iph.daddr;
 270                 const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
 271                 __be16 vport = 0;
 272
 273                 if (dst_port == svc->port) {
 274                         /* non-FTP template:
 275                          * <protocol, caddr, 0, vaddr, vport, daddr, dport>
 276                          * FTP template:
 277                          * <protocol, caddr, 0, vaddr, 0, daddr, 0>
 278                          */
 279                         if (svc->port != FTPPORT)
 280                                 vport = dst_port;
 281                 } else {
 282                         /* Note: persistent fwmark-based services and
 283                          * persistent port zero service are handled here.
 284                          * fwmark template:
 285                          * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
 286                          * port zero template:
 287                          * <protocol,caddr,0,vaddr,0,daddr,0>
 288                          */
 289                         if (svc->fwmark) {
 290                                 protocol = IPPROTO_IP;
 291                                 vaddr = &fwmark;
 292                         }
 293                 }
 294                 /* return *ignored = -1 so NF_DROP can be used */
 295                 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
 296                                                   vaddr, vport, &param) < 0) {
 297                         *ignored = -1;
 298                         return NULL;
 299                 }
 300         }
 301
 302         /* Check if a template already exists */
 303         ct = ip_vs_ct_in_get(&param);
 304         if (!ct || !ip_vs_check_template(ct)) {
 305                 /*
 306                  * No template found or the dest of the connection
 307                  * template is not available.
 308                  * return *ignored=0 i.e. ICMP and NF_DROP
 309                  */
 310                 dest = svc->scheduler->schedule(svc, skb);
 311                 if (!dest) {
 312                         IP_VS_DBG(1, "p-schedule: no dest found.\n");
 313                         kfree(param.pe_data);
 314                         *ignored = 0;
 315                         return NULL;
 316                 }
 317
 318                 if (dst_port == svc->port && svc->port != FTPPORT)
 319                         dport = dest->port;
 320
 321                 /* Create a template
 322                  * This adds param.pe_data to the template,
 323                  * and thus param.pe_data will be destroyed
 324                  * when the template expires */
 325                 ct = ip_vs_conn_new(&param, &dest->addr, dport,
 326                                     IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
 327                 if (ct == NULL) {
 328                         kfree(param.pe_data);
 329                         *ignored = -1;
 330                         return NULL;
 331                 }
 332
 333                 ct->timeout = svc->timeout;
 334         } else {
 335                 /* set destination with the found template */
 336                 dest = ct->dest;
 337                 kfree(param.pe_data);
 338         }
 339
 340         dport = dst_port;
 341         if (dport == svc->port && dest->port)
 342                 dport = dest->port;
 343
 344         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 345                  && iph.protocol == IPPROTO_UDP)?
 346                 IP_VS_CONN_F_ONE_PACKET : 0;
 347
 348         /*
 349          *    Create a new connection according to the template
 350          */
 351         ip_vs_conn_fill_param(svc->net, svc->af, iph.protocol, &iph.saddr,
 352                               src_port, &iph.daddr, dst_port, &param);
 353
 354         cp = ip_vs_conn_new(&param, &dest->addr, dport, flags, dest, skb->mark);
 355         if (cp == NULL) {
 356                 ip_vs_conn_put(ct);
 357                 *ignored = -1;
 358                 return NULL;
 359         }
 360
 361         /*
 362          *    Add its control
 363          */
 364         ip_vs_control_add(cp, ct);
 365         ip_vs_conn_put(ct);
 366
 367         ip_vs_conn_stats(cp, svc);
 368         return cp;
 369 }
 370
 371
 372 /*
 373  *  IPVS main scheduling function
 374  *  It selects a server according to the virtual service, and
 375  *  creates a connection entry.
 376  *  Protocols supported: TCP, UDP
 377  *
 378  *  Usage of *ignored
 379  *
 380  * 1 :   protocol tried to schedule (eg. on SYN), found svc but the
 381  *       svc/scheduler decides that this packet should be accepted with
 382  *       NF_ACCEPT because it must not be scheduled.
 383  *
 384  * 0 :   scheduler can not find destination, so try bypass or
 385  *       return ICMP and then NF_DROP (ip_vs_leave).
 386  *
 387  * -1 :  scheduler tried to schedule but fatal error occurred, eg.
 388  *       ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
 389  *       failure such as missing Call-ID, ENOMEM on skb_linearize
 390  *       or pe_data. In this case we should return NF_DROP without
 391  *       any attempts to send ICMP with ip_vs_leave.
 392  */
 393 struct ip_vs_conn *
 394 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
 395                struct ip_vs_proto_data *pd, int *ignored)
 396 {
 397         struct ip_vs_protocol *pp = pd->pp;
 398         struct ip_vs_conn *cp = NULL;
 399         struct ip_vs_iphdr iph;
 400         struct ip_vs_dest *dest;
 401         __be16 _ports[2], *pptr;
 402         unsigned int flags;
 403
 404         *ignored = 1;
 405         ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph);
 406         pptr = skb_header_pointer(skb, iph.len, sizeof(_ports), _ports);
 407         if (pptr == NULL)
 408                 return NULL;
 409
 410         /*
 411          * FTPDATA needs this check when using local real server.
 412          * Never schedule Active FTPDATA connections from real server.
 413          * For LVS-NAT they must be already created. For other methods
 414          * with persistence the connection is created on SYN+ACK.
 415          */
 416         if (pptr[0] == FTPDATA) {
 417                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 418                               "Not scheduling FTPDATA");
 419                 return NULL;
 420         }
 421
 422         /*
 423          *    Do not schedule replies from local real server.
 424          */
 425         if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
 426             (cp = pp->conn_in_get(svc->af, skb, &iph, iph.len, 1))) {
 427                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 428                               "Not scheduling reply for existing connection");
 429                 __ip_vs_conn_put(cp);
 430                 return NULL;
 431         }
 432
 433         /*
 434          *    Persistent service
 435          */
 436         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 437                 return ip_vs_sched_persist(svc, skb, pptr[0], pptr[1], ignored);
 438
 439         *ignored = 0;
 440
 441         /*
 442          *    Non-persistent service
 443          */
 444         if (!svc->fwmark && pptr[1] != svc->port) {
 445                 if (!svc->port)
 446                         pr_err("Schedule: port zero only supported "
 447                                "in persistent services, "
 448                                "check your ipvs configuration\n");
 449                 return NULL;
 450         }
 451
 452         dest = svc->scheduler->schedule(svc, skb);
 453         if (dest == NULL) {
 454                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 455                 return NULL;
 456         }
 457
 458         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 459                  && iph.protocol == IPPROTO_UDP)?
 460                 IP_VS_CONN_F_ONE_PACKET : 0;
 461
 462         /*
 463          *    Create a connection entry.
 464          */
 465         {
 466                 struct ip_vs_conn_param p;
 467
 468                 ip_vs_conn_fill_param(svc->net, svc->af, iph.protocol,
 469                                       &iph.saddr, pptr[0], &iph.daddr, pptr[1],
 470                                       &p);
 471                 cp = ip_vs_conn_new(&p, &dest->addr,
 472                                     dest->port ? dest->port : pptr[1],
 473                                     flags, dest, skb->mark);
 474                 if (!cp) {
 475                         *ignored = -1;
 476                         return NULL;
 477                 }
 478         }
 479
 480         IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
 481                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
 482                       ip_vs_fwd_tag(cp),
 483                       IP_VS_DBG_ADDR(svc->af, &cp->caddr), ntohs(cp->cport),
 484                       IP_VS_DBG_ADDR(svc->af, &cp->vaddr), ntohs(cp->vport),
 485                       IP_VS_DBG_ADDR(svc->af, &cp->daddr), ntohs(cp->dport),
 486                       cp->flags, atomic_read(&cp->refcnt));
 487
 488         ip_vs_conn_stats(cp, svc);
 489         return cp;
 490 }
 491
 492
 493 /*
 494  *  Pass or drop the packet.
 495  *  Called by ip_vs_in, when the virtual service is available but
 496  *  no destination is available for a new connection.
 497  */
 498 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 499                 struct ip_vs_proto_data *pd)
 500 {
 501         __be16 _ports[2], *pptr;
 502         struct ip_vs_iphdr iph;
 503 #ifdef CONFIG_SYSCTL
 504         struct net *net;
 505         struct netns_ipvs *ipvs;
 506         int unicast;
 507 #endif
 508
 509         ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph);
 510
 511         pptr = skb_header_pointer(skb, iph.len, sizeof(_ports), _ports);
 512         if (pptr == NULL) {
 513                 ip_vs_service_put(svc);
 514                 return NF_DROP;
 515         }
 516
 517 #ifdef CONFIG_SYSCTL
 518         net = skb_net(skb);
 519
 520 #ifdef CONFIG_IP_VS_IPV6
 521         if (svc->af == AF_INET6)
 522                 unicast = ipv6_addr_type(&iph.daddr.in6) & IPV6_ADDR_UNICAST;
 523         else
 524 #endif
 525                 unicast = (inet_addr_type(net, iph.daddr.ip) == RTN_UNICAST);
 526
 527         /* if it is fwmark-based service, the cache_bypass sysctl is up
 528            and the destination is a non-local unicast, then create
 529            a cache_bypass connection entry */
 530         ipvs = net_ipvs(net);
 531         if (ipvs->sysctl_cache_bypass && svc->fwmark && unicast) {
 532                 int ret;
 533                 struct ip_vs_conn *cp;
 534                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
 535                                       iph.protocol == IPPROTO_UDP)?
 536                                       IP_VS_CONN_F_ONE_PACKET : 0;
 537                 union nf_inet_addr daddr =  { .all = { 0, 0, 0, 0 } };
 538
 539                 ip_vs_service_put(svc);
 540
 541                 /* create a new connection entry */
 542                 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
 543                 {
 544                         struct ip_vs_conn_param p;
 545                         ip_vs_conn_fill_param(svc->net, svc->af, iph.protocol,
 546                                               &iph.saddr, pptr[0],
 547                                               &iph.daddr, pptr[1], &p);
 548                         cp = ip_vs_conn_new(&p, &daddr, 0,
 549                                             IP_VS_CONN_F_BYPASS | flags,
 550                                             NULL, skb->mark);
 551                         if (!cp)
 552                                 return NF_DROP;
 553                 }
 554
 555                 /* statistics */
 556                 ip_vs_in_stats(cp, skb);
 557
 558                 /* set state */
 559                 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
 560
 561                 /* transmit the first SYN packet */
 562                 ret = cp->packet_xmit(skb, cp, pd->pp);
 563                 /* do not touch skb anymore */
 564
 565                 atomic_inc(&cp->in_pkts);
 566                 ip_vs_conn_put(cp);
 567                 return ret;
 568         }
 569 #endif
 570
 571         /*
 572          * When the virtual ftp service is presented, packets destined
 573          * for other services on the VIP may get here (except services
 574          * listed in the ipvs table), pass the packets, because it is
 575          * not ipvs job to decide to drop the packets.
 576          */
 577         if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT)) {
 578                 ip_vs_service_put(svc);
 579                 return NF_ACCEPT;
 580         }
 581
 582         ip_vs_service_put(svc);
 583
 584         /*
 585          * Notify the client that the destination is unreachable, and
 586          * release the socket buffer.
 587          * Since it is in IP layer, the TCP socket is not actually
 588          * created, the TCP RST packet cannot be sent, instead that
 589          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 590          */
 591 #ifdef CONFIG_IP_VS_IPV6
 592         if (svc->af == AF_INET6) {
 593                 if (!skb->dev) {
 594                         struct net *net = dev_net(skb_dst(skb)->dev);
 595
 596                         skb->dev = net->loopback_dev;
 597                 }
 598                 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 599         } else
 600 #endif
 601                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 602
 603         return NF_DROP;
 604 }
 605
 606 #ifdef CONFIG_SYSCTL
 607
 608 static int sysctl_snat_reroute(struct sk_buff *skb)
 609 {
 610         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 611         return ipvs->sysctl_snat_reroute;
 612 }
 613
 614 static int sysctl_nat_icmp_send(struct net *net)
 615 {
 616         struct netns_ipvs *ipvs = net_ipvs(net);
 617         return ipvs->sysctl_nat_icmp_send;
 618 }
 619
 620 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
 621 {
 622         return ipvs->sysctl_expire_nodest_conn;
 623 }
 624
 625 #else
 626
 627 static int sysctl_snat_reroute(struct sk_buff *skb) { return 0; }
 628 static int sysctl_nat_icmp_send(struct net *net) { return 0; }
 629 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
 630
 631 #endif
 632
 633 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 634 {
 635         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 636 }
 637
 638 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
 639 {
 640         if (NF_INET_LOCAL_IN == hooknum)
 641                 return IP_DEFRAG_VS_IN;
 642         if (NF_INET_FORWARD == hooknum)
 643                 return IP_DEFRAG_VS_FWD;
 644         return IP_DEFRAG_VS_OUT;
 645 }
 646
 647 static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user)
 648 {
 649         int err = ip_defrag(skb, user);
 650
 651         if (!err)
 652                 ip_send_check(ip_hdr(skb));
 653
 654         return err;
 655 }
 656
 657 #ifdef CONFIG_IP_VS_IPV6
 658 static inline int ip_vs_gather_frags_v6(struct sk_buff *skb, u_int32_t user)
 659 {
 660         /* TODO IPv6: Find out what to do here for IPv6 */
 661         return 0;
 662 }
 663 #endif
 664
 665 static int ip_vs_route_me_harder(int af, struct sk_buff *skb)
 666 {
 667 #ifdef CONFIG_IP_VS_IPV6
 668         if (af == AF_INET6) {
 669                 if (sysctl_snat_reroute(skb) && ip6_route_me_harder(skb) != 0)
 670                         return 1;
 671         } else
 672 #endif
 673                 if ((sysctl_snat_reroute(skb) ||
 674                      skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
 675                     ip_route_me_harder(skb, RTN_LOCAL) != 0)
 676                         return 1;
 677
 678         return 0;
 679 }
 680
 681 /*
 682  * Packet has been made sufficiently writable in caller
 683  * - inout: 1=in->out, 0=out->in
 684  */
 685 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 686                     struct ip_vs_conn *cp, int inout)
 687 {
 688         struct iphdr *iph        = ip_hdr(skb);
 689         unsigned int icmp_offset = iph->ihl*4;
 690         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
 691                                                       icmp_offset);
 692         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 693
 694         if (inout) {
 695                 iph->saddr = cp->vaddr.ip;
 696                 ip_send_check(iph);
 697                 ciph->daddr = cp->vaddr.ip;
 698                 ip_send_check(ciph);
 699         } else {
 700                 iph->daddr = cp->daddr.ip;
 701                 ip_send_check(iph);
 702                 ciph->saddr = cp->daddr.ip;
 703                 ip_send_check(ciph);
 704         }
 705
 706         /* the TCP/UDP/SCTP port */
 707         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
 708             IPPROTO_SCTP == ciph->protocol) {
 709                 __be16 *ports = (void *)ciph + ciph->ihl*4;
 710
 711                 if (inout)
 712                         ports[1] = cp->vport;
 713                 else
 714                         ports[0] = cp->dport;
 715         }
 716
 717         /* And finally the ICMP checksum */
 718         icmph->checksum = 0;
 719         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 720         skb->ip_summed = CHECKSUM_UNNECESSARY;
 721
 722         if (inout)
 723                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 724                         "Forwarding altered outgoing ICMP");
 725         else
 726                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 727                         "Forwarding altered incoming ICMP");
 728 }
 729
 730 #ifdef CONFIG_IP_VS_IPV6
 731 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
 732                     struct ip_vs_conn *cp, int inout)
 733 {
 734         struct ipv6hdr *iph      = ipv6_hdr(skb);
 735         unsigned int icmp_offset = sizeof(struct ipv6hdr);
 736         struct icmp6hdr *icmph   = (struct icmp6hdr *)(skb_network_header(skb) +
 737                                                       icmp_offset);
 738         struct ipv6hdr *ciph     = (struct ipv6hdr *)(icmph + 1);
 739
 740         if (inout) {
 741                 iph->saddr = cp->vaddr.in6;
 742                 ciph->daddr = cp->vaddr.in6;
 743         } else {
 744                 iph->daddr = cp->daddr.in6;
 745                 ciph->saddr = cp->daddr.in6;
 746         }
 747
 748         /* the TCP/UDP/SCTP port */
 749         if (IPPROTO_TCP == ciph->nexthdr || IPPROTO_UDP == ciph->nexthdr ||
 750             IPPROTO_SCTP == ciph->nexthdr) {
 751                 __be16 *ports = (void *)ciph + sizeof(struct ipv6hdr);
 752
 753                 if (inout)
 754                         ports[1] = cp->vport;
 755                 else
 756                         ports[0] = cp->dport;
 757         }
 758
 759         /* And finally the ICMP checksum */
 760         icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
 761                                               skb->len - icmp_offset,
 762                                               IPPROTO_ICMPV6, 0);
 763         skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
 764         skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
 765         skb->ip_summed = CHECKSUM_PARTIAL;
 766
 767         if (inout)
 768                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 769                               (void *)ciph - (void *)iph,
 770                               "Forwarding altered outgoing ICMPv6");
 771         else
 772                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 773                               (void *)ciph - (void *)iph,
 774                               "Forwarding altered incoming ICMPv6");
 775 }
 776 #endif
 777
 778 /* Handle relevant response ICMP messages - forward to the right
 779  * destination host.
 780  */
 781 static int handle_response_icmp(int af, struct sk_buff *skb,
 782                                 union nf_inet_addr *snet,
 783                                 __u8 protocol, struct ip_vs_conn *cp,
 784                                 struct ip_vs_protocol *pp,
 785                                 unsigned int offset, unsigned int ihl)
 786 {
 787         unsigned int verdict = NF_DROP;
 788
 789         if (IP_VS_FWD_METHOD(cp) != 0) {
 790                 pr_err("shouldn't reach here, because the box is on the "
 791                        "half connection in the tun/dr module.\n");
 792         }
 793
 794         /* Ensure the checksum is correct */
 795         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 796                 /* Failed checksum! */
 797                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
 798                               IP_VS_DBG_ADDR(af, snet));
 799                 goto out;
 800         }
 801
 802         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 803             IPPROTO_SCTP == protocol)
 804                 offset += 2 * sizeof(__u16);
 805         if (!skb_make_writable(skb, offset))
 806                 goto out;
 807
 808 #ifdef CONFIG_IP_VS_IPV6
 809         if (af == AF_INET6)
 810                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
 811         else
 812 #endif
 813                 ip_vs_nat_icmp(skb, pp, cp, 1);
 814
 815         if (ip_vs_route_me_harder(af, skb))
 816                 goto out;
 817
 818         /* do the statistics and put it back */
 819         ip_vs_out_stats(cp, skb);
 820
 821         skb->ipvs_property = 1;
 822         if (!(cp->flags & IP_VS_CONN_F_NFCT))
 823                 ip_vs_notrack(skb);
 824         else
 825                 ip_vs_update_conntrack(skb, cp, 0);
 826         verdict = NF_ACCEPT;
 827
 828 out:
 829         __ip_vs_conn_put(cp);
 830
 831         return verdict;
 832 }
 833
 834 /*
 835  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 836  *      Find any that might be relevant, check against existing connections.
 837  *      Currently handles error types - unreachable, quench, ttl exceeded.
 838  */
 839 static int ip_vs_out_icmp(struct sk_buff *skb, int *related,
 840                           unsigned int hooknum)
 841 {
 842         struct iphdr *iph;
 843         struct icmphdr  _icmph, *ic;
 844         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 845         struct ip_vs_iphdr ciph;
 846         struct ip_vs_conn *cp;
 847         struct ip_vs_protocol *pp;
 848         unsigned int offset, ihl;
 849         union nf_inet_addr snet;
 850
 851         *related = 1;
 852
 853         /* reassemble IP fragments */
 854         if (ip_is_fragment(ip_hdr(skb))) {
 855                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
 856                         return NF_STOLEN;
 857         }
 858
 859         iph = ip_hdr(skb);
 860         offset = ihl = iph->ihl * 4;
 861         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 862         if (ic == NULL)
 863                 return NF_DROP;
 864
 865         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
 866                   ic->type, ntohs(icmp_id(ic)),
 867                   &iph->saddr, &iph->daddr);
 868
 869         /*
 870          * Work through seeing if this is for us.
 871          * These checks are supposed to be in an order that means easy
 872          * things are checked first to speed up processing.... however
 873          * this means that some packets will manage to get a long way
 874          * down this stack and then be rejected, but that's life.
 875          */
 876         if ((ic->type != ICMP_DEST_UNREACH) &&
 877             (ic->type != ICMP_SOURCE_QUENCH) &&
 878             (ic->type != ICMP_TIME_EXCEEDED)) {
 879                 *related = 0;
 880                 return NF_ACCEPT;
 881         }
 882
 883         /* Now find the contained IP header */
 884         offset += sizeof(_icmph);
 885         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 886         if (cih == NULL)
 887                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 888
 889         pp = ip_vs_proto_get(cih->protocol);
 890         if (!pp)
 891                 return NF_ACCEPT;
 892
 893         /* Is the embedded protocol header present? */
 894         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 895                      pp->dont_defrag))
 896                 return NF_ACCEPT;
 897
 898         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
 899                       "Checking outgoing ICMP for");
 900
 901         offset += cih->ihl * 4;
 902
 903         ip_vs_fill_iphdr(AF_INET, cih, &ciph);
 904         /* The embedded headers contain source and dest in reverse order */
 905         cp = pp->conn_out_get(AF_INET, skb, &ciph, offset, 1);
 906         if (!cp)
 907                 return NF_ACCEPT;
 908
 909         snet.ip = iph->saddr;
 910         return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
 911                                     pp, offset, ihl);
 912 }
 913
 914 #ifdef CONFIG_IP_VS_IPV6
 915 static int ip_vs_out_icmp_v6(struct sk_buff *skb, int *related,
 916                              unsigned int hooknum)
 917 {
 918         struct ipv6hdr *iph;
 919         struct icmp6hdr _icmph, *ic;
 920         struct ipv6hdr  _ciph, *cih;    /* The ip header contained
 921                                            within the ICMP */
 922         struct ip_vs_iphdr ciph;
 923         struct ip_vs_conn *cp;
 924         struct ip_vs_protocol *pp;
 925         unsigned int offset;
 926         union nf_inet_addr snet;
 927
 928         *related = 1;
 929
 930         /* reassemble IP fragments */
 931         if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) {
 932                 if (ip_vs_gather_frags_v6(skb, ip_vs_defrag_user(hooknum)))
 933                         return NF_STOLEN;
 934         }
 935
 936         iph = ipv6_hdr(skb);
 937         offset = sizeof(struct ipv6hdr);
 938         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 939         if (ic == NULL)
 940                 return NF_DROP;
 941
 942         IP_VS_DBG(12, "Outgoing ICMPv6 (%d,%d) %pI6->%pI6\n",
 943                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
 944                   &iph->saddr, &iph->daddr);
 945
 946         /*
 947          * Work through seeing if this is for us.
 948          * These checks are supposed to be in an order that means easy
 949          * things are checked first to speed up processing.... however
 950          * this means that some packets will manage to get a long way
 951          * down this stack and then be rejected, but that's life.
 952          */
 953         if ((ic->icmp6_type != ICMPV6_DEST_UNREACH) &&
 954             (ic->icmp6_type != ICMPV6_PKT_TOOBIG) &&
 955             (ic->icmp6_type != ICMPV6_TIME_EXCEED)) {
 956                 *related = 0;
 957                 return NF_ACCEPT;
 958         }
 959
 960         /* Now find the contained IP header */
 961         offset += sizeof(_icmph);
 962         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 963         if (cih == NULL)
 964                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 965
 966         pp = ip_vs_proto_get(cih->nexthdr);
 967         if (!pp)
 968                 return NF_ACCEPT;
 969
 970         /* Is the embedded protocol header present? */
 971         /* TODO: we don't support fragmentation at the moment anyways */
 972         if (unlikely(cih->nexthdr == IPPROTO_FRAGMENT && pp->dont_defrag))
 973                 return NF_ACCEPT;
 974
 975         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
 976                       "Checking outgoing ICMPv6 for");
 977
 978         offset += sizeof(struct ipv6hdr);
 979
 980         ip_vs_fill_iphdr(AF_INET6, cih, &ciph);
 981         /* The embedded headers contain source and dest in reverse order */
 982         cp = pp->conn_out_get(AF_INET6, skb, &ciph, offset, 1);
 983         if (!cp)
 984                 return NF_ACCEPT;
 985
 986         ipv6_addr_copy(&snet.in6, &iph->saddr);
 987         return handle_response_icmp(AF_INET6, skb, &snet, cih->nexthdr, cp,
 988                                     pp, offset, sizeof(struct ipv6hdr));
 989 }
 990 #endif
 991
 992 /*
 993  * Check if sctp chunc is ABORT chunk
 994  */
 995 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
 996 {
 997         sctp_chunkhdr_t *sch, schunk;
 998         sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
 999                         sizeof(schunk), &schunk);
1000         if (sch == NULL)
1001                 return 0;
1002         if (sch->type == SCTP_CID_ABORT)
1003                 return 1;
1004         return 0;
1005 }
1006
1007 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
1008 {
1009         struct tcphdr _tcph, *th;
1010
1011         th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
1012         if (th == NULL)
1013                 return 0;
1014         return th->rst;
1015 }
1016
1017 /* Handle response packets: rewrite addresses and send away...
1018  */
1019 static unsigned int
1020 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1021                 struct ip_vs_conn *cp, int ihl)
1022 {
1023         struct ip_vs_protocol *pp = pd->pp;
1024
1025         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Outgoing packet");
1026
1027         if (!skb_make_writable(skb, ihl))
1028                 goto drop;
1029
1030         /* mangle the packet */
1031         if (pp->snat_handler && !pp->snat_handler(skb, pp, cp))
1032                 goto drop;
1033
1034 #ifdef CONFIG_IP_VS_IPV6
1035         if (af == AF_INET6)
1036                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1037         else
1038 #endif
1039         {
1040                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1041                 ip_send_check(ip_hdr(skb));
1042         }
1043
1044         /*
1045          * nf_iterate does not expect change in the skb->dst->dev.
1046          * It looks like it is not fatal to enable this code for hooks
1047          * where our handlers are at the end of the chain list and
1048          * when all next handlers use skb->dst->dev and not outdev.
1049          * It will definitely route properly the inout NAT traffic
1050          * when multiple paths are used.
1051          */
1052
1053         /* For policy routing, packets originating from this
1054          * machine itself may be routed differently to packets
1055          * passing through.  We want this packet to be routed as
1056          * if it came from this machine itself.  So re-compute
1057          * the routing information.
1058          */
1059         if (ip_vs_route_me_harder(af, skb))
1060                 goto drop;
1061
1062         IP_VS_DBG_PKT(10, af, pp, skb, 0, "After SNAT");
1063
1064         ip_vs_out_stats(cp, skb);
1065         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1066         skb->ipvs_property = 1;
1067         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1068                 ip_vs_notrack(skb);
1069         else
1070                 ip_vs_update_conntrack(skb, cp, 0);
1071         ip_vs_conn_put(cp);
1072
1073         LeaveFunction(11);
1074         return NF_ACCEPT;
1075
1076 drop:
1077         ip_vs_conn_put(cp);
1078         kfree_skb(skb);
1079         LeaveFunction(11);
1080         return NF_STOLEN;
1081 }
1082
1083 /*
1084  *      Check if outgoing packet belongs to the established ip_vs_conn.
1085  */
1086 static unsigned int
1087 ip_vs_out(unsigned int hooknum, struct sk_buff *skb, int af)
1088 {
1089         struct net *net = NULL;
1090         struct ip_vs_iphdr iph;
1091         struct ip_vs_protocol *pp;
1092         struct ip_vs_proto_data *pd;
1093         struct ip_vs_conn *cp;
1094
1095         EnterFunction(11);
1096
1097         /* Already marked as IPVS request or reply? */
1098         if (skb->ipvs_property)
1099                 return NF_ACCEPT;
1100
1101         /* Bad... Do not break raw sockets */
1102         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1103                      af == AF_INET)) {
1104                 struct sock *sk = skb->sk;
1105                 struct inet_sock *inet = inet_sk(skb->sk);
1106
1107                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1108                         return NF_ACCEPT;
1109         }
1110
1111         if (unlikely(!skb_dst(skb)))
1112                 return NF_ACCEPT;
1113
1114         net = skb_net(skb);
1115         if (!net_ipvs(net)->enable)
1116                 return NF_ACCEPT;
1117
1118         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1119 #ifdef CONFIG_IP_VS_IPV6
1120         if (af == AF_INET6) {
1121                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1122                         int related;
1123                         int verdict = ip_vs_out_icmp_v6(skb, &related,
1124                                                         hooknum);
1125
1126                         if (related)
1127                                 return verdict;
1128                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1129                 }
1130         } else
1131 #endif
1132                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1133                         int related;
1134                         int verdict = ip_vs_out_icmp(skb, &related, hooknum);
1135
1136                         if (related)
1137                                 return verdict;
1138                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1139                 }
1140
1141         pd = ip_vs_proto_data_get(net, iph.protocol);
1142         if (unlikely(!pd))
1143                 return NF_ACCEPT;
1144         pp = pd->pp;
1145
1146         /* reassemble IP fragments */
1147 #ifdef CONFIG_IP_VS_IPV6
1148         if (af == AF_INET6) {
1149                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) {
1150                         if (ip_vs_gather_frags_v6(skb,
1151                                                   ip_vs_defrag_user(hooknum)))
1152                                 return NF_STOLEN;
1153                 }
1154
1155                 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1156         } else
1157 #endif
1158                 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1159                         if (ip_vs_gather_frags(skb,
1160                                                ip_vs_defrag_user(hooknum)))
1161                                 return NF_STOLEN;
1162
1163                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1164                 }
1165
1166         /*
1167          * Check if the packet belongs to an existing entry
1168          */
1169         cp = pp->conn_out_get(af, skb, &iph, iph.len, 0);
1170
1171         if (likely(cp))
1172                 return handle_response(af, skb, pd, cp, iph.len);
1173         if (sysctl_nat_icmp_send(net) &&
1174             (pp->protocol == IPPROTO_TCP ||
1175              pp->protocol == IPPROTO_UDP ||
1176              pp->protocol == IPPROTO_SCTP)) {
1177                 __be16 _ports[2], *pptr;
1178
1179                 pptr = skb_header_pointer(skb, iph.len,
1180                                           sizeof(_ports), _ports);
1181                 if (pptr == NULL)
1182                         return NF_ACCEPT;       /* Not for me */
1183                 if (ip_vs_lookup_real_service(net, af, iph.protocol,
1184                                               &iph.saddr,
1185                                               pptr[0])) {
1186                         /*
1187                          * Notify the real server: there is no
1188                          * existing entry if it is not RST
1189                          * packet or not TCP packet.
1190                          */
1191                         if ((iph.protocol != IPPROTO_TCP &&
1192                              iph.protocol != IPPROTO_SCTP)
1193                              || ((iph.protocol == IPPROTO_TCP
1194                                   && !is_tcp_reset(skb, iph.len))
1195                                  || (iph.protocol == IPPROTO_SCTP
1196                                         && !is_sctp_abort(skb,
1197                                                 iph.len)))) {
1198 #ifdef CONFIG_IP_VS_IPV6
1199                                 if (af == AF_INET6) {
1200                                         struct net *net =
1201                                                 dev_net(skb_dst(skb)->dev);
1202
1203                                         if (!skb->dev)
1204                                                 skb->dev = net->loopback_dev;
1205                                         icmpv6_send(skb,
1206                                                     ICMPV6_DEST_UNREACH,
1207                                                     ICMPV6_PORT_UNREACH,
1208                                                     0);
1209                                 } else
1210 #endif
1211                                         icmp_send(skb,
1212                                                   ICMP_DEST_UNREACH,
1213                                                   ICMP_PORT_UNREACH, 0);
1214                                 return NF_DROP;
1215                         }
1216                 }
1217         }
1218         IP_VS_DBG_PKT(12, af, pp, skb, 0,
1219                       "ip_vs_out: packet continues traversal as normal");
1220         return NF_ACCEPT;
1221 }
1222
1223 /*
1224  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1225  *      used only for VS/NAT.
1226  *      Check if packet is reply for established ip_vs_conn.
1227  */
1228 static unsigned int
1229 ip_vs_reply4(unsigned int hooknum, struct sk_buff *skb,
1230              const struct net_device *in, const struct net_device *out,
1231              int (*okfn)(struct sk_buff *))
1232 {
1233         return ip_vs_out(hooknum, skb, AF_INET);
1234 }
1235
1236 /*
1237  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1238  *      Check if packet is reply for established ip_vs_conn.
1239  */
1240 static unsigned int
1241 ip_vs_local_reply4(unsigned int hooknum, struct sk_buff *skb,
1242                    const struct net_device *in, const struct net_device *out,
1243                    int (*okfn)(struct sk_buff *))
1244 {
1245         unsigned int verdict;
1246
1247         /* Disable BH in LOCAL_OUT until all places are fixed */
1248         local_bh_disable();
1249         verdict = ip_vs_out(hooknum, skb, AF_INET);
1250         local_bh_enable();
1251         return verdict;
1252 }
1253
1254 #ifdef CONFIG_IP_VS_IPV6
1255
1256 /*
1257  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1258  *      used only for VS/NAT.
1259  *      Check if packet is reply for established ip_vs_conn.
1260  */
1261 static unsigned int
1262 ip_vs_reply6(unsigned int hooknum, struct sk_buff *skb,
1263              const struct net_device *in, const struct net_device *out,
1264              int (*okfn)(struct sk_buff *))
1265 {
1266         return ip_vs_out(hooknum, skb, AF_INET6);
1267 }
1268
1269 /*
1270  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1271  *      Check if packet is reply for established ip_vs_conn.
1272  */
1273 static unsigned int
1274 ip_vs_local_reply6(unsigned int hooknum, struct sk_buff *skb,
1275                    const struct net_device *in, const struct net_device *out,
1276                    int (*okfn)(struct sk_buff *))
1277 {
1278         unsigned int verdict;
1279
1280         /* Disable BH in LOCAL_OUT until all places are fixed */
1281         local_bh_disable();
1282         verdict = ip_vs_out(hooknum, skb, AF_INET6);
1283         local_bh_enable();
1284         return verdict;
1285 }
1286
1287 #endif
1288
1289 /*
1290  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1291  *      Find any that might be relevant, check against existing connections,
1292  *      forward to the right destination host if relevant.
1293  *      Currently handles error types - unreachable, quench, ttl exceeded.
1294  */
1295 static int
1296 ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum)
1297 {
1298         struct net *net = NULL;
1299         struct iphdr *iph;
1300         struct icmphdr  _icmph, *ic;
1301         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1302         struct ip_vs_iphdr ciph;
1303         struct ip_vs_conn *cp;
1304         struct ip_vs_protocol *pp;
1305         struct ip_vs_proto_data *pd;
1306         unsigned int offset, ihl, verdict;
1307
1308         *related = 1;
1309
1310         /* reassemble IP fragments */
1311         if (ip_is_fragment(ip_hdr(skb))) {
1312                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
1313                         return NF_STOLEN;
1314         }
1315
1316         iph = ip_hdr(skb);
1317         offset = ihl = iph->ihl * 4;
1318         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1319         if (ic == NULL)
1320                 return NF_DROP;
1321
1322         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1323                   ic->type, ntohs(icmp_id(ic)),
1324                   &iph->saddr, &iph->daddr);
1325
1326         /*
1327          * Work through seeing if this is for us.
1328          * These checks are supposed to be in an order that means easy
1329          * things are checked first to speed up processing.... however
1330          * this means that some packets will manage to get a long way
1331          * down this stack and then be rejected, but that's life.
1332          */
1333         if ((ic->type != ICMP_DEST_UNREACH) &&
1334             (ic->type != ICMP_SOURCE_QUENCH) &&
1335             (ic->type != ICMP_TIME_EXCEEDED)) {
1336                 *related = 0;
1337                 return NF_ACCEPT;
1338         }
1339
1340         /* Now find the contained IP header */
1341         offset += sizeof(_icmph);
1342         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1343         if (cih == NULL)
1344                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1345
1346         net = skb_net(skb);
1347
1348         pd = ip_vs_proto_data_get(net, cih->protocol);
1349         if (!pd)
1350                 return NF_ACCEPT;
1351         pp = pd->pp;
1352
1353         /* Is the embedded protocol header present? */
1354         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1355                      pp->dont_defrag))
1356                 return NF_ACCEPT;
1357
1358         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1359                       "Checking incoming ICMP for");
1360
1361         offset += cih->ihl * 4;
1362
1363         ip_vs_fill_iphdr(AF_INET, cih, &ciph);
1364         /* The embedded headers contain source and dest in reverse order */
1365         cp = pp->conn_in_get(AF_INET, skb, &ciph, offset, 1);
1366         if (!cp)
1367                 return NF_ACCEPT;
1368
1369         verdict = NF_DROP;
1370
1371         /* Ensure the checksum is correct */
1372         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1373                 /* Failed checksum! */
1374                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1375                           &iph->saddr);
1376                 goto out;
1377         }
1378
1379         /* do the statistics and put it back */
1380         ip_vs_in_stats(cp, skb);
1381         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol)
1382                 offset += 2 * sizeof(__u16);
1383         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum);
1384
1385 out:
1386         __ip_vs_conn_put(cp);
1387
1388         return verdict;
1389 }
1390
1391 #ifdef CONFIG_IP_VS_IPV6
1392 static int
1393 ip_vs_in_icmp_v6(struct sk_buff *skb, int *related, unsigned int hooknum)
1394 {
1395         struct net *net = NULL;
1396         struct ipv6hdr *iph;
1397         struct icmp6hdr _icmph, *ic;
1398         struct ipv6hdr  _ciph, *cih;    /* The ip header contained
1399                                            within the ICMP */
1400         struct ip_vs_iphdr ciph;
1401         struct ip_vs_conn *cp;
1402         struct ip_vs_protocol *pp;
1403         struct ip_vs_proto_data *pd;
1404         unsigned int offset, verdict;
1405
1406         *related = 1;
1407
1408         /* reassemble IP fragments */
1409         if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) {
1410                 if (ip_vs_gather_frags_v6(skb, ip_vs_defrag_user(hooknum)))
1411                         return NF_STOLEN;
1412         }
1413
1414         iph = ipv6_hdr(skb);
1415         offset = sizeof(struct ipv6hdr);
1416         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1417         if (ic == NULL)
1418                 return NF_DROP;
1419
1420         IP_VS_DBG(12, "Incoming ICMPv6 (%d,%d) %pI6->%pI6\n",
1421                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1422                   &iph->saddr, &iph->daddr);
1423
1424         /*
1425          * Work through seeing if this is for us.
1426          * These checks are supposed to be in an order that means easy
1427          * things are checked first to speed up processing.... however
1428          * this means that some packets will manage to get a long way
1429          * down this stack and then be rejected, but that's life.
1430          */
1431         if ((ic->icmp6_type != ICMPV6_DEST_UNREACH) &&
1432             (ic->icmp6_type != ICMPV6_PKT_TOOBIG) &&
1433             (ic->icmp6_type != ICMPV6_TIME_EXCEED)) {
1434                 *related = 0;
1435                 return NF_ACCEPT;
1436         }
1437
1438         /* Now find the contained IP header */
1439         offset += sizeof(_icmph);
1440         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1441         if (cih == NULL)
1442                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1443
1444         net = skb_net(skb);
1445         pd = ip_vs_proto_data_get(net, cih->nexthdr);
1446         if (!pd)
1447                 return NF_ACCEPT;
1448         pp = pd->pp;
1449
1450         /* Is the embedded protocol header present? */
1451         /* TODO: we don't support fragmentation at the moment anyways */
1452         if (unlikely(cih->nexthdr == IPPROTO_FRAGMENT && pp->dont_defrag))
1453                 return NF_ACCEPT;
1454
1455         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
1456                       "Checking incoming ICMPv6 for");
1457
1458         offset += sizeof(struct ipv6hdr);
1459
1460         ip_vs_fill_iphdr(AF_INET6, cih, &ciph);
1461         /* The embedded headers contain source and dest in reverse order */
1462         cp = pp->conn_in_get(AF_INET6, skb, &ciph, offset, 1);
1463         if (!cp)
1464                 return NF_ACCEPT;
1465
1466         /* do the statistics and put it back */
1467         ip_vs_in_stats(cp, skb);
1468         if (IPPROTO_TCP == cih->nexthdr || IPPROTO_UDP == cih->nexthdr ||
1469             IPPROTO_SCTP == cih->nexthdr)
1470                 offset += 2 * sizeof(__u16);
1471         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset, hooknum);
1472
1473         __ip_vs_conn_put(cp);
1474
1475         return verdict;
1476 }
1477 #endif
1478
1479
1480 /*
1481  *      Check if it's for virtual services, look it up,
1482  *      and send it on its way...
1483  */
1484 static unsigned int
1485 ip_vs_in(unsigned int hooknum, struct sk_buff *skb, int af)
1486 {
1487         struct net *net;
1488         struct ip_vs_iphdr iph;
1489         struct ip_vs_protocol *pp;
1490         struct ip_vs_proto_data *pd;
1491         struct ip_vs_conn *cp;
1492         int ret, pkts;
1493         struct netns_ipvs *ipvs;
1494
1495         /* Already marked as IPVS request or reply? */
1496         if (skb->ipvs_property)
1497                 return NF_ACCEPT;
1498
1499         /*
1500          *      Big tappo:
1501          *      - remote client: only PACKET_HOST
1502          *      - route: used for struct net when skb->dev is unset
1503          */
1504         if (unlikely((skb->pkt_type != PACKET_HOST &&
1505                       hooknum != NF_INET_LOCAL_OUT) ||
1506                      !skb_dst(skb))) {
1507                 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1508                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1509                               " ignored in hook %u\n",
1510                               skb->pkt_type, iph.protocol,
1511                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1512                 return NF_ACCEPT;
1513         }
1514         /* ipvs enabled in this netns ? */
1515         net = skb_net(skb);
1516         if (!net_ipvs(net)->enable)
1517                 return NF_ACCEPT;
1518
1519         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1520
1521         /* Bad... Do not break raw sockets */
1522         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1523                      af == AF_INET)) {
1524                 struct sock *sk = skb->sk;
1525                 struct inet_sock *inet = inet_sk(skb->sk);
1526
1527                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1528                         return NF_ACCEPT;
1529         }
1530
1531 #ifdef CONFIG_IP_VS_IPV6
1532         if (af == AF_INET6) {
1533                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1534                         int related;
1535                         int verdict = ip_vs_in_icmp_v6(skb, &related, hooknum);
1536
1537                         if (related)
1538                                 return verdict;
1539                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1540                 }
1541         } else
1542 #endif
1543                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1544                         int related;
1545                         int verdict = ip_vs_in_icmp(skb, &related, hooknum);
1546
1547                         if (related)
1548                                 return verdict;
1549                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1550                 }
1551
1552         /* Protocol supported? */
1553         pd = ip_vs_proto_data_get(net, iph.protocol);
1554         if (unlikely(!pd))
1555                 return NF_ACCEPT;
1556         pp = pd->pp;
1557         /*
1558          * Check if the packet belongs to an existing connection entry
1559          */
1560         cp = pp->conn_in_get(af, skb, &iph, iph.len, 0);
1561
1562         if (unlikely(!cp)) {
1563                 int v;
1564
1565                 if (!pp->conn_schedule(af, skb, pd, &v, &cp))
1566                         return v;
1567         }
1568
1569         if (unlikely(!cp)) {
1570                 /* sorry, all this trouble for a no-hit :) */
1571                 IP_VS_DBG_PKT(12, af, pp, skb, 0,
1572                               "ip_vs_in: packet continues traversal as normal");
1573                 return NF_ACCEPT;
1574         }
1575
1576         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Incoming packet");
1577         ipvs = net_ipvs(net);
1578         /* Check the server status */
1579         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1580                 /* the destination server is not available */
1581
1582                 if (sysctl_expire_nodest_conn(ipvs)) {
1583                         /* try to expire the connection immediately */
1584                         ip_vs_conn_expire_now(cp);
1585                 }
1586                 /* don't restart its timer, and silently
1587                    drop the packet. */
1588                 __ip_vs_conn_put(cp);
1589                 return NF_DROP;
1590         }
1591
1592         ip_vs_in_stats(cp, skb);
1593         ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1594         if (cp->packet_xmit)
1595                 ret = cp->packet_xmit(skb, cp, pp);
1596                 /* do not touch skb anymore */
1597         else {
1598                 IP_VS_DBG_RL("warning: packet_xmit is null");
1599                 ret = NF_ACCEPT;
1600         }
1601
1602         /* Increase its packet counter and check if it is needed
1603          * to be synchronized
1604          *
1605          * Sync connection if it is about to close to
1606          * encorage the standby servers to update the connections timeout
1607          *
1608          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1609          */
1610
1611         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
1612                 pkts = sysctl_sync_threshold(ipvs);
1613         else
1614                 pkts = atomic_add_return(1, &cp->in_pkts);
1615
1616         if ((ipvs->sync_state & IP_VS_STATE_MASTER) &&
1617             cp->protocol == IPPROTO_SCTP) {
1618                 if ((cp->state == IP_VS_SCTP_S_ESTABLISHED &&
1619                         (pkts % sysctl_sync_period(ipvs)
1620                          == sysctl_sync_threshold(ipvs))) ||
1621                                 (cp->old_state != cp->state &&
1622                                  ((cp->state == IP_VS_SCTP_S_CLOSED) ||
1623                                   (cp->state == IP_VS_SCTP_S_SHUT_ACK_CLI) ||
1624                                   (cp->state == IP_VS_SCTP_S_SHUT_ACK_SER)))) {
1625                         ip_vs_sync_conn(net, cp);
1626                         goto out;
1627                 }
1628         }
1629
1630         /* Keep this block last: TCP and others with pp->num_states <= 1 */
1631         else if ((ipvs->sync_state & IP_VS_STATE_MASTER) &&
1632             (((cp->protocol != IPPROTO_TCP ||
1633                cp->state == IP_VS_TCP_S_ESTABLISHED) &&
1634               (pkts % sysctl_sync_period(ipvs)
1635                == sysctl_sync_threshold(ipvs))) ||
1636              ((cp->protocol == IPPROTO_TCP) && (cp->old_state != cp->state) &&
1637               ((cp->state == IP_VS_TCP_S_FIN_WAIT) ||
1638                (cp->state == IP_VS_TCP_S_CLOSE) ||
1639                (cp->state == IP_VS_TCP_S_CLOSE_WAIT) ||
1640                (cp->state == IP_VS_TCP_S_TIME_WAIT)))))
1641                 ip_vs_sync_conn(net, cp);
1642 out:
1643         cp->old_state = cp->state;
1644
1645         ip_vs_conn_put(cp);
1646         return ret;
1647 }
1648
1649 /*
1650  *      AF_INET handler in NF_INET_LOCAL_IN chain
1651  *      Schedule and forward packets from remote clients
1652  */
1653 static unsigned int
1654 ip_vs_remote_request4(unsigned int hooknum, struct sk_buff *skb,
1655                       const struct net_device *in,
1656                       const struct net_device *out,
1657                       int (*okfn)(struct sk_buff *))
1658 {
1659         return ip_vs_in(hooknum, skb, AF_INET);
1660 }
1661
1662 /*
1663  *      AF_INET handler in NF_INET_LOCAL_OUT chain
1664  *      Schedule and forward packets from local clients
1665  */
1666 static unsigned int
1667 ip_vs_local_request4(unsigned int hooknum, struct sk_buff *skb,
1668                      const struct net_device *in, const struct net_device *out,
1669                      int (*okfn)(struct sk_buff *))
1670 {
1671         unsigned int verdict;
1672
1673         /* Disable BH in LOCAL_OUT until all places are fixed */
1674         local_bh_disable();
1675         verdict = ip_vs_in(hooknum, skb, AF_INET);
1676         local_bh_enable();
1677         return verdict;
1678 }
1679
1680 #ifdef CONFIG_IP_VS_IPV6
1681
1682 /*
1683  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
1684  *      Schedule and forward packets from remote clients
1685  */
1686 static unsigned int
1687 ip_vs_remote_request6(unsigned int hooknum, struct sk_buff *skb,
1688                       const struct net_device *in,
1689                       const struct net_device *out,
1690                       int (*okfn)(struct sk_buff *))
1691 {
1692         return ip_vs_in(hooknum, skb, AF_INET6);
1693 }
1694
1695 /*
1696  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
1697  *      Schedule and forward packets from local clients
1698  */
1699 static unsigned int
1700 ip_vs_local_request6(unsigned int hooknum, struct sk_buff *skb,
1701                      const struct net_device *in, const struct net_device *out,
1702                      int (*okfn)(struct sk_buff *))
1703 {
1704         unsigned int verdict;
1705
1706         /* Disable BH in LOCAL_OUT until all places are fixed */
1707         local_bh_disable();
1708         verdict = ip_vs_in(hooknum, skb, AF_INET6);
1709         local_bh_enable();
1710         return verdict;
1711 }
1712
1713 #endif
1714
1715
1716 /*
1717  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1718  *      related packets destined for 0.0.0.0/0.
1719  *      When fwmark-based virtual service is used, such as transparent
1720  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
1721  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1722  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1723  *      and send them to ip_vs_in_icmp.
1724  */
1725 static unsigned int
1726 ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff *skb,
1727                    const struct net_device *in, const struct net_device *out,
1728                    int (*okfn)(struct sk_buff *))
1729 {
1730         int r;
1731         struct net *net;
1732
1733         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1734                 return NF_ACCEPT;
1735
1736         /* ipvs enabled in this netns ? */
1737         net = skb_net(skb);
1738         if (!net_ipvs(net)->enable)
1739                 return NF_ACCEPT;
1740
1741         return ip_vs_in_icmp(skb, &r, hooknum);
1742 }
1743
1744 #ifdef CONFIG_IP_VS_IPV6
1745 static unsigned int
1746 ip_vs_forward_icmp_v6(unsigned int hooknum, struct sk_buff *skb,
1747                       const struct net_device *in, const struct net_device *out,
1748                       int (*okfn)(struct sk_buff *))
1749 {
1750         int r;
1751         struct net *net;
1752
1753         if (ipv6_hdr(skb)->nexthdr != IPPROTO_ICMPV6)
1754                 return NF_ACCEPT;
1755
1756         /* ipvs enabled in this netns ? */
1757         net = skb_net(skb);
1758         if (!net_ipvs(net)->enable)
1759                 return NF_ACCEPT;
1760
1761         return ip_vs_in_icmp_v6(skb, &r, hooknum);
1762 }
1763 #endif
1764
1765
1766 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
1767         /* After packet filtering, change source only for VS/NAT */
1768         {
1769                 .hook           = ip_vs_reply4,
1770                 .owner          = THIS_MODULE,
1771                 .pf             = PF_INET,
1772                 .hooknum        = NF_INET_LOCAL_IN,
1773                 .priority       = NF_IP_PRI_NAT_SRC - 2,
1774         },
1775         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1776          * or VS/NAT(change destination), so that filtering rules can be
1777          * applied to IPVS. */
1778         {
1779                 .hook           = ip_vs_remote_request4,
1780                 .owner          = THIS_MODULE,
1781                 .pf             = PF_INET,
1782                 .hooknum        = NF_INET_LOCAL_IN,
1783                 .priority       = NF_IP_PRI_NAT_SRC - 1,
1784         },
1785         /* Before ip_vs_in, change source only for VS/NAT */
1786         {
1787                 .hook           = ip_vs_local_reply4,
1788                 .owner          = THIS_MODULE,
1789                 .pf             = PF_INET,
1790                 .hooknum        = NF_INET_LOCAL_OUT,
1791                 .priority       = NF_IP_PRI_NAT_DST + 1,
1792         },
1793         /* After mangle, schedule and forward local requests */
1794         {
1795                 .hook           = ip_vs_local_request4,
1796                 .owner          = THIS_MODULE,
1797                 .pf             = PF_INET,
1798                 .hooknum        = NF_INET_LOCAL_OUT,
1799                 .priority       = NF_IP_PRI_NAT_DST + 2,
1800         },
1801         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1802          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1803         {
1804                 .hook           = ip_vs_forward_icmp,
1805                 .owner          = THIS_MODULE,
1806                 .pf             = PF_INET,
1807                 .hooknum        = NF_INET_FORWARD,
1808                 .priority       = 99,
1809         },
1810         /* After packet filtering, change source only for VS/NAT */
1811         {
1812                 .hook           = ip_vs_reply4,
1813                 .owner          = THIS_MODULE,
1814                 .pf             = PF_INET,
1815                 .hooknum        = NF_INET_FORWARD,
1816                 .priority       = 100,
1817         },
1818 #ifdef CONFIG_IP_VS_IPV6
1819         /* After packet filtering, change source only for VS/NAT */
1820         {
1821                 .hook           = ip_vs_reply6,
1822                 .owner          = THIS_MODULE,
1823                 .pf             = PF_INET6,
1824                 .hooknum        = NF_INET_LOCAL_IN,
1825                 .priority       = NF_IP6_PRI_NAT_SRC - 2,
1826         },
1827         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1828          * or VS/NAT(change destination), so that filtering rules can be
1829          * applied to IPVS. */
1830         {
1831                 .hook           = ip_vs_remote_request6,
1832                 .owner          = THIS_MODULE,
1833                 .pf             = PF_INET6,
1834                 .hooknum        = NF_INET_LOCAL_IN,
1835                 .priority       = NF_IP6_PRI_NAT_SRC - 1,
1836         },
1837         /* Before ip_vs_in, change source only for VS/NAT */
1838         {
1839                 .hook           = ip_vs_local_reply6,
1840                 .owner          = THIS_MODULE,
1841                 .pf             = PF_INET,
1842                 .hooknum        = NF_INET_LOCAL_OUT,
1843                 .priority       = NF_IP6_PRI_NAT_DST + 1,
1844         },
1845         /* After mangle, schedule and forward local requests */
1846         {
1847                 .hook           = ip_vs_local_request6,
1848                 .owner          = THIS_MODULE,
1849                 .pf             = PF_INET6,
1850                 .hooknum        = NF_INET_LOCAL_OUT,
1851                 .priority       = NF_IP6_PRI_NAT_DST + 2,
1852         },
1853         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1854          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1855         {
1856                 .hook           = ip_vs_forward_icmp_v6,
1857                 .owner          = THIS_MODULE,
1858                 .pf             = PF_INET6,
1859                 .hooknum        = NF_INET_FORWARD,
1860                 .priority       = 99,
1861         },
1862         /* After packet filtering, change source only for VS/NAT */
1863         {
1864                 .hook           = ip_vs_reply6,
1865                 .owner          = THIS_MODULE,
1866                 .pf             = PF_INET6,
1867                 .hooknum        = NF_INET_FORWARD,
1868                 .priority       = 100,
1869         },
1870 #endif
1871 };
1872 /*
1873  *      Initialize IP Virtual Server netns mem.
1874  */
1875 static int __net_init __ip_vs_init(struct net *net)
1876 {
1877         struct netns_ipvs *ipvs;
1878
1879         ipvs = net_generic(net, ip_vs_net_id);
1880         if (ipvs == NULL)
1881                 return -ENOMEM;
1882
1883         /* Hold the beast until a service is registerd */
1884         ipvs->enable = 0;
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
1891         if (ip_vs_estimator_net_init(net) < 0)
1892                 goto estimator_fail;
1893
1894         if (ip_vs_control_net_init(net) < 0)
1895                 goto control_fail;
1896
1897         if (ip_vs_protocol_net_init(net) < 0)
1898                 goto protocol_fail;
1899
1900         if (ip_vs_app_net_init(net) < 0)
1901                 goto app_fail;
1902
1903         if (ip_vs_conn_net_init(net) < 0)
1904                 goto conn_fail;
1905
1906         if (ip_vs_sync_net_init(net) < 0)
1907                 goto sync_fail;
1908
1909         printk(KERN_INFO "IPVS: Creating netns size=%zu id=%d\n",
1910                          sizeof(struct netns_ipvs), ipvs->gen);
1911         return 0;
1912 /*
1913  * Error handling
1914  */
1915
1916 sync_fail:
1917         ip_vs_conn_net_cleanup(net);
1918 conn_fail:
1919         ip_vs_app_net_cleanup(net);
1920 app_fail:
1921         ip_vs_protocol_net_cleanup(net);
1922 protocol_fail:
1923         ip_vs_control_net_cleanup(net);
1924 control_fail:
1925         ip_vs_estimator_net_cleanup(net);
1926 estimator_fail:
1927         return -ENOMEM;
1928 }
1929
1930 static void __net_exit __ip_vs_cleanup(struct net *net)
1931 {
1932         ip_vs_service_net_cleanup(net); /* ip_vs_flush() with locks */
1933         ip_vs_conn_net_cleanup(net);
1934         ip_vs_app_net_cleanup(net);
1935         ip_vs_protocol_net_cleanup(net);
1936         ip_vs_control_net_cleanup(net);
1937         ip_vs_estimator_net_cleanup(net);
1938         IP_VS_DBG(2, "ipvs netns %d released\n", net_ipvs(net)->gen);
1939 }
1940
1941 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
1942 {
1943         EnterFunction(2);
1944         net_ipvs(net)->enable = 0;      /* Disable packet reception */
1945         smp_wmb();
1946         ip_vs_sync_net_cleanup(net);
1947         LeaveFunction(2);
1948 }
1949
1950 static struct pernet_operations ipvs_core_ops = {
1951         .init = __ip_vs_init,
1952         .exit = __ip_vs_cleanup,
1953         .id   = &ip_vs_net_id,
1954         .size = sizeof(struct netns_ipvs),
1955 };
1956
1957 static struct pernet_operations ipvs_core_dev_ops = {
1958         .exit = __ip_vs_dev_cleanup,
1959 };
1960
1961 /*
1962  *      Initialize IP Virtual Server
1963  */
1964 static int __init ip_vs_init(void)
1965 {
1966         int ret;
1967
1968         ret = ip_vs_control_init();
1969         if (ret < 0) {
1970                 pr_err("can't setup control.\n");
1971                 goto exit;
1972         }
1973
1974         ip_vs_protocol_init();
1975
1976         ret = ip_vs_conn_init();
1977         if (ret < 0) {
1978                 pr_err("can't setup connection table.\n");
1979                 goto cleanup_protocol;
1980         }
1981
1982         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
1983         if (ret < 0)
1984                 goto cleanup_conn;
1985
1986         ret = register_pernet_device(&ipvs_core_dev_ops);
1987         if (ret < 0)
1988                 goto cleanup_sub;
1989
1990         ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
1991         if (ret < 0) {
1992                 pr_err("can't register hooks.\n");
1993                 goto cleanup_dev;
1994         }
1995
1996         pr_info("ipvs loaded.\n");
1997
1998         return ret;
1999
2000 cleanup_dev:
2001         unregister_pernet_device(&ipvs_core_dev_ops);
2002 cleanup_sub:
2003         unregister_pernet_subsys(&ipvs_core_ops);
2004 cleanup_conn:
2005         ip_vs_conn_cleanup();
2006 cleanup_protocol:
2007         ip_vs_protocol_cleanup();
2008         ip_vs_control_cleanup();
2009 exit:
2010         return ret;
2011 }
2012
2013 static void __exit ip_vs_cleanup(void)
2014 {
2015         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2016         unregister_pernet_device(&ipvs_core_dev_ops);
2017         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
2018         ip_vs_conn_cleanup();
2019         ip_vs_protocol_cleanup();
2020         ip_vs_control_cleanup();
2021         pr_info("ipvs unloaded.\n");
2022 }
2023
2024 module_init(ip_vs_init);
2025 module_exit(ip_vs_cleanup);
2026 MODULE_LICENSE("GPL");