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 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         __be16 _ports[2], *pptr;
 503         struct ip_vs_iphdr iph;
 504 #ifdef CONFIG_SYSCTL
 505         struct net *net;
 506         struct netns_ipvs *ipvs;
 507         int unicast;
 508 #endif
 509
 510         ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph);
 511
 512         pptr = skb_header_pointer(skb, iph.len, sizeof(_ports), _ports);
 513         if (pptr == NULL) {
 514                 ip_vs_service_put(svc);
 515                 return NF_DROP;
 516         }
 517
 518 #ifdef CONFIG_SYSCTL
 519         net = skb_net(skb);
 520
 521 #ifdef CONFIG_IP_VS_IPV6
 522         if (svc->af == AF_INET6)
 523                 unicast = ipv6_addr_type(&iph.daddr.in6) & IPV6_ADDR_UNICAST;
 524         else
 525 #endif
 526                 unicast = (inet_addr_type(net, iph.daddr.ip) == RTN_UNICAST);
 527
 528         /* if it is fwmark-based service, the cache_bypass sysctl is up
 529            and the destination is a non-local unicast, then create
 530            a cache_bypass connection entry */
 531         ipvs = net_ipvs(net);
 532         if (ipvs->sysctl_cache_bypass && svc->fwmark && unicast) {
 533                 int ret, cs;
 534                 struct ip_vs_conn *cp;
 535                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
 536                                       iph.protocol == IPPROTO_UDP)?
 537                                       IP_VS_CONN_F_ONE_PACKET : 0;
 538                 union nf_inet_addr daddr =  { .all = { 0, 0, 0, 0 } };
 539
 540                 ip_vs_service_put(svc);
 541
 542                 /* create a new connection entry */
 543                 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
 544                 {
 545                         struct ip_vs_conn_param p;
 546                         ip_vs_conn_fill_param(svc->net, svc->af, iph.protocol,
 547                                               &iph.saddr, pptr[0],
 548                                               &iph.daddr, pptr[1], &p);
 549                         cp = ip_vs_conn_new(&p, &daddr, 0,
 550                                             IP_VS_CONN_F_BYPASS | flags,
 551                                             NULL, skb->mark);
 552                         if (!cp)
 553                                 return NF_DROP;
 554                 }
 555
 556                 /* statistics */
 557                 ip_vs_in_stats(cp, skb);
 558
 559                 /* set state */
 560                 cs = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
 561
 562                 /* transmit the first SYN packet */
 563                 ret = cp->packet_xmit(skb, cp, pd->pp);
 564                 /* do not touch skb anymore */
 565
 566                 atomic_inc(&cp->in_pkts);
 567                 ip_vs_conn_put(cp);
 568                 return ret;
 569         }
 570 #endif
 571
 572         /*
 573          * When the virtual ftp service is presented, packets destined
 574          * for other services on the VIP may get here (except services
 575          * listed in the ipvs table), pass the packets, because it is
 576          * not ipvs job to decide to drop the packets.
 577          */
 578         if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT)) {
 579                 ip_vs_service_put(svc);
 580                 return NF_ACCEPT;
 581         }
 582
 583         ip_vs_service_put(svc);
 584
 585         /*
 586          * Notify the client that the destination is unreachable, and
 587          * release the socket buffer.
 588          * Since it is in IP layer, the TCP socket is not actually
 589          * created, the TCP RST packet cannot be sent, instead that
 590          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 591          */
 592 #ifdef CONFIG_IP_VS_IPV6
 593         if (svc->af == AF_INET6) {
 594                 if (!skb->dev) {
 595                         struct net *net = dev_net(skb_dst(skb)->dev);
 596
 597                         skb->dev = net->loopback_dev;
 598                 }
 599                 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 600         } else
 601 #endif
 602                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 603
 604         return NF_DROP;
 605 }
 606
 607 #ifdef CONFIG_SYSCTL
 608
 609 static int sysctl_snat_reroute(struct sk_buff *skb)
 610 {
 611         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 612         return ipvs->sysctl_snat_reroute;
 613 }
 614
 615 static int sysctl_nat_icmp_send(struct net *net)
 616 {
 617         struct netns_ipvs *ipvs = net_ipvs(net);
 618         return ipvs->sysctl_nat_icmp_send;
 619 }
 620
 621 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
 622 {
 623         return ipvs->sysctl_expire_nodest_conn;
 624 }
 625
 626 #else
 627
 628 static int sysctl_snat_reroute(struct sk_buff *skb) { return 0; }
 629 static int sysctl_nat_icmp_send(struct net *net) { return 0; }
 630 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
 631
 632 #endif
 633
 634 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 635 {
 636         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 637 }
 638
 639 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
 640 {
 641         if (NF_INET_LOCAL_IN == hooknum)
 642                 return IP_DEFRAG_VS_IN;
 643         if (NF_INET_FORWARD == hooknum)
 644                 return IP_DEFRAG_VS_FWD;
 645         return IP_DEFRAG_VS_OUT;
 646 }
 647
 648 static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user)
 649 {
 650         int err = ip_defrag(skb, user);
 651
 652         if (!err)
 653                 ip_send_check(ip_hdr(skb));
 654
 655         return err;
 656 }
 657
 658 #ifdef CONFIG_IP_VS_IPV6
 659 static inline int ip_vs_gather_frags_v6(struct sk_buff *skb, u_int32_t user)
 660 {
 661         /* TODO IPv6: Find out what to do here for IPv6 */
 662         return 0;
 663 }
 664 #endif
 665
 666 static int ip_vs_route_me_harder(int af, struct sk_buff *skb)
 667 {
 668 #ifdef CONFIG_IP_VS_IPV6
 669         if (af == AF_INET6) {
 670                 if (sysctl_snat_reroute(skb) && ip6_route_me_harder(skb) != 0)
 671                         return 1;
 672         } else
 673 #endif
 674                 if ((sysctl_snat_reroute(skb) ||
 675                      skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
 676                     ip_route_me_harder(skb, RTN_LOCAL) != 0)
 677                         return 1;
 678
 679         return 0;
 680 }
 681
 682 /*
 683  * Packet has been made sufficiently writable in caller
 684  * - inout: 1=in->out, 0=out->in
 685  */
 686 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 687                     struct ip_vs_conn *cp, int inout)
 688 {
 689         struct iphdr *iph        = ip_hdr(skb);
 690         unsigned int icmp_offset = iph->ihl*4;
 691         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
 692                                                       icmp_offset);
 693         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 694
 695         if (inout) {
 696                 iph->saddr = cp->vaddr.ip;
 697                 ip_send_check(iph);
 698                 ciph->daddr = cp->vaddr.ip;
 699                 ip_send_check(ciph);
 700         } else {
 701                 iph->daddr = cp->daddr.ip;
 702                 ip_send_check(iph);
 703                 ciph->saddr = cp->daddr.ip;
 704                 ip_send_check(ciph);
 705         }
 706
 707         /* the TCP/UDP/SCTP port */
 708         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
 709             IPPROTO_SCTP == ciph->protocol) {
 710                 __be16 *ports = (void *)ciph + ciph->ihl*4;
 711
 712                 if (inout)
 713                         ports[1] = cp->vport;
 714                 else
 715                         ports[0] = cp->dport;
 716         }
 717
 718         /* And finally the ICMP checksum */
 719         icmph->checksum = 0;
 720         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 721         skb->ip_summed = CHECKSUM_UNNECESSARY;
 722
 723         if (inout)
 724                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 725                         "Forwarding altered outgoing ICMP");
 726         else
 727                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 728                         "Forwarding altered incoming ICMP");
 729 }
 730
 731 #ifdef CONFIG_IP_VS_IPV6
 732 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
 733                     struct ip_vs_conn *cp, int inout)
 734 {
 735         struct ipv6hdr *iph      = ipv6_hdr(skb);
 736         unsigned int icmp_offset = sizeof(struct ipv6hdr);
 737         struct icmp6hdr *icmph   = (struct icmp6hdr *)(skb_network_header(skb) +
 738                                                       icmp_offset);
 739         struct ipv6hdr *ciph     = (struct ipv6hdr *)(icmph + 1);
 740
 741         if (inout) {
 742                 iph->saddr = cp->vaddr.in6;
 743                 ciph->daddr = cp->vaddr.in6;
 744         } else {
 745                 iph->daddr = cp->daddr.in6;
 746                 ciph->saddr = cp->daddr.in6;
 747         }
 748
 749         /* the TCP/UDP/SCTP port */
 750         if (IPPROTO_TCP == ciph->nexthdr || IPPROTO_UDP == ciph->nexthdr ||
 751             IPPROTO_SCTP == ciph->nexthdr) {
 752                 __be16 *ports = (void *)ciph + sizeof(struct ipv6hdr);
 753
 754                 if (inout)
 755                         ports[1] = cp->vport;
 756                 else
 757                         ports[0] = cp->dport;
 758         }
 759
 760         /* And finally the ICMP checksum */
 761         icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
 762                                               skb->len - icmp_offset,
 763                                               IPPROTO_ICMPV6, 0);
 764         skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
 765         skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
 766         skb->ip_summed = CHECKSUM_PARTIAL;
 767
 768         if (inout)
 769                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 770                               (void *)ciph - (void *)iph,
 771                               "Forwarding altered outgoing ICMPv6");
 772         else
 773                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 774                               (void *)ciph - (void *)iph,
 775                               "Forwarding altered incoming ICMPv6");
 776 }
 777 #endif
 778
 779 /* Handle relevant response ICMP messages - forward to the right
 780  * destination host.
 781  */
 782 static int handle_response_icmp(int af, struct sk_buff *skb,
 783                                 union nf_inet_addr *snet,
 784                                 __u8 protocol, struct ip_vs_conn *cp,
 785                                 struct ip_vs_protocol *pp,
 786                                 unsigned int offset, unsigned int ihl)
 787 {
 788         unsigned int verdict = NF_DROP;
 789
 790         if (IP_VS_FWD_METHOD(cp) != 0) {
 791                 pr_err("shouldn't reach here, because the box is on the "
 792                        "half connection in the tun/dr module.\n");
 793         }
 794
 795         /* Ensure the checksum is correct */
 796         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 797                 /* Failed checksum! */
 798                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
 799                               IP_VS_DBG_ADDR(af, snet));
 800                 goto out;
 801         }
 802
 803         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 804             IPPROTO_SCTP == protocol)
 805                 offset += 2 * sizeof(__u16);
 806         if (!skb_make_writable(skb, offset))
 807                 goto out;
 808
 809 #ifdef CONFIG_IP_VS_IPV6
 810         if (af == AF_INET6)
 811                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
 812         else
 813 #endif
 814                 ip_vs_nat_icmp(skb, pp, cp, 1);
 815
 816         if (ip_vs_route_me_harder(af, skb))
 817                 goto out;
 818
 819         /* do the statistics and put it back */
 820         ip_vs_out_stats(cp, skb);
 821
 822         skb->ipvs_property = 1;
 823         if (!(cp->flags & IP_VS_CONN_F_NFCT))
 824                 ip_vs_notrack(skb);
 825         else
 826                 ip_vs_update_conntrack(skb, cp, 0);
 827         verdict = NF_ACCEPT;
 828
 829 out:
 830         __ip_vs_conn_put(cp);
 831
 832         return verdict;
 833 }
 834
 835 /*
 836  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 837  *      Find any that might be relevant, check against existing connections.
 838  *      Currently handles error types - unreachable, quench, ttl exceeded.
 839  */
 840 static int ip_vs_out_icmp(struct sk_buff *skb, int *related,
 841                           unsigned int hooknum)
 842 {
 843         struct iphdr *iph;
 844         struct icmphdr  _icmph, *ic;
 845         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 846         struct ip_vs_iphdr ciph;
 847         struct ip_vs_conn *cp;
 848         struct ip_vs_protocol *pp;
 849         unsigned int offset, ihl;
 850         union nf_inet_addr snet;
 851
 852         *related = 1;
 853
 854         /* reassemble IP fragments */
 855         if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
 856                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
 857                         return NF_STOLEN;
 858         }
 859
 860         iph = ip_hdr(skb);
 861         offset = ihl = iph->ihl * 4;
 862         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 863         if (ic == NULL)
 864                 return NF_DROP;
 865
 866         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
 867                   ic->type, ntohs(icmp_id(ic)),
 868                   &iph->saddr, &iph->daddr);
 869
 870         /*
 871          * Work through seeing if this is for us.
 872          * These checks are supposed to be in an order that means easy
 873          * things are checked first to speed up processing.... however
 874          * this means that some packets will manage to get a long way
 875          * down this stack and then be rejected, but that's life.
 876          */
 877         if ((ic->type != ICMP_DEST_UNREACH) &&
 878             (ic->type != ICMP_SOURCE_QUENCH) &&
 879             (ic->type != ICMP_TIME_EXCEEDED)) {
 880                 *related = 0;
 881                 return NF_ACCEPT;
 882         }
 883
 884         /* Now find the contained IP header */
 885         offset += sizeof(_icmph);
 886         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 887         if (cih == NULL)
 888                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 889
 890         pp = ip_vs_proto_get(cih->protocol);
 891         if (!pp)
 892                 return NF_ACCEPT;
 893
 894         /* Is the embedded protocol header present? */
 895         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 896                      pp->dont_defrag))
 897                 return NF_ACCEPT;
 898
 899         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
 900                       "Checking outgoing ICMP for");
 901
 902         offset += cih->ihl * 4;
 903
 904         ip_vs_fill_iphdr(AF_INET, cih, &ciph);
 905         /* The embedded headers contain source and dest in reverse order */
 906         cp = pp->conn_out_get(AF_INET, skb, &ciph, offset, 1);
 907         if (!cp)
 908                 return NF_ACCEPT;
 909
 910         snet.ip = iph->saddr;
 911         return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
 912                                     pp, offset, ihl);
 913 }
 914
 915 #ifdef CONFIG_IP_VS_IPV6
 916 static int ip_vs_out_icmp_v6(struct sk_buff *skb, int *related,
 917                              unsigned int hooknum)
 918 {
 919         struct ipv6hdr *iph;
 920         struct icmp6hdr _icmph, *ic;
 921         struct ipv6hdr  _ciph, *cih;    /* The ip header contained
 922                                            within the ICMP */
 923         struct ip_vs_iphdr ciph;
 924         struct ip_vs_conn *cp;
 925         struct ip_vs_protocol *pp;
 926         unsigned int offset;
 927         union nf_inet_addr snet;
 928
 929         *related = 1;
 930
 931         /* reassemble IP fragments */
 932         if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) {
 933                 if (ip_vs_gather_frags_v6(skb, ip_vs_defrag_user(hooknum)))
 934                         return NF_STOLEN;
 935         }
 936
 937         iph = ipv6_hdr(skb);
 938         offset = sizeof(struct ipv6hdr);
 939         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 940         if (ic == NULL)
 941                 return NF_DROP;
 942
 943         IP_VS_DBG(12, "Outgoing ICMPv6 (%d,%d) %pI6->%pI6\n",
 944                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
 945                   &iph->saddr, &iph->daddr);
 946
 947         /*
 948          * Work through seeing if this is for us.
 949          * These checks are supposed to be in an order that means easy
 950          * things are checked first to speed up processing.... however
 951          * this means that some packets will manage to get a long way
 952          * down this stack and then be rejected, but that's life.
 953          */
 954         if ((ic->icmp6_type != ICMPV6_DEST_UNREACH) &&
 955             (ic->icmp6_type != ICMPV6_PKT_TOOBIG) &&
 956             (ic->icmp6_type != ICMPV6_TIME_EXCEED)) {
 957                 *related = 0;
 958                 return NF_ACCEPT;
 959         }
 960
 961         /* Now find the contained IP header */
 962         offset += sizeof(_icmph);
 963         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 964         if (cih == NULL)
 965                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 966
 967         pp = ip_vs_proto_get(cih->nexthdr);
 968         if (!pp)
 969                 return NF_ACCEPT;
 970
 971         /* Is the embedded protocol header present? */
 972         /* TODO: we don't support fragmentation at the moment anyways */
 973         if (unlikely(cih->nexthdr == IPPROTO_FRAGMENT && pp->dont_defrag))
 974                 return NF_ACCEPT;
 975
 976         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
 977                       "Checking outgoing ICMPv6 for");
 978
 979         offset += sizeof(struct ipv6hdr);
 980
 981         ip_vs_fill_iphdr(AF_INET6, cih, &ciph);
 982         /* The embedded headers contain source and dest in reverse order */
 983         cp = pp->conn_out_get(AF_INET6, skb, &ciph, offset, 1);
 984         if (!cp)
 985                 return NF_ACCEPT;
 986
 987         ipv6_addr_copy(&snet.in6, &iph->saddr);
 988         return handle_response_icmp(AF_INET6, skb, &snet, cih->nexthdr, cp,
 989                                     pp, offset, sizeof(struct ipv6hdr));
 990 }
 991 #endif
 992
 993 /*
 994  * Check if sctp chunc is ABORT chunk
 995  */
 996 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
 997 {
 998         sctp_chunkhdr_t *sch, schunk;
 999         sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
1000                         sizeof(schunk), &schunk);
1001         if (sch == NULL)
1002                 return 0;
1003         if (sch->type == SCTP_CID_ABORT)
1004                 return 1;
1005         return 0;
1006 }
1007
1008 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
1009 {
1010         struct tcphdr _tcph, *th;
1011
1012         th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
1013         if (th == NULL)
1014                 return 0;
1015         return th->rst;
1016 }
1017
1018 /* Handle response packets: rewrite addresses and send away...
1019  */
1020 static unsigned int
1021 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1022                 struct ip_vs_conn *cp, int ihl)
1023 {
1024         struct ip_vs_protocol *pp = pd->pp;
1025
1026         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Outgoing packet");
1027
1028         if (!skb_make_writable(skb, ihl))
1029                 goto drop;
1030
1031         /* mangle the packet */
1032         if (pp->snat_handler && !pp->snat_handler(skb, pp, cp))
1033                 goto drop;
1034
1035 #ifdef CONFIG_IP_VS_IPV6
1036         if (af == AF_INET6)
1037                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1038         else
1039 #endif
1040         {
1041                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1042                 ip_send_check(ip_hdr(skb));
1043         }
1044
1045         /*
1046          * nf_iterate does not expect change in the skb->dst->dev.
1047          * It looks like it is not fatal to enable this code for hooks
1048          * where our handlers are at the end of the chain list and
1049          * when all next handlers use skb->dst->dev and not outdev.
1050          * It will definitely route properly the inout NAT traffic
1051          * when multiple paths are used.
1052          */
1053
1054         /* For policy routing, packets originating from this
1055          * machine itself may be routed differently to packets
1056          * passing through.  We want this packet to be routed as
1057          * if it came from this machine itself.  So re-compute
1058          * the routing information.
1059          */
1060         if (ip_vs_route_me_harder(af, skb))
1061                 goto drop;
1062
1063         IP_VS_DBG_PKT(10, af, pp, skb, 0, "After SNAT");
1064
1065         ip_vs_out_stats(cp, skb);
1066         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1067         skb->ipvs_property = 1;
1068         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1069                 ip_vs_notrack(skb);
1070         else
1071                 ip_vs_update_conntrack(skb, cp, 0);
1072         ip_vs_conn_put(cp);
1073
1074         LeaveFunction(11);
1075         return NF_ACCEPT;
1076
1077 drop:
1078         ip_vs_conn_put(cp);
1079         kfree_skb(skb);
1080         LeaveFunction(11);
1081         return NF_STOLEN;
1082 }
1083
1084 /*
1085  *      Check if outgoing packet belongs to the established ip_vs_conn.
1086  */
1087 static unsigned int
1088 ip_vs_out(unsigned int hooknum, struct sk_buff *skb, int af)
1089 {
1090         struct net *net = NULL;
1091         struct ip_vs_iphdr iph;
1092         struct ip_vs_protocol *pp;
1093         struct ip_vs_proto_data *pd;
1094         struct ip_vs_conn *cp;
1095
1096         EnterFunction(11);
1097
1098         /* Already marked as IPVS request or reply? */
1099         if (skb->ipvs_property)
1100                 return NF_ACCEPT;
1101
1102         /* Bad... Do not break raw sockets */
1103         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1104                      af == AF_INET)) {
1105                 struct sock *sk = skb->sk;
1106                 struct inet_sock *inet = inet_sk(skb->sk);
1107
1108                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1109                         return NF_ACCEPT;
1110         }
1111
1112         if (unlikely(!skb_dst(skb)))
1113                 return NF_ACCEPT;
1114
1115         net = skb_net(skb);
1116         if (!net_ipvs(net)->enable)
1117                 return NF_ACCEPT;
1118
1119         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1120 #ifdef CONFIG_IP_VS_IPV6
1121         if (af == AF_INET6) {
1122                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1123                         int related;
1124                         int verdict = ip_vs_out_icmp_v6(skb, &related,
1125                                                         hooknum);
1126
1127                         if (related)
1128                                 return verdict;
1129                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1130                 }
1131         } else
1132 #endif
1133                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1134                         int related;
1135                         int verdict = ip_vs_out_icmp(skb, &related, hooknum);
1136
1137                         if (related)
1138                                 return verdict;
1139                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1140                 }
1141
1142         pd = ip_vs_proto_data_get(net, iph.protocol);
1143         if (unlikely(!pd))
1144                 return NF_ACCEPT;
1145         pp = pd->pp;
1146
1147         /* reassemble IP fragments */
1148 #ifdef CONFIG_IP_VS_IPV6
1149         if (af == AF_INET6) {
1150                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) {
1151                         if (ip_vs_gather_frags_v6(skb,
1152                                                   ip_vs_defrag_user(hooknum)))
1153                                 return NF_STOLEN;
1154                 }
1155
1156                 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1157         } else
1158 #endif
1159                 if (unlikely(ip_hdr(skb)->frag_off & htons(IP_MF|IP_OFFSET) &&
1160                              !pp->dont_defrag)) {
1161                         if (ip_vs_gather_frags(skb,
1162                                                ip_vs_defrag_user(hooknum)))
1163                                 return NF_STOLEN;
1164
1165                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1166                 }
1167
1168         /*
1169          * Check if the packet belongs to an existing entry
1170          */
1171         cp = pp->conn_out_get(af, skb, &iph, iph.len, 0);
1172
1173         if (likely(cp))
1174                 return handle_response(af, skb, pd, cp, iph.len);
1175         if (sysctl_nat_icmp_send(net) &&
1176             (pp->protocol == IPPROTO_TCP ||
1177              pp->protocol == IPPROTO_UDP ||
1178              pp->protocol == IPPROTO_SCTP)) {
1179                 __be16 _ports[2], *pptr;
1180
1181                 pptr = skb_header_pointer(skb, iph.len,
1182                                           sizeof(_ports), _ports);
1183                 if (pptr == NULL)
1184                         return NF_ACCEPT;       /* Not for me */
1185                 if (ip_vs_lookup_real_service(net, af, iph.protocol,
1186                                               &iph.saddr,
1187                                               pptr[0])) {
1188                         /*
1189                          * Notify the real server: there is no
1190                          * existing entry if it is not RST
1191                          * packet or not TCP packet.
1192                          */
1193                         if ((iph.protocol != IPPROTO_TCP &&
1194                              iph.protocol != IPPROTO_SCTP)
1195                              || ((iph.protocol == IPPROTO_TCP
1196                                   && !is_tcp_reset(skb, iph.len))
1197                                  || (iph.protocol == IPPROTO_SCTP
1198                                         && !is_sctp_abort(skb,
1199                                                 iph.len)))) {
1200 #ifdef CONFIG_IP_VS_IPV6
1201                                 if (af == AF_INET6) {
1202                                         struct net *net =
1203                                                 dev_net(skb_dst(skb)->dev);
1204
1205                                         if (!skb->dev)
1206                                                 skb->dev = net->loopback_dev;
1207                                         icmpv6_send(skb,
1208                                                     ICMPV6_DEST_UNREACH,
1209                                                     ICMPV6_PORT_UNREACH,
1210                                                     0);
1211                                 } else
1212 #endif
1213                                         icmp_send(skb,
1214                                                   ICMP_DEST_UNREACH,
1215                                                   ICMP_PORT_UNREACH, 0);
1216                                 return NF_DROP;
1217                         }
1218                 }
1219         }
1220         IP_VS_DBG_PKT(12, af, pp, skb, 0,
1221                       "ip_vs_out: packet continues traversal as normal");
1222         return NF_ACCEPT;
1223 }
1224
1225 /*
1226  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1227  *      used only for VS/NAT.
1228  *      Check if packet is reply for established ip_vs_conn.
1229  */
1230 static unsigned int
1231 ip_vs_reply4(unsigned int hooknum, struct sk_buff *skb,
1232              const struct net_device *in, const struct net_device *out,
1233              int (*okfn)(struct sk_buff *))
1234 {
1235         return ip_vs_out(hooknum, skb, AF_INET);
1236 }
1237
1238 /*
1239  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1240  *      Check if packet is reply for established ip_vs_conn.
1241  */
1242 static unsigned int
1243 ip_vs_local_reply4(unsigned int hooknum, struct sk_buff *skb,
1244                    const struct net_device *in, const struct net_device *out,
1245                    int (*okfn)(struct sk_buff *))
1246 {
1247         unsigned int verdict;
1248
1249         /* Disable BH in LOCAL_OUT until all places are fixed */
1250         local_bh_disable();
1251         verdict = ip_vs_out(hooknum, skb, AF_INET);
1252         local_bh_enable();
1253         return verdict;
1254 }
1255
1256 #ifdef CONFIG_IP_VS_IPV6
1257
1258 /*
1259  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1260  *      used only for VS/NAT.
1261  *      Check if packet is reply for established ip_vs_conn.
1262  */
1263 static unsigned int
1264 ip_vs_reply6(unsigned int hooknum, struct sk_buff *skb,
1265              const struct net_device *in, const struct net_device *out,
1266              int (*okfn)(struct sk_buff *))
1267 {
1268         return ip_vs_out(hooknum, skb, AF_INET6);
1269 }
1270
1271 /*
1272  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1273  *      Check if packet is reply for established ip_vs_conn.
1274  */
1275 static unsigned int
1276 ip_vs_local_reply6(unsigned int hooknum, struct sk_buff *skb,
1277                    const struct net_device *in, const struct net_device *out,
1278                    int (*okfn)(struct sk_buff *))
1279 {
1280         unsigned int verdict;
1281
1282         /* Disable BH in LOCAL_OUT until all places are fixed */
1283         local_bh_disable();
1284         verdict = ip_vs_out(hooknum, skb, AF_INET6);
1285         local_bh_enable();
1286         return verdict;
1287 }
1288
1289 #endif
1290
1291 /*
1292  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1293  *      Find any that might be relevant, check against existing connections,
1294  *      forward to the right destination host if relevant.
1295  *      Currently handles error types - unreachable, quench, ttl exceeded.
1296  */
1297 static int
1298 ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum)
1299 {
1300         struct net *net = NULL;
1301         struct iphdr *iph;
1302         struct icmphdr  _icmph, *ic;
1303         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1304         struct ip_vs_iphdr ciph;
1305         struct ip_vs_conn *cp;
1306         struct ip_vs_protocol *pp;
1307         struct ip_vs_proto_data *pd;
1308         unsigned int offset, ihl, verdict;
1309
1310         *related = 1;
1311
1312         /* reassemble IP fragments */
1313         if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
1314                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
1315                         return NF_STOLEN;
1316         }
1317
1318         iph = ip_hdr(skb);
1319         offset = ihl = iph->ihl * 4;
1320         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1321         if (ic == NULL)
1322                 return NF_DROP;
1323
1324         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1325                   ic->type, ntohs(icmp_id(ic)),
1326                   &iph->saddr, &iph->daddr);
1327
1328         /*
1329          * Work through seeing if this is for us.
1330          * These checks are supposed to be in an order that means easy
1331          * things are checked first to speed up processing.... however
1332          * this means that some packets will manage to get a long way
1333          * down this stack and then be rejected, but that's life.
1334          */
1335         if ((ic->type != ICMP_DEST_UNREACH) &&
1336             (ic->type != ICMP_SOURCE_QUENCH) &&
1337             (ic->type != ICMP_TIME_EXCEEDED)) {
1338                 *related = 0;
1339                 return NF_ACCEPT;
1340         }
1341
1342         /* Now find the contained IP header */
1343         offset += sizeof(_icmph);
1344         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1345         if (cih == NULL)
1346                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1347
1348         net = skb_net(skb);
1349
1350         pd = ip_vs_proto_data_get(net, cih->protocol);
1351         if (!pd)
1352                 return NF_ACCEPT;
1353         pp = pd->pp;
1354
1355         /* Is the embedded protocol header present? */
1356         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1357                      pp->dont_defrag))
1358                 return NF_ACCEPT;
1359
1360         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1361                       "Checking incoming ICMP for");
1362
1363         offset += cih->ihl * 4;
1364
1365         ip_vs_fill_iphdr(AF_INET, cih, &ciph);
1366         /* The embedded headers contain source and dest in reverse order */
1367         cp = pp->conn_in_get(AF_INET, skb, &ciph, offset, 1);
1368         if (!cp)
1369                 return NF_ACCEPT;
1370
1371         verdict = NF_DROP;
1372
1373         /* Ensure the checksum is correct */
1374         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1375                 /* Failed checksum! */
1376                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1377                           &iph->saddr);
1378                 goto out;
1379         }
1380
1381         /* do the statistics and put it back */
1382         ip_vs_in_stats(cp, skb);
1383         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol)
1384                 offset += 2 * sizeof(__u16);
1385         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum);
1386
1387   out:
1388         __ip_vs_conn_put(cp);
1389
1390         return verdict;
1391 }
1392
1393 #ifdef CONFIG_IP_VS_IPV6
1394 static int
1395 ip_vs_in_icmp_v6(struct sk_buff *skb, int *related, unsigned int hooknum)
1396 {
1397         struct net *net = NULL;
1398         struct ipv6hdr *iph;
1399         struct icmp6hdr _icmph, *ic;
1400         struct ipv6hdr  _ciph, *cih;    /* The ip header contained
1401                                            within the ICMP */
1402         struct ip_vs_iphdr ciph;
1403         struct ip_vs_conn *cp;
1404         struct ip_vs_protocol *pp;
1405         struct ip_vs_proto_data *pd;
1406         unsigned int offset, verdict;
1407
1408         *related = 1;
1409
1410         /* reassemble IP fragments */
1411         if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) {
1412                 if (ip_vs_gather_frags_v6(skb, ip_vs_defrag_user(hooknum)))
1413                         return NF_STOLEN;
1414         }
1415
1416         iph = ipv6_hdr(skb);
1417         offset = sizeof(struct ipv6hdr);
1418         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1419         if (ic == NULL)
1420                 return NF_DROP;
1421
1422         IP_VS_DBG(12, "Incoming ICMPv6 (%d,%d) %pI6->%pI6\n",
1423                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1424                   &iph->saddr, &iph->daddr);
1425
1426         /*
1427          * Work through seeing if this is for us.
1428          * These checks are supposed to be in an order that means easy
1429          * things are checked first to speed up processing.... however
1430          * this means that some packets will manage to get a long way
1431          * down this stack and then be rejected, but that's life.
1432          */
1433         if ((ic->icmp6_type != ICMPV6_DEST_UNREACH) &&
1434             (ic->icmp6_type != ICMPV6_PKT_TOOBIG) &&
1435             (ic->icmp6_type != ICMPV6_TIME_EXCEED)) {
1436                 *related = 0;
1437                 return NF_ACCEPT;
1438         }
1439
1440         /* Now find the contained IP header */
1441         offset += sizeof(_icmph);
1442         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1443         if (cih == NULL)
1444                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1445
1446         net = skb_net(skb);
1447         pd = ip_vs_proto_data_get(net, cih->nexthdr);
1448         if (!pd)
1449                 return NF_ACCEPT;
1450         pp = pd->pp;
1451
1452         /* Is the embedded protocol header present? */
1453         /* TODO: we don't support fragmentation at the moment anyways */
1454         if (unlikely(cih->nexthdr == IPPROTO_FRAGMENT && pp->dont_defrag))
1455                 return NF_ACCEPT;
1456
1457         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
1458                       "Checking incoming ICMPv6 for");
1459
1460         offset += sizeof(struct ipv6hdr);
1461
1462         ip_vs_fill_iphdr(AF_INET6, cih, &ciph);
1463         /* The embedded headers contain source and dest in reverse order */
1464         cp = pp->conn_in_get(AF_INET6, skb, &ciph, offset, 1);
1465         if (!cp)
1466                 return NF_ACCEPT;
1467
1468         /* do the statistics and put it back */
1469         ip_vs_in_stats(cp, skb);
1470         if (IPPROTO_TCP == cih->nexthdr || IPPROTO_UDP == cih->nexthdr ||
1471             IPPROTO_SCTP == cih->nexthdr)
1472                 offset += 2 * sizeof(__u16);
1473         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset, hooknum);
1474
1475         __ip_vs_conn_put(cp);
1476
1477         return verdict;
1478 }
1479 #endif
1480
1481
1482 /*
1483  *      Check if it's for virtual services, look it up,
1484  *      and send it on its way...
1485  */
1486 static unsigned int
1487 ip_vs_in(unsigned int hooknum, struct sk_buff *skb, int af)
1488 {
1489         struct net *net;
1490         struct ip_vs_iphdr iph;
1491         struct ip_vs_protocol *pp;
1492         struct ip_vs_proto_data *pd;
1493         struct ip_vs_conn *cp;
1494         int ret, restart, pkts;
1495         struct netns_ipvs *ipvs;
1496
1497         /* Already marked as IPVS request or reply? */
1498         if (skb->ipvs_property)
1499                 return NF_ACCEPT;
1500
1501         /*
1502          *      Big tappo:
1503          *      - remote client: only PACKET_HOST
1504          *      - route: used for struct net when skb->dev is unset
1505          */
1506         if (unlikely((skb->pkt_type != PACKET_HOST &&
1507                       hooknum != NF_INET_LOCAL_OUT) ||
1508                      !skb_dst(skb))) {
1509                 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1510                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1511                               " ignored in hook %u\n",
1512                               skb->pkt_type, iph.protocol,
1513                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1514                 return NF_ACCEPT;
1515         }
1516         /* ipvs enabled in this netns ? */
1517         net = skb_net(skb);
1518         if (!net_ipvs(net)->enable)
1519                 return NF_ACCEPT;
1520
1521         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1522
1523         /* Bad... Do not break raw sockets */
1524         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1525                      af == AF_INET)) {
1526                 struct sock *sk = skb->sk;
1527                 struct inet_sock *inet = inet_sk(skb->sk);
1528
1529                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1530                         return NF_ACCEPT;
1531         }
1532
1533 #ifdef CONFIG_IP_VS_IPV6
1534         if (af == AF_INET6) {
1535                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1536                         int related;
1537                         int verdict = ip_vs_in_icmp_v6(skb, &related, hooknum);
1538
1539                         if (related)
1540                                 return verdict;
1541                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1542                 }
1543         } else
1544 #endif
1545                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1546                         int related;
1547                         int verdict = ip_vs_in_icmp(skb, &related, hooknum);
1548
1549                         if (related)
1550                                 return verdict;
1551                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1552                 }
1553
1554         /* Protocol supported? */
1555         pd = ip_vs_proto_data_get(net, iph.protocol);
1556         if (unlikely(!pd))
1557                 return NF_ACCEPT;
1558         pp = pd->pp;
1559         /*
1560          * Check if the packet belongs to an existing connection entry
1561          */
1562         cp = pp->conn_in_get(af, skb, &iph, iph.len, 0);
1563
1564         if (unlikely(!cp)) {
1565                 int v;
1566
1567                 if (!pp->conn_schedule(af, skb, pd, &v, &cp))
1568                         return v;
1569         }
1570
1571         if (unlikely(!cp)) {
1572                 /* sorry, all this trouble for a no-hit :) */
1573                 IP_VS_DBG_PKT(12, af, pp, skb, 0,
1574                               "ip_vs_in: packet continues traversal as normal");
1575                 return NF_ACCEPT;
1576         }
1577
1578         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Incoming packet");
1579         ipvs = net_ipvs(net);
1580         /* Check the server status */
1581         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1582                 /* the destination server is not available */
1583
1584                 if (sysctl_expire_nodest_conn(ipvs)) {
1585                         /* try to expire the connection immediately */
1586                         ip_vs_conn_expire_now(cp);
1587                 }
1588                 /* don't restart its timer, and silently
1589                    drop the packet. */
1590                 __ip_vs_conn_put(cp);
1591                 return NF_DROP;
1592         }
1593
1594         ip_vs_in_stats(cp, skb);
1595         restart = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1596         if (cp->packet_xmit)
1597                 ret = cp->packet_xmit(skb, cp, pp);
1598                 /* do not touch skb anymore */
1599         else {
1600                 IP_VS_DBG_RL("warning: packet_xmit is null");
1601                 ret = NF_ACCEPT;
1602         }
1603
1604         /* Increase its packet counter and check if it is needed
1605          * to be synchronized
1606          *
1607          * Sync connection if it is about to close to
1608          * encorage the standby servers to update the connections timeout
1609          *
1610          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1611          */
1612
1613         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
1614                 pkts = sysctl_sync_threshold(ipvs);
1615         else
1616                 pkts = atomic_add_return(1, &cp->in_pkts);
1617
1618         if ((ipvs->sync_state & IP_VS_STATE_MASTER) &&
1619             cp->protocol == IPPROTO_SCTP) {
1620                 if ((cp->state == IP_VS_SCTP_S_ESTABLISHED &&
1621                         (pkts % sysctl_sync_period(ipvs)
1622                          == sysctl_sync_threshold(ipvs))) ||
1623                                 (cp->old_state != cp->state &&
1624                                  ((cp->state == IP_VS_SCTP_S_CLOSED) ||
1625                                   (cp->state == IP_VS_SCTP_S_SHUT_ACK_CLI) ||
1626                                   (cp->state == IP_VS_SCTP_S_SHUT_ACK_SER)))) {
1627                         ip_vs_sync_conn(net, cp);
1628                         goto out;
1629                 }
1630         }
1631
1632         /* Keep this block last: TCP and others with pp->num_states <= 1 */
1633         else if ((ipvs->sync_state & IP_VS_STATE_MASTER) &&
1634             (((cp->protocol != IPPROTO_TCP ||
1635                cp->state == IP_VS_TCP_S_ESTABLISHED) &&
1636               (pkts % sysctl_sync_period(ipvs)
1637                == sysctl_sync_threshold(ipvs))) ||
1638              ((cp->protocol == IPPROTO_TCP) && (cp->old_state != cp->state) &&
1639               ((cp->state == IP_VS_TCP_S_FIN_WAIT) ||
1640                (cp->state == IP_VS_TCP_S_CLOSE) ||
1641                (cp->state == IP_VS_TCP_S_CLOSE_WAIT) ||
1642                (cp->state == IP_VS_TCP_S_TIME_WAIT)))))
1643                 ip_vs_sync_conn(net, cp);
1644 out:
1645         cp->old_state = cp->state;
1646
1647         ip_vs_conn_put(cp);
1648         return ret;
1649 }
1650
1651 /*
1652  *      AF_INET handler in NF_INET_LOCAL_IN chain
1653  *      Schedule and forward packets from remote clients
1654  */
1655 static unsigned int
1656 ip_vs_remote_request4(unsigned int hooknum, struct sk_buff *skb,
1657                       const struct net_device *in,
1658                       const struct net_device *out,
1659                       int (*okfn)(struct sk_buff *))
1660 {
1661         return ip_vs_in(hooknum, skb, AF_INET);
1662 }
1663
1664 /*
1665  *      AF_INET handler in NF_INET_LOCAL_OUT chain
1666  *      Schedule and forward packets from local clients
1667  */
1668 static unsigned int
1669 ip_vs_local_request4(unsigned int hooknum, struct sk_buff *skb,
1670                      const struct net_device *in, const struct net_device *out,
1671                      int (*okfn)(struct sk_buff *))
1672 {
1673         unsigned int verdict;
1674
1675         /* Disable BH in LOCAL_OUT until all places are fixed */
1676         local_bh_disable();
1677         verdict = ip_vs_in(hooknum, skb, AF_INET);
1678         local_bh_enable();
1679         return verdict;
1680 }
1681
1682 #ifdef CONFIG_IP_VS_IPV6
1683
1684 /*
1685  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
1686  *      Schedule and forward packets from remote clients
1687  */
1688 static unsigned int
1689 ip_vs_remote_request6(unsigned int hooknum, struct sk_buff *skb,
1690                       const struct net_device *in,
1691                       const struct net_device *out,
1692                       int (*okfn)(struct sk_buff *))
1693 {
1694         return ip_vs_in(hooknum, skb, AF_INET6);
1695 }
1696
1697 /*
1698  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
1699  *      Schedule and forward packets from local clients
1700  */
1701 static unsigned int
1702 ip_vs_local_request6(unsigned int hooknum, struct sk_buff *skb,
1703                      const struct net_device *in, const struct net_device *out,
1704                      int (*okfn)(struct sk_buff *))
1705 {
1706         unsigned int verdict;
1707
1708         /* Disable BH in LOCAL_OUT until all places are fixed */
1709         local_bh_disable();
1710         verdict = ip_vs_in(hooknum, skb, AF_INET6);
1711         local_bh_enable();
1712         return verdict;
1713 }
1714
1715 #endif
1716
1717
1718 /*
1719  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1720  *      related packets destined for 0.0.0.0/0.
1721  *      When fwmark-based virtual service is used, such as transparent
1722  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
1723  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1724  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1725  *      and send them to ip_vs_in_icmp.
1726  */
1727 static unsigned int
1728 ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff *skb,
1729                    const struct net_device *in, const struct net_device *out,
1730                    int (*okfn)(struct sk_buff *))
1731 {
1732         int r;
1733         struct net *net;
1734
1735         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1736                 return NF_ACCEPT;
1737
1738         /* ipvs enabled in this netns ? */
1739         net = skb_net(skb);
1740         if (!net_ipvs(net)->enable)
1741                 return NF_ACCEPT;
1742
1743         return ip_vs_in_icmp(skb, &r, hooknum);
1744 }
1745
1746 #ifdef CONFIG_IP_VS_IPV6
1747 static unsigned int
1748 ip_vs_forward_icmp_v6(unsigned int hooknum, struct sk_buff *skb,
1749                       const struct net_device *in, const struct net_device *out,
1750                       int (*okfn)(struct sk_buff *))
1751 {
1752         int r;
1753         struct net *net;
1754
1755         if (ipv6_hdr(skb)->nexthdr != IPPROTO_ICMPV6)
1756                 return NF_ACCEPT;
1757
1758         /* ipvs enabled in this netns ? */
1759         net = skb_net(skb);
1760         if (!net_ipvs(net)->enable)
1761                 return NF_ACCEPT;
1762
1763         return ip_vs_in_icmp_v6(skb, &r, hooknum);
1764 }
1765 #endif
1766
1767
1768 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
1769         /* After packet filtering, change source only for VS/NAT */
1770         {
1771                 .hook           = ip_vs_reply4,
1772                 .owner          = THIS_MODULE,
1773                 .pf             = PF_INET,
1774                 .hooknum        = NF_INET_LOCAL_IN,
1775                 .priority       = NF_IP_PRI_NAT_SRC - 2,
1776         },
1777         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1778          * or VS/NAT(change destination), so that filtering rules can be
1779          * applied to IPVS. */
1780         {
1781                 .hook           = ip_vs_remote_request4,
1782                 .owner          = THIS_MODULE,
1783                 .pf             = PF_INET,
1784                 .hooknum        = NF_INET_LOCAL_IN,
1785                 .priority       = NF_IP_PRI_NAT_SRC - 1,
1786         },
1787         /* Before ip_vs_in, change source only for VS/NAT */
1788         {
1789                 .hook           = ip_vs_local_reply4,
1790                 .owner          = THIS_MODULE,
1791                 .pf             = PF_INET,
1792                 .hooknum        = NF_INET_LOCAL_OUT,
1793                 .priority       = NF_IP_PRI_NAT_DST + 1,
1794         },
1795         /* After mangle, schedule and forward local requests */
1796         {
1797                 .hook           = ip_vs_local_request4,
1798                 .owner          = THIS_MODULE,
1799                 .pf             = PF_INET,
1800                 .hooknum        = NF_INET_LOCAL_OUT,
1801                 .priority       = NF_IP_PRI_NAT_DST + 2,
1802         },
1803         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1804          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1805         {
1806                 .hook           = ip_vs_forward_icmp,
1807                 .owner          = THIS_MODULE,
1808                 .pf             = PF_INET,
1809                 .hooknum        = NF_INET_FORWARD,
1810                 .priority       = 99,
1811         },
1812         /* After packet filtering, change source only for VS/NAT */
1813         {
1814                 .hook           = ip_vs_reply4,
1815                 .owner          = THIS_MODULE,
1816                 .pf             = PF_INET,
1817                 .hooknum        = NF_INET_FORWARD,
1818                 .priority       = 100,
1819         },
1820 #ifdef CONFIG_IP_VS_IPV6
1821         /* After packet filtering, change source only for VS/NAT */
1822         {
1823                 .hook           = ip_vs_reply6,
1824                 .owner          = THIS_MODULE,
1825                 .pf             = PF_INET6,
1826                 .hooknum        = NF_INET_LOCAL_IN,
1827                 .priority       = NF_IP6_PRI_NAT_SRC - 2,
1828         },
1829         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1830          * or VS/NAT(change destination), so that filtering rules can be
1831          * applied to IPVS. */
1832         {
1833                 .hook           = ip_vs_remote_request6,
1834                 .owner          = THIS_MODULE,
1835                 .pf             = PF_INET6,
1836                 .hooknum        = NF_INET_LOCAL_IN,
1837                 .priority       = NF_IP6_PRI_NAT_SRC - 1,
1838         },
1839         /* Before ip_vs_in, change source only for VS/NAT */
1840         {
1841                 .hook           = ip_vs_local_reply6,
1842                 .owner          = THIS_MODULE,
1843                 .pf             = PF_INET,
1844                 .hooknum        = NF_INET_LOCAL_OUT,
1845                 .priority       = NF_IP6_PRI_NAT_DST + 1,
1846         },
1847         /* After mangle, schedule and forward local requests */
1848         {
1849                 .hook           = ip_vs_local_request6,
1850                 .owner          = THIS_MODULE,
1851                 .pf             = PF_INET6,
1852                 .hooknum        = NF_INET_LOCAL_OUT,
1853                 .priority       = NF_IP6_PRI_NAT_DST + 2,
1854         },
1855         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1856          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1857         {
1858                 .hook           = ip_vs_forward_icmp_v6,
1859                 .owner          = THIS_MODULE,
1860                 .pf             = PF_INET6,
1861                 .hooknum        = NF_INET_FORWARD,
1862                 .priority       = 99,
1863         },
1864         /* After packet filtering, change source only for VS/NAT */
1865         {
1866                 .hook           = ip_vs_reply6,
1867                 .owner          = THIS_MODULE,
1868                 .pf             = PF_INET6,
1869                 .hooknum        = NF_INET_FORWARD,
1870                 .priority       = 100,
1871         },
1872 #endif
1873 };
1874 /*
1875  *      Initialize IP Virtual Server netns mem.
1876  */
1877 static int __net_init __ip_vs_init(struct net *net)
1878 {
1879         struct netns_ipvs *ipvs;
1880
1881         ipvs = net_generic(net, ip_vs_net_id);
1882         if (ipvs == NULL) {
1883                 pr_err("%s(): no memory.\n", __func__);
1884                 return -ENOMEM;
1885         }
1886         /* Hold the beast until a service is registerd */
1887         ipvs->enable = 0;
1888         ipvs->net = net;
1889         /* Counters used for creating unique names */
1890         ipvs->gen = atomic_read(&ipvs_netns_cnt);
1891         atomic_inc(&ipvs_netns_cnt);
1892         net->ipvs = ipvs;
1893
1894         if (__ip_vs_estimator_init(net) < 0)
1895                 goto estimator_fail;
1896
1897         if (__ip_vs_control_init(net) < 0)
1898                 goto control_fail;
1899
1900         if (__ip_vs_protocol_init(net) < 0)
1901                 goto protocol_fail;
1902
1903         if (__ip_vs_app_init(net) < 0)
1904                 goto app_fail;
1905
1906         if (__ip_vs_conn_init(net) < 0)
1907                 goto conn_fail;
1908
1909         if (__ip_vs_sync_init(net) < 0)
1910                 goto sync_fail;
1911
1912         printk(KERN_INFO "IPVS: Creating netns size=%zu id=%d\n",
1913                          sizeof(struct netns_ipvs), ipvs->gen);
1914         return 0;
1915 /*
1916  * Error handling
1917  */
1918
1919 sync_fail:
1920         __ip_vs_conn_cleanup(net);
1921 conn_fail:
1922         __ip_vs_app_cleanup(net);
1923 app_fail:
1924         __ip_vs_protocol_cleanup(net);
1925 protocol_fail:
1926         __ip_vs_control_cleanup(net);
1927 control_fail:
1928         __ip_vs_estimator_cleanup(net);
1929 estimator_fail:
1930         return -ENOMEM;
1931 }
1932
1933 static void __net_exit __ip_vs_cleanup(struct net *net)
1934 {
1935         __ip_vs_service_cleanup(net);   /* ip_vs_flush() with locks */
1936         __ip_vs_conn_cleanup(net);
1937         __ip_vs_app_cleanup(net);
1938         __ip_vs_protocol_cleanup(net);
1939         __ip_vs_control_cleanup(net);
1940         __ip_vs_estimator_cleanup(net);
1941         IP_VS_DBG(2, "ipvs netns %d released\n", net_ipvs(net)->gen);
1942 }
1943
1944 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
1945 {
1946         EnterFunction(2);
1947         net_ipvs(net)->enable = 0;      /* Disable packet reception */
1948         smp_wmb();
1949         __ip_vs_sync_cleanup(net);
1950         LeaveFunction(2);
1951 }
1952
1953 static struct pernet_operations ipvs_core_ops = {
1954         .init = __ip_vs_init,
1955         .exit = __ip_vs_cleanup,
1956         .id   = &ip_vs_net_id,
1957         .size = sizeof(struct netns_ipvs),
1958 };
1959
1960 static struct pernet_operations ipvs_core_dev_ops = {
1961         .exit = __ip_vs_dev_cleanup,
1962 };
1963
1964 /*
1965  *      Initialize IP Virtual Server
1966  */
1967 static int __init ip_vs_init(void)
1968 {
1969         int ret;
1970
1971         ip_vs_estimator_init();
1972         ret = ip_vs_control_init();
1973         if (ret < 0) {
1974                 pr_err("can't setup control.\n");
1975                 goto cleanup_estimator;
1976         }
1977
1978         ip_vs_protocol_init();
1979
1980         ret = ip_vs_app_init();
1981         if (ret < 0) {
1982                 pr_err("can't setup application helper.\n");
1983                 goto cleanup_protocol;
1984         }
1985
1986         ret = ip_vs_conn_init();
1987         if (ret < 0) {
1988                 pr_err("can't setup connection table.\n");
1989                 goto cleanup_app;
1990         }
1991
1992         ret = ip_vs_sync_init();
1993         if (ret < 0) {
1994                 pr_err("can't setup sync data.\n");
1995                 goto cleanup_conn;
1996         }
1997
1998         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
1999         if (ret < 0)
2000                 goto cleanup_sync;
2001
2002         ret = register_pernet_device(&ipvs_core_dev_ops);
2003         if (ret < 0)
2004                 goto cleanup_sub;
2005
2006         ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2007         if (ret < 0) {
2008                 pr_err("can't register hooks.\n");
2009                 goto cleanup_dev;
2010         }
2011
2012         pr_info("ipvs loaded.\n");
2013
2014         return ret;
2015
2016 cleanup_dev:
2017         unregister_pernet_device(&ipvs_core_dev_ops);
2018 cleanup_sub:
2019         unregister_pernet_subsys(&ipvs_core_ops);
2020 cleanup_sync:
2021         ip_vs_sync_cleanup();
2022   cleanup_conn:
2023         ip_vs_conn_cleanup();
2024   cleanup_app:
2025         ip_vs_app_cleanup();
2026   cleanup_protocol:
2027         ip_vs_protocol_cleanup();
2028         ip_vs_control_cleanup();
2029   cleanup_estimator:
2030         ip_vs_estimator_cleanup();
2031         return ret;
2032 }
2033
2034 static void __exit ip_vs_cleanup(void)
2035 {
2036         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2037         unregister_pernet_device(&ipvs_core_dev_ops);
2038         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
2039         ip_vs_sync_cleanup();
2040         ip_vs_conn_cleanup();
2041         ip_vs_app_cleanup();
2042         ip_vs_protocol_cleanup();
2043         ip_vs_control_cleanup();
2044         ip_vs_estimator_cleanup();
2045         pr_info("ipvs unloaded.\n");
2046 }
2047
2048 module_init(ip_vs_init);
2049 module_exit(ip_vs_cleanup);
2050 MODULE_LICENSE("GPL");