net/netfilter/ipvs/ip_vs_core.c

   1 /*
   2  * IPVS         An implementation of the IP virtual server support for the
   3  *              LINUX operating system.  IPVS is now implemented as a module
   4  *              over the Netfilter framework. IPVS can be used to build a
   5  *              high-performance and highly available server based on a
   6  *              cluster of servers.
   7  *
   8  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
   9  *              Peter Kese <peter.kese@ijs.si>
  10  *              Julian Anastasov <ja@ssi.bg>
  11  *
  12  *              This program is free software; you can redistribute it and/or
  13  *              modify it under the terms of the GNU General Public License
  14  *              as published by the Free Software Foundation; either version
  15  *              2 of the License, or (at your option) any later version.
  16  *
  17  * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
  18  * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
  19  * and others.
  20  *
  21  * Changes:
  22  *      Paul `Rusty' Russell            properly handle non-linear skbs
  23  *      Harald Welte                    don't use nfcache
  24  *
  25  */
  26
  27 #define KMSG_COMPONENT "IPVS"
  28 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  29
  30 #include <linux/module.h>
  31 #include <linux/kernel.h>
  32 #include <linux/ip.h>
  33 #include <linux/tcp.h>
  34 #include <linux/sctp.h>
  35 #include <linux/icmp.h>
  36 #include <linux/slab.h>
  37
  38 #include <net/ip.h>
  39 #include <net/tcp.h>
  40 #include <net/udp.h>
  41 #include <net/icmp.h>                   /* for icmp_send */
  42 #include <net/route.h>
  43 #include <net/ip6_checksum.h>
  44 #include <net/netns/generic.h>          /* net_generic() */
  45
  46 #include <linux/netfilter.h>
  47 #include <linux/netfilter_ipv4.h>
  48
  49 #ifdef CONFIG_IP_VS_IPV6
  50 #include <net/ipv6.h>
  51 #include <linux/netfilter_ipv6.h>
  52 #include <net/ip6_route.h>
  53 #endif
  54
  55 #include <net/ip_vs.h>
  56
  57
  58 EXPORT_SYMBOL(register_ip_vs_scheduler);
  59 EXPORT_SYMBOL(unregister_ip_vs_scheduler);
  60 EXPORT_SYMBOL(ip_vs_proto_name);
  61 EXPORT_SYMBOL(ip_vs_conn_new);
  62 EXPORT_SYMBOL(ip_vs_conn_in_get);
  63 EXPORT_SYMBOL(ip_vs_conn_out_get);
  64 #ifdef CONFIG_IP_VS_PROTO_TCP
  65 EXPORT_SYMBOL(ip_vs_tcp_conn_listen);
  66 #endif
  67 EXPORT_SYMBOL(ip_vs_conn_put);
  68 #ifdef CONFIG_IP_VS_DEBUG
  69 EXPORT_SYMBOL(ip_vs_get_debug_level);
  70 #endif
  71
  72 int ip_vs_net_id __read_mostly;
  73 #ifdef IP_VS_GENERIC_NETNS
  74 EXPORT_SYMBOL(ip_vs_net_id);
  75 #endif
  76 /* netns cnt used for uniqueness */
  77 static atomic_t ipvs_netns_cnt = ATOMIC_INIT(0);
  78
  79 /* ID used in ICMP lookups */
  80 #define icmp_id(icmph)          (((icmph)->un).echo.id)
  81 #define icmpv6_id(icmph)        (icmph->icmp6_dataun.u_echo.identifier)
  82
  83 const char *ip_vs_proto_name(unsigned proto)
  84 {
  85         static char buf[20];
  86
  87         switch (proto) {
  88         case IPPROTO_IP:
  89                 return "IP";
  90         case IPPROTO_UDP:
  91                 return "UDP";
  92         case IPPROTO_TCP:
  93                 return "TCP";
  94         case IPPROTO_SCTP:
  95                 return "SCTP";
  96         case IPPROTO_ICMP:
  97                 return "ICMP";
  98 #ifdef CONFIG_IP_VS_IPV6
  99         case IPPROTO_ICMPV6:
 100                 return "ICMPv6";
 101 #endif
 102         default:
 103                 sprintf(buf, "IP_%d", proto);
 104                 return buf;
 105         }
 106 }
 107
 108 void ip_vs_init_hash_table(struct list_head *table, int rows)
 109 {
 110         while (--rows >= 0)
 111                 INIT_LIST_HEAD(&table[rows]);
 112 }
 113
 114 static inline void
 115 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 116 {
 117         struct ip_vs_dest *dest = cp->dest;
 118         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 119
 120         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 121                 struct ip_vs_cpu_stats *s;
 122
 123                 s = this_cpu_ptr(dest->stats.cpustats);
 124                 s->ustats.inpkts++;
 125                 u64_stats_update_begin(&s->syncp);
 126                 s->ustats.inbytes += skb->len;
 127                 u64_stats_update_end(&s->syncp);
 128
 129                 s = this_cpu_ptr(dest->svc->stats.cpustats);
 130                 s->ustats.inpkts++;
 131                 u64_stats_update_begin(&s->syncp);
 132                 s->ustats.inbytes += skb->len;
 133                 u64_stats_update_end(&s->syncp);
 134
 135                 s = this_cpu_ptr(ipvs->cpustats);
 136                 s->ustats.inpkts++;
 137                 u64_stats_update_begin(&s->syncp);
 138                 s->ustats.inbytes += skb->len;
 139                 u64_stats_update_end(&s->syncp);
 140         }
 141 }
 142
 143
 144 static inline void
 145 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 146 {
 147         struct ip_vs_dest *dest = cp->dest;
 148         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 149
 150         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 151                 struct ip_vs_cpu_stats *s;
 152
 153                 s = this_cpu_ptr(dest->stats.cpustats);
 154                 s->ustats.outpkts++;
 155                 u64_stats_update_begin(&s->syncp);
 156                 s->ustats.outbytes += skb->len;
 157                 u64_stats_update_end(&s->syncp);
 158
 159                 s = this_cpu_ptr(dest->svc->stats.cpustats);
 160                 s->ustats.outpkts++;
 161                 u64_stats_update_begin(&s->syncp);
 162                 s->ustats.outbytes += skb->len;
 163                 u64_stats_update_end(&s->syncp);
 164
 165                 s = this_cpu_ptr(ipvs->cpustats);
 166                 s->ustats.outpkts++;
 167                 u64_stats_update_begin(&s->syncp);
 168                 s->ustats.outbytes += skb->len;
 169                 u64_stats_update_end(&s->syncp);
 170         }
 171 }
 172
 173
 174 static inline void
 175 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
 176 {
 177         struct netns_ipvs *ipvs = net_ipvs(svc->net);
 178         struct ip_vs_cpu_stats *s;
 179
 180         s = this_cpu_ptr(cp->dest->stats.cpustats);
 181         s->ustats.conns++;
 182
 183         s = this_cpu_ptr(svc->stats.cpustats);
 184         s->ustats.conns++;
 185
 186         s = this_cpu_ptr(ipvs->cpustats);
 187         s->ustats.conns++;
 188 }
 189
 190
 191 static inline int
 192 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
 193                 const struct sk_buff *skb,
 194                 struct ip_vs_proto_data *pd)
 195 {
 196         if (unlikely(!pd->pp->state_transition))
 197                 return 0;
 198         return pd->pp->state_transition(cp, direction, skb, pd);
 199 }
 200
 201 static inline int
 202 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
 203                               struct sk_buff *skb, int protocol,
 204                               const union nf_inet_addr *caddr, __be16 cport,
 205                               const union nf_inet_addr *vaddr, __be16 vport,
 206                               struct ip_vs_conn_param *p)
 207 {
 208         ip_vs_conn_fill_param(svc->net, svc->af, protocol, caddr, cport, vaddr,
 209                               vport, p);
 210         p->pe = svc->pe;
 211         if (p->pe && p->pe->fill_param)
 212                 return p->pe->fill_param(p, skb);
 213
 214         return 0;
 215 }
 216
 217 /*
 218  *  IPVS persistent scheduling function
 219  *  It creates a connection entry according to its template if exists,
 220  *  or selects a server and creates a connection entry plus a template.
 221  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
 222  *  Protocols supported: TCP, UDP
 223  */
 224 static struct ip_vs_conn *
 225 ip_vs_sched_persist(struct ip_vs_service *svc,
 226                     struct sk_buff *skb,
 227                     __be16 src_port, __be16 dst_port, int *ignored)
 228 {
 229         struct ip_vs_conn *cp = NULL;
 230         struct ip_vs_iphdr iph;
 231         struct ip_vs_dest *dest;
 232         struct ip_vs_conn *ct;
 233         __be16 dport = 0;               /* destination port to forward */
 234         unsigned int flags;
 235         struct ip_vs_conn_param param;
 236         union nf_inet_addr snet;        /* source network of the client,
 237                                            after masking */
 238
 239         ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph);
 240
 241         /* Mask saddr with the netmask to adjust template granularity */
 242 #ifdef CONFIG_IP_VS_IPV6
 243         if (svc->af == AF_INET6)
 244                 ipv6_addr_prefix(&snet.in6, &iph.saddr.in6, svc->netmask);
 245         else
 246 #endif
 247                 snet.ip = iph.saddr.ip & svc->netmask;
 248
 249         IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
 250                       "mnet %s\n",
 251                       IP_VS_DBG_ADDR(svc->af, &iph.saddr), ntohs(src_port),
 252                       IP_VS_DBG_ADDR(svc->af, &iph.daddr), ntohs(dst_port),
 253                       IP_VS_DBG_ADDR(svc->af, &snet));
 254
 255         /*
 256          * As far as we know, FTP is a very complicated network protocol, and
 257          * it uses control connection and data connections. For active FTP,
 258          * FTP server initialize data connection to the client, its source port
 259          * is often 20. For passive FTP, FTP server tells the clients the port
 260          * that it passively listens to,  and the client issues the data
 261          * connection. In the tunneling or direct routing mode, the load
 262          * balancer is on the client-to-server half of connection, the port
 263          * number is unknown to the load balancer. So, a conn template like
 264          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
 265          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
 266          * is created for other persistent services.
 267          */
 268         {
 269                 int protocol = iph.protocol;
 270                 const union nf_inet_addr *vaddr = &iph.daddr;
 271                 const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
 272                 __be16 vport = 0;
 273
 274                 if (dst_port == svc->port) {
 275                         /* non-FTP template:
 276                          * <protocol, caddr, 0, vaddr, vport, daddr, dport>
 277                          * FTP template:
 278                          * <protocol, caddr, 0, vaddr, 0, daddr, 0>
 279                          */
 280                         if (svc->port != FTPPORT)
 281                                 vport = dst_port;
 282                 } else {
 283                         /* Note: persistent fwmark-based services and
 284                          * persistent port zero service are handled here.
 285                          * fwmark template:
 286                          * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
 287                          * port zero template:
 288                          * <protocol,caddr,0,vaddr,0,daddr,0>
 289                          */
 290                         if (svc->fwmark) {
 291                                 protocol = IPPROTO_IP;
 292                                 vaddr = &fwmark;
 293                         }
 294                 }
 295                 /* return *ignored = -1 so NF_DROP can be used */
 296                 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
 297                                                   vaddr, vport, &param) < 0) {
 298                         *ignored = -1;
 299                         return NULL;
 300                 }
 301         }
 302
 303         /* Check if a template already exists */
 304         ct = ip_vs_ct_in_get(&param);
 305         if (!ct || !ip_vs_check_template(ct)) {
 306                 /*
 307                  * No template found or the dest of the connection
 308                  * template is not available.
 309                  * return *ignored=0 i.e. ICMP and NF_DROP
 310                  */
 311                 dest = svc->scheduler->schedule(svc, skb);
 312                 if (!dest) {
 313                         IP_VS_DBG(1, "p-schedule: no dest found.\n");
 314                         kfree(param.pe_data);
 315                         *ignored = 0;
 316                         return NULL;
 317                 }
 318
 319                 if (dst_port == svc->port && svc->port != FTPPORT)
 320                         dport = dest->port;
 321
 322                 /* Create a template
 323                  * This adds param.pe_data to the template,
 324                  * and thus param.pe_data will be destroyed
 325                  * when the template expires */
 326                 ct = ip_vs_conn_new(&param, &dest->addr, dport,
 327                                     IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
 328                 if (ct == NULL) {
 329                         kfree(param.pe_data);
 330                         *ignored = -1;
 331                         return NULL;
 332                 }
 333
 334                 ct->timeout = svc->timeout;
 335         } else {
 336                 /* set destination with the found template */
 337                 dest = ct->dest;
 338                 kfree(param.pe_data);
 339         }
 340
 341         dport = dst_port;
 342         if (dport == svc->port && dest->port)
 343                 dport = dest->port;
 344
 345         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 346                  && iph.protocol == IPPROTO_UDP)?
 347                 IP_VS_CONN_F_ONE_PACKET : 0;
 348
 349         /*
 350          *    Create a new connection according to the template
 351          */
 352         ip_vs_conn_fill_param(svc->net, svc->af, iph.protocol, &iph.saddr,
 353                               src_port, &iph.daddr, dst_port, &param);
 354
 355         cp = ip_vs_conn_new(&param, &dest->addr, dport, flags, dest, skb->mark);
 356         if (cp == NULL) {
 357                 ip_vs_conn_put(ct);
 358                 *ignored = -1;
 359                 return NULL;
 360         }
 361
 362         /*
 363          *    Add its control
 364          */
 365         ip_vs_control_add(cp, ct);
 366         ip_vs_conn_put(ct);
 367
 368         ip_vs_conn_stats(cp, svc);
 369         return cp;
 370 }
 371
 372
 373 /*
 374  *  IPVS main scheduling function
 375  *  It selects a server according to the virtual service, and
 376  *  creates a connection entry.
 377  *  Protocols supported: TCP, UDP
 378  *
 379  *  Usage of *ignored
 380  *
 381  * 1 :   protocol tried to schedule (eg. on SYN), found svc but the
 382  *       svc/scheduler decides that this packet should be accepted with
 383  *       NF_ACCEPT because it must not be scheduled.
 384  *
 385  * 0 :   scheduler can not find destination, so try bypass or
 386  *       return ICMP and then NF_DROP (ip_vs_leave).
 387  *
 388  * -1 :  scheduler tried to schedule but fatal error occurred, eg.
 389  *       ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
 390  *       failure such as missing Call-ID, ENOMEM on skb_linearize
 391  *       or pe_data. In this case we should return NF_DROP without
 392  *       any attempts to send ICMP with ip_vs_leave.
 393  */
 394 struct ip_vs_conn *
 395 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
 396                struct ip_vs_proto_data *pd, int *ignored)
 397 {
 398         struct ip_vs_protocol *pp = pd->pp;
 399         struct ip_vs_conn *cp = NULL;
 400         struct ip_vs_iphdr iph;
 401         struct ip_vs_dest *dest;
 402         __be16 _ports[2], *pptr;
 403         unsigned int flags;
 404
 405         *ignored = 1;
 406         ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph);
 407         pptr = skb_header_pointer(skb, iph.len, sizeof(_ports), _ports);
 408         if (pptr == NULL)
 409                 return NULL;
 410
 411         /*
 412          * FTPDATA needs this check when using local real server.
 413          * Never schedule Active FTPDATA connections from real server.
 414          * For LVS-NAT they must be already created. For other methods
 415          * with persistence the connection is created on SYN+ACK.
 416          */
 417         if (pptr[0] == FTPDATA) {
 418                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 419                               "Not scheduling FTPDATA");
 420                 return NULL;
 421         }
 422
 423         /*
 424          *    Do not schedule replies from local real server.
 425          */
 426         if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
 427             (cp = pp->conn_in_get(svc->af, skb, &iph, iph.len, 1))) {
 428                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 429                               "Not scheduling reply for existing connection");
 430                 __ip_vs_conn_put(cp);
 431                 return NULL;
 432         }
 433
 434         /*
 435          *    Persistent service
 436          */
 437         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 438                 return ip_vs_sched_persist(svc, skb, pptr[0], pptr[1], ignored);
 439
 440         *ignored = 0;
 441
 442         /*
 443          *    Non-persistent service
 444          */
 445         if (!svc->fwmark && pptr[1] != svc->port) {
 446                 if (!svc->port)
 447                         pr_err("Schedule: port zero only supported "
 448                                "in persistent services, "
 449                                "check your ipvs configuration\n");
 450                 return NULL;
 451         }
 452
 453         dest = svc->scheduler->schedule(svc, skb);
 454         if (dest == NULL) {
 455                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 456                 return NULL;
 457         }
 458
 459         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 460                  && iph.protocol == IPPROTO_UDP)?
 461                 IP_VS_CONN_F_ONE_PACKET : 0;
 462
 463         /*
 464          *    Create a connection entry.
 465          */
 466         {
 467                 struct ip_vs_conn_param p;
 468
 469                 ip_vs_conn_fill_param(svc->net, svc->af, iph.protocol,
 470                                       &iph.saddr, pptr[0], &iph.daddr, pptr[1],
 471                                       &p);
 472                 cp = ip_vs_conn_new(&p, &dest->addr,
 473                                     dest->port ? dest->port : pptr[1],
 474                                     flags, dest, skb->mark);
 475                 if (!cp) {
 476                         *ignored = -1;
 477                         return NULL;
 478                 }
 479         }
 480
 481         IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
 482                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
 483                       ip_vs_fwd_tag(cp),
 484                       IP_VS_DBG_ADDR(svc->af, &cp->caddr), ntohs(cp->cport),
 485                       IP_VS_DBG_ADDR(svc->af, &cp->vaddr), ntohs(cp->vport),
 486                       IP_VS_DBG_ADDR(svc->af, &cp->daddr), ntohs(cp->dport),
 487                       cp->flags, atomic_read(&cp->refcnt));
 488
 489         ip_vs_conn_stats(cp, svc);
 490         return cp;
 491 }
 492
 493
 494 /*
 495  *  Pass or drop the packet.
 496  *  Called by ip_vs_in, when the virtual service is available but
 497  *  no destination is available for a new connection.
 498  */
 499 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 500                 struct ip_vs_proto_data *pd)
 501 {
 502         struct netns_ipvs *ipvs;
 503         __be16 _ports[2], *pptr;
 504         struct ip_vs_iphdr iph;
 505         int unicast;
 506
 507         ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph);
 508
 509         pptr = skb_header_pointer(skb, iph.len, sizeof(_ports), _ports);
 510         if (pptr == NULL) {
 511                 ip_vs_service_put(svc);
 512                 return NF_DROP;
 513         }
 514
 515 #ifdef CONFIG_IP_VS_IPV6
 516         if (svc->af == AF_INET6)
 517                 unicast = ipv6_addr_type(&iph.daddr.in6) & IPV6_ADDR_UNICAST;
 518         else
 519 #endif
 520                 unicast = (inet_addr_type(&init_net, iph.daddr.ip) == RTN_UNICAST);
 521
 522         /* if it is fwmark-based service, the cache_bypass sysctl is up
 523            and the destination is a non-local unicast, then create
 524            a cache_bypass connection entry */
 525         ipvs = net_ipvs(skb_net(skb));
 526         if (ipvs->sysctl_cache_bypass && svc->fwmark && unicast) {
 527                 int ret, cs;
 528                 struct ip_vs_conn *cp;
 529                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
 530                                       iph.protocol == IPPROTO_UDP)?
 531                                       IP_VS_CONN_F_ONE_PACKET : 0;
 532                 union nf_inet_addr daddr =  { .all = { 0, 0, 0, 0 } };
 533
 534                 ip_vs_service_put(svc);
 535
 536                 /* create a new connection entry */
 537                 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
 538                 {
 539                         struct ip_vs_conn_param p;
 540                         ip_vs_conn_fill_param(svc->net, svc->af, iph.protocol,
 541                                               &iph.saddr, pptr[0],
 542                                               &iph.daddr, pptr[1], &p);
 543                         cp = ip_vs_conn_new(&p, &daddr, 0,
 544                                             IP_VS_CONN_F_BYPASS | flags,
 545                                             NULL, skb->mark);
 546                         if (!cp)
 547                                 return NF_DROP;
 548                 }
 549
 550                 /* statistics */
 551                 ip_vs_in_stats(cp, skb);
 552
 553                 /* set state */
 554                 cs = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
 555
 556                 /* transmit the first SYN packet */
 557                 ret = cp->packet_xmit(skb, cp, pd->pp);
 558                 /* do not touch skb anymore */
 559
 560                 atomic_inc(&cp->in_pkts);
 561                 ip_vs_conn_put(cp);
 562                 return ret;
 563         }
 564
 565         /*
 566          * When the virtual ftp service is presented, packets destined
 567          * for other services on the VIP may get here (except services
 568          * listed in the ipvs table), pass the packets, because it is
 569          * not ipvs job to decide to drop the packets.
 570          */
 571         if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT)) {
 572                 ip_vs_service_put(svc);
 573                 return NF_ACCEPT;
 574         }
 575
 576         ip_vs_service_put(svc);
 577
 578         /*
 579          * Notify the client that the destination is unreachable, and
 580          * release the socket buffer.
 581          * Since it is in IP layer, the TCP socket is not actually
 582          * created, the TCP RST packet cannot be sent, instead that
 583          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 584          */
 585 #ifdef CONFIG_IP_VS_IPV6
 586         if (svc->af == AF_INET6) {
 587                 if (!skb->dev) {
 588                         struct net *net = dev_net(skb_dst(skb)->dev);
 589
 590                         skb->dev = net->loopback_dev;
 591                 }
 592                 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 593         } else
 594 #endif
 595                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 596
 597         return NF_DROP;
 598 }
 599
 600 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 601 {
 602         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 603 }
 604
 605 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
 606 {
 607         if (NF_INET_LOCAL_IN == hooknum)
 608                 return IP_DEFRAG_VS_IN;
 609         if (NF_INET_FORWARD == hooknum)
 610                 return IP_DEFRAG_VS_FWD;
 611         return IP_DEFRAG_VS_OUT;
 612 }
 613
 614 static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user)
 615 {
 616         int err = ip_defrag(skb, user);
 617
 618         if (!err)
 619                 ip_send_check(ip_hdr(skb));
 620
 621         return err;
 622 }
 623
 624 #ifdef CONFIG_IP_VS_IPV6
 625 static inline int ip_vs_gather_frags_v6(struct sk_buff *skb, u_int32_t user)
 626 {
 627         /* TODO IPv6: Find out what to do here for IPv6 */
 628         return 0;
 629 }
 630 #endif
 631
 632 /*
 633  * Packet has been made sufficiently writable in caller
 634  * - inout: 1=in->out, 0=out->in
 635  */
 636 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 637                     struct ip_vs_conn *cp, int inout)
 638 {
 639         struct iphdr *iph        = ip_hdr(skb);
 640         unsigned int icmp_offset = iph->ihl*4;
 641         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
 642                                                       icmp_offset);
 643         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 644
 645         if (inout) {
 646                 iph->saddr = cp->vaddr.ip;
 647                 ip_send_check(iph);
 648                 ciph->daddr = cp->vaddr.ip;
 649                 ip_send_check(ciph);
 650         } else {
 651                 iph->daddr = cp->daddr.ip;
 652                 ip_send_check(iph);
 653                 ciph->saddr = cp->daddr.ip;
 654                 ip_send_check(ciph);
 655         }
 656
 657         /* the TCP/UDP/SCTP port */
 658         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
 659             IPPROTO_SCTP == ciph->protocol) {
 660                 __be16 *ports = (void *)ciph + ciph->ihl*4;
 661
 662                 if (inout)
 663                         ports[1] = cp->vport;
 664                 else
 665                         ports[0] = cp->dport;
 666         }
 667
 668         /* And finally the ICMP checksum */
 669         icmph->checksum = 0;
 670         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 671         skb->ip_summed = CHECKSUM_UNNECESSARY;
 672
 673         if (inout)
 674                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 675                         "Forwarding altered outgoing ICMP");
 676         else
 677                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 678                         "Forwarding altered incoming ICMP");
 679 }
 680
 681 #ifdef CONFIG_IP_VS_IPV6
 682 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
 683                     struct ip_vs_conn *cp, int inout)
 684 {
 685         struct ipv6hdr *iph      = ipv6_hdr(skb);
 686         unsigned int icmp_offset = sizeof(struct ipv6hdr);
 687         struct icmp6hdr *icmph   = (struct icmp6hdr *)(skb_network_header(skb) +
 688                                                       icmp_offset);
 689         struct ipv6hdr *ciph     = (struct ipv6hdr *)(icmph + 1);
 690
 691         if (inout) {
 692                 iph->saddr = cp->vaddr.in6;
 693                 ciph->daddr = cp->vaddr.in6;
 694         } else {
 695                 iph->daddr = cp->daddr.in6;
 696                 ciph->saddr = cp->daddr.in6;
 697         }
 698
 699         /* the TCP/UDP/SCTP port */
 700         if (IPPROTO_TCP == ciph->nexthdr || IPPROTO_UDP == ciph->nexthdr ||
 701             IPPROTO_SCTP == ciph->nexthdr) {
 702                 __be16 *ports = (void *)ciph + sizeof(struct ipv6hdr);
 703
 704                 if (inout)
 705                         ports[1] = cp->vport;
 706                 else
 707                         ports[0] = cp->dport;
 708         }
 709
 710         /* And finally the ICMP checksum */
 711         icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
 712                                               skb->len - icmp_offset,
 713                                               IPPROTO_ICMPV6, 0);
 714         skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
 715         skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
 716         skb->ip_summed = CHECKSUM_PARTIAL;
 717
 718         if (inout)
 719                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 720                               (void *)ciph - (void *)iph,
 721                               "Forwarding altered outgoing ICMPv6");
 722         else
 723                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 724                               (void *)ciph - (void *)iph,
 725                               "Forwarding altered incoming ICMPv6");
 726 }
 727 #endif
 728
 729 /* Handle relevant response ICMP messages - forward to the right
 730  * destination host. Used for NAT and local client.
 731  */
 732 static int handle_response_icmp(int af, struct sk_buff *skb,
 733                                 union nf_inet_addr *snet,
 734                                 __u8 protocol, struct ip_vs_conn *cp,
 735                                 struct ip_vs_protocol *pp,
 736                                 unsigned int offset, unsigned int ihl)
 737 {
 738         struct netns_ipvs *ipvs;
 739         unsigned int verdict = NF_DROP;
 740
 741         if (IP_VS_FWD_METHOD(cp) != 0) {
 742                 pr_err("shouldn't reach here, because the box is on the "
 743                        "half connection in the tun/dr module.\n");
 744         }
 745
 746         /* Ensure the checksum is correct */
 747         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 748                 /* Failed checksum! */
 749                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
 750                               IP_VS_DBG_ADDR(af, snet));
 751                 goto out;
 752         }
 753
 754         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 755             IPPROTO_SCTP == protocol)
 756                 offset += 2 * sizeof(__u16);
 757         if (!skb_make_writable(skb, offset))
 758                 goto out;
 759
 760         ipvs = net_ipvs(skb_net(skb));
 761
 762 #ifdef CONFIG_IP_VS_IPV6
 763         if (af == AF_INET6)
 764                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
 765         else
 766 #endif
 767                 ip_vs_nat_icmp(skb, pp, cp, 1);
 768
 769 #ifdef CONFIG_IP_VS_IPV6
 770         if (af == AF_INET6) {
 771                 if (ipvs->sysctl_snat_reroute && ip6_route_me_harder(skb) != 0)
 772                         goto out;
 773         } else
 774 #endif
 775                 if ((ipvs->sysctl_snat_reroute ||
 776                      skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
 777                     ip_route_me_harder(skb, RTN_LOCAL) != 0)
 778                         goto out;
 779
 780         /* do the statistics and put it back */
 781         ip_vs_out_stats(cp, skb);
 782
 783         skb->ipvs_property = 1;
 784         if (!(cp->flags & IP_VS_CONN_F_NFCT))
 785                 ip_vs_notrack(skb);
 786         else
 787                 ip_vs_update_conntrack(skb, cp, 0);
 788         verdict = NF_ACCEPT;
 789
 790 out:
 791         __ip_vs_conn_put(cp);
 792
 793         return verdict;
 794 }
 795
 796 /*
 797  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 798  *      Find any that might be relevant, check against existing connections.
 799  *      Currently handles error types - unreachable, quench, ttl exceeded.
 800  */
 801 static int ip_vs_out_icmp(struct sk_buff *skb, int *related,
 802                           unsigned int hooknum)
 803 {
 804         struct iphdr *iph;
 805         struct icmphdr  _icmph, *ic;
 806         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 807         struct ip_vs_iphdr ciph;
 808         struct ip_vs_conn *cp;
 809         struct ip_vs_protocol *pp;
 810         unsigned int offset, ihl;
 811         union nf_inet_addr snet;
 812
 813         *related = 1;
 814
 815         /* reassemble IP fragments */
 816         if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
 817                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
 818                         return NF_STOLEN;
 819         }
 820
 821         iph = ip_hdr(skb);
 822         offset = ihl = iph->ihl * 4;
 823         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 824         if (ic == NULL)
 825                 return NF_DROP;
 826
 827         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
 828                   ic->type, ntohs(icmp_id(ic)),
 829                   &iph->saddr, &iph->daddr);
 830
 831         /*
 832          * Work through seeing if this is for us.
 833          * These checks are supposed to be in an order that means easy
 834          * things are checked first to speed up processing.... however
 835          * this means that some packets will manage to get a long way
 836          * down this stack and then be rejected, but that's life.
 837          */
 838         if ((ic->type != ICMP_DEST_UNREACH) &&
 839             (ic->type != ICMP_SOURCE_QUENCH) &&
 840             (ic->type != ICMP_TIME_EXCEEDED)) {
 841                 *related = 0;
 842                 return NF_ACCEPT;
 843         }
 844
 845         /* Now find the contained IP header */
 846         offset += sizeof(_icmph);
 847         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 848         if (cih == NULL)
 849                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 850
 851         pp = ip_vs_proto_get(cih->protocol);
 852         if (!pp)
 853                 return NF_ACCEPT;
 854
 855         /* Is the embedded protocol header present? */
 856         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 857                      pp->dont_defrag))
 858                 return NF_ACCEPT;
 859
 860         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
 861                       "Checking outgoing ICMP for");
 862
 863         offset += cih->ihl * 4;
 864
 865         ip_vs_fill_iphdr(AF_INET, cih, &ciph);
 866         /* The embedded headers contain source and dest in reverse order */
 867         cp = pp->conn_out_get(AF_INET, skb, &ciph, offset, 1);
 868         if (!cp)
 869                 return NF_ACCEPT;
 870
 871         snet.ip = iph->saddr;
 872         return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
 873                                     pp, offset, ihl);
 874 }
 875
 876 #ifdef CONFIG_IP_VS_IPV6
 877 static int ip_vs_out_icmp_v6(struct sk_buff *skb, int *related,
 878                              unsigned int hooknum)
 879 {
 880         struct ipv6hdr *iph;
 881         struct icmp6hdr _icmph, *ic;
 882         struct ipv6hdr  _ciph, *cih;    /* The ip header contained
 883                                            within the ICMP */
 884         struct ip_vs_iphdr ciph;
 885         struct ip_vs_conn *cp;
 886         struct ip_vs_protocol *pp;
 887         unsigned int offset;
 888         union nf_inet_addr snet;
 889
 890         *related = 1;
 891
 892         /* reassemble IP fragments */
 893         if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) {
 894                 if (ip_vs_gather_frags_v6(skb, ip_vs_defrag_user(hooknum)))
 895                         return NF_STOLEN;
 896         }
 897
 898         iph = ipv6_hdr(skb);
 899         offset = sizeof(struct ipv6hdr);
 900         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 901         if (ic == NULL)
 902                 return NF_DROP;
 903
 904         IP_VS_DBG(12, "Outgoing ICMPv6 (%d,%d) %pI6->%pI6\n",
 905                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
 906                   &iph->saddr, &iph->daddr);
 907
 908         /*
 909          * Work through seeing if this is for us.
 910          * These checks are supposed to be in an order that means easy
 911          * things are checked first to speed up processing.... however
 912          * this means that some packets will manage to get a long way
 913          * down this stack and then be rejected, but that's life.
 914          */
 915         if ((ic->icmp6_type != ICMPV6_DEST_UNREACH) &&
 916             (ic->icmp6_type != ICMPV6_PKT_TOOBIG) &&
 917             (ic->icmp6_type != ICMPV6_TIME_EXCEED)) {
 918                 *related = 0;
 919                 return NF_ACCEPT;
 920         }
 921
 922         /* Now find the contained IP header */
 923         offset += sizeof(_icmph);
 924         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 925         if (cih == NULL)
 926                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 927
 928         pp = ip_vs_proto_get(cih->nexthdr);
 929         if (!pp)
 930                 return NF_ACCEPT;
 931
 932         /* Is the embedded protocol header present? */
 933         /* TODO: we don't support fragmentation at the moment anyways */
 934         if (unlikely(cih->nexthdr == IPPROTO_FRAGMENT && pp->dont_defrag))
 935                 return NF_ACCEPT;
 936
 937         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
 938                       "Checking outgoing ICMPv6 for");
 939
 940         offset += sizeof(struct ipv6hdr);
 941
 942         ip_vs_fill_iphdr(AF_INET6, cih, &ciph);
 943         /* The embedded headers contain source and dest in reverse order */
 944         cp = pp->conn_out_get(AF_INET6, skb, &ciph, offset, 1);
 945         if (!cp)
 946                 return NF_ACCEPT;
 947
 948         ipv6_addr_copy(&snet.in6, &iph->saddr);
 949         return handle_response_icmp(AF_INET6, skb, &snet, cih->nexthdr, cp,
 950                                     pp, offset, sizeof(struct ipv6hdr));
 951 }
 952 #endif
 953
 954 /*
 955  * Check if sctp chunc is ABORT chunk
 956  */
 957 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
 958 {
 959         sctp_chunkhdr_t *sch, schunk;
 960         sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
 961                         sizeof(schunk), &schunk);
 962         if (sch == NULL)
 963                 return 0;
 964         if (sch->type == SCTP_CID_ABORT)
 965                 return 1;
 966         return 0;
 967 }
 968
 969 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
 970 {
 971         struct tcphdr _tcph, *th;
 972
 973         th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
 974         if (th == NULL)
 975                 return 0;
 976         return th->rst;
 977 }
 978
 979 /* Handle response packets: rewrite addresses and send away...
 980  * Used for NAT and local client.
 981  */
 982 static unsigned int
 983 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
 984                 struct ip_vs_conn *cp, int ihl)
 985 {
 986         struct ip_vs_protocol *pp = pd->pp;
 987         struct netns_ipvs *ipvs;
 988
 989         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Outgoing packet");
 990
 991         if (!skb_make_writable(skb, ihl))
 992                 goto drop;
 993
 994         /* mangle the packet */
 995         if (pp->snat_handler && !pp->snat_handler(skb, pp, cp))
 996                 goto drop;
 997
 998 #ifdef CONFIG_IP_VS_IPV6
 999         if (af == AF_INET6)
1000                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1001         else
1002 #endif
1003         {
1004                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1005                 ip_send_check(ip_hdr(skb));
1006         }
1007
1008         /*
1009          * nf_iterate does not expect change in the skb->dst->dev.
1010          * It looks like it is not fatal to enable this code for hooks
1011          * where our handlers are at the end of the chain list and
1012          * when all next handlers use skb->dst->dev and not outdev.
1013          * It will definitely route properly the inout NAT traffic
1014          * when multiple paths are used.
1015          */
1016
1017         /* For policy routing, packets originating from this
1018          * machine itself may be routed differently to packets
1019          * passing through.  We want this packet to be routed as
1020          * if it came from this machine itself.  So re-compute
1021          * the routing information.
1022          */
1023         ipvs = net_ipvs(skb_net(skb));
1024
1025 #ifdef CONFIG_IP_VS_IPV6
1026         if (af == AF_INET6) {
1027                 if (ipvs->sysctl_snat_reroute && ip6_route_me_harder(skb) != 0)
1028                         goto drop;
1029         } else
1030 #endif
1031                 if ((ipvs->sysctl_snat_reroute ||
1032                      skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
1033                     ip_route_me_harder(skb, RTN_LOCAL) != 0)
1034                         goto drop;
1035
1036         IP_VS_DBG_PKT(10, af, pp, skb, 0, "After SNAT");
1037
1038         ip_vs_out_stats(cp, skb);
1039         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1040         skb->ipvs_property = 1;
1041         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1042                 ip_vs_notrack(skb);
1043         else
1044                 ip_vs_update_conntrack(skb, cp, 0);
1045         ip_vs_conn_put(cp);
1046
1047         LeaveFunction(11);
1048         return NF_ACCEPT;
1049
1050 drop:
1051         ip_vs_conn_put(cp);
1052         kfree_skb(skb);
1053         LeaveFunction(11);
1054         return NF_STOLEN;
1055 }
1056
1057 /*
1058  *      Check if outgoing packet belongs to the established ip_vs_conn.
1059  */
1060 static unsigned int
1061 ip_vs_out(unsigned int hooknum, struct sk_buff *skb, int af)
1062 {
1063         struct net *net = NULL;
1064         struct ip_vs_iphdr iph;
1065         struct ip_vs_protocol *pp;
1066         struct ip_vs_proto_data *pd;
1067         struct ip_vs_conn *cp;
1068         struct netns_ipvs *ipvs;
1069
1070         EnterFunction(11);
1071
1072         /* Already marked as IPVS request or reply? */
1073         if (skb->ipvs_property)
1074                 return NF_ACCEPT;
1075
1076         /* Bad... Do not break raw sockets */
1077         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1078                      af == AF_INET)) {
1079                 struct sock *sk = skb->sk;
1080                 struct inet_sock *inet = inet_sk(skb->sk);
1081
1082                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1083                         return NF_ACCEPT;
1084         }
1085
1086         if (unlikely(!skb_dst(skb)))
1087                 return NF_ACCEPT;
1088
1089         net = skb_net(skb);
1090         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1091 #ifdef CONFIG_IP_VS_IPV6
1092         if (af == AF_INET6) {
1093                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1094                         int related;
1095                         int verdict = ip_vs_out_icmp_v6(skb, &related,
1096                                                         hooknum);
1097
1098                         if (related)
1099                                 return verdict;
1100                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1101                 }
1102         } else
1103 #endif
1104                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1105                         int related;
1106                         int verdict = ip_vs_out_icmp(skb, &related, hooknum);
1107
1108                         if (related)
1109                                 return verdict;
1110                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1111                 }
1112
1113         pd = ip_vs_proto_data_get(net, iph.protocol);
1114         if (unlikely(!pd))
1115                 return NF_ACCEPT;
1116         pp = pd->pp;
1117
1118         /* reassemble IP fragments */
1119 #ifdef CONFIG_IP_VS_IPV6
1120         if (af == AF_INET6) {
1121                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) {
1122                         if (ip_vs_gather_frags_v6(skb,
1123                                                   ip_vs_defrag_user(hooknum)))
1124                                 return NF_STOLEN;
1125                 }
1126
1127                 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1128         } else
1129 #endif
1130                 if (unlikely(ip_hdr(skb)->frag_off & htons(IP_MF|IP_OFFSET) &&
1131                              !pp->dont_defrag)) {
1132                         if (ip_vs_gather_frags(skb,
1133                                                ip_vs_defrag_user(hooknum)))
1134                                 return NF_STOLEN;
1135
1136                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1137                 }
1138
1139         /*
1140          * Check if the packet belongs to an existing entry
1141          */
1142         cp = pp->conn_out_get(af, skb, &iph, iph.len, 0);
1143         ipvs = net_ipvs(net);
1144
1145         if (likely(cp))
1146                 return handle_response(af, skb, pd, cp, iph.len);
1147         if (ipvs->sysctl_nat_icmp_send &&
1148             (pp->protocol == IPPROTO_TCP ||
1149              pp->protocol == IPPROTO_UDP ||
1150              pp->protocol == IPPROTO_SCTP)) {
1151                 __be16 _ports[2], *pptr;
1152
1153                 pptr = skb_header_pointer(skb, iph.len,
1154                                           sizeof(_ports), _ports);
1155                 if (pptr == NULL)
1156                         return NF_ACCEPT;       /* Not for me */
1157                 if (ip_vs_lookup_real_service(net, af, iph.protocol,
1158                                               &iph.saddr,
1159                                               pptr[0])) {
1160                         /*
1161                          * Notify the real server: there is no
1162                          * existing entry if it is not RST
1163                          * packet or not TCP packet.
1164                          */
1165                         if ((iph.protocol != IPPROTO_TCP &&
1166                              iph.protocol != IPPROTO_SCTP)
1167                              || ((iph.protocol == IPPROTO_TCP
1168                                   && !is_tcp_reset(skb, iph.len))
1169                                  || (iph.protocol == IPPROTO_SCTP
1170                                         && !is_sctp_abort(skb,
1171                                                 iph.len)))) {
1172 #ifdef CONFIG_IP_VS_IPV6
1173                                 if (af == AF_INET6) {
1174                                         struct net *net =
1175                                                 dev_net(skb_dst(skb)->dev);
1176
1177                                         if (!skb->dev)
1178                                                 skb->dev = net->loopback_dev;
1179                                         icmpv6_send(skb,
1180                                                     ICMPV6_DEST_UNREACH,
1181                                                     ICMPV6_PORT_UNREACH,
1182                                                     0);
1183                                 } else
1184 #endif
1185                                         icmp_send(skb,
1186                                                   ICMP_DEST_UNREACH,
1187                                                   ICMP_PORT_UNREACH, 0);
1188                                 return NF_DROP;
1189                         }
1190                 }
1191         }
1192         IP_VS_DBG_PKT(12, af, pp, skb, 0,
1193                       "ip_vs_out: packet continues traversal as normal");
1194         return NF_ACCEPT;
1195 }
1196
1197 /*
1198  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1199  *      used only for VS/NAT.
1200  *      Check if packet is reply for established ip_vs_conn.
1201  */
1202 static unsigned int
1203 ip_vs_reply4(unsigned int hooknum, struct sk_buff *skb,
1204              const struct net_device *in, const struct net_device *out,
1205              int (*okfn)(struct sk_buff *))
1206 {
1207         return ip_vs_out(hooknum, skb, AF_INET);
1208 }
1209
1210 /*
1211  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1212  *      Check if packet is reply for established ip_vs_conn.
1213  */
1214 static unsigned int
1215 ip_vs_local_reply4(unsigned int hooknum, struct sk_buff *skb,
1216                    const struct net_device *in, const struct net_device *out,
1217                    int (*okfn)(struct sk_buff *))
1218 {
1219         unsigned int verdict;
1220
1221         /* Disable BH in LOCAL_OUT until all places are fixed */
1222         local_bh_disable();
1223         verdict = ip_vs_out(hooknum, skb, AF_INET);
1224         local_bh_enable();
1225         return verdict;
1226 }
1227
1228 #ifdef CONFIG_IP_VS_IPV6
1229
1230 /*
1231  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1232  *      used only for VS/NAT.
1233  *      Check if packet is reply for established ip_vs_conn.
1234  */
1235 static unsigned int
1236 ip_vs_reply6(unsigned int hooknum, struct sk_buff *skb,
1237              const struct net_device *in, const struct net_device *out,
1238              int (*okfn)(struct sk_buff *))
1239 {
1240         return ip_vs_out(hooknum, skb, AF_INET6);
1241 }
1242
1243 /*
1244  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1245  *      Check if packet is reply for established ip_vs_conn.
1246  */
1247 static unsigned int
1248 ip_vs_local_reply6(unsigned int hooknum, struct sk_buff *skb,
1249                    const struct net_device *in, const struct net_device *out,
1250                    int (*okfn)(struct sk_buff *))
1251 {
1252         unsigned int verdict;
1253
1254         /* Disable BH in LOCAL_OUT until all places are fixed */
1255         local_bh_disable();
1256         verdict = ip_vs_out(hooknum, skb, AF_INET6);
1257         local_bh_enable();
1258         return verdict;
1259 }
1260
1261 #endif
1262
1263 /*
1264  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1265  *      Find any that might be relevant, check against existing connections,
1266  *      forward to the right destination host if relevant.
1267  *      Currently handles error types - unreachable, quench, ttl exceeded.
1268  */
1269 static int
1270 ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum)
1271 {
1272         struct net *net = NULL;
1273         struct iphdr *iph;
1274         struct icmphdr  _icmph, *ic;
1275         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1276         struct ip_vs_iphdr ciph;
1277         struct ip_vs_conn *cp;
1278         struct ip_vs_protocol *pp;
1279         struct ip_vs_proto_data *pd;
1280         unsigned int offset, ihl, verdict;
1281         union nf_inet_addr snet;
1282
1283         *related = 1;
1284
1285         /* reassemble IP fragments */
1286         if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
1287                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
1288                         return NF_STOLEN;
1289         }
1290
1291         iph = ip_hdr(skb);
1292         offset = ihl = iph->ihl * 4;
1293         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1294         if (ic == NULL)
1295                 return NF_DROP;
1296
1297         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1298                   ic->type, ntohs(icmp_id(ic)),
1299                   &iph->saddr, &iph->daddr);
1300
1301         /*
1302          * Work through seeing if this is for us.
1303          * These checks are supposed to be in an order that means easy
1304          * things are checked first to speed up processing.... however
1305          * this means that some packets will manage to get a long way
1306          * down this stack and then be rejected, but that's life.
1307          */
1308         if ((ic->type != ICMP_DEST_UNREACH) &&
1309             (ic->type != ICMP_SOURCE_QUENCH) &&
1310             (ic->type != ICMP_TIME_EXCEEDED)) {
1311                 *related = 0;
1312                 return NF_ACCEPT;
1313         }
1314
1315         /* Now find the contained IP header */
1316         offset += sizeof(_icmph);
1317         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1318         if (cih == NULL)
1319                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1320
1321         net = skb_net(skb);
1322         pd = ip_vs_proto_data_get(net, cih->protocol);
1323         if (!pd)
1324                 return NF_ACCEPT;
1325         pp = pd->pp;
1326
1327         /* Is the embedded protocol header present? */
1328         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1329                      pp->dont_defrag))
1330                 return NF_ACCEPT;
1331
1332         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1333                       "Checking incoming ICMP for");
1334
1335         offset += cih->ihl * 4;
1336
1337         ip_vs_fill_iphdr(AF_INET, cih, &ciph);
1338         /* The embedded headers contain source and dest in reverse order */
1339         cp = pp->conn_in_get(AF_INET, skb, &ciph, offset, 1);
1340         if (!cp) {
1341                 /* The packet could also belong to a local client */
1342                 cp = pp->conn_out_get(AF_INET, skb, &ciph, offset, 1);
1343                 if (cp) {
1344                         snet.ip = iph->saddr;
1345                         return handle_response_icmp(AF_INET, skb, &snet,
1346                                                     cih->protocol, cp, pp,
1347                                                     offset, ihl);
1348                 }
1349                 return NF_ACCEPT;
1350         }
1351
1352         verdict = NF_DROP;
1353
1354         /* Ensure the checksum is correct */
1355         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1356                 /* Failed checksum! */
1357                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1358                           &iph->saddr);
1359                 goto out;
1360         }
1361
1362         /* do the statistics and put it back */
1363         ip_vs_in_stats(cp, skb);
1364         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol)
1365                 offset += 2 * sizeof(__u16);
1366         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset);
1367         /* LOCALNODE from FORWARD hook is not supported */
1368         if (verdict == NF_ACCEPT && hooknum == NF_INET_FORWARD &&
1369             skb_rtable(skb)->rt_flags & RTCF_LOCAL) {
1370                 IP_VS_DBG(1, "%s(): "
1371                           "local delivery to %pI4 but in FORWARD\n",
1372                           __func__, &skb_rtable(skb)->rt_dst);
1373                 verdict = NF_DROP;
1374         }
1375
1376   out:
1377         __ip_vs_conn_put(cp);
1378
1379         return verdict;
1380 }
1381
1382 #ifdef CONFIG_IP_VS_IPV6
1383 static int
1384 ip_vs_in_icmp_v6(struct sk_buff *skb, int *related, unsigned int hooknum)
1385 {
1386         struct net *net = NULL;
1387         struct ipv6hdr *iph;
1388         struct icmp6hdr _icmph, *ic;
1389         struct ipv6hdr  _ciph, *cih;    /* The ip header contained
1390                                            within the ICMP */
1391         struct ip_vs_iphdr ciph;
1392         struct ip_vs_conn *cp;
1393         struct ip_vs_protocol *pp;
1394         struct ip_vs_proto_data *pd;
1395         unsigned int offset, verdict;
1396         union nf_inet_addr snet;
1397         struct rt6_info *rt;
1398
1399         *related = 1;
1400
1401         /* reassemble IP fragments */
1402         if (ipv6_hdr(skb)->nexthdr == IPPROTO_FRAGMENT) {
1403                 if (ip_vs_gather_frags_v6(skb, ip_vs_defrag_user(hooknum)))
1404                         return NF_STOLEN;
1405         }
1406
1407         iph = ipv6_hdr(skb);
1408         offset = sizeof(struct ipv6hdr);
1409         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1410         if (ic == NULL)
1411                 return NF_DROP;
1412
1413         IP_VS_DBG(12, "Incoming ICMPv6 (%d,%d) %pI6->%pI6\n",
1414                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1415                   &iph->saddr, &iph->daddr);
1416
1417         /*
1418          * Work through seeing if this is for us.
1419          * These checks are supposed to be in an order that means easy
1420          * things are checked first to speed up processing.... however
1421          * this means that some packets will manage to get a long way
1422          * down this stack and then be rejected, but that's life.
1423          */
1424         if ((ic->icmp6_type != ICMPV6_DEST_UNREACH) &&
1425             (ic->icmp6_type != ICMPV6_PKT_TOOBIG) &&
1426             (ic->icmp6_type != ICMPV6_TIME_EXCEED)) {
1427                 *related = 0;
1428                 return NF_ACCEPT;
1429         }
1430
1431         /* Now find the contained IP header */
1432         offset += sizeof(_icmph);
1433         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1434         if (cih == NULL)
1435                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1436
1437         net = skb_net(skb);
1438         pd = ip_vs_proto_data_get(net, cih->nexthdr);
1439         if (!pd)
1440                 return NF_ACCEPT;
1441         pp = pd->pp;
1442
1443         /* Is the embedded protocol header present? */
1444         /* TODO: we don't support fragmentation at the moment anyways */
1445         if (unlikely(cih->nexthdr == IPPROTO_FRAGMENT && pp->dont_defrag))
1446                 return NF_ACCEPT;
1447
1448         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
1449                       "Checking incoming ICMPv6 for");
1450
1451         offset += sizeof(struct ipv6hdr);
1452
1453         ip_vs_fill_iphdr(AF_INET6, cih, &ciph);
1454         /* The embedded headers contain source and dest in reverse order */
1455         cp = pp->conn_in_get(AF_INET6, skb, &ciph, offset, 1);
1456         if (!cp) {
1457                 /* The packet could also belong to a local client */
1458                 cp = pp->conn_out_get(AF_INET6, skb, &ciph, offset, 1);
1459                 if (cp) {
1460                         ipv6_addr_copy(&snet.in6, &iph->saddr);
1461                         return handle_response_icmp(AF_INET6, skb, &snet,
1462                                                     cih->nexthdr,
1463                                                     cp, pp, offset,
1464                                                     sizeof(struct ipv6hdr));
1465                 }
1466                 return NF_ACCEPT;
1467         }
1468
1469         verdict = NF_DROP;
1470
1471         /* do the statistics and put it back */
1472         ip_vs_in_stats(cp, skb);
1473         if (IPPROTO_TCP == cih->nexthdr || IPPROTO_UDP == cih->nexthdr ||
1474             IPPROTO_SCTP == cih->nexthdr)
1475                 offset += 2 * sizeof(__u16);
1476         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset);
1477         /* LOCALNODE from FORWARD hook is not supported */
1478         if (verdict == NF_ACCEPT && hooknum == NF_INET_FORWARD &&
1479             (rt = (struct rt6_info *) skb_dst(skb)) &&
1480             rt->rt6i_dev && rt->rt6i_dev->flags & IFF_LOOPBACK) {
1481                 IP_VS_DBG(1, "%s(): "
1482                           "local delivery to %pI6 but in FORWARD\n",
1483                           __func__, &rt->rt6i_dst);
1484                 verdict = NF_DROP;
1485         }
1486
1487         __ip_vs_conn_put(cp);
1488
1489         return verdict;
1490 }
1491 #endif
1492
1493
1494 /*
1495  *      Check if it's for virtual services, look it up,
1496  *      and send it on its way...
1497  */
1498 static unsigned int
1499 ip_vs_in(unsigned int hooknum, struct sk_buff *skb, int af)
1500 {
1501         struct net *net;
1502         struct ip_vs_iphdr iph;
1503         struct ip_vs_protocol *pp;
1504         struct ip_vs_proto_data *pd;
1505         struct ip_vs_conn *cp;
1506         int ret, restart, pkts;
1507         struct netns_ipvs *ipvs;
1508
1509         /* Already marked as IPVS request or reply? */
1510         if (skb->ipvs_property)
1511                 return NF_ACCEPT;
1512
1513         /*
1514          *      Big tappo:
1515          *      - remote client: only PACKET_HOST
1516          *      - route: used for struct net when skb->dev is unset
1517          */
1518         if (unlikely((skb->pkt_type != PACKET_HOST &&
1519                       hooknum != NF_INET_LOCAL_OUT) ||
1520                      !skb_dst(skb))) {
1521                 ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1522                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1523                               " ignored in hook %u\n",
1524                               skb->pkt_type, iph.protocol,
1525                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1526                 return NF_ACCEPT;
1527         }
1528         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1529
1530         /* Bad... Do not break raw sockets */
1531         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1532                      af == AF_INET)) {
1533                 struct sock *sk = skb->sk;
1534                 struct inet_sock *inet = inet_sk(skb->sk);
1535
1536                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1537                         return NF_ACCEPT;
1538         }
1539
1540 #ifdef CONFIG_IP_VS_IPV6
1541         if (af == AF_INET6) {
1542                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1543                         int related;
1544                         int verdict = ip_vs_in_icmp_v6(skb, &related, hooknum);
1545
1546                         if (related)
1547                                 return verdict;
1548                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1549                 }
1550         } else
1551 #endif
1552                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1553                         int related;
1554                         int verdict = ip_vs_in_icmp(skb, &related, hooknum);
1555
1556                         if (related)
1557                                 return verdict;
1558                         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
1559                 }
1560
1561         net = skb_net(skb);
1562         /* Protocol supported? */
1563         pd = ip_vs_proto_data_get(net, iph.protocol);
1564         if (unlikely(!pd))
1565                 return NF_ACCEPT;
1566         pp = pd->pp;
1567         /*
1568          * Check if the packet belongs to an existing connection entry
1569          */
1570         cp = pp->conn_in_get(af, skb, &iph, iph.len, 0);
1571
1572         if (unlikely(!cp)) {
1573                 int v;
1574
1575                 if (!pp->conn_schedule(af, skb, pd, &v, &cp))
1576                         return v;
1577         }
1578
1579         if (unlikely(!cp)) {
1580                 /* sorry, all this trouble for a no-hit :) */
1581                 IP_VS_DBG_PKT(12, af, pp, skb, 0,
1582                               "ip_vs_in: packet continues traversal as normal");
1583                 return NF_ACCEPT;
1584         }
1585
1586         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Incoming packet");
1587         net = skb_net(skb);
1588         ipvs = net_ipvs(net);
1589         /* Check the server status */
1590         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1591                 /* the destination server is not available */
1592
1593                 if (ipvs->sysctl_expire_nodest_conn) {
1594                         /* try to expire the connection immediately */
1595                         ip_vs_conn_expire_now(cp);
1596                 }
1597                 /* don't restart its timer, and silently
1598                    drop the packet. */
1599                 __ip_vs_conn_put(cp);
1600                 return NF_DROP;
1601         }
1602
1603         ip_vs_in_stats(cp, skb);
1604         restart = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1605         if (cp->packet_xmit)
1606                 ret = cp->packet_xmit(skb, cp, pp);
1607                 /* do not touch skb anymore */
1608         else {
1609                 IP_VS_DBG_RL("warning: packet_xmit is null");
1610                 ret = NF_ACCEPT;
1611         }
1612
1613         /* Increase its packet counter and check if it is needed
1614          * to be synchronized
1615          *
1616          * Sync connection if it is about to close to
1617          * encorage the standby servers to update the connections timeout
1618          *
1619          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1620          */
1621
1622         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
1623                 pkts = ipvs->sysctl_sync_threshold[0];
1624         else
1625                 pkts = atomic_add_return(1, &cp->in_pkts);
1626
1627         if ((ipvs->sync_state & IP_VS_STATE_MASTER) &&
1628             cp->protocol == IPPROTO_SCTP) {
1629                 if ((cp->state == IP_VS_SCTP_S_ESTABLISHED &&
1630                         (pkts % ipvs->sysctl_sync_threshold[1]
1631                          == ipvs->sysctl_sync_threshold[0])) ||
1632                                 (cp->old_state != cp->state &&
1633                                  ((cp->state == IP_VS_SCTP_S_CLOSED) ||
1634                                   (cp->state == IP_VS_SCTP_S_SHUT_ACK_CLI) ||
1635                                   (cp->state == IP_VS_SCTP_S_SHUT_ACK_SER)))) {
1636                         ip_vs_sync_conn(net, cp);
1637                         goto out;
1638                 }
1639         }
1640
1641         /* Keep this block last: TCP and others with pp->num_states <= 1 */
1642         else if ((ipvs->sync_state & IP_VS_STATE_MASTER) &&
1643             (((cp->protocol != IPPROTO_TCP ||
1644                cp->state == IP_VS_TCP_S_ESTABLISHED) &&
1645               (pkts % ipvs->sysctl_sync_threshold[1]
1646                == ipvs->sysctl_sync_threshold[0])) ||
1647              ((cp->protocol == IPPROTO_TCP) && (cp->old_state != cp->state) &&
1648               ((cp->state == IP_VS_TCP_S_FIN_WAIT) ||
1649                (cp->state == IP_VS_TCP_S_CLOSE) ||
1650                (cp->state == IP_VS_TCP_S_CLOSE_WAIT) ||
1651                (cp->state == IP_VS_TCP_S_TIME_WAIT)))))
1652                 ip_vs_sync_conn(net, cp);
1653 out:
1654         cp->old_state = cp->state;
1655
1656         ip_vs_conn_put(cp);
1657         return ret;
1658 }
1659
1660 /*
1661  *      AF_INET handler in NF_INET_LOCAL_IN chain
1662  *      Schedule and forward packets from remote clients
1663  */
1664 static unsigned int
1665 ip_vs_remote_request4(unsigned int hooknum, struct sk_buff *skb,
1666                       const struct net_device *in,
1667                       const struct net_device *out,
1668                       int (*okfn)(struct sk_buff *))
1669 {
1670         return ip_vs_in(hooknum, skb, AF_INET);
1671 }
1672
1673 /*
1674  *      AF_INET handler in NF_INET_LOCAL_OUT chain
1675  *      Schedule and forward packets from local clients
1676  */
1677 static unsigned int
1678 ip_vs_local_request4(unsigned int hooknum, struct sk_buff *skb,
1679                      const struct net_device *in, const struct net_device *out,
1680                      int (*okfn)(struct sk_buff *))
1681 {
1682         unsigned int verdict;
1683
1684         /* Disable BH in LOCAL_OUT until all places are fixed */
1685         local_bh_disable();
1686         verdict = ip_vs_in(hooknum, skb, AF_INET);
1687         local_bh_enable();
1688         return verdict;
1689 }
1690
1691 #ifdef CONFIG_IP_VS_IPV6
1692
1693 /*
1694  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
1695  *      Schedule and forward packets from remote clients
1696  */
1697 static unsigned int
1698 ip_vs_remote_request6(unsigned int hooknum, struct sk_buff *skb,
1699                       const struct net_device *in,
1700                       const struct net_device *out,
1701                       int (*okfn)(struct sk_buff *))
1702 {
1703         return ip_vs_in(hooknum, skb, AF_INET6);
1704 }
1705
1706 /*
1707  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
1708  *      Schedule and forward packets from local clients
1709  */
1710 static unsigned int
1711 ip_vs_local_request6(unsigned int hooknum, struct sk_buff *skb,
1712                      const struct net_device *in, const struct net_device *out,
1713                      int (*okfn)(struct sk_buff *))
1714 {
1715         unsigned int verdict;
1716
1717         /* Disable BH in LOCAL_OUT until all places are fixed */
1718         local_bh_disable();
1719         verdict = ip_vs_in(hooknum, skb, AF_INET6);
1720         local_bh_enable();
1721         return verdict;
1722 }
1723
1724 #endif
1725
1726
1727 /*
1728  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1729  *      related packets destined for 0.0.0.0/0.
1730  *      When fwmark-based virtual service is used, such as transparent
1731  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
1732  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1733  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1734  *      and send them to ip_vs_in_icmp.
1735  */
1736 static unsigned int
1737 ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff *skb,
1738                    const struct net_device *in, const struct net_device *out,
1739                    int (*okfn)(struct sk_buff *))
1740 {
1741         int r;
1742
1743         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1744                 return NF_ACCEPT;
1745
1746         return ip_vs_in_icmp(skb, &r, hooknum);
1747 }
1748
1749 #ifdef CONFIG_IP_VS_IPV6
1750 static unsigned int
1751 ip_vs_forward_icmp_v6(unsigned int hooknum, struct sk_buff *skb,
1752                       const struct net_device *in, const struct net_device *out,
1753                       int (*okfn)(struct sk_buff *))
1754 {
1755         int r;
1756
1757         if (ipv6_hdr(skb)->nexthdr != IPPROTO_ICMPV6)
1758                 return NF_ACCEPT;
1759
1760         return ip_vs_in_icmp_v6(skb, &r, hooknum);
1761 }
1762 #endif
1763
1764
1765 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
1766         /* After packet filtering, change source only for VS/NAT */
1767         {
1768                 .hook           = ip_vs_reply4,
1769                 .owner          = THIS_MODULE,
1770                 .pf             = PF_INET,
1771                 .hooknum        = NF_INET_LOCAL_IN,
1772                 .priority       = 99,
1773         },
1774         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1775          * or VS/NAT(change destination), so that filtering rules can be
1776          * applied to IPVS. */
1777         {
1778                 .hook           = ip_vs_remote_request4,
1779                 .owner          = THIS_MODULE,
1780                 .pf             = PF_INET,
1781                 .hooknum        = NF_INET_LOCAL_IN,
1782                 .priority       = 101,
1783         },
1784         /* Before ip_vs_in, change source only for VS/NAT */
1785         {
1786                 .hook           = ip_vs_local_reply4,
1787                 .owner          = THIS_MODULE,
1788                 .pf             = PF_INET,
1789                 .hooknum        = NF_INET_LOCAL_OUT,
1790                 .priority       = -99,
1791         },
1792         /* After mangle, schedule and forward local requests */
1793         {
1794                 .hook           = ip_vs_local_request4,
1795                 .owner          = THIS_MODULE,
1796                 .pf             = PF_INET,
1797                 .hooknum        = NF_INET_LOCAL_OUT,
1798                 .priority       = -98,
1799         },
1800         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1801          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1802         {
1803                 .hook           = ip_vs_forward_icmp,
1804                 .owner          = THIS_MODULE,
1805                 .pf             = PF_INET,
1806                 .hooknum        = NF_INET_FORWARD,
1807                 .priority       = 99,
1808         },
1809         /* After packet filtering, change source only for VS/NAT */
1810         {
1811                 .hook           = ip_vs_reply4,
1812                 .owner          = THIS_MODULE,
1813                 .pf             = PF_INET,
1814                 .hooknum        = NF_INET_FORWARD,
1815                 .priority       = 100,
1816         },
1817 #ifdef CONFIG_IP_VS_IPV6
1818         /* After packet filtering, change source only for VS/NAT */
1819         {
1820                 .hook           = ip_vs_reply6,
1821                 .owner          = THIS_MODULE,
1822                 .pf             = PF_INET6,
1823                 .hooknum        = NF_INET_LOCAL_IN,
1824                 .priority       = 99,
1825         },
1826         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1827          * or VS/NAT(change destination), so that filtering rules can be
1828          * applied to IPVS. */
1829         {
1830                 .hook           = ip_vs_remote_request6,
1831                 .owner          = THIS_MODULE,
1832                 .pf             = PF_INET6,
1833                 .hooknum        = NF_INET_LOCAL_IN,
1834                 .priority       = 101,
1835         },
1836         /* Before ip_vs_in, change source only for VS/NAT */
1837         {
1838                 .hook           = ip_vs_local_reply6,
1839                 .owner          = THIS_MODULE,
1840                 .pf             = PF_INET,
1841                 .hooknum        = NF_INET_LOCAL_OUT,
1842                 .priority       = -99,
1843         },
1844         /* After mangle, schedule and forward local requests */
1845         {
1846                 .hook           = ip_vs_local_request6,
1847                 .owner          = THIS_MODULE,
1848                 .pf             = PF_INET6,
1849                 .hooknum        = NF_INET_LOCAL_OUT,
1850                 .priority       = -98,
1851         },
1852         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1853          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1854         {
1855                 .hook           = ip_vs_forward_icmp_v6,
1856                 .owner          = THIS_MODULE,
1857                 .pf             = PF_INET6,
1858                 .hooknum        = NF_INET_FORWARD,
1859                 .priority       = 99,
1860         },
1861         /* After packet filtering, change source only for VS/NAT */
1862         {
1863                 .hook           = ip_vs_reply6,
1864                 .owner          = THIS_MODULE,
1865                 .pf             = PF_INET6,
1866                 .hooknum        = NF_INET_FORWARD,
1867                 .priority       = 100,
1868         },
1869 #endif
1870 };
1871 /*
1872  *      Initialize IP Virtual Server netns mem.
1873  */
1874 static int __net_init __ip_vs_init(struct net *net)
1875 {
1876         struct netns_ipvs *ipvs;
1877
1878         if (!net_eq(net, &init_net)) {
1879                 pr_err("The final patch for enabling netns is missing\n");
1880                 return -EPERM;
1881         }
1882         ipvs = net_generic(net, ip_vs_net_id);
1883         if (ipvs == NULL) {
1884                 pr_err("%s(): no memory.\n", __func__);
1885                 return -ENOMEM;
1886         }
1887         ipvs->net = net;
1888         /* Counters used for creating unique names */
1889         ipvs->gen = atomic_read(&ipvs_netns_cnt);
1890         atomic_inc(&ipvs_netns_cnt);
1891         net->ipvs = ipvs;
1892         printk(KERN_INFO "IPVS: Creating netns size=%lu id=%d\n",
1893                          sizeof(struct netns_ipvs), ipvs->gen);
1894         return 0;
1895 }
1896
1897 static void __net_exit __ip_vs_cleanup(struct net *net)
1898 {
1899         struct netns_ipvs *ipvs = net_ipvs(net);
1900
1901         IP_VS_DBG(10, "ipvs netns %d released\n", ipvs->gen);
1902 }
1903
1904 static struct pernet_operations ipvs_core_ops = {
1905         .init = __ip_vs_init,
1906         .exit = __ip_vs_cleanup,
1907         .id   = &ip_vs_net_id,
1908         .size = sizeof(struct netns_ipvs),
1909 };
1910
1911 /*
1912  *      Initialize IP Virtual Server
1913  */
1914 static int __init ip_vs_init(void)
1915 {
1916         int ret;
1917
1918         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
1919         if (ret < 0)
1920                 return ret;
1921
1922         ip_vs_estimator_init();
1923         ret = ip_vs_control_init();
1924         if (ret < 0) {
1925                 pr_err("can't setup control.\n");
1926                 goto cleanup_estimator;
1927         }
1928
1929         ip_vs_protocol_init();
1930
1931         ret = ip_vs_app_init();
1932         if (ret < 0) {
1933                 pr_err("can't setup application helper.\n");
1934                 goto cleanup_protocol;
1935         }
1936
1937         ret = ip_vs_conn_init();
1938         if (ret < 0) {
1939                 pr_err("can't setup connection table.\n");
1940                 goto cleanup_app;
1941         }
1942
1943         ret = ip_vs_sync_init();
1944         if (ret < 0) {
1945                 pr_err("can't setup sync data.\n");
1946                 goto cleanup_conn;
1947         }
1948
1949         ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
1950         if (ret < 0) {
1951                 pr_err("can't register hooks.\n");
1952                 goto cleanup_sync;
1953         }
1954
1955         pr_info("ipvs loaded.\n");
1956         return ret;
1957
1958 cleanup_sync:
1959         ip_vs_sync_cleanup();
1960   cleanup_conn:
1961         ip_vs_conn_cleanup();
1962   cleanup_app:
1963         ip_vs_app_cleanup();
1964   cleanup_protocol:
1965         ip_vs_protocol_cleanup();
1966         ip_vs_control_cleanup();
1967   cleanup_estimator:
1968         ip_vs_estimator_cleanup();
1969         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
1970         return ret;
1971 }
1972
1973 static void __exit ip_vs_cleanup(void)
1974 {
1975         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
1976         ip_vs_sync_cleanup();
1977         ip_vs_conn_cleanup();
1978         ip_vs_app_cleanup();
1979         ip_vs_protocol_cleanup();
1980         ip_vs_control_cleanup();
1981         ip_vs_estimator_cleanup();
1982         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
1983         pr_info("ipvs unloaded.\n");
1984 }
1985
1986 module_init(ip_vs_init);
1987 module_exit(ip_vs_cleanup);
1988 MODULE_LICENSE("GPL");