net/ipv4/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  * Version:     $Id: ip_vs_core.c,v 1.31.2.5 2003/07/29 14:37:12 wensong Exp $
   9  *
  10  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
  11  *              Peter Kese <peter.kese@ijs.si>
  12  *              Julian Anastasov <ja@ssi.bg>
  13  *
  14  *              This program is free software; you can redistribute it and/or
  15  *              modify it under the terms of the GNU General Public License
  16  *              as published by the Free Software Foundation; either version
  17  *              2 of the License, or (at your option) any later version.
  18  *
  19  * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
  20  * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
  21  * and others.
  22  *
  23  * Changes:
  24  *
  25  */
  26
  27 #include <linux/config.h>
  28 #include <linux/module.h>
  29 #include <linux/types.h>
  30 #include <linux/kernel.h>
  31 #include <linux/errno.h>
  32 #include <linux/ip.h>
  33 #include <linux/tcp.h>
  34 #include <linux/icmp.h>
  35
  36 #include <net/ip.h>
  37 #include <net/tcp.h>
  38 #include <net/udp.h>
  39 #include <net/icmp.h>                   /* for icmp_send */
  40 #include <net/route.h>
  41
  42 #include <linux/netfilter.h>
  43 #include <linux/netfilter_ipv4.h>
  44
  45 #include <net/ip_vs.h>
  46
  47
  48 EXPORT_SYMBOL(register_ip_vs_scheduler);
  49 EXPORT_SYMBOL(unregister_ip_vs_scheduler);
  50 EXPORT_SYMBOL(ip_vs_skb_replace);
  51 EXPORT_SYMBOL(ip_vs_proto_name);
  52 EXPORT_SYMBOL(ip_vs_conn_new);
  53 EXPORT_SYMBOL(ip_vs_conn_in_get);
  54 EXPORT_SYMBOL(ip_vs_conn_out_get);
  55 EXPORT_SYMBOL(ip_vs_conn_listen);
  56 EXPORT_SYMBOL(ip_vs_conn_put);
  57 #ifdef CONFIG_IP_VS_DEBUG
  58 EXPORT_SYMBOL(ip_vs_get_debug_level);
  59 #endif
  60 EXPORT_SYMBOL(check_for_ip_vs_out);
  61
  62
  63 /* ID used in ICMP lookups */
  64 #define icmp_id(icmph)          ((icmph->un).echo.id)
  65
  66 const char *ip_vs_proto_name(unsigned proto)
  67 {
  68         static char buf[20];
  69
  70         switch (proto) {
  71         case IPPROTO_IP:
  72                 return "IP";
  73         case IPPROTO_UDP:
  74                 return "UDP";
  75         case IPPROTO_TCP:
  76                 return "TCP";
  77         case IPPROTO_ICMP:
  78                 return "ICMP";
  79         default:
  80                 sprintf(buf, "IP_%d", proto);
  81                 return buf;
  82         }
  83 }
  84
  85
  86 static inline void
  87 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
  88 {
  89         struct ip_vs_dest *dest = cp->dest;
  90         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
  91                 spin_lock(&dest->stats.lock);
  92                 dest->stats.inpkts++;
  93                 dest->stats.inbytes += skb->len;
  94                 spin_unlock(&dest->stats.lock);
  95
  96                 spin_lock(&dest->svc->stats.lock);
  97                 dest->svc->stats.inpkts++;
  98                 dest->svc->stats.inbytes += skb->len;
  99                 spin_unlock(&dest->svc->stats.lock);
 100
 101                 spin_lock(&ip_vs_stats.lock);
 102                 ip_vs_stats.inpkts++;
 103                 ip_vs_stats.inbytes += skb->len;
 104                 spin_unlock(&ip_vs_stats.lock);
 105         }
 106 }
 107
 108
 109 static inline void
 110 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 111 {
 112         struct ip_vs_dest *dest = cp->dest;
 113         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 114                 spin_lock(&dest->stats.lock);
 115                 dest->stats.outpkts++;
 116                 dest->stats.outbytes += skb->len;
 117                 spin_unlock(&dest->stats.lock);
 118
 119                 spin_lock(&dest->svc->stats.lock);
 120                 dest->svc->stats.outpkts++;
 121                 dest->svc->stats.outbytes += skb->len;
 122                 spin_unlock(&dest->svc->stats.lock);
 123
 124                 spin_lock(&ip_vs_stats.lock);
 125                 ip_vs_stats.outpkts++;
 126                 ip_vs_stats.outbytes += skb->len;
 127                 spin_unlock(&ip_vs_stats.lock);
 128         }
 129 }
 130
 131
 132 static inline void
 133 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
 134 {
 135         spin_lock(&cp->dest->stats.lock);
 136         cp->dest->stats.conns++;
 137         spin_unlock(&cp->dest->stats.lock);
 138
 139         spin_lock(&svc->stats.lock);
 140         svc->stats.conns++;
 141         spin_unlock(&svc->stats.lock);
 142
 143         spin_lock(&ip_vs_stats.lock);
 144         ip_vs_stats.conns++;
 145         spin_unlock(&ip_vs_stats.lock);
 146 }
 147
 148 /*
 149  *  IPVS persistent scheduling function
 150  *  It creates a connection entry according to its template if exists,
 151  *  or selects a server and creates a connection entry plus a template.
 152  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
 153  */
 154 static struct ip_vs_conn *
 155 ip_vs_sched_persist(struct ip_vs_service *svc, struct iphdr *iph)
 156 {
 157         struct ip_vs_conn *cp = NULL;
 158         struct ip_vs_dest *dest;
 159         const __u16 *portp;
 160         struct ip_vs_conn *ct;
 161         __u16  dport;    /* destination port to forward */
 162         __u32  snet;     /* source network of the client, after masking */
 163
 164         portp = (__u16 *)&(((char *)iph)[iph->ihl*4]);
 165
 166         /* Mask saddr with the netmask to adjust template granularity */
 167         snet = iph->saddr & svc->netmask;
 168
 169         IP_VS_DBG(6, "P-schedule: src %u.%u.%u.%u:%u dest %u.%u.%u.%u:%u "
 170                   "mnet %u.%u.%u.%u\n",
 171                   NIPQUAD(iph->saddr), ntohs(portp[0]),
 172                   NIPQUAD(iph->daddr), ntohs(portp[1]),
 173                   NIPQUAD(snet));
 174
 175         /*
 176          * As far as we know, FTP is a very complicated network protocol, and
 177          * it uses control connection and data connections. For active FTP,
 178          * FTP server initialize data connection to the client, its source port
 179          * is often 20. For passive FTP, FTP server tells the clients the port
 180          * that it passively listens to,  and the client issues the data
 181          * connection. In the tunneling or direct routing mode, the load
 182          * balancer is on the client-to-server half of connection, the port
 183          * number is unknown to the load balancer. So, a conn template like
 184          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
 185          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
 186          * is created for other persistent services.
 187          */
 188         if (portp[1] == svc->port) {
 189                 /* Check if a template already exists */
 190                 if (svc->port != FTPPORT)
 191                         ct = ip_vs_ct_in_get(iph->protocol, snet, 0,
 192                                                iph->daddr, portp[1]);
 193                 else
 194                         ct = ip_vs_ct_in_get(iph->protocol, snet, 0,
 195                                                iph->daddr, 0);
 196
 197                 if (!ct || !ip_vs_check_template(ct)) {
 198                         /*
 199                          * No template found or the dest of the connection
 200                          * template is not available.
 201                          */
 202                         dest = svc->scheduler->schedule(svc, iph);
 203                         if (dest == NULL) {
 204                                 IP_VS_DBG(1, "P-schedule: no dest found.\n");
 205                                 return NULL;
 206                         }
 207
 208                         /*
 209                          * Create a template like <protocol,caddr,0,
 210                          * vaddr,vport,daddr,dport> for non-ftp service,
 211                          * and <protocol,caddr,0,vaddr,0,daddr,0>
 212                          * for ftp service.
 213                          */
 214                         if (svc->port != FTPPORT)
 215                                 ct = ip_vs_conn_new(iph->protocol,
 216                                                     snet, 0,
 217                                                     iph->daddr, portp[1],
 218                                                     dest->addr, dest->port,
 219                                                     IP_VS_CONN_F_TEMPLATE,
 220                                                     dest);
 221                         else
 222                                 ct = ip_vs_conn_new(iph->protocol,
 223                                                     snet, 0,
 224                                                     iph->daddr, 0,
 225                                                     dest->addr, 0,
 226                                                     IP_VS_CONN_F_TEMPLATE,
 227                                                     dest);
 228                         if (ct == NULL)
 229                                 return NULL;
 230
 231                         ct->timeout = svc->timeout;
 232                 } else {
 233                         /* set destination with the found template */
 234                         dest = ct->dest;
 235                 }
 236                 dport = dest->port;
 237         } else {
 238                 /*
 239                  * Note: persistent fwmark-based services and persistent
 240                  * port zero service are handled here.
 241                  * fwmark template: <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
 242                  * port zero template: <protocol,caddr,0,vaddr,0,daddr,0>
 243                  */
 244                 if (svc->fwmark)
 245                         ct = ip_vs_ct_in_get(IPPROTO_IP, snet, 0,
 246                                                htonl(svc->fwmark), 0);
 247                 else
 248                         ct = ip_vs_ct_in_get(iph->protocol, snet, 0,
 249                                                iph->daddr, 0);
 250
 251                 if (!ct || !ip_vs_check_template(ct)) {
 252                         /*
 253                          * If it is not persistent port zero, return NULL,
 254                          * otherwise create a connection template.
 255                          */
 256                         if (svc->port)
 257                                 return NULL;
 258
 259                         dest = svc->scheduler->schedule(svc, iph);
 260                         if (dest == NULL) {
 261                                 IP_VS_DBG(1, "P-schedule: no dest found.\n");
 262                                 return NULL;
 263                         }
 264
 265                         /*
 266                          * Create a template according to the service
 267                          */
 268                         if (svc->fwmark)
 269                                 ct = ip_vs_conn_new(IPPROTO_IP,
 270                                                     snet, 0,
 271                                                     htonl(svc->fwmark), 0,
 272                                                     dest->addr, 0,
 273                                                     IP_VS_CONN_F_TEMPLATE,
 274                                                     dest);
 275                         else
 276                                 ct = ip_vs_conn_new(iph->protocol,
 277                                                     snet, 0,
 278                                                     iph->daddr, 0,
 279                                                     dest->addr, 0,
 280                                                     IP_VS_CONN_F_TEMPLATE,
 281                                                     dest);
 282                         if (ct == NULL)
 283                                 return NULL;
 284
 285                         ct->timeout = svc->timeout;
 286                 } else {
 287                         /* set destination with the found template */
 288                         dest = ct->dest;
 289                 }
 290                 dport = portp[1];
 291         }
 292
 293         /*
 294          *    Create a new connection according to the template
 295          */
 296         cp = ip_vs_conn_new(iph->protocol,
 297                             iph->saddr, portp[0],
 298                             iph->daddr, portp[1],
 299                             dest->addr, dport,
 300                             0,
 301                             dest);
 302         if (cp == NULL) {
 303                 ip_vs_conn_put(ct);
 304                 return NULL;
 305         }
 306
 307         /*
 308          *    Increase the inactive connection counter
 309          *    because it is in Syn-Received
 310          *    state (inactive) when the connection is created.
 311          */
 312         atomic_inc(&dest->inactconns);
 313
 314         /*
 315          *    Add its control
 316          */
 317         ip_vs_control_add(cp, ct);
 318
 319         ip_vs_conn_put(ct);
 320         return cp;
 321 }
 322
 323
 324 /*
 325  *  IPVS main scheduling function
 326  *  It selects a server according to the virtual service, and
 327  *  creates a connection entry.
 328  */
 329 static struct ip_vs_conn *
 330 ip_vs_schedule(struct ip_vs_service *svc, struct iphdr *iph)
 331 {
 332         struct ip_vs_conn *cp = NULL;
 333         struct ip_vs_dest *dest;
 334         const __u16 *portp;
 335
 336         /*
 337          *    Persistent service
 338          */
 339         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 340                 return ip_vs_sched_persist(svc, iph);
 341
 342         /*
 343          *    Non-persistent service
 344          */
 345         portp = (__u16 *)&(((char *)iph)[iph->ihl*4]);
 346         if (!svc->fwmark && portp[1] != svc->port) {
 347                 if (!svc->port)
 348                         IP_VS_ERR("Schedule: port zero only supported "
 349                                   "in persistent services, "
 350                                   "check your ipvs configuration\n");
 351                 return NULL;
 352         }
 353
 354         dest = svc->scheduler->schedule(svc, iph);
 355         if (dest == NULL) {
 356                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 357                 return NULL;
 358         }
 359
 360         /*
 361          *    Create a connection entry.
 362          */
 363         cp = ip_vs_conn_new(iph->protocol,
 364                             iph->saddr, portp[0],
 365                             iph->daddr, portp[1],
 366                             dest->addr, dest->port?dest->port:portp[1],
 367                             0,
 368                             dest);
 369         if (cp == NULL)
 370                 return NULL;
 371
 372         /*
 373          *    Increase the inactive connection counter because it is in
 374          *    Syn-Received state (inactive) when the connection is created.
 375          */
 376         atomic_inc(&dest->inactconns);
 377
 378         IP_VS_DBG(6, "Schedule fwd:%c s:%s c:%u.%u.%u.%u:%u v:%u.%u.%u.%u:%u "
 379                   "d:%u.%u.%u.%u:%u flg:%X cnt:%d\n",
 380                   ip_vs_fwd_tag(cp), ip_vs_state_name(cp->state),
 381                   NIPQUAD(cp->caddr), ntohs(cp->cport),
 382                   NIPQUAD(cp->vaddr), ntohs(cp->vport),
 383                   NIPQUAD(cp->daddr), ntohs(cp->dport),
 384                   cp->flags, atomic_read(&cp->refcnt));
 385
 386         return cp;
 387 }
 388
 389
 390 /*
 391  *  Pass or drop the packet.
 392  *  Called by ip_vs_in, when the virtual service is available but
 393  *  no destination is available for a new connection.
 394  */
 395 static int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb)
 396 {
 397         struct iphdr *iph = skb->nh.iph;
 398         __u16 *portp = (__u16 *)&(((char *)iph)[iph->ihl*4]);
 399
 400         /* if it is fwmark-based service, the cache_bypass sysctl is up
 401            and the destination is RTN_UNICAST (and not local), then create
 402            a cache_bypass connection entry */
 403         if (sysctl_ip_vs_cache_bypass && svc->fwmark
 404             && (inet_addr_type(iph->daddr) == RTN_UNICAST)) {
 405                 int ret;
 406                 struct ip_vs_conn *cp;
 407
 408                 ip_vs_service_put(svc);
 409
 410                 /* create a new connection entry */
 411                 IP_VS_DBG(6, "ip_vs_leave: create a cache_bypass entry\n");
 412                 cp = ip_vs_conn_new(iph->protocol,
 413                                     iph->saddr, portp[0],
 414                                     iph->daddr, portp[1],
 415                                     0, 0,
 416                                     IP_VS_CONN_F_BYPASS,
 417                                     NULL);
 418                 if (cp == NULL) {
 419                         kfree_skb(skb);
 420                         return NF_STOLEN;
 421                 }
 422
 423                 /* statistics */
 424                 ip_vs_in_stats(cp, skb);
 425
 426                 /* set state */
 427                 ip_vs_set_state(cp, VS_STATE_INPUT, iph, portp);
 428
 429                 /* transmit the first SYN packet */
 430                 ret = cp->packet_xmit(skb, cp);
 431
 432                 atomic_inc(&cp->in_pkts);
 433                 ip_vs_conn_put(cp);
 434                 return ret;
 435         }
 436
 437         /*
 438          * When the virtual ftp service is presented, packets destined
 439          * for other services on the VIP may get here (except services
 440          * listed in the ipvs table), pass the packets, because it is
 441          * not ipvs job to decide to drop the packets.
 442          */
 443         if ((svc->port == FTPPORT) && (portp[1] != FTPPORT)) {
 444                 ip_vs_service_put(svc);
 445                 return NF_ACCEPT;
 446         }
 447
 448         ip_vs_service_put(svc);
 449
 450         /*
 451          * Notify the client that the destination is unreachable, and
 452          * release the socket buffer.
 453          * Since it is in IP layer, the TCP socket is not actually
 454          * created, the TCP RST packet cannot be sent, instead that
 455          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 456          */
 457         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 458         kfree_skb(skb);
 459         return NF_STOLEN;
 460 }
 461
 462
 463 /*
 464  *      It is hooked before NF_IP_PRI_NAT_SRC at the NF_IP_POST_ROUTING
 465  *      chain, and is used for VS/NAT.
 466  *      It detects packets for VS/NAT connections and sends the packets
 467  *      immediately. This can avoid that iptable_nat mangles the packets
 468  *      for VS/NAT.
 469  */
 470 static unsigned int ip_vs_post_routing(unsigned int hooknum,
 471                                        struct sk_buff **skb_p,
 472                                        const struct net_device *in,
 473                                        const struct net_device *out,
 474                                        int (*okfn)(struct sk_buff *))
 475 {
 476         struct sk_buff  *skb = *skb_p;
 477
 478         if (!(skb->nfcache & NFC_IPVS_PROPERTY))
 479                 return NF_ACCEPT;
 480
 481         /* The packet was sent from IPVS, exit this chain */
 482         (*okfn)(skb);
 483
 484         return NF_STOLEN;
 485 }
 486
 487
 488 /*
 489  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 490  *      Find any that might be relevant, check against existing connections,
 491  *      forward to the right destination host if relevant.
 492  *      Currently handles error types - unreachable, quench, ttl exceeded.
 493  *      (Only used in VS/NAT)
 494  */
 495 static int ip_vs_out_icmp(struct sk_buff **skb_p)
 496 {
 497         struct sk_buff  *skb   = *skb_p;
 498         struct iphdr    *iph;
 499         struct icmphdr  *icmph;
 500         struct iphdr    *ciph;  /* The ip header contained within the ICMP */
 501         __u16           *pptr;  /* port numbers from TCP/UDP contained header */
 502         unsigned short  ihl;
 503         unsigned short  len;
 504         unsigned short  clen, csize;
 505         struct ip_vs_conn *cp;
 506
 507         /* reassemble IP fragments, but will it happen in ICMP packets?? */
 508         if (skb->nh.iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) {
 509                 skb = ip_defrag(skb, IP_DEFRAG_VS_OUT);
 510                 if (!skb)
 511                         return NF_STOLEN;
 512                 *skb_p = skb;
 513         }
 514
 515         if (skb_is_nonlinear(skb)) {
 516                 if (skb_linearize(skb, GFP_ATOMIC) != 0)
 517                         return NF_DROP;
 518                 ip_send_check(skb->nh.iph);
 519         }
 520
 521         iph = skb->nh.iph;
 522         ihl = iph->ihl << 2;
 523         icmph = (struct icmphdr *)((char *)iph + ihl);
 524         len   = ntohs(iph->tot_len) - ihl;
 525         if (len < sizeof(struct icmphdr))
 526                 return NF_DROP;
 527
 528         IP_VS_DBG(12, "outgoing ICMP (%d,%d) %u.%u.%u.%u->%u.%u.%u.%u\n",
 529                   icmph->type, ntohs(icmp_id(icmph)),
 530                   NIPQUAD(iph->saddr), NIPQUAD(iph->daddr));
 531
 532         /*
 533          * Work through seeing if this is for us.
 534          * These checks are supposed to be in an order that means easy
 535          * things are checked first to speed up processing.... however
 536          * this means that some packets will manage to get a long way
 537          * down this stack and then be rejected, but that's life.
 538          */
 539         if ((icmph->type != ICMP_DEST_UNREACH) &&
 540             (icmph->type != ICMP_SOURCE_QUENCH) &&
 541             (icmph->type != ICMP_TIME_EXCEEDED))
 542                 return NF_ACCEPT;
 543
 544         /* Now find the contained IP header */
 545         clen = len - sizeof(struct icmphdr);
 546         if (clen < sizeof(struct iphdr))
 547                 return NF_DROP;
 548         ciph = (struct iphdr *) (icmph + 1);
 549         csize = ciph->ihl << 2;
 550         if (clen < csize)
 551                 return NF_DROP;
 552
 553         /* We are only interested ICMPs generated from TCP or UDP packets */
 554         if (ciph->protocol != IPPROTO_UDP && ciph->protocol != IPPROTO_TCP)
 555                 return NF_ACCEPT;
 556
 557         /* Skip non-first embedded TCP/UDP fragments */
 558         if (ciph->frag_off & __constant_htons(IP_OFFSET))
 559                 return NF_ACCEPT;
 560
 561         /* We need at least TCP/UDP ports here */
 562         if (clen < csize + sizeof(struct udphdr))
 563                 return NF_DROP;
 564
 565         /*
 566          * Find the ports involved - this packet was
 567          * incoming so the ports are right way round
 568          * (but reversed relative to outer IP header!)
 569          */
 570         pptr = (__u16 *)&(((char *)ciph)[csize]);
 571
 572         /* Ensure the checksum is correct */
 573         if (ip_compute_csum((unsigned char *) icmph, len)) {
 574                 /* Failed checksum! */
 575                 IP_VS_DBG(1, "forward ICMP: failed checksum from %d.%d.%d.%d!\n",
 576                           NIPQUAD(iph->saddr));
 577                 return NF_DROP;
 578         }
 579
 580         IP_VS_DBG(11, "Handling outgoing ICMP for "
 581                   "%u.%u.%u.%u:%d -> %u.%u.%u.%u:%d\n",
 582                   NIPQUAD(ciph->saddr), ntohs(pptr[0]),
 583                   NIPQUAD(ciph->daddr), ntohs(pptr[1]));
 584
 585         /* ciph content is actually <protocol, caddr, cport, daddr, dport> */
 586         cp = ip_vs_conn_out_get(ciph->protocol, ciph->daddr, pptr[1],
 587                                 ciph->saddr, pptr[0]);
 588         if (!cp)
 589                 return NF_ACCEPT;
 590
 591         if (IP_VS_FWD_METHOD(cp) != 0) {
 592                 IP_VS_ERR("shouldn't reach here, because the box is on the"
 593                           "half connection in the tun/dr module.\n");
 594         }
 595
 596         /* Now we do real damage to this packet...! */
 597         /* First change the source IP address, and recalc checksum */
 598         iph->saddr = cp->vaddr;
 599         ip_send_check(iph);
 600
 601         /* Now change the *dest* address in the contained IP */
 602         ciph->daddr = cp->vaddr;
 603         ip_send_check(ciph);
 604
 605         /* the TCP/UDP dest port - cannot redo check */
 606         pptr[1] = cp->vport;
 607
 608         /* And finally the ICMP checksum */
 609         icmph->checksum = 0;
 610         icmph->checksum = ip_compute_csum((unsigned char *) icmph, len);
 611         skb->ip_summed = CHECKSUM_UNNECESSARY;
 612
 613         /* do the statistics and put it back */
 614         ip_vs_out_stats(cp, skb);
 615         ip_vs_conn_put(cp);
 616
 617         IP_VS_DBG(11, "Forwarding correct outgoing ICMP to "
 618                   "%u.%u.%u.%u:%d -> %u.%u.%u.%u:%d\n",
 619                   NIPQUAD(ciph->saddr), ntohs(pptr[0]),
 620                   NIPQUAD(ciph->daddr), ntohs(pptr[1]));
 621
 622         skb->nfcache |= NFC_IPVS_PROPERTY;
 623
 624         return NF_ACCEPT;
 625 }
 626
 627
 628 /*
 629  *      It is hooked at the NF_IP_FORWARD chain, used only for VS/NAT.
 630  *      Check if outgoing packet belongs to the established ip_vs_conn,
 631  *      rewrite addresses of the packet and send it on its way...
 632  */
 633 static unsigned int ip_vs_out(unsigned int hooknum,
 634                               struct sk_buff **skb_p,
 635                               const struct net_device *in,
 636                               const struct net_device *out,
 637                               int (*okfn)(struct sk_buff *))
 638 {
 639         struct sk_buff  *skb = *skb_p;
 640         struct iphdr    *iph;
 641         union ip_vs_tphdr h;
 642         struct ip_vs_conn *cp;
 643         int size;
 644         int ihl;
 645
 646         EnterFunction(11);
 647
 648         if (skb->nfcache & NFC_IPVS_PROPERTY)
 649                 return NF_ACCEPT;
 650
 651         iph = skb->nh.iph;
 652         if (iph->protocol == IPPROTO_ICMP)
 653                 return ip_vs_out_icmp(skb_p);
 654
 655         /* let it go if other IP protocols */
 656         if (iph->protocol != IPPROTO_TCP && iph->protocol != IPPROTO_UDP)
 657                 return NF_ACCEPT;
 658
 659         /* reassemble IP fragments */
 660         if (iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) {
 661                 skb = ip_defrag(skb, IP_DEFRAG_VS_OUT);
 662                 if (!skb)
 663                         return NF_STOLEN;
 664                 iph = skb->nh.iph;
 665                 *skb_p = skb;
 666         }
 667
 668         /* make sure that protocol header available in skb data area,
 669            note that skb data area may be reallocated. */
 670         ihl = iph->ihl << 2;
 671         if (ip_vs_header_check(skb, iph->protocol, ihl) == -1)
 672                 return NF_DROP;
 673
 674         iph = skb->nh.iph;
 675         h.raw = (char*) iph + ihl;
 676
 677         /*
 678          *      Check if the packet belongs to an old entry
 679          */
 680         cp = ip_vs_conn_out_get(iph->protocol, iph->saddr, h.portp[0],
 681                                 iph->daddr, h.portp[1]);
 682         if (!cp) {
 683                 if (sysctl_ip_vs_nat_icmp_send &&
 684                     ip_vs_lookup_real_service(iph->protocol,
 685                                               iph->saddr, h.portp[0])) {
 686                         /*
 687                          * Notify the real server: there is no existing
 688                          * entry if it is not RST packet or not TCP packet.
 689                          */
 690                         if (!h.th->rst || iph->protocol != IPPROTO_TCP) {
 691                                 icmp_send(skb, ICMP_DEST_UNREACH,
 692                                           ICMP_PORT_UNREACH, 0);
 693                                 kfree_skb(skb);
 694                                 return NF_STOLEN;
 695                         }
 696                 }
 697                 IP_VS_DBG(12, "packet for %s %d.%d.%d.%d:%d "
 698                           "continue traversal as normal.\n",
 699                           ip_vs_proto_name(iph->protocol),
 700                           NIPQUAD(iph->daddr),
 701                           ntohs(h.portp[1]));
 702                 if (skb_is_nonlinear(skb))
 703                         ip_send_check(iph);
 704                 return NF_ACCEPT;
 705         }
 706
 707         /*
 708          * If it has ip_vs_app helper, the helper may change the payload,
 709          * so it needs full checksum checking and checksum calculation.
 710          * If not, only the header (addr/port) is changed, so it is fast
 711          * to do incremental checksum update, and let the destination host
 712          * do final checksum checking.
 713          */
 714
 715         if (cp->app && skb_is_nonlinear(skb)) {
 716                 if (skb_linearize(skb, GFP_ATOMIC) != 0) {
 717                         ip_vs_conn_put(cp);
 718                         return NF_DROP;
 719                 }
 720                 iph = skb->nh.iph;
 721                 h.raw = (char*) iph + ihl;
 722         }
 723
 724         size = skb->len - ihl;
 725         IP_VS_DBG(11, "O-pkt: %s size=%d\n",
 726                   ip_vs_proto_name(iph->protocol), size);
 727
 728         /* do TCP/UDP checksum checking if it has application helper */
 729         if (cp->app && (iph->protocol != IPPROTO_UDP || h.uh->check != 0)) {
 730                 switch (skb->ip_summed) {
 731                 case CHECKSUM_NONE:
 732                         skb->csum = csum_partial(h.raw, size, 0);
 733                 case CHECKSUM_HW:
 734                         if (csum_tcpudp_magic(iph->saddr, iph->daddr, size,
 735                                               iph->protocol, skb->csum)) {
 736                                 ip_vs_conn_put(cp);
 737                                 IP_VS_DBG_RL("Outgoing failed %s checksum "
 738                                              "from %d.%d.%d.%d (size=%d)!\n",
 739                                              ip_vs_proto_name(iph->protocol),
 740                                              NIPQUAD(iph->saddr),
 741                                              size);
 742                                 return NF_DROP;
 743                         }
 744                         break;
 745                 default:
 746                         /* CHECKSUM_UNNECESSARY */
 747                         break;
 748                 }
 749         }
 750
 751         IP_VS_DBG(11, "Outgoing %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d\n",
 752                   ip_vs_proto_name(iph->protocol),
 753                   NIPQUAD(iph->saddr), ntohs(h.portp[0]),
 754                   NIPQUAD(iph->daddr), ntohs(h.portp[1]));
 755
 756         /* mangle the packet */
 757         iph->saddr = cp->vaddr;
 758         h.portp[0] = cp->vport;
 759
 760         /*
 761          *      Call application helper if needed
 762          */
 763         if (ip_vs_app_pkt_out(cp, skb) != 0) {
 764                 /* skb data has probably changed, update pointers */
 765                 iph = skb->nh.iph;
 766                 h.raw = (char*)iph + ihl;
 767                 size = skb->len - ihl;
 768         }
 769
 770         /*
 771          *      Adjust TCP/UDP checksums
 772          */
 773         if (!cp->app && (iph->protocol != IPPROTO_UDP || h.uh->check != 0)) {
 774                 /* Only port and addr are changed, do fast csum update */
 775                 ip_vs_fast_check_update(&h, cp->daddr, cp->vaddr,
 776                                         cp->dport, cp->vport, iph->protocol);
 777                 if (skb->ip_summed == CHECKSUM_HW)
 778                         skb->ip_summed = CHECKSUM_NONE;
 779         } else {
 780                 /* full checksum calculation */
 781                 switch (iph->protocol) {
 782                 case IPPROTO_TCP:
 783                         h.th->check = 0;
 784                         skb->csum = csum_partial(h.raw, size, 0);
 785                         h.th->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 786                                                         size, iph->protocol,
 787                                                         skb->csum);
 788                         IP_VS_DBG(11, "O-pkt: %s O-csum=%d (+%d)\n",
 789                                   ip_vs_proto_name(iph->protocol), h.th->check,
 790                                   (char*)&(h.th->check) - (char*)h.raw);
 791                         break;
 792                 case IPPROTO_UDP:
 793                         h.uh->check = 0;
 794                         skb->csum = csum_partial(h.raw, size, 0);
 795                         h.uh->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 796                                                         size, iph->protocol,
 797                                                         skb->csum);
 798                         if (h.uh->check == 0)
 799                                 h.uh->check = 0xFFFF;
 800                         IP_VS_DBG(11, "O-pkt: %s O-csum=%d (+%d)\n",
 801                                   ip_vs_proto_name(iph->protocol), h.uh->check,
 802                                   (char*)&(h.uh->check) - (char*)h.raw);
 803                         break;
 804                 }
 805         }
 806         ip_send_check(iph);
 807
 808         ip_vs_out_stats(cp, skb);
 809         ip_vs_set_state(cp, VS_STATE_OUTPUT, iph, h.portp);
 810         ip_vs_conn_put(cp);
 811
 812         skb->nfcache |= NFC_IPVS_PROPERTY;
 813
 814         LeaveFunction(11);
 815         return NF_ACCEPT;
 816 }
 817
 818
 819 /*
 820  *      Check if the packet is for VS/NAT connections, then send it
 821  *      immediately.
 822  *      Called by ip_fw_compact to detect packets for VS/NAT before
 823  *      they are changed by ipchains masquerading code.
 824  */
 825 unsigned int check_for_ip_vs_out(struct sk_buff **skb_p,
 826                                  int (*okfn)(struct sk_buff *))
 827 {
 828         unsigned int ret;
 829
 830         ret = ip_vs_out(NF_IP_FORWARD, skb_p, NULL, NULL, NULL);
 831         if (ret != NF_ACCEPT) {
 832                 return ret;
 833         } else {
 834                 /* send the packet immediately if it is already mangled
 835                    by ip_vs_out */
 836                 if ((*skb_p)->nfcache & NFC_IPVS_PROPERTY) {
 837                         (*okfn)(*skb_p);
 838                         return NF_STOLEN;
 839                 }
 840         }
 841         return NF_ACCEPT;
 842 }
 843
 844
 845 /*
 846  *      Handle ICMP messages in the outside-to-inside direction (incoming)
 847  *      and sometimes in outgoing direction from ip_vs_forward_icmp.
 848  *      Find any that might be relevant, check against existing connections,
 849  *      forward to the right destination host if relevant.
 850  *      Currently handles error types - unreachable, quench, ttl exceeded.
 851  */
 852 static int ip_vs_in_icmp(struct sk_buff **skb_p)
 853 {
 854         struct sk_buff  *skb   = *skb_p;
 855         struct iphdr    *iph;
 856         struct icmphdr  *icmph;
 857         struct iphdr    *ciph;  /* The ip header contained within the ICMP */
 858         __u16           *pptr;  /* port numbers from TCP/UDP contained header */
 859         unsigned short   len;
 860         unsigned short  clen, csize;
 861         struct ip_vs_conn *cp;
 862         struct rtable *rt;                      /* Route to the other host */
 863         int    mtu;
 864
 865         if (skb_is_nonlinear(skb)) {
 866                 if (skb_linearize(skb, GFP_ATOMIC) != 0)
 867                         return NF_DROP;
 868         }
 869
 870         iph = skb->nh.iph;
 871         ip_send_check(iph);
 872         icmph = (struct icmphdr *)((char *)iph + (iph->ihl << 2));
 873         len = ntohs(iph->tot_len) - (iph->ihl<<2);
 874         if (len < sizeof(struct icmphdr))
 875                 return NF_DROP;
 876
 877         IP_VS_DBG(12, "icmp in (%d,%d) %u.%u.%u.%u -> %u.%u.%u.%u\n",
 878                   icmph->type, ntohs(icmp_id(icmph)),
 879                   NIPQUAD(iph->saddr), NIPQUAD(iph->daddr));
 880
 881         if ((icmph->type != ICMP_DEST_UNREACH) &&
 882             (icmph->type != ICMP_SOURCE_QUENCH) &&
 883             (icmph->type != ICMP_TIME_EXCEEDED))
 884                 return NF_ACCEPT;
 885
 886         /*
 887          * If we get here we have an ICMP error of one of the above 3 types
 888          * Now find the contained IP header
 889          */
 890         clen = len - sizeof(struct icmphdr);
 891         if (clen < sizeof(struct iphdr))
 892                 return NF_DROP;
 893         ciph = (struct iphdr *) (icmph + 1);
 894         csize = ciph->ihl << 2;
 895         if (clen < csize)
 896                 return NF_DROP;
 897
 898         /* We are only interested ICMPs generated from TCP or UDP packets */
 899         if (ciph->protocol != IPPROTO_UDP && ciph->protocol != IPPROTO_TCP)
 900                 return NF_ACCEPT;
 901
 902         /* Skip non-first embedded TCP/UDP fragments */
 903         if (ciph->frag_off & __constant_htons(IP_OFFSET))
 904                 return NF_ACCEPT;
 905
 906         /* We need at least TCP/UDP ports here */
 907         if (clen < csize + sizeof(struct udphdr))
 908                 return NF_DROP;
 909
 910         /* Ensure the checksum is correct */
 911         if (ip_compute_csum((unsigned char *) icmph, len)) {
 912                 /* Failed checksum! */
 913                 IP_VS_ERR_RL("incoming ICMP: failed checksum from "
 914                              "%d.%d.%d.%d!\n", NIPQUAD(iph->saddr));
 915                 return NF_DROP;
 916         }
 917
 918         pptr = (__u16 *)&(((char *)ciph)[csize]);
 919
 920         IP_VS_DBG(11, "Handling incoming ICMP for "
 921                   "%u.%u.%u.%u:%d -> %u.%u.%u.%u:%d\n",
 922                   NIPQUAD(ciph->saddr), ntohs(pptr[0]),
 923                   NIPQUAD(ciph->daddr), ntohs(pptr[1]));
 924
 925         /* This is pretty much what ip_vs_conn_in_get() does,
 926            except parameters are in the reverse order */
 927         cp = ip_vs_conn_in_get(ciph->protocol,
 928                                ciph->daddr, pptr[1],
 929                                ciph->saddr, pptr[0]);
 930         if (cp == NULL)
 931                 return NF_ACCEPT;
 932
 933         ip_vs_in_stats(cp, skb);
 934
 935         /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
 936            forwarded directly here, because there is no need to
 937            translate address/port back */
 938         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
 939                 int ret;
 940                 if (cp->packet_xmit)
 941                         ret = cp->packet_xmit(skb, cp);
 942                 else
 943                         ret = NF_ACCEPT;
 944                 atomic_inc(&cp->in_pkts);
 945                 ip_vs_conn_put(cp);
 946                 return ret;
 947         }
 948
 949         /*
 950          * mangle and send the packet here
 951          */
 952         if (!(rt = __ip_vs_get_out_rt(cp, RT_TOS(iph->tos))))
 953                 goto tx_error_icmp;
 954
 955         /* MTU checking */
 956         mtu = rt->u.dst.pmtu;
 957         if ((skb->len > mtu) && (iph->frag_off&__constant_htons(IP_DF))) {
 958                 ip_rt_put(rt);
 959                 icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu));
 960                 IP_VS_DBG_RL("ip_vs_in_icmp(): frag needed\n");
 961                 goto tx_error;
 962         }
 963
 964         /* drop old route */
 965         dst_release(skb->dst);
 966         skb->dst = &rt->u.dst;
 967
 968         /* copy-on-write the packet before mangling it */
 969         if (ip_vs_skb_cow(skb, rt->u.dst.dev->hard_header_len,
 970                           &iph, (unsigned char**)&icmph)) {
 971                 ip_vs_conn_put(cp);
 972                 return NF_DROP;
 973         }
 974         ciph = (struct iphdr *) (icmph + 1);
 975         pptr = (__u16 *)&(((char *)ciph)[csize]);
 976
 977         /* The ICMP packet for VS/NAT must be written to correct addresses
 978            before being forwarded to the right server */
 979
 980         /* First change the dest IP address, and recalc checksum */
 981         iph->daddr = cp->daddr;
 982         ip_send_check(iph);
 983
 984         /* Now change the *source* address in the contained IP */
 985         ciph->saddr = cp->daddr;
 986         ip_send_check(ciph);
 987
 988         /* the TCP/UDP source port - cannot redo check */
 989         pptr[0] = cp->dport;
 990
 991         /* And finally the ICMP checksum */
 992         icmph->checksum = 0;
 993         icmph->checksum = ip_compute_csum((unsigned char *) icmph, len);
 994         skb->ip_summed = CHECKSUM_UNNECESSARY;
 995
 996         IP_VS_DBG(11, "Forwarding incoming ICMP to "
 997                   "%u.%u.%u.%u:%d -> %u.%u.%u.%u:%d\n",
 998                   NIPQUAD(ciph->saddr), ntohs(pptr[0]),
 999                   NIPQUAD(ciph->daddr), ntohs(pptr[1]));
1000
1001 #ifdef CONFIG_NETFILTER_DEBUG
1002         skb->nf_debug = 1 << NF_IP_LOCAL_OUT;
1003 #endif /* CONFIG_NETFILTER_DEBUG */
1004         ip_send(skb);
1005         ip_vs_conn_put(cp);
1006         return NF_STOLEN;
1007
1008   tx_error_icmp:
1009         dst_link_failure(skb);
1010   tx_error:
1011         dev_kfree_skb(skb);
1012         ip_vs_conn_put(cp);
1013         return NF_STOLEN;
1014 }
1015
1016
1017 /*
1018  *      Check if it's for virtual services, look it up,
1019  *      and send it on its way...
1020  */
1021 static unsigned int ip_vs_in(unsigned int hooknum,
1022                              struct sk_buff **skb_p,
1023                              const struct net_device *in,
1024                              const struct net_device *out,
1025                              int (*okfn)(struct sk_buff *))
1026 {
1027         struct sk_buff  *skb = *skb_p;
1028         struct iphdr    *iph = skb->nh.iph;
1029         union ip_vs_tphdr h;
1030         struct ip_vs_conn *cp;
1031         struct ip_vs_service *svc;
1032         int ihl;
1033         int ret;
1034
1035         /*
1036          *      Big tappo: only PACKET_HOST (nor loopback neither mcasts)
1037          *      ... don't know why 1st test DOES NOT include 2nd (?)
1038          */
1039         if (skb->pkt_type != PACKET_HOST || skb->dev == &loopback_dev) {
1040                 IP_VS_DBG(12, "packet type=%d proto=%d daddr=%d.%d.%d.%d ignored\n",
1041                           skb->pkt_type,
1042                           iph->protocol,
1043                           NIPQUAD(iph->daddr));
1044                 return NF_ACCEPT;
1045         }
1046
1047         if (iph->protocol == IPPROTO_ICMP)
1048                 return ip_vs_in_icmp(skb_p);
1049
1050         /* let it go if other IP protocols */
1051         if (iph->protocol != IPPROTO_TCP && iph->protocol != IPPROTO_UDP)
1052                 return NF_ACCEPT;
1053
1054         /* make sure that protocol header available in skb data area,
1055            note that skb data area may be reallocated. */
1056         ihl = iph->ihl << 2;
1057         if (ip_vs_header_check(skb, iph->protocol, ihl) == -1)
1058                 return NF_DROP;
1059         iph = skb->nh.iph;
1060         h.raw = (char*) iph + ihl;
1061
1062         /*
1063          * Check if the packet belongs to an existing connection entry
1064          */
1065         cp = ip_vs_conn_in_get(iph->protocol, iph->saddr, h.portp[0],
1066                                iph->daddr, h.portp[1]);
1067
1068         if (!cp &&
1069             (h.th->syn || (iph->protocol!=IPPROTO_TCP)) &&
1070             (svc = ip_vs_service_get(skb->nfmark, iph->protocol,
1071                                      iph->daddr, h.portp[1]))) {
1072                 if (ip_vs_todrop()) {
1073                         /*
1074                          * It seems that we are very loaded.
1075                          * We have to drop this packet :(
1076                          */
1077                         ip_vs_service_put(svc);
1078                         return NF_DROP;
1079                 }
1080
1081                 /*
1082                  * Let the virtual server select a real server for the
1083                  * incoming connection, and create a connection entry.
1084                  */
1085                 cp = ip_vs_schedule(svc, iph);
1086                 if (!cp)
1087                         return ip_vs_leave(svc, skb);
1088                 ip_vs_conn_stats(cp, svc);
1089                 ip_vs_service_put(svc);
1090         }
1091
1092         if (!cp) {
1093                 /* sorry, all this trouble for a no-hit :) */
1094                 IP_VS_DBG(12, "packet for %s %d.%d.%d.%d:%d continue "
1095                           "traversal as normal.\n",
1096                           ip_vs_proto_name(iph->protocol),
1097                           NIPQUAD(iph->daddr),
1098                           ntohs(h.portp[1]));
1099                 return NF_ACCEPT;
1100         }
1101
1102         IP_VS_DBG(11, "Incoming %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d\n",
1103                   ip_vs_proto_name(iph->protocol),
1104                   NIPQUAD(iph->saddr), ntohs(h.portp[0]),
1105                   NIPQUAD(iph->daddr), ntohs(h.portp[1]));
1106
1107         /* Check the server status */
1108         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1109                 /* the destination server is not available */
1110
1111                 if (sysctl_ip_vs_expire_nodest_conn) {
1112                         /* try to expire the connection immediately */
1113                         ip_vs_conn_expire_now(cp);
1114                 }
1115                 /* don't restart its timer, and silently
1116                    drop the packet. */
1117                 __ip_vs_conn_put(cp);
1118                 return NF_DROP;
1119         }
1120
1121         ip_vs_in_stats(cp, skb);
1122         ip_vs_set_state(cp, VS_STATE_INPUT, iph, h.portp);
1123         if (cp->packet_xmit)
1124                 ret = cp->packet_xmit(skb, cp);
1125         else {
1126                 IP_VS_DBG_RL("warning: packet_xmit is null");
1127                 ret = NF_ACCEPT;
1128         }
1129
1130         /* increase its packet counter and check if it is needed
1131            to be synchronized */
1132         atomic_inc(&cp->in_pkts);
1133         if (ip_vs_sync_state & IP_VS_STATE_MASTER &&
1134             (cp->protocol != IPPROTO_TCP ||
1135              cp->state == IP_VS_S_ESTABLISHED) &&
1136             (atomic_read(&cp->in_pkts) % 50 == sysctl_ip_vs_sync_threshold))
1137                 ip_vs_sync_conn(cp);
1138
1139         ip_vs_conn_put(cp);
1140         return ret;
1141 }
1142
1143
1144 /*
1145  *      It is hooked at the NF_IP_FORWARD chain, in order to catch ICMP
1146  *      packets destined for 0.0.0.0/0.
1147  *      When fwmark-based virtual service is used, such as transparent
1148  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
1149  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1150  *      sent to ip_vs_in_icmp. So, catch them at the NF_IP_FORWARD chain
1151  *      and send them to ip_vs_in_icmp.
1152  */
1153 static unsigned int ip_vs_forward_icmp(unsigned int hooknum,
1154                                        struct sk_buff **skb_p,
1155                                        const struct net_device *in,
1156                                        const struct net_device *out,
1157                                        int (*okfn)(struct sk_buff *))
1158 {
1159         struct sk_buff  *skb = *skb_p;
1160         struct iphdr    *iph = skb->nh.iph;
1161
1162         if (iph->protocol != IPPROTO_ICMP)
1163                 return NF_ACCEPT;
1164
1165         if (iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) {
1166                 skb = ip_defrag(skb, IP_DEFRAG_VS_FWD);
1167                 if (!skb)
1168                         return NF_STOLEN;
1169                 *skb_p = skb;
1170         }
1171
1172         return ip_vs_in_icmp(skb_p);
1173 }
1174
1175
1176 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1177    or VS/NAT(change destination), so that filtering rules can be
1178    applied to IPVS. */
1179 static struct nf_hook_ops ip_vs_in_ops = {
1180         { NULL, NULL },
1181         ip_vs_in, PF_INET, NF_IP_LOCAL_IN, 100
1182 };
1183
1184 /* After packet filtering, change source only for VS/NAT */
1185 static struct nf_hook_ops ip_vs_out_ops = {
1186         { NULL, NULL },
1187         ip_vs_out, PF_INET, NF_IP_FORWARD, 100
1188 };
1189
1190 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1191    destined for 0.0.0.0/0, which is for incoming IPVS connections */
1192 static struct nf_hook_ops ip_vs_forward_icmp_ops = {
1193         { NULL, NULL },
1194         ip_vs_forward_icmp, PF_INET, NF_IP_FORWARD, 99
1195 };
1196
1197 /* Before the netfilter connection tracking, exit from POST_ROUTING */
1198 static struct nf_hook_ops ip_vs_post_routing_ops = {
1199         { NULL, NULL },
1200         ip_vs_post_routing, PF_INET, NF_IP_POST_ROUTING, NF_IP_PRI_NAT_SRC-1
1201 };
1202
1203
1204 /*
1205  *      Initialize IP Virtual Server
1206  */
1207 static int __init ip_vs_init(void)
1208 {
1209         int ret;
1210
1211         ret = ip_vs_control_init();
1212         if (ret < 0) {
1213                 IP_VS_ERR("can't setup control.\n");
1214                 goto cleanup_nothing;
1215         }
1216
1217         ret = ip_vs_conn_init();
1218         if (ret < 0) {
1219                 IP_VS_ERR("can't setup connection table.\n");
1220                 goto cleanup_control;
1221         }
1222
1223         ret = ip_vs_app_init();
1224         if (ret < 0) {
1225                 IP_VS_ERR("can't setup application helper.\n");
1226                 goto cleanup_conn;
1227         }
1228
1229         ret = nf_register_hook(&ip_vs_in_ops);
1230         if (ret < 0) {
1231                 IP_VS_ERR("can't register in hook.\n");
1232                 goto cleanup_app;
1233         }
1234         ret = nf_register_hook(&ip_vs_out_ops);
1235         if (ret < 0) {
1236                 IP_VS_ERR("can't register out hook.\n");
1237                 goto cleanup_inops;
1238         }
1239         ret = nf_register_hook(&ip_vs_post_routing_ops);
1240         if (ret < 0) {
1241                 IP_VS_ERR("can't register post_routing hook.\n");
1242                 goto cleanup_outops;
1243         }
1244         ret = nf_register_hook(&ip_vs_forward_icmp_ops);
1245         if (ret < 0) {
1246                 IP_VS_ERR("can't register forward_icmp hook.\n");
1247                 goto cleanup_postroutingops;
1248         }
1249
1250         IP_VS_INFO("ipvs loaded.\n");
1251         return ret;
1252
1253   cleanup_postroutingops:
1254         nf_unregister_hook(&ip_vs_post_routing_ops);
1255   cleanup_outops:
1256         nf_unregister_hook(&ip_vs_out_ops);
1257   cleanup_inops:
1258         nf_unregister_hook(&ip_vs_in_ops);
1259   cleanup_app:
1260         ip_vs_app_cleanup();
1261   cleanup_conn:
1262         ip_vs_conn_cleanup();
1263   cleanup_control:
1264         ip_vs_control_cleanup();
1265   cleanup_nothing:
1266         return ret;
1267 }
1268
1269 static void __exit ip_vs_cleanup(void)
1270 {
1271         nf_unregister_hook(&ip_vs_forward_icmp_ops);
1272         nf_unregister_hook(&ip_vs_post_routing_ops);
1273         nf_unregister_hook(&ip_vs_out_ops);
1274         nf_unregister_hook(&ip_vs_in_ops);
1275         ip_vs_app_cleanup();
1276         ip_vs_conn_cleanup();
1277         ip_vs_control_cleanup();
1278         IP_VS_INFO("ipvs unloaded.\n");
1279 }
1280
1281 module_init(ip_vs_init);
1282 module_exit(ip_vs_cleanup);
1283 MODULE_LICENSE("GPL");