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 1.34 2003/05/10 03:05:23 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  *      Paul `Rusty' Russell            properly handle non-linear skbs
  25  *
  26  */
  27
  28 #include <linux/module.h>
  29 #include <linux/kernel.h>
  30 #include <linux/ip.h>
  31 #include <linux/tcp.h>
  32 #include <linux/icmp.h>
  33
  34 #include <net/ip.h>
  35 #include <net/tcp.h>
  36 #include <net/udp.h>
  37 #include <net/icmp.h>                   /* for icmp_send */
  38 #include <net/route.h>
  39
  40 #include <linux/netfilter.h>
  41 #include <linux/netfilter_ipv4.h>
  42
  43 #include <net/ip_vs.h>
  44
  45
  46 EXPORT_SYMBOL(register_ip_vs_scheduler);
  47 EXPORT_SYMBOL(unregister_ip_vs_scheduler);
  48 EXPORT_SYMBOL(ip_vs_skb_replace);
  49 EXPORT_SYMBOL(ip_vs_proto_name);
  50 EXPORT_SYMBOL(ip_vs_conn_new);
  51 EXPORT_SYMBOL(ip_vs_conn_in_get);
  52 EXPORT_SYMBOL(ip_vs_conn_out_get);
  53 #ifdef CONFIG_IP_VS_PROTO_TCP
  54 EXPORT_SYMBOL(ip_vs_tcp_conn_listen);
  55 #endif
  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 EXPORT_SYMBOL(ip_vs_make_skb_writable);
  62
  63
  64 /* ID used in ICMP lookups */
  65 #define icmp_id(icmph)          (((icmph)->un).echo.id)
  66
  67 const char *ip_vs_proto_name(unsigned proto)
  68 {
  69         static char buf[20];
  70
  71         switch (proto) {
  72         case IPPROTO_IP:
  73                 return "IP";
  74         case IPPROTO_UDP:
  75                 return "UDP";
  76         case IPPROTO_TCP:
  77                 return "TCP";
  78         case IPPROTO_ICMP:
  79                 return "ICMP";
  80         default:
  81                 sprintf(buf, "IP_%d", proto);
  82                 return buf;
  83         }
  84 }
  85
  86 void ip_vs_init_hash_table(struct list_head *table, int rows)
  87 {
  88         while (--rows >= 0)
  89                 INIT_LIST_HEAD(&table[rows]);
  90 }
  91
  92 static inline void
  93 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
  94 {
  95         struct ip_vs_dest *dest = cp->dest;
  96         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
  97                 spin_lock(&dest->stats.lock);
  98                 dest->stats.inpkts++;
  99                 dest->stats.inbytes += skb->len;
 100                 spin_unlock(&dest->stats.lock);
 101
 102                 spin_lock(&dest->svc->stats.lock);
 103                 dest->svc->stats.inpkts++;
 104                 dest->svc->stats.inbytes += skb->len;
 105                 spin_unlock(&dest->svc->stats.lock);
 106
 107                 spin_lock(&ip_vs_stats.lock);
 108                 ip_vs_stats.inpkts++;
 109                 ip_vs_stats.inbytes += skb->len;
 110                 spin_unlock(&ip_vs_stats.lock);
 111         }
 112 }
 113
 114
 115 static inline void
 116 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 117 {
 118         struct ip_vs_dest *dest = cp->dest;
 119         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 120                 spin_lock(&dest->stats.lock);
 121                 dest->stats.outpkts++;
 122                 dest->stats.outbytes += skb->len;
 123                 spin_unlock(&dest->stats.lock);
 124
 125                 spin_lock(&dest->svc->stats.lock);
 126                 dest->svc->stats.outpkts++;
 127                 dest->svc->stats.outbytes += skb->len;
 128                 spin_unlock(&dest->svc->stats.lock);
 129
 130                 spin_lock(&ip_vs_stats.lock);
 131                 ip_vs_stats.outpkts++;
 132                 ip_vs_stats.outbytes += skb->len;
 133                 spin_unlock(&ip_vs_stats.lock);
 134         }
 135 }
 136
 137
 138 static inline void
 139 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
 140 {
 141         spin_lock(&cp->dest->stats.lock);
 142         cp->dest->stats.conns++;
 143         spin_unlock(&cp->dest->stats.lock);
 144
 145         spin_lock(&svc->stats.lock);
 146         svc->stats.conns++;
 147         spin_unlock(&svc->stats.lock);
 148
 149         spin_lock(&ip_vs_stats.lock);
 150         ip_vs_stats.conns++;
 151         spin_unlock(&ip_vs_stats.lock);
 152 }
 153
 154
 155 static inline int
 156 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
 157                 const struct sk_buff *skb,
 158                 struct ip_vs_protocol *pp)
 159 {
 160         if (unlikely(!pp->state_transition))
 161                 return 0;
 162         return pp->state_transition(cp, direction, skb, pp);
 163 }
 164
 165
 166 int ip_vs_make_skb_writable(struct sk_buff **pskb, int writable_len)
 167 {
 168         struct sk_buff *skb = *pskb;
 169
 170         /* skb is already used, better copy skb and its payload */
 171         if (unlikely(skb_shared(skb) || skb->sk))
 172                 goto copy_skb;
 173
 174         /* skb data is already used, copy it */
 175         if (unlikely(skb_cloned(skb)))
 176                 goto copy_data;
 177
 178         return pskb_may_pull(skb, writable_len);
 179
 180   copy_data:
 181         if (unlikely(writable_len > skb->len))
 182                 return 0;
 183         return !pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
 184
 185   copy_skb:
 186         if (unlikely(writable_len > skb->len))
 187                 return 0;
 188         skb = skb_copy(skb, GFP_ATOMIC);
 189         if (!skb)
 190                 return 0;
 191         BUG_ON(skb_is_nonlinear(skb));
 192
 193         /* Rest of kernel will get very unhappy if we pass it a
 194            suddenly-orphaned skbuff */
 195         if ((*pskb)->sk)
 196                 skb_set_owner_w(skb, (*pskb)->sk);
 197         kfree_skb(*pskb);
 198         *pskb = skb;
 199         return 1;
 200 }
 201
 202 /*
 203  *  IPVS persistent scheduling function
 204  *  It creates a connection entry according to its template if exists,
 205  *  or selects a server and creates a connection entry plus a template.
 206  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
 207  *  Protocols supported: TCP, UDP
 208  */
 209 static struct ip_vs_conn *
 210 ip_vs_sched_persist(struct ip_vs_service *svc,
 211                     const struct sk_buff *skb,
 212                     __u16 ports[2])
 213 {
 214         struct ip_vs_conn *cp = NULL;
 215         struct iphdr *iph = skb->nh.iph;
 216         struct ip_vs_dest *dest;
 217         struct ip_vs_conn *ct;
 218         __u16  dport;    /* destination port to forward */
 219         __u32  snet;     /* source network of the client, after masking */
 220
 221         /* Mask saddr with the netmask to adjust template granularity */
 222         snet = iph->saddr & svc->netmask;
 223
 224         IP_VS_DBG(6, "p-schedule: src %u.%u.%u.%u:%u dest %u.%u.%u.%u:%u "
 225                   "mnet %u.%u.%u.%u\n",
 226                   NIPQUAD(iph->saddr), ntohs(ports[0]),
 227                   NIPQUAD(iph->daddr), ntohs(ports[1]),
 228                   NIPQUAD(snet));
 229
 230         /*
 231          * As far as we know, FTP is a very complicated network protocol, and
 232          * it uses control connection and data connections. For active FTP,
 233          * FTP server initialize data connection to the client, its source port
 234          * is often 20. For passive FTP, FTP server tells the clients the port
 235          * that it passively listens to,  and the client issues the data
 236          * connection. In the tunneling or direct routing mode, the load
 237          * balancer is on the client-to-server half of connection, the port
 238          * number is unknown to the load balancer. So, a conn template like
 239          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
 240          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
 241          * is created for other persistent services.
 242          */
 243         if (ports[1] == svc->port) {
 244                 /* Check if a template already exists */
 245                 if (svc->port != FTPPORT)
 246                         ct = ip_vs_conn_in_get(iph->protocol, snet, 0,
 247                                                iph->daddr, ports[1]);
 248                 else
 249                         ct = ip_vs_conn_in_get(iph->protocol, snet, 0,
 250                                                iph->daddr, 0);
 251
 252                 if (!ct || !ip_vs_check_template(ct)) {
 253                         /*
 254                          * No template found or the dest of the connection
 255                          * template is not available.
 256                          */
 257                         dest = svc->scheduler->schedule(svc, skb);
 258                         if (dest == NULL) {
 259                                 IP_VS_DBG(1, "p-schedule: no dest found.\n");
 260                                 return NULL;
 261                         }
 262
 263                         /*
 264                          * Create a template like <protocol,caddr,0,
 265                          * vaddr,vport,daddr,dport> for non-ftp service,
 266                          * and <protocol,caddr,0,vaddr,0,daddr,0>
 267                          * for ftp service.
 268                          */
 269                         if (svc->port != FTPPORT)
 270                                 ct = ip_vs_conn_new(iph->protocol,
 271                                                     snet, 0,
 272                                                     iph->daddr,
 273                                                     ports[1],
 274                                                     dest->addr, dest->port,
 275                                                     0,
 276                                                     dest);
 277                         else
 278                                 ct = ip_vs_conn_new(iph->protocol,
 279                                                     snet, 0,
 280                                                     iph->daddr, 0,
 281                                                     dest->addr, 0,
 282                                                     0,
 283                                                     dest);
 284                         if (ct == NULL)
 285                                 return NULL;
 286
 287                         ct->timeout = svc->timeout;
 288                 } else {
 289                         /* set destination with the found template */
 290                         dest = ct->dest;
 291                 }
 292                 dport = dest->port;
 293         } else {
 294                 /*
 295                  * Note: persistent fwmark-based services and persistent
 296                  * port zero service are handled here.
 297                  * fwmark template: <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
 298                  * port zero template: <protocol,caddr,0,vaddr,0,daddr,0>
 299                  */
 300                 if (svc->fwmark)
 301                         ct = ip_vs_conn_in_get(IPPROTO_IP, snet, 0,
 302                                                htonl(svc->fwmark), 0);
 303                 else
 304                         ct = ip_vs_conn_in_get(iph->protocol, snet, 0,
 305                                                iph->daddr, 0);
 306
 307                 if (!ct || !ip_vs_check_template(ct)) {
 308                         /*
 309                          * If it is not persistent port zero, return NULL,
 310                          * otherwise create a connection template.
 311                          */
 312                         if (svc->port)
 313                                 return NULL;
 314
 315                         dest = svc->scheduler->schedule(svc, skb);
 316                         if (dest == NULL) {
 317                                 IP_VS_DBG(1, "p-schedule: no dest found.\n");
 318                                 return NULL;
 319                         }
 320
 321                         /*
 322                          * Create a template according to the service
 323                          */
 324                         if (svc->fwmark)
 325                                 ct = ip_vs_conn_new(IPPROTO_IP,
 326                                                     snet, 0,
 327                                                     htonl(svc->fwmark), 0,
 328                                                     dest->addr, 0,
 329                                                     0,
 330                                                     dest);
 331                         else
 332                                 ct = ip_vs_conn_new(iph->protocol,
 333                                                     snet, 0,
 334                                                     iph->daddr, 0,
 335                                                     dest->addr, 0,
 336                                                     0,
 337                                                     dest);
 338                         if (ct == NULL)
 339                                 return NULL;
 340
 341                         ct->timeout = svc->timeout;
 342                 } else {
 343                         /* set destination with the found template */
 344                         dest = ct->dest;
 345                 }
 346                 dport = ports[1];
 347         }
 348
 349         /*
 350          *    Create a new connection according to the template
 351          */
 352         cp = ip_vs_conn_new(iph->protocol,
 353                             iph->saddr, ports[0],
 354                             iph->daddr, ports[1],
 355                             dest->addr, dport,
 356                             0,
 357                             dest);
 358         if (cp == NULL) {
 359                 ip_vs_conn_put(ct);
 360                 return NULL;
 361         }
 362
 363         /*
 364          *    Add its control
 365          */
 366         ip_vs_control_add(cp, ct);
 367         ip_vs_conn_put(ct);
 368
 369         ip_vs_conn_stats(cp, svc);
 370         return cp;
 371 }
 372
 373
 374 /*
 375  *  IPVS main scheduling function
 376  *  It selects a server according to the virtual service, and
 377  *  creates a connection entry.
 378  *  Protocols supported: TCP, UDP
 379  */
 380 struct ip_vs_conn *
 381 ip_vs_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
 382 {
 383         struct ip_vs_conn *cp = NULL;
 384         struct iphdr *iph = skb->nh.iph;
 385         struct ip_vs_dest *dest;
 386         __u16 ports[2];
 387
 388         if (skb_copy_bits(skb, iph->ihl*4, ports, sizeof(ports)) < 0)
 389                 return NULL;
 390
 391         /*
 392          *    Persistent service
 393          */
 394         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 395                 return ip_vs_sched_persist(svc, skb, ports);
 396
 397         /*
 398          *    Non-persistent service
 399          */
 400         if (!svc->fwmark && ports[1] != svc->port) {
 401                 if (!svc->port)
 402                         IP_VS_ERR("Schedule: port zero only supported "
 403                                   "in persistent services, "
 404                                   "check your ipvs configuration\n");
 405                 return NULL;
 406         }
 407
 408         dest = svc->scheduler->schedule(svc, skb);
 409         if (dest == NULL) {
 410                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 411                 return NULL;
 412         }
 413
 414         /*
 415          *    Create a connection entry.
 416          */
 417         cp = ip_vs_conn_new(iph->protocol,
 418                             iph->saddr, ports[0],
 419                             iph->daddr, ports[1],
 420                             dest->addr, dest->port?dest->port:ports[1],
 421                             0,
 422                             dest);
 423         if (cp == NULL)
 424                 return NULL;
 425
 426         IP_VS_DBG(6, "Schedule fwd:%c c:%u.%u.%u.%u:%u v:%u.%u.%u.%u:%u "
 427                   "d:%u.%u.%u.%u:%u flg:%X cnt:%d\n",
 428                   ip_vs_fwd_tag(cp),
 429                   NIPQUAD(cp->caddr), ntohs(cp->cport),
 430                   NIPQUAD(cp->vaddr), ntohs(cp->vport),
 431                   NIPQUAD(cp->daddr), ntohs(cp->dport),
 432                   cp->flags, atomic_read(&cp->refcnt));
 433
 434         ip_vs_conn_stats(cp, svc);
 435         return cp;
 436 }
 437
 438
 439 /*
 440  *  Pass or drop the packet.
 441  *  Called by ip_vs_in, when the virtual service is available but
 442  *  no destination is available for a new connection.
 443  */
 444 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 445                 struct ip_vs_protocol *pp)
 446 {
 447         __u16 ports[2];
 448         struct iphdr *iph = skb->nh.iph;
 449
 450         if (skb_copy_bits(skb, iph->ihl*4, ports, sizeof(ports)) < 0) {
 451                 ip_vs_service_put(svc);
 452                 return NF_DROP;
 453         }
 454
 455         /* if it is fwmark-based service, the cache_bypass sysctl is up
 456            and the destination is RTN_UNICAST (and not local), then create
 457            a cache_bypass connection entry */
 458         if (sysctl_ip_vs_cache_bypass && svc->fwmark
 459             && (inet_addr_type(iph->daddr) == RTN_UNICAST)) {
 460                 int ret, cs;
 461                 struct ip_vs_conn *cp;
 462
 463                 ip_vs_service_put(svc);
 464
 465                 /* create a new connection entry */
 466                 IP_VS_DBG(6, "ip_vs_leave: create a cache_bypass entry\n");
 467                 cp = ip_vs_conn_new(iph->protocol,
 468                                     iph->saddr, ports[0],
 469                                     iph->daddr, ports[1],
 470                                     0, 0,
 471                                     IP_VS_CONN_F_BYPASS,
 472                                     NULL);
 473                 if (cp == NULL)
 474                         return NF_DROP;
 475
 476                 /* statistics */
 477                 ip_vs_in_stats(cp, skb);
 478
 479                 /* set state */
 480                 cs = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pp);
 481
 482                 /* transmit the first SYN packet */
 483                 ret = cp->packet_xmit(skb, cp, pp);
 484                 /* do not touch skb anymore */
 485
 486                 atomic_inc(&cp->in_pkts);
 487                 ip_vs_conn_put(cp);
 488                 return ret;
 489         }
 490
 491         /*
 492          * When the virtual ftp service is presented, packets destined
 493          * for other services on the VIP may get here (except services
 494          * listed in the ipvs table), pass the packets, because it is
 495          * not ipvs job to decide to drop the packets.
 496          */
 497         if ((svc->port == FTPPORT) && (ports[1] != FTPPORT)) {
 498                 ip_vs_service_put(svc);
 499                 return NF_ACCEPT;
 500         }
 501
 502         ip_vs_service_put(svc);
 503
 504         /*
 505          * Notify the client that the destination is unreachable, and
 506          * release the socket buffer.
 507          * Since it is in IP layer, the TCP socket is not actually
 508          * created, the TCP RST packet cannot be sent, instead that
 509          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 510          */
 511         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 512         return NF_DROP;
 513 }
 514
 515
 516 /*
 517  *      It is hooked before NF_IP_PRI_NAT_SRC at the NF_IP_POST_ROUTING
 518  *      chain, and is used for VS/NAT.
 519  *      It detects packets for VS/NAT connections and sends the packets
 520  *      immediately. This can avoid that iptable_nat mangles the packets
 521  *      for VS/NAT.
 522  */
 523 static unsigned int ip_vs_post_routing(unsigned int hooknum,
 524                                        struct sk_buff **pskb,
 525                                        const struct net_device *in,
 526                                        const struct net_device *out,
 527                                        int (*okfn)(struct sk_buff *))
 528 {
 529         if (!((*pskb)->nfcache & NFC_IPVS_PROPERTY))
 530                 return NF_ACCEPT;
 531
 532         /* The packet was sent from IPVS, exit this chain */
 533         (*okfn)(*pskb);
 534
 535         return NF_STOLEN;
 536 }
 537
 538 u16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 539 {
 540         return (u16) csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 541 }
 542
 543 static inline struct sk_buff *
 544 ip_vs_gather_frags(struct sk_buff *skb)
 545 {
 546         skb = ip_defrag(skb);
 547         if (skb)
 548                 ip_send_check(skb->nh.iph);
 549         return skb;
 550 }
 551
 552 /*
 553  * Packet has been made sufficiently writable in caller
 554  * - inout: 1=in->out, 0=out->in
 555  */
 556 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 557                     struct ip_vs_conn *cp, int inout)
 558 {
 559         struct iphdr *iph        = skb->nh.iph;
 560         unsigned int icmp_offset = iph->ihl*4;
 561         struct icmphdr *icmph    = (struct icmphdr *)(skb->nh.raw + icmp_offset);
 562         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 563
 564         if (inout) {
 565                 iph->saddr = cp->vaddr;
 566                 ip_send_check(iph);
 567                 ciph->daddr = cp->vaddr;
 568                 ip_send_check(ciph);
 569         } else {
 570                 iph->daddr = cp->daddr;
 571                 ip_send_check(iph);
 572                 ciph->saddr = cp->daddr;
 573                 ip_send_check(ciph);
 574         }
 575
 576         /* the TCP/UDP port */
 577         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol) {
 578                 __u16 *ports = (void *)ciph + ciph->ihl*4;
 579
 580                 if (inout)
 581                         ports[1] = cp->vport;
 582                 else
 583                         ports[0] = cp->dport;
 584         }
 585
 586         /* And finally the ICMP checksum */
 587         icmph->checksum = 0;
 588         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 589         skb->ip_summed = CHECKSUM_UNNECESSARY;
 590
 591         if (inout)
 592                 IP_VS_DBG_PKT(11, pp, skb, (void *)ciph - (void *)iph,
 593                         "Forwarding altered outgoing ICMP");
 594         else
 595                 IP_VS_DBG_PKT(11, pp, skb, (void *)ciph - (void *)iph,
 596                         "Forwarding altered incoming ICMP");
 597 }
 598
 599 /*
 600  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 601  *      Find any that might be relevant, check against existing connections,
 602  *      forward to the right destination host if relevant.
 603  *      Currently handles error types - unreachable, quench, ttl exceeded.
 604  *      (Only used in VS/NAT)
 605  */
 606 static int ip_vs_out_icmp(struct sk_buff **pskb, int *related)
 607 {
 608         struct sk_buff *skb = *pskb;
 609         struct iphdr *iph;
 610         struct icmphdr  icmph;
 611         struct iphdr    ciph;   /* The ip header contained within the ICMP */
 612         struct ip_vs_conn *cp;
 613         struct ip_vs_protocol *pp;
 614         unsigned int offset, ihl, verdict;
 615
 616         *related = 1;
 617
 618         /* reassemble IP fragments */
 619         if (skb->nh.iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) {
 620                 skb = ip_vs_gather_frags(skb);
 621                 if (!skb)
 622                         return NF_STOLEN;
 623                 *pskb = skb;
 624         }
 625
 626         iph = skb->nh.iph;
 627         offset = ihl = iph->ihl * 4;
 628         if (skb_copy_bits(skb, offset, &icmph, sizeof(icmph)) < 0)
 629                 return NF_DROP;
 630
 631         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %u.%u.%u.%u->%u.%u.%u.%u\n",
 632                   icmph.type, ntohs(icmp_id(&icmph)),
 633                   NIPQUAD(iph->saddr), NIPQUAD(iph->daddr));
 634
 635         /*
 636          * Work through seeing if this is for us.
 637          * These checks are supposed to be in an order that means easy
 638          * things are checked first to speed up processing.... however
 639          * this means that some packets will manage to get a long way
 640          * down this stack and then be rejected, but that's life.
 641          */
 642         if ((icmph.type != ICMP_DEST_UNREACH) &&
 643             (icmph.type != ICMP_SOURCE_QUENCH) &&
 644             (icmph.type != ICMP_TIME_EXCEEDED)) {
 645                 *related = 0;
 646                 return NF_ACCEPT;
 647         }
 648
 649         /* Now find the contained IP header */
 650         offset += sizeof(icmph);
 651         if (skb_copy_bits(skb, offset, &ciph, sizeof(ciph)) < 0)
 652                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 653
 654         pp = ip_vs_proto_get(ciph.protocol);
 655         if (!pp)
 656                 return NF_ACCEPT;
 657
 658         /* Is the embedded protocol header present? */
 659         if (unlikely(ciph.frag_off & __constant_htons(IP_OFFSET) &&
 660                      pp->dont_defrag))
 661                 return NF_ACCEPT;
 662
 663         IP_VS_DBG_PKT(11, pp, skb, offset, "Checking outgoing ICMP for");
 664
 665         offset += ciph.ihl * 4;
 666
 667         /* The embedded headers contain source and dest in reverse order */
 668         cp = pp->conn_out_get(skb, pp, &ciph, offset, 1);
 669         if (!cp)
 670                 return NF_ACCEPT;
 671
 672         verdict = NF_DROP;
 673
 674         if (IP_VS_FWD_METHOD(cp) != 0) {
 675                 IP_VS_ERR("shouldn't reach here, because the box is on the"
 676                           "half connection in the tun/dr module.\n");
 677         }
 678
 679         /* Ensure the checksum is correct */
 680         if (skb->ip_summed != CHECKSUM_UNNECESSARY &&
 681             ip_vs_checksum_complete(skb, ihl)) {
 682                 /* Failed checksum! */
 683                 IP_VS_DBG(1, "Forward ICMP: failed checksum from %d.%d.%d.%d!\n",
 684                           NIPQUAD(iph->saddr));
 685                 goto out;
 686         }
 687
 688         if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol)
 689                 offset += 2 * sizeof(__u16);
 690         if (!ip_vs_make_skb_writable(pskb, offset))
 691                 goto out;
 692         skb = *pskb;
 693
 694         ip_vs_nat_icmp(skb, pp, cp, 1);
 695
 696         /* do the statistics and put it back */
 697         ip_vs_out_stats(cp, skb);
 698
 699         skb->nfcache |= NFC_IPVS_PROPERTY;
 700         verdict = NF_ACCEPT;
 701
 702   out:
 703         __ip_vs_conn_put(cp);
 704
 705         return verdict;
 706 }
 707
 708 static inline int is_tcp_reset(const struct sk_buff *skb)
 709 {
 710         struct tcphdr tcph;
 711
 712         if (skb_copy_bits(skb, skb->nh.iph->ihl * 4, &tcph, sizeof(tcph)) < 0)
 713                 return 0;
 714         return tcph.rst;
 715 }
 716
 717 /*
 718  *      It is hooked at the NF_IP_FORWARD chain, used only for VS/NAT.
 719  *      Check if outgoing packet belongs to the established ip_vs_conn,
 720  *      rewrite addresses of the packet and send it on its way...
 721  */
 722 static unsigned int
 723 ip_vs_out(unsigned int hooknum, struct sk_buff **pskb,
 724           const struct net_device *in, const struct net_device *out,
 725           int (*okfn)(struct sk_buff *))
 726 {
 727         struct sk_buff  *skb = *pskb;
 728         struct iphdr    *iph;
 729         struct ip_vs_protocol *pp;
 730         struct ip_vs_conn *cp;
 731         int ihl;
 732
 733         EnterFunction(11);
 734
 735         if (skb->nfcache & NFC_IPVS_PROPERTY)
 736                 return NF_ACCEPT;
 737
 738         iph = skb->nh.iph;
 739         if (unlikely(iph->protocol == IPPROTO_ICMP)) {
 740                 int related, verdict = ip_vs_out_icmp(pskb, &related);
 741
 742                 if (related)
 743                         return verdict;
 744                 skb = *pskb;
 745                 iph = skb->nh.iph;
 746         }
 747
 748         pp = ip_vs_proto_get(iph->protocol);
 749         if (unlikely(!pp))
 750                 return NF_ACCEPT;
 751
 752         /* reassemble IP fragments */
 753         if (unlikely(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET) &&
 754                      !pp->dont_defrag)) {
 755                 skb = ip_vs_gather_frags(skb);
 756                 if (!skb)
 757                         return NF_STOLEN;
 758                 iph = skb->nh.iph;
 759                 *pskb = skb;
 760         }
 761
 762         ihl = iph->ihl << 2;
 763
 764         /*
 765          * Check if the packet belongs to an existing entry
 766          */
 767         cp = pp->conn_out_get(skb, pp, iph, ihl, 0);
 768
 769         if (unlikely(!cp)) {
 770                 if (sysctl_ip_vs_nat_icmp_send &&
 771                     (pp->protocol == IPPROTO_TCP ||
 772                      pp->protocol == IPPROTO_UDP)) {
 773                         __u16 ports[2];
 774
 775                         if (skb_copy_bits(skb, ihl, ports, sizeof(ports)) < 0)
 776                                 return NF_ACCEPT;       /* Not for me */
 777                         if (ip_vs_lookup_real_service(iph->protocol,
 778                                                       iph->saddr, ports[0])) {
 779                                 /*
 780                                  * Notify the real server: there is no
 781                                  * existing entry if it is not RST
 782                                  * packet or not TCP packet.
 783                                  */
 784                                 if (iph->protocol != IPPROTO_TCP
 785                                     || !is_tcp_reset(skb)) {
 786                                         icmp_send(skb,ICMP_DEST_UNREACH,
 787                                                   ICMP_PORT_UNREACH, 0);
 788                                         return NF_DROP;
 789                                 }
 790                         }
 791                 }
 792                 IP_VS_DBG_PKT(12, pp, skb, 0,
 793                               "packet continues traversal as normal");
 794                 return NF_ACCEPT;
 795         }
 796
 797         IP_VS_DBG_PKT(11, pp, skb, 0, "Outgoing packet");
 798
 799         if (!ip_vs_make_skb_writable(pskb, ihl))
 800                 goto drop;
 801
 802         /* mangle the packet */
 803         if (pp->snat_handler && !pp->snat_handler(pskb, pp, cp))
 804                 goto drop;
 805         skb = *pskb;
 806         skb->nh.iph->saddr = cp->vaddr;
 807         ip_send_check(skb->nh.iph);
 808
 809         IP_VS_DBG_PKT(10, pp, skb, 0, "After SNAT");
 810
 811         ip_vs_out_stats(cp, skb);
 812         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pp);
 813         ip_vs_conn_put(cp);
 814
 815         skb->nfcache |= NFC_IPVS_PROPERTY;
 816
 817         LeaveFunction(11);
 818         return NF_ACCEPT;
 819
 820   drop:
 821         ip_vs_conn_put(cp);
 822         kfree_skb(*pskb);
 823         return NF_STOLEN;
 824 }
 825
 826
 827 /*
 828  *      Check if the packet is for VS/NAT connections, then send it
 829  *      immediately.
 830  *      Called by ip_fw_compact to detect packets for VS/NAT before
 831  *      they are changed by ipchains masquerading code.
 832  */
 833 unsigned int
 834 check_for_ip_vs_out(struct sk_buff **pskb, int (*okfn)(struct sk_buff *))
 835 {
 836         unsigned int ret;
 837
 838         ret = ip_vs_out(NF_IP_FORWARD, pskb, NULL, NULL, NULL);
 839         if (ret != NF_ACCEPT) {
 840                 return ret;
 841         } else {
 842                 /* send the packet immediately if it is already mangled
 843                    by ip_vs_out */
 844                 if ((*pskb)->nfcache & NFC_IPVS_PROPERTY) {
 845                         (*okfn)(*pskb);
 846                         return NF_STOLEN;
 847                 }
 848         }
 849         return NF_ACCEPT;
 850 }
 851
 852 /*
 853  *      Handle ICMP messages in the outside-to-inside direction (incoming).
 854  *      Find any that might be relevant, check against existing connections,
 855  *      forward to the right destination host if relevant.
 856  *      Currently handles error types - unreachable, quench, ttl exceeded.
 857  */
 858 static int ip_vs_in_icmp(struct sk_buff **pskb, int *related)
 859 {
 860         struct sk_buff *skb = *pskb;
 861         struct iphdr *iph;
 862         struct icmphdr  icmph;
 863         struct iphdr    ciph;   /* The ip header contained within the ICMP */
 864         struct ip_vs_conn *cp;
 865         struct ip_vs_protocol *pp;
 866         unsigned int offset, ihl, verdict;
 867
 868         *related = 1;
 869
 870         /* reassemble IP fragments */
 871         if (skb->nh.iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) {
 872                 skb = ip_vs_gather_frags(skb);
 873                 if (!skb)
 874                         return NF_STOLEN;
 875                 *pskb = skb;
 876         }
 877
 878         iph = skb->nh.iph;
 879         offset = ihl = iph->ihl * 4;
 880         if (skb_copy_bits(skb, offset, &icmph, sizeof(icmph)) < 0)
 881                 return NF_DROP;
 882
 883         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %u.%u.%u.%u->%u.%u.%u.%u\n",
 884                   icmph.type, ntohs(icmp_id(&icmph)),
 885                   NIPQUAD(iph->saddr), NIPQUAD(iph->daddr));
 886
 887         /*
 888          * Work through seeing if this is for us.
 889          * These checks are supposed to be in an order that means easy
 890          * things are checked first to speed up processing.... however
 891          * this means that some packets will manage to get a long way
 892          * down this stack and then be rejected, but that's life.
 893          */
 894         if ((icmph.type != ICMP_DEST_UNREACH) &&
 895             (icmph.type != ICMP_SOURCE_QUENCH) &&
 896             (icmph.type != ICMP_TIME_EXCEEDED)) {
 897                 *related = 0;
 898                 return NF_ACCEPT;
 899         }
 900
 901         /* Now find the contained IP header */
 902         offset += sizeof(icmph);
 903         if (skb_copy_bits(skb, offset, &ciph, sizeof(ciph)) < 0)
 904                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 905
 906         pp = ip_vs_proto_get(ciph.protocol);
 907         if (!pp)
 908                 return NF_ACCEPT;
 909
 910         /* Is the embedded protocol header present? */
 911         if (unlikely(ciph.frag_off & __constant_htons(IP_OFFSET) &&
 912                      pp->dont_defrag))
 913                 return NF_ACCEPT;
 914
 915         IP_VS_DBG_PKT(11, pp, skb, offset, "Checking incoming ICMP for");
 916
 917         offset += ciph.ihl * 4;
 918
 919         /* The embedded headers contain source and dest in reverse order */
 920         cp = pp->conn_in_get(skb, pp, &ciph, offset, 1);
 921         if (!cp)
 922                 return NF_ACCEPT;
 923
 924         verdict = NF_DROP;
 925
 926         /* Ensure the checksum is correct */
 927         if (skb->ip_summed != CHECKSUM_UNNECESSARY &&
 928             ip_vs_checksum_complete(skb, ihl)) {
 929                 /* Failed checksum! */
 930                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %d.%d.%d.%d!\n",
 931                           NIPQUAD(iph->saddr));
 932                 goto out;
 933         }
 934
 935         /* do the statistics and put it back */
 936         ip_vs_in_stats(cp, skb);
 937         if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol)
 938                 offset += 2 * sizeof(__u16);
 939         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset);
 940         /* do not touch skb anymore */
 941
 942   out:
 943         __ip_vs_conn_put(cp);
 944
 945         return verdict;
 946 }
 947
 948 /*
 949  *      Check if it's for virtual services, look it up,
 950  *      and send it on its way...
 951  */
 952 static unsigned int
 953 ip_vs_in(unsigned int hooknum, struct sk_buff **pskb,
 954          const struct net_device *in, const struct net_device *out,
 955          int (*okfn)(struct sk_buff *))
 956 {
 957         struct sk_buff  *skb = *pskb;
 958         struct iphdr    *iph;
 959         struct ip_vs_protocol *pp;
 960         struct ip_vs_conn *cp;
 961         int ret, restart;
 962         int ihl;
 963
 964         /*
 965          *      Big tappo: only PACKET_HOST (neither loopback nor mcasts)
 966          *      ... don't know why 1st test DOES NOT include 2nd (?)
 967          */
 968         if (unlikely(skb->pkt_type != PACKET_HOST
 969                      || skb->dev == &loopback_dev || skb->sk)) {
 970                 IP_VS_DBG(12, "packet type=%d proto=%d daddr=%d.%d.%d.%d ignored\n",
 971                           skb->pkt_type,
 972                           skb->nh.iph->protocol,
 973                           NIPQUAD(skb->nh.iph->daddr));
 974                 return NF_ACCEPT;
 975         }
 976
 977         iph = skb->nh.iph;
 978         if (unlikely(iph->protocol == IPPROTO_ICMP)) {
 979                 int related, verdict = ip_vs_in_icmp(pskb, &related);
 980
 981                 if (related)
 982                         return verdict;
 983                 skb = *pskb;
 984                 iph = skb->nh.iph;
 985         }
 986
 987         /* Protocol supported? */
 988         pp = ip_vs_proto_get(iph->protocol);
 989         if (unlikely(!pp))
 990                 return NF_ACCEPT;
 991
 992         ihl = iph->ihl << 2;
 993
 994         /*
 995          * Check if the packet belongs to an existing connection entry
 996          */
 997         cp = pp->conn_in_get(skb, pp, iph, ihl, 0);
 998
 999         if (unlikely(!cp)) {
1000                 int v;
1001
1002                 if (!pp->conn_schedule(skb, pp, &v, &cp))
1003                         return v;
1004         }
1005
1006         if (unlikely(!cp)) {
1007                 /* sorry, all this trouble for a no-hit :) */
1008                 IP_VS_DBG_PKT(12, pp, skb, 0,
1009                               "packet continues traversal as normal");
1010                 return NF_ACCEPT;
1011         }
1012
1013         IP_VS_DBG_PKT(11, pp, skb, 0, "Incoming packet");
1014
1015         /* Check the server status */
1016         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1017                 /* the destination server is not availabe */
1018
1019                 if (sysctl_ip_vs_expire_nodest_conn) {
1020                         /* try to expire the connection immediately */
1021                         ip_vs_conn_expire_now(cp);
1022                 } else {
1023                         /* don't restart its timer, and silently
1024                            drop the packet. */
1025                         __ip_vs_conn_put(cp);
1026                 }
1027                 return NF_DROP;
1028         }
1029
1030         ip_vs_in_stats(cp, skb);
1031         restart = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pp);
1032         if (cp->packet_xmit)
1033                 ret = cp->packet_xmit(skb, cp, pp);
1034                 /* do not touch skb anymore */
1035         else {
1036                 IP_VS_DBG_RL("warning: packet_xmit is null");
1037                 ret = NF_ACCEPT;
1038         }
1039
1040         /* increase its packet counter and check if it is needed
1041            to be synchronized */
1042         atomic_inc(&cp->in_pkts);
1043         if ((ip_vs_sync_state & IP_VS_STATE_MASTER) &&
1044             (cp->protocol != IPPROTO_TCP ||
1045              cp->state == IP_VS_TCP_S_ESTABLISHED) &&
1046             (atomic_read(&cp->in_pkts) % sysctl_ip_vs_sync_threshold[1]
1047              == sysctl_ip_vs_sync_threshold[0]))
1048                 ip_vs_sync_conn(cp);
1049
1050         ip_vs_conn_put(cp);
1051         return ret;
1052 }
1053
1054
1055 /*
1056  *      It is hooked at the NF_IP_FORWARD chain, in order to catch ICMP
1057  *      related packets destined for 0.0.0.0/0.
1058  *      When fwmark-based virtual service is used, such as transparent
1059  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
1060  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1061  *      sent to ip_vs_in_icmp. So, catch them at the NF_IP_FORWARD chain
1062  *      and send them to ip_vs_in_icmp.
1063  */
1064 static unsigned int
1065 ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff **pskb,
1066                    const struct net_device *in, const struct net_device *out,
1067                    int (*okfn)(struct sk_buff *))
1068 {
1069         int r;
1070
1071         if ((*pskb)->nh.iph->protocol != IPPROTO_ICMP)
1072                 return NF_ACCEPT;
1073
1074         return ip_vs_in_icmp(pskb, &r);
1075 }
1076
1077
1078 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1079    or VS/NAT(change destination), so that filtering rules can be
1080    applied to IPVS. */
1081 static struct nf_hook_ops ip_vs_in_ops = {
1082         .hook           = ip_vs_in,
1083         .owner          = THIS_MODULE,
1084         .pf             = PF_INET,
1085         .hooknum        = NF_IP_LOCAL_IN,
1086         .priority       = 100,
1087 };
1088
1089 /* After packet filtering, change source only for VS/NAT */
1090 static struct nf_hook_ops ip_vs_out_ops = {
1091         .hook           = ip_vs_out,
1092         .owner          = THIS_MODULE,
1093         .pf             = PF_INET,
1094         .hooknum        = NF_IP_FORWARD,
1095         .priority       = 100,
1096 };
1097
1098 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1099    destined for 0.0.0.0/0, which is for incoming IPVS connections */
1100 static struct nf_hook_ops ip_vs_forward_icmp_ops = {
1101         .hook           = ip_vs_forward_icmp,
1102         .owner          = THIS_MODULE,
1103         .pf             = PF_INET,
1104         .hooknum        = NF_IP_FORWARD,
1105         .priority       = 99,
1106 };
1107
1108 /* Before the netfilter connection tracking, exit from POST_ROUTING */
1109 static struct nf_hook_ops ip_vs_post_routing_ops = {
1110         .hook           = ip_vs_post_routing,
1111         .owner          = THIS_MODULE,
1112         .pf             = PF_INET,
1113         .hooknum        = NF_IP_POST_ROUTING,
1114         .priority       = NF_IP_PRI_NAT_SRC-1,
1115 };
1116
1117
1118 /*
1119  *      Initialize IP Virtual Server
1120  */
1121 static int __init ip_vs_init(void)
1122 {
1123         int ret;
1124
1125         ret = ip_vs_control_init();
1126         if (ret < 0) {
1127                 IP_VS_ERR("can't setup control.\n");
1128                 goto cleanup_nothing;
1129         }
1130
1131         ip_vs_protocol_init();
1132
1133         ret = ip_vs_app_init();
1134         if (ret < 0) {
1135                 IP_VS_ERR("can't setup application helper.\n");
1136                 goto cleanup_protocol;
1137         }
1138
1139         ret = ip_vs_conn_init();
1140         if (ret < 0) {
1141                 IP_VS_ERR("can't setup connection table.\n");
1142                 goto cleanup_app;
1143         }
1144
1145         ret = nf_register_hook(&ip_vs_in_ops);
1146         if (ret < 0) {
1147                 IP_VS_ERR("can't register in hook.\n");
1148                 goto cleanup_conn;
1149         }
1150
1151         ret = nf_register_hook(&ip_vs_out_ops);
1152         if (ret < 0) {
1153                 IP_VS_ERR("can't register out hook.\n");
1154                 goto cleanup_inops;
1155         }
1156         ret = nf_register_hook(&ip_vs_post_routing_ops);
1157         if (ret < 0) {
1158                 IP_VS_ERR("can't register post_routing hook.\n");
1159                 goto cleanup_outops;
1160         }
1161         ret = nf_register_hook(&ip_vs_forward_icmp_ops);
1162         if (ret < 0) {
1163                 IP_VS_ERR("can't register forward_icmp hook.\n");
1164                 goto cleanup_postroutingops;
1165         }
1166
1167         IP_VS_INFO("ipvs loaded.\n");
1168         return ret;
1169
1170   cleanup_postroutingops:
1171         nf_unregister_hook(&ip_vs_post_routing_ops);
1172   cleanup_outops:
1173         nf_unregister_hook(&ip_vs_out_ops);
1174   cleanup_inops:
1175         nf_unregister_hook(&ip_vs_in_ops);
1176   cleanup_conn:
1177         ip_vs_conn_cleanup();
1178   cleanup_app:
1179         ip_vs_app_cleanup();
1180   cleanup_protocol:
1181         ip_vs_protocol_cleanup();
1182         ip_vs_control_cleanup();
1183   cleanup_nothing:
1184         return ret;
1185 }
1186
1187 static void __exit ip_vs_cleanup(void)
1188 {
1189         nf_unregister_hook(&ip_vs_forward_icmp_ops);
1190         nf_unregister_hook(&ip_vs_post_routing_ops);
1191         nf_unregister_hook(&ip_vs_out_ops);
1192         nf_unregister_hook(&ip_vs_in_ops);
1193         ip_vs_conn_cleanup();
1194         ip_vs_app_cleanup();
1195         ip_vs_protocol_cleanup();
1196         ip_vs_control_cleanup();
1197         IP_VS_INFO("ipvs unloaded.\n");
1198 }
1199
1200 module_init(ip_vs_init);
1201 module_exit(ip_vs_cleanup);
1202 MODULE_LICENSE("GPL");