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  *      Harald Welte                    don't use nfcache
  26  *
  27  */
  28
  29 #include <linux/module.h>
  30 #include <linux/kernel.h>
  31 #include <linux/ip.h>
  32 #include <linux/tcp.h>
  33 #include <linux/icmp.h>
  34
  35 #include <net/ip.h>
  36 #include <net/tcp.h>
  37 #include <net/udp.h>
  38 #include <net/icmp.h>                   /* for icmp_send */
  39 #include <net/route.h>
  40
  41 #include <linux/netfilter.h>
  42 #include <linux/netfilter_ipv4.h>
  43
  44 #include <net/ip_vs.h>
  45
  46
  47 EXPORT_SYMBOL(register_ip_vs_scheduler);
  48 EXPORT_SYMBOL(unregister_ip_vs_scheduler);
  49 EXPORT_SYMBOL(ip_vs_skb_replace);
  50 EXPORT_SYMBOL(ip_vs_proto_name);
  51 EXPORT_SYMBOL(ip_vs_conn_new);
  52 EXPORT_SYMBOL(ip_vs_conn_in_get);
  53 EXPORT_SYMBOL(ip_vs_conn_out_get);
  54 #ifdef CONFIG_IP_VS_PROTO_TCP
  55 EXPORT_SYMBOL(ip_vs_tcp_conn_listen);
  56 #endif
  57 EXPORT_SYMBOL(ip_vs_conn_put);
  58 #ifdef CONFIG_IP_VS_DEBUG
  59 EXPORT_SYMBOL(ip_vs_get_debug_level);
  60 #endif
  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                     __be16 ports[2])
 213 {
 214         struct ip_vs_conn *cp = NULL;
 215         struct iphdr *iph = ip_hdr(skb);
 216         struct ip_vs_dest *dest;
 217         struct ip_vs_conn *ct;
 218         __be16  dport;   /* destination port to forward */
 219         __be32  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_ct_in_get(iph->protocol, snet, 0,
 247                                                iph->daddr, ports[1]);
 248                 else
 249                         ct = ip_vs_ct_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                                                     IP_VS_CONN_F_TEMPLATE,
 276                                                     dest);
 277                         else
 278                                 ct = ip_vs_conn_new(iph->protocol,
 279                                                     snet, 0,
 280                                                     iph->daddr, 0,
 281                                                     dest->addr, 0,
 282                                                     IP_VS_CONN_F_TEMPLATE,
 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_ct_in_get(IPPROTO_IP, snet, 0,
 302                                                htonl(svc->fwmark), 0);
 303                 else
 304                         ct = ip_vs_ct_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                                                     IP_VS_CONN_F_TEMPLATE,
 330                                                     dest);
 331                         else
 332                                 ct = ip_vs_conn_new(iph->protocol,
 333                                                     snet, 0,
 334                                                     iph->daddr, 0,
 335                                                     dest->addr, 0,
 336                                                     IP_VS_CONN_F_TEMPLATE,
 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 = ip_hdr(skb);
 385         struct ip_vs_dest *dest;
 386         __be16 _ports[2], *pptr;
 387
 388         pptr = skb_header_pointer(skb, iph->ihl*4,
 389                                   sizeof(_ports), _ports);
 390         if (pptr == NULL)
 391                 return NULL;
 392
 393         /*
 394          *    Persistent service
 395          */
 396         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 397                 return ip_vs_sched_persist(svc, skb, pptr);
 398
 399         /*
 400          *    Non-persistent service
 401          */
 402         if (!svc->fwmark && pptr[1] != svc->port) {
 403                 if (!svc->port)
 404                         IP_VS_ERR("Schedule: port zero only supported "
 405                                   "in persistent services, "
 406                                   "check your ipvs configuration\n");
 407                 return NULL;
 408         }
 409
 410         dest = svc->scheduler->schedule(svc, skb);
 411         if (dest == NULL) {
 412                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 413                 return NULL;
 414         }
 415
 416         /*
 417          *    Create a connection entry.
 418          */
 419         cp = ip_vs_conn_new(iph->protocol,
 420                             iph->saddr, pptr[0],
 421                             iph->daddr, pptr[1],
 422                             dest->addr, dest->port?dest->port:pptr[1],
 423                             0,
 424                             dest);
 425         if (cp == NULL)
 426                 return NULL;
 427
 428         IP_VS_DBG(6, "Schedule fwd:%c c:%u.%u.%u.%u:%u v:%u.%u.%u.%u:%u "
 429                   "d:%u.%u.%u.%u:%u conn->flags:%X conn->refcnt:%d\n",
 430                   ip_vs_fwd_tag(cp),
 431                   NIPQUAD(cp->caddr), ntohs(cp->cport),
 432                   NIPQUAD(cp->vaddr), ntohs(cp->vport),
 433                   NIPQUAD(cp->daddr), ntohs(cp->dport),
 434                   cp->flags, atomic_read(&cp->refcnt));
 435
 436         ip_vs_conn_stats(cp, svc);
 437         return cp;
 438 }
 439
 440
 441 /*
 442  *  Pass or drop the packet.
 443  *  Called by ip_vs_in, when the virtual service is available but
 444  *  no destination is available for a new connection.
 445  */
 446 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 447                 struct ip_vs_protocol *pp)
 448 {
 449         __be16 _ports[2], *pptr;
 450         struct iphdr *iph = ip_hdr(skb);
 451
 452         pptr = skb_header_pointer(skb, iph->ihl*4,
 453                                   sizeof(_ports), _ports);
 454         if (pptr == NULL) {
 455                 ip_vs_service_put(svc);
 456                 return NF_DROP;
 457         }
 458
 459         /* if it is fwmark-based service, the cache_bypass sysctl is up
 460            and the destination is RTN_UNICAST (and not local), then create
 461            a cache_bypass connection entry */
 462         if (sysctl_ip_vs_cache_bypass && svc->fwmark
 463             && (inet_addr_type(iph->daddr) == RTN_UNICAST)) {
 464                 int ret, cs;
 465                 struct ip_vs_conn *cp;
 466
 467                 ip_vs_service_put(svc);
 468
 469                 /* create a new connection entry */
 470                 IP_VS_DBG(6, "ip_vs_leave: create a cache_bypass entry\n");
 471                 cp = ip_vs_conn_new(iph->protocol,
 472                                     iph->saddr, pptr[0],
 473                                     iph->daddr, pptr[1],
 474                                     0, 0,
 475                                     IP_VS_CONN_F_BYPASS,
 476                                     NULL);
 477                 if (cp == NULL)
 478                         return NF_DROP;
 479
 480                 /* statistics */
 481                 ip_vs_in_stats(cp, skb);
 482
 483                 /* set state */
 484                 cs = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pp);
 485
 486                 /* transmit the first SYN packet */
 487                 ret = cp->packet_xmit(skb, cp, pp);
 488                 /* do not touch skb anymore */
 489
 490                 atomic_inc(&cp->in_pkts);
 491                 ip_vs_conn_put(cp);
 492                 return ret;
 493         }
 494
 495         /*
 496          * When the virtual ftp service is presented, packets destined
 497          * for other services on the VIP may get here (except services
 498          * listed in the ipvs table), pass the packets, because it is
 499          * not ipvs job to decide to drop the packets.
 500          */
 501         if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT)) {
 502                 ip_vs_service_put(svc);
 503                 return NF_ACCEPT;
 504         }
 505
 506         ip_vs_service_put(svc);
 507
 508         /*
 509          * Notify the client that the destination is unreachable, and
 510          * release the socket buffer.
 511          * Since it is in IP layer, the TCP socket is not actually
 512          * created, the TCP RST packet cannot be sent, instead that
 513          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 514          */
 515         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 516         return NF_DROP;
 517 }
 518
 519
 520 /*
 521  *      It is hooked before NF_IP_PRI_NAT_SRC at the NF_IP_POST_ROUTING
 522  *      chain, and is used for VS/NAT.
 523  *      It detects packets for VS/NAT connections and sends the packets
 524  *      immediately. This can avoid that iptable_nat mangles the packets
 525  *      for VS/NAT.
 526  */
 527 static unsigned int ip_vs_post_routing(unsigned int hooknum,
 528                                        struct sk_buff **pskb,
 529                                        const struct net_device *in,
 530                                        const struct net_device *out,
 531                                        int (*okfn)(struct sk_buff *))
 532 {
 533         if (!((*pskb)->ipvs_property))
 534                 return NF_ACCEPT;
 535         /* The packet was sent from IPVS, exit this chain */
 536         return NF_STOP;
 537 }
 538
 539 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 540 {
 541         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 542 }
 543
 544 static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user)
 545 {
 546         int err = ip_defrag(skb, user);
 547
 548         if (!err)
 549                 ip_send_check(ip_hdr(skb));
 550
 551         return err;
 552 }
 553
 554 /*
 555  * Packet has been made sufficiently writable in caller
 556  * - inout: 1=in->out, 0=out->in
 557  */
 558 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 559                     struct ip_vs_conn *cp, int inout)
 560 {
 561         struct iphdr *iph        = ip_hdr(skb);
 562         unsigned int icmp_offset = iph->ihl*4;
 563         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
 564                                                       icmp_offset);
 565         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 566
 567         if (inout) {
 568                 iph->saddr = cp->vaddr;
 569                 ip_send_check(iph);
 570                 ciph->daddr = cp->vaddr;
 571                 ip_send_check(ciph);
 572         } else {
 573                 iph->daddr = cp->daddr;
 574                 ip_send_check(iph);
 575                 ciph->saddr = cp->daddr;
 576                 ip_send_check(ciph);
 577         }
 578
 579         /* the TCP/UDP port */
 580         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol) {
 581                 __be16 *ports = (void *)ciph + ciph->ihl*4;
 582
 583                 if (inout)
 584                         ports[1] = cp->vport;
 585                 else
 586                         ports[0] = cp->dport;
 587         }
 588
 589         /* And finally the ICMP checksum */
 590         icmph->checksum = 0;
 591         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 592         skb->ip_summed = CHECKSUM_UNNECESSARY;
 593
 594         if (inout)
 595                 IP_VS_DBG_PKT(11, pp, skb, (void *)ciph - (void *)iph,
 596                         "Forwarding altered outgoing ICMP");
 597         else
 598                 IP_VS_DBG_PKT(11, pp, skb, (void *)ciph - (void *)iph,
 599                         "Forwarding altered incoming ICMP");
 600 }
 601
 602 /*
 603  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 604  *      Find any that might be relevant, check against existing connections,
 605  *      forward to the right destination host if relevant.
 606  *      Currently handles error types - unreachable, quench, ttl exceeded.
 607  *      (Only used in VS/NAT)
 608  */
 609 static int ip_vs_out_icmp(struct sk_buff **pskb, int *related)
 610 {
 611         struct sk_buff *skb = *pskb;
 612         struct iphdr *iph;
 613         struct icmphdr  _icmph, *ic;
 614         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 615         struct ip_vs_conn *cp;
 616         struct ip_vs_protocol *pp;
 617         unsigned int offset, ihl, verdict;
 618
 619         *related = 1;
 620
 621         /* reassemble IP fragments */
 622         if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
 623                 if (ip_vs_gather_frags(skb, IP_DEFRAG_VS_OUT))
 624                         return NF_STOLEN;
 625         }
 626
 627         iph = ip_hdr(skb);
 628         offset = ihl = iph->ihl * 4;
 629         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 630         if (ic == NULL)
 631                 return NF_DROP;
 632
 633         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %u.%u.%u.%u->%u.%u.%u.%u\n",
 634                   ic->type, ntohs(icmp_id(ic)),
 635                   NIPQUAD(iph->saddr), NIPQUAD(iph->daddr));
 636
 637         /*
 638          * Work through seeing if this is for us.
 639          * These checks are supposed to be in an order that means easy
 640          * things are checked first to speed up processing.... however
 641          * this means that some packets will manage to get a long way
 642          * down this stack and then be rejected, but that's life.
 643          */
 644         if ((ic->type != ICMP_DEST_UNREACH) &&
 645             (ic->type != ICMP_SOURCE_QUENCH) &&
 646             (ic->type != ICMP_TIME_EXCEEDED)) {
 647                 *related = 0;
 648                 return NF_ACCEPT;
 649         }
 650
 651         /* Now find the contained IP header */
 652         offset += sizeof(_icmph);
 653         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 654         if (cih == NULL)
 655                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 656
 657         pp = ip_vs_proto_get(cih->protocol);
 658         if (!pp)
 659                 return NF_ACCEPT;
 660
 661         /* Is the embedded protocol header present? */
 662         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 663                      pp->dont_defrag))
 664                 return NF_ACCEPT;
 665
 666         IP_VS_DBG_PKT(11, pp, skb, offset, "Checking outgoing ICMP for");
 667
 668         offset += cih->ihl * 4;
 669
 670         /* The embedded headers contain source and dest in reverse order */
 671         cp = pp->conn_out_get(skb, pp, cih, offset, 1);
 672         if (!cp)
 673                 return NF_ACCEPT;
 674
 675         verdict = NF_DROP;
 676
 677         if (IP_VS_FWD_METHOD(cp) != 0) {
 678                 IP_VS_ERR("shouldn't reach here, because the box is on the"
 679                           "half connection in the tun/dr module.\n");
 680         }
 681
 682         /* Ensure the checksum is correct */
 683         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 684                 /* Failed checksum! */
 685                 IP_VS_DBG(1, "Forward ICMP: failed checksum from %d.%d.%d.%d!\n",
 686                           NIPQUAD(iph->saddr));
 687                 goto out;
 688         }
 689
 690         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol)
 691                 offset += 2 * sizeof(__u16);
 692         if (!ip_vs_make_skb_writable(pskb, offset))
 693                 goto out;
 694         skb = *pskb;
 695
 696         ip_vs_nat_icmp(skb, pp, cp, 1);
 697
 698         /* do the statistics and put it back */
 699         ip_vs_out_stats(cp, skb);
 700
 701         skb->ipvs_property = 1;
 702         verdict = NF_ACCEPT;
 703
 704   out:
 705         __ip_vs_conn_put(cp);
 706
 707         return verdict;
 708 }
 709
 710 static inline int is_tcp_reset(const struct sk_buff *skb)
 711 {
 712         struct tcphdr _tcph, *th;
 713
 714         th = skb_header_pointer(skb, ip_hdrlen(skb), sizeof(_tcph), &_tcph);
 715         if (th == NULL)
 716                 return 0;
 717         return th->rst;
 718 }
 719
 720 /*
 721  *      It is hooked at the NF_IP_FORWARD chain, used only for VS/NAT.
 722  *      Check if outgoing packet belongs to the established ip_vs_conn,
 723  *      rewrite addresses of the packet and send it on its way...
 724  */
 725 static unsigned int
 726 ip_vs_out(unsigned int hooknum, struct sk_buff **pskb,
 727           const struct net_device *in, const struct net_device *out,
 728           int (*okfn)(struct sk_buff *))
 729 {
 730         struct sk_buff  *skb = *pskb;
 731         struct iphdr    *iph;
 732         struct ip_vs_protocol *pp;
 733         struct ip_vs_conn *cp;
 734         int ihl;
 735
 736         EnterFunction(11);
 737
 738         if (skb->ipvs_property)
 739                 return NF_ACCEPT;
 740
 741         iph = ip_hdr(skb);
 742         if (unlikely(iph->protocol == IPPROTO_ICMP)) {
 743                 int related, verdict = ip_vs_out_icmp(pskb, &related);
 744
 745                 if (related)
 746                         return verdict;
 747                 skb = *pskb;
 748                 iph = ip_hdr(skb);
 749         }
 750
 751         pp = ip_vs_proto_get(iph->protocol);
 752         if (unlikely(!pp))
 753                 return NF_ACCEPT;
 754
 755         /* reassemble IP fragments */
 756         if (unlikely(iph->frag_off & htons(IP_MF|IP_OFFSET) &&
 757                      !pp->dont_defrag)) {
 758                 if (ip_vs_gather_frags(skb, IP_DEFRAG_VS_OUT))
 759                         return NF_STOLEN;
 760                 iph = ip_hdr(skb);
 761         }
 762
 763         ihl = iph->ihl << 2;
 764
 765         /*
 766          * Check if the packet belongs to an existing entry
 767          */
 768         cp = pp->conn_out_get(skb, pp, iph, ihl, 0);
 769
 770         if (unlikely(!cp)) {
 771                 if (sysctl_ip_vs_nat_icmp_send &&
 772                     (pp->protocol == IPPROTO_TCP ||
 773                      pp->protocol == IPPROTO_UDP)) {
 774                         __be16 _ports[2], *pptr;
 775
 776                         pptr = skb_header_pointer(skb, ihl,
 777                                                   sizeof(_ports), _ports);
 778                         if (pptr == NULL)
 779                                 return NF_ACCEPT;       /* Not for me */
 780                         if (ip_vs_lookup_real_service(iph->protocol,
 781                                                       iph->saddr, pptr[0])) {
 782                                 /*
 783                                  * Notify the real server: there is no
 784                                  * existing entry if it is not RST
 785                                  * packet or not TCP packet.
 786                                  */
 787                                 if (iph->protocol != IPPROTO_TCP
 788                                     || !is_tcp_reset(skb)) {
 789                                         icmp_send(skb,ICMP_DEST_UNREACH,
 790                                                   ICMP_PORT_UNREACH, 0);
 791                                         return NF_DROP;
 792                                 }
 793                         }
 794                 }
 795                 IP_VS_DBG_PKT(12, pp, skb, 0,
 796                               "packet continues traversal as normal");
 797                 return NF_ACCEPT;
 798         }
 799
 800         IP_VS_DBG_PKT(11, pp, skb, 0, "Outgoing packet");
 801
 802         if (!ip_vs_make_skb_writable(pskb, ihl))
 803                 goto drop;
 804
 805         /* mangle the packet */
 806         if (pp->snat_handler && !pp->snat_handler(pskb, pp, cp))
 807                 goto drop;
 808         skb = *pskb;
 809         ip_hdr(skb)->saddr = cp->vaddr;
 810         ip_send_check(ip_hdr(skb));
 811
 812         /* For policy routing, packets originating from this
 813          * machine itself may be routed differently to packets
 814          * passing through.  We want this packet to be routed as
 815          * if it came from this machine itself.  So re-compute
 816          * the routing information.
 817          */
 818         if (ip_route_me_harder(pskb, RTN_LOCAL) != 0)
 819                 goto drop;
 820         skb = *pskb;
 821
 822         IP_VS_DBG_PKT(10, pp, skb, 0, "After SNAT");
 823
 824         ip_vs_out_stats(cp, skb);
 825         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pp);
 826         ip_vs_conn_put(cp);
 827
 828         skb->ipvs_property = 1;
 829
 830         LeaveFunction(11);
 831         return NF_ACCEPT;
 832
 833   drop:
 834         ip_vs_conn_put(cp);
 835         kfree_skb(*pskb);
 836         return NF_STOLEN;
 837 }
 838
 839
 840 /*
 841  *      Handle ICMP messages in the outside-to-inside direction (incoming).
 842  *      Find any that might be relevant, check against existing connections,
 843  *      forward to the right destination host if relevant.
 844  *      Currently handles error types - unreachable, quench, ttl exceeded.
 845  */
 846 static int
 847 ip_vs_in_icmp(struct sk_buff **pskb, int *related, unsigned int hooknum)
 848 {
 849         struct sk_buff *skb = *pskb;
 850         struct iphdr *iph;
 851         struct icmphdr  _icmph, *ic;
 852         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 853         struct ip_vs_conn *cp;
 854         struct ip_vs_protocol *pp;
 855         unsigned int offset, ihl, verdict;
 856
 857         *related = 1;
 858
 859         /* reassemble IP fragments */
 860         if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
 861                 if (ip_vs_gather_frags(skb, hooknum == NF_IP_LOCAL_IN ?
 862                                             IP_DEFRAG_VS_IN : IP_DEFRAG_VS_FWD))
 863                         return NF_STOLEN;
 864         }
 865
 866         iph = ip_hdr(skb);
 867         offset = ihl = iph->ihl * 4;
 868         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 869         if (ic == NULL)
 870                 return NF_DROP;
 871
 872         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %u.%u.%u.%u->%u.%u.%u.%u\n",
 873                   ic->type, ntohs(icmp_id(ic)),
 874                   NIPQUAD(iph->saddr), NIPQUAD(iph->daddr));
 875
 876         /*
 877          * Work through seeing if this is for us.
 878          * These checks are supposed to be in an order that means easy
 879          * things are checked first to speed up processing.... however
 880          * this means that some packets will manage to get a long way
 881          * down this stack and then be rejected, but that's life.
 882          */
 883         if ((ic->type != ICMP_DEST_UNREACH) &&
 884             (ic->type != ICMP_SOURCE_QUENCH) &&
 885             (ic->type != ICMP_TIME_EXCEEDED)) {
 886                 *related = 0;
 887                 return NF_ACCEPT;
 888         }
 889
 890         /* Now find the contained IP header */
 891         offset += sizeof(_icmph);
 892         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 893         if (cih == NULL)
 894                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 895
 896         pp = ip_vs_proto_get(cih->protocol);
 897         if (!pp)
 898                 return NF_ACCEPT;
 899
 900         /* Is the embedded protocol header present? */
 901         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 902                      pp->dont_defrag))
 903                 return NF_ACCEPT;
 904
 905         IP_VS_DBG_PKT(11, pp, skb, offset, "Checking incoming ICMP for");
 906
 907         offset += cih->ihl * 4;
 908
 909         /* The embedded headers contain source and dest in reverse order */
 910         cp = pp->conn_in_get(skb, pp, cih, offset, 1);
 911         if (!cp)
 912                 return NF_ACCEPT;
 913
 914         verdict = NF_DROP;
 915
 916         /* Ensure the checksum is correct */
 917         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 918                 /* Failed checksum! */
 919                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %d.%d.%d.%d!\n",
 920                           NIPQUAD(iph->saddr));
 921                 goto out;
 922         }
 923
 924         /* do the statistics and put it back */
 925         ip_vs_in_stats(cp, skb);
 926         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol)
 927                 offset += 2 * sizeof(__u16);
 928         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset);
 929         /* do not touch skb anymore */
 930
 931   out:
 932         __ip_vs_conn_put(cp);
 933
 934         return verdict;
 935 }
 936
 937 /*
 938  *      Check if it's for virtual services, look it up,
 939  *      and send it on its way...
 940  */
 941 static unsigned int
 942 ip_vs_in(unsigned int hooknum, struct sk_buff **pskb,
 943          const struct net_device *in, const struct net_device *out,
 944          int (*okfn)(struct sk_buff *))
 945 {
 946         struct sk_buff  *skb = *pskb;
 947         struct iphdr    *iph;
 948         struct ip_vs_protocol *pp;
 949         struct ip_vs_conn *cp;
 950         int ret, restart;
 951         int ihl;
 952
 953         /*
 954          *      Big tappo: only PACKET_HOST (neither loopback nor mcasts)
 955          *      ... don't know why 1st test DOES NOT include 2nd (?)
 956          */
 957         if (unlikely(skb->pkt_type != PACKET_HOST
 958                      || skb->dev->flags & IFF_LOOPBACK || skb->sk)) {
 959                 IP_VS_DBG(12, "packet type=%d proto=%d daddr=%d.%d.%d.%d ignored\n",
 960                           skb->pkt_type,
 961                           ip_hdr(skb)->protocol,
 962                           NIPQUAD(ip_hdr(skb)->daddr));
 963                 return NF_ACCEPT;
 964         }
 965
 966         iph = ip_hdr(skb);
 967         if (unlikely(iph->protocol == IPPROTO_ICMP)) {
 968                 int related, verdict = ip_vs_in_icmp(pskb, &related, hooknum);
 969
 970                 if (related)
 971                         return verdict;
 972                 skb = *pskb;
 973                 iph = ip_hdr(skb);
 974         }
 975
 976         /* Protocol supported? */
 977         pp = ip_vs_proto_get(iph->protocol);
 978         if (unlikely(!pp))
 979                 return NF_ACCEPT;
 980
 981         ihl = iph->ihl << 2;
 982
 983         /*
 984          * Check if the packet belongs to an existing connection entry
 985          */
 986         cp = pp->conn_in_get(skb, pp, iph, ihl, 0);
 987
 988         if (unlikely(!cp)) {
 989                 int v;
 990
 991                 if (!pp->conn_schedule(skb, pp, &v, &cp))
 992                         return v;
 993         }
 994
 995         if (unlikely(!cp)) {
 996                 /* sorry, all this trouble for a no-hit :) */
 997                 IP_VS_DBG_PKT(12, pp, skb, 0,
 998                               "packet continues traversal as normal");
 999                 return NF_ACCEPT;
1000         }
1001
1002         IP_VS_DBG_PKT(11, pp, skb, 0, "Incoming packet");
1003
1004         /* Check the server status */
1005         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1006                 /* the destination server is not available */
1007
1008                 if (sysctl_ip_vs_expire_nodest_conn) {
1009                         /* try to expire the connection immediately */
1010                         ip_vs_conn_expire_now(cp);
1011                 }
1012                 /* don't restart its timer, and silently
1013                    drop the packet. */
1014                 __ip_vs_conn_put(cp);
1015                 return NF_DROP;
1016         }
1017
1018         ip_vs_in_stats(cp, skb);
1019         restart = ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pp);
1020         if (cp->packet_xmit)
1021                 ret = cp->packet_xmit(skb, cp, pp);
1022                 /* do not touch skb anymore */
1023         else {
1024                 IP_VS_DBG_RL("warning: packet_xmit is null");
1025                 ret = NF_ACCEPT;
1026         }
1027
1028         /* increase its packet counter and check if it is needed
1029            to be synchronized */
1030         atomic_inc(&cp->in_pkts);
1031         if ((ip_vs_sync_state & IP_VS_STATE_MASTER) &&
1032             (cp->protocol != IPPROTO_TCP ||
1033              cp->state == IP_VS_TCP_S_ESTABLISHED) &&
1034             (atomic_read(&cp->in_pkts) % sysctl_ip_vs_sync_threshold[1]
1035              == sysctl_ip_vs_sync_threshold[0]))
1036                 ip_vs_sync_conn(cp);
1037
1038         ip_vs_conn_put(cp);
1039         return ret;
1040 }
1041
1042
1043 /*
1044  *      It is hooked at the NF_IP_FORWARD chain, in order to catch ICMP
1045  *      related packets destined for 0.0.0.0/0.
1046  *      When fwmark-based virtual service is used, such as transparent
1047  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
1048  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1049  *      sent to ip_vs_in_icmp. So, catch them at the NF_IP_FORWARD chain
1050  *      and send them to ip_vs_in_icmp.
1051  */
1052 static unsigned int
1053 ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff **pskb,
1054                    const struct net_device *in, const struct net_device *out,
1055                    int (*okfn)(struct sk_buff *))
1056 {
1057         int r;
1058
1059         if (ip_hdr(*pskb)->protocol != IPPROTO_ICMP)
1060                 return NF_ACCEPT;
1061
1062         return ip_vs_in_icmp(pskb, &r, hooknum);
1063 }
1064
1065
1066 /* After packet filtering, forward packet through VS/DR, VS/TUN,
1067    or VS/NAT(change destination), so that filtering rules can be
1068    applied to IPVS. */
1069 static struct nf_hook_ops ip_vs_in_ops = {
1070         .hook           = ip_vs_in,
1071         .owner          = THIS_MODULE,
1072         .pf             = PF_INET,
1073         .hooknum        = NF_IP_LOCAL_IN,
1074         .priority       = 100,
1075 };
1076
1077 /* After packet filtering, change source only for VS/NAT */
1078 static struct nf_hook_ops ip_vs_out_ops = {
1079         .hook           = ip_vs_out,
1080         .owner          = THIS_MODULE,
1081         .pf             = PF_INET,
1082         .hooknum        = NF_IP_FORWARD,
1083         .priority       = 100,
1084 };
1085
1086 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1087    destined for 0.0.0.0/0, which is for incoming IPVS connections */
1088 static struct nf_hook_ops ip_vs_forward_icmp_ops = {
1089         .hook           = ip_vs_forward_icmp,
1090         .owner          = THIS_MODULE,
1091         .pf             = PF_INET,
1092         .hooknum        = NF_IP_FORWARD,
1093         .priority       = 99,
1094 };
1095
1096 /* Before the netfilter connection tracking, exit from POST_ROUTING */
1097 static struct nf_hook_ops ip_vs_post_routing_ops = {
1098         .hook           = ip_vs_post_routing,
1099         .owner          = THIS_MODULE,
1100         .pf             = PF_INET,
1101         .hooknum        = NF_IP_POST_ROUTING,
1102         .priority       = NF_IP_PRI_NAT_SRC-1,
1103 };
1104
1105
1106 /*
1107  *      Initialize IP Virtual Server
1108  */
1109 static int __init ip_vs_init(void)
1110 {
1111         int ret;
1112
1113         ret = ip_vs_control_init();
1114         if (ret < 0) {
1115                 IP_VS_ERR("can't setup control.\n");
1116                 goto cleanup_nothing;
1117         }
1118
1119         ip_vs_protocol_init();
1120
1121         ret = ip_vs_app_init();
1122         if (ret < 0) {
1123                 IP_VS_ERR("can't setup application helper.\n");
1124                 goto cleanup_protocol;
1125         }
1126
1127         ret = ip_vs_conn_init();
1128         if (ret < 0) {
1129                 IP_VS_ERR("can't setup connection table.\n");
1130                 goto cleanup_app;
1131         }
1132
1133         ret = nf_register_hook(&ip_vs_in_ops);
1134         if (ret < 0) {
1135                 IP_VS_ERR("can't register in hook.\n");
1136                 goto cleanup_conn;
1137         }
1138
1139         ret = nf_register_hook(&ip_vs_out_ops);
1140         if (ret < 0) {
1141                 IP_VS_ERR("can't register out hook.\n");
1142                 goto cleanup_inops;
1143         }
1144         ret = nf_register_hook(&ip_vs_post_routing_ops);
1145         if (ret < 0) {
1146                 IP_VS_ERR("can't register post_routing hook.\n");
1147                 goto cleanup_outops;
1148         }
1149         ret = nf_register_hook(&ip_vs_forward_icmp_ops);
1150         if (ret < 0) {
1151                 IP_VS_ERR("can't register forward_icmp hook.\n");
1152                 goto cleanup_postroutingops;
1153         }
1154
1155         IP_VS_INFO("ipvs loaded.\n");
1156         return ret;
1157
1158   cleanup_postroutingops:
1159         nf_unregister_hook(&ip_vs_post_routing_ops);
1160   cleanup_outops:
1161         nf_unregister_hook(&ip_vs_out_ops);
1162   cleanup_inops:
1163         nf_unregister_hook(&ip_vs_in_ops);
1164   cleanup_conn:
1165         ip_vs_conn_cleanup();
1166   cleanup_app:
1167         ip_vs_app_cleanup();
1168   cleanup_protocol:
1169         ip_vs_protocol_cleanup();
1170         ip_vs_control_cleanup();
1171   cleanup_nothing:
1172         return ret;
1173 }
1174
1175 static void __exit ip_vs_cleanup(void)
1176 {
1177         nf_unregister_hook(&ip_vs_forward_icmp_ops);
1178         nf_unregister_hook(&ip_vs_post_routing_ops);
1179         nf_unregister_hook(&ip_vs_out_ops);
1180         nf_unregister_hook(&ip_vs_in_ops);
1181         ip_vs_conn_cleanup();
1182         ip_vs_app_cleanup();
1183         ip_vs_protocol_cleanup();
1184         ip_vs_control_cleanup();
1185         IP_VS_INFO("ipvs unloaded.\n");
1186 }
1187
1188 module_init(ip_vs_init);
1189 module_exit(ip_vs_cleanup);
1190 MODULE_LICENSE("GPL");