net/netfilter/ipvs/ip_vs_core.c

   1 /*
   2  * IPVS         An implementation of the IP virtual server support for the
   3  *              LINUX operating system.  IPVS is now implemented as a module
   4  *              over the Netfilter framework. IPVS can be used to build a
   5  *              high-performance and highly available server based on a
   6  *              cluster of servers.
   7  *
   8  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
   9  *              Peter Kese <peter.kese@ijs.si>
  10  *              Julian Anastasov <ja@ssi.bg>
  11  *
  12  *              This program is free software; you can redistribute it and/or
  13  *              modify it under the terms of the GNU General Public License
  14  *              as published by the Free Software Foundation; either version
  15  *              2 of the License, or (at your option) any later version.
  16  *
  17  * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
  18  * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
  19  * and others.
  20  *
  21  * Changes:
  22  *      Paul `Rusty' Russell            properly handle non-linear skbs
  23  *      Harald Welte                    don't use nfcache
  24  *
  25  */
  26
  27 #define KMSG_COMPONENT "IPVS"
  28 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  29
  30 #include <linux/module.h>
  31 #include <linux/kernel.h>
  32 #include <linux/ip.h>
  33 #include <linux/tcp.h>
  34 #include <linux/sctp.h>
  35 #include <linux/icmp.h>
  36 #include <linux/slab.h>
  37
  38 #include <net/ip.h>
  39 #include <net/tcp.h>
  40 #include <net/udp.h>
  41 #include <net/icmp.h>                   /* for icmp_send */
  42 #include <net/route.h>
  43 #include <net/ip6_checksum.h>
  44 #include <net/netns/generic.h>          /* net_generic() */
  45
  46 #include <linux/netfilter.h>
  47 #include <linux/netfilter_ipv4.h>
  48
  49 #ifdef CONFIG_IP_VS_IPV6
  50 #include <net/ipv6.h>
  51 #include <linux/netfilter_ipv6.h>
  52 #include <net/ip6_route.h>
  53 #endif
  54
  55 #include <net/ip_vs.h>
  56
  57
  58 EXPORT_SYMBOL(register_ip_vs_scheduler);
  59 EXPORT_SYMBOL(unregister_ip_vs_scheduler);
  60 EXPORT_SYMBOL(ip_vs_proto_name);
  61 EXPORT_SYMBOL(ip_vs_conn_new);
  62 EXPORT_SYMBOL(ip_vs_conn_in_get);
  63 EXPORT_SYMBOL(ip_vs_conn_out_get);
  64 #ifdef CONFIG_IP_VS_PROTO_TCP
  65 EXPORT_SYMBOL(ip_vs_tcp_conn_listen);
  66 #endif
  67 EXPORT_SYMBOL(ip_vs_conn_put);
  68 #ifdef CONFIG_IP_VS_DEBUG
  69 EXPORT_SYMBOL(ip_vs_get_debug_level);
  70 #endif
  71
  72 static int ip_vs_net_id __read_mostly;
  73 /* netns cnt used for uniqueness */
  74 static atomic_t ipvs_netns_cnt = ATOMIC_INIT(0);
  75
  76 /* ID used in ICMP lookups */
  77 #define icmp_id(icmph)          (((icmph)->un).echo.id)
  78 #define icmpv6_id(icmph)        (icmph->icmp6_dataun.u_echo.identifier)
  79
  80 const char *ip_vs_proto_name(unsigned int proto)
  81 {
  82         static char buf[20];
  83
  84         switch (proto) {
  85         case IPPROTO_IP:
  86                 return "IP";
  87         case IPPROTO_UDP:
  88                 return "UDP";
  89         case IPPROTO_TCP:
  90                 return "TCP";
  91         case IPPROTO_SCTP:
  92                 return "SCTP";
  93         case IPPROTO_ICMP:
  94                 return "ICMP";
  95 #ifdef CONFIG_IP_VS_IPV6
  96         case IPPROTO_ICMPV6:
  97                 return "ICMPv6";
  98 #endif
  99         default:
 100                 sprintf(buf, "IP_%d", proto);
 101                 return buf;
 102         }
 103 }
 104
 105 void ip_vs_init_hash_table(struct list_head *table, int rows)
 106 {
 107         while (--rows >= 0)
 108                 INIT_LIST_HEAD(&table[rows]);
 109 }
 110
 111 static inline void
 112 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 113 {
 114         struct ip_vs_dest *dest = cp->dest;
 115         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 116
 117         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 118                 struct ip_vs_cpu_stats *s;
 119
 120                 s = this_cpu_ptr(dest->stats.cpustats);
 121                 s->ustats.inpkts++;
 122                 u64_stats_update_begin(&s->syncp);
 123                 s->ustats.inbytes += skb->len;
 124                 u64_stats_update_end(&s->syncp);
 125
 126                 s = this_cpu_ptr(dest->svc->stats.cpustats);
 127                 s->ustats.inpkts++;
 128                 u64_stats_update_begin(&s->syncp);
 129                 s->ustats.inbytes += skb->len;
 130                 u64_stats_update_end(&s->syncp);
 131
 132                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 133                 s->ustats.inpkts++;
 134                 u64_stats_update_begin(&s->syncp);
 135                 s->ustats.inbytes += skb->len;
 136                 u64_stats_update_end(&s->syncp);
 137         }
 138 }
 139
 140
 141 static inline void
 142 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 143 {
 144         struct ip_vs_dest *dest = cp->dest;
 145         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 146
 147         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 148                 struct ip_vs_cpu_stats *s;
 149
 150                 s = this_cpu_ptr(dest->stats.cpustats);
 151                 s->ustats.outpkts++;
 152                 u64_stats_update_begin(&s->syncp);
 153                 s->ustats.outbytes += skb->len;
 154                 u64_stats_update_end(&s->syncp);
 155
 156                 s = this_cpu_ptr(dest->svc->stats.cpustats);
 157                 s->ustats.outpkts++;
 158                 u64_stats_update_begin(&s->syncp);
 159                 s->ustats.outbytes += skb->len;
 160                 u64_stats_update_end(&s->syncp);
 161
 162                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 163                 s->ustats.outpkts++;
 164                 u64_stats_update_begin(&s->syncp);
 165                 s->ustats.outbytes += skb->len;
 166                 u64_stats_update_end(&s->syncp);
 167         }
 168 }
 169
 170
 171 static inline void
 172 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
 173 {
 174         struct netns_ipvs *ipvs = net_ipvs(svc->net);
 175         struct ip_vs_cpu_stats *s;
 176
 177         s = this_cpu_ptr(cp->dest->stats.cpustats);
 178         s->ustats.conns++;
 179
 180         s = this_cpu_ptr(svc->stats.cpustats);
 181         s->ustats.conns++;
 182
 183         s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 184         s->ustats.conns++;
 185 }
 186
 187
 188 static inline void
 189 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
 190                 const struct sk_buff *skb,
 191                 struct ip_vs_proto_data *pd)
 192 {
 193         if (likely(pd->pp->state_transition))
 194                 pd->pp->state_transition(cp, direction, skb, pd);
 195 }
 196
 197 static inline int
 198 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
 199                               struct sk_buff *skb, int protocol,
 200                               const union nf_inet_addr *caddr, __be16 cport,
 201                               const union nf_inet_addr *vaddr, __be16 vport,
 202                               struct ip_vs_conn_param *p)
 203 {
 204         ip_vs_conn_fill_param(svc->net, svc->af, protocol, caddr, cport, vaddr,
 205                               vport, p);
 206         p->pe = rcu_dereference(svc->pe);
 207         if (p->pe && p->pe->fill_param)
 208                 return p->pe->fill_param(p, skb);
 209
 210         return 0;
 211 }
 212
 213 /*
 214  *  IPVS persistent scheduling function
 215  *  It creates a connection entry according to its template if exists,
 216  *  or selects a server and creates a connection entry plus a template.
 217  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
 218  *  Protocols supported: TCP, UDP
 219  */
 220 static struct ip_vs_conn *
 221 ip_vs_sched_persist(struct ip_vs_service *svc,
 222                     struct sk_buff *skb, __be16 src_port, __be16 dst_port,
 223                     int *ignored, struct ip_vs_iphdr *iph)
 224 {
 225         struct ip_vs_conn *cp = NULL;
 226         struct ip_vs_dest *dest;
 227         struct ip_vs_conn *ct;
 228         __be16 dport = 0;               /* destination port to forward */
 229         unsigned int flags;
 230         struct ip_vs_conn_param param;
 231         const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
 232         union nf_inet_addr snet;        /* source network of the client,
 233                                            after masking */
 234
 235         /* Mask saddr with the netmask to adjust template granularity */
 236 #ifdef CONFIG_IP_VS_IPV6
 237         if (svc->af == AF_INET6)
 238                 ipv6_addr_prefix(&snet.in6, &iph->saddr.in6,
 239                                  (__force __u32) svc->netmask);
 240         else
 241 #endif
 242                 snet.ip = iph->saddr.ip & svc->netmask;
 243
 244         IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
 245                       "mnet %s\n",
 246                       IP_VS_DBG_ADDR(svc->af, &iph->saddr), ntohs(src_port),
 247                       IP_VS_DBG_ADDR(svc->af, &iph->daddr), ntohs(dst_port),
 248                       IP_VS_DBG_ADDR(svc->af, &snet));
 249
 250         /*
 251          * As far as we know, FTP is a very complicated network protocol, and
 252          * it uses control connection and data connections. For active FTP,
 253          * FTP server initialize data connection to the client, its source port
 254          * is often 20. For passive FTP, FTP server tells the clients the port
 255          * that it passively listens to,  and the client issues the data
 256          * connection. In the tunneling or direct routing mode, the load
 257          * balancer is on the client-to-server half of connection, the port
 258          * number is unknown to the load balancer. So, a conn template like
 259          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
 260          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
 261          * is created for other persistent services.
 262          */
 263         {
 264                 int protocol = iph->protocol;
 265                 const union nf_inet_addr *vaddr = &iph->daddr;
 266                 __be16 vport = 0;
 267
 268                 if (dst_port == svc->port) {
 269                         /* non-FTP template:
 270                          * <protocol, caddr, 0, vaddr, vport, daddr, dport>
 271                          * FTP template:
 272                          * <protocol, caddr, 0, vaddr, 0, daddr, 0>
 273                          */
 274                         if (svc->port != FTPPORT)
 275                                 vport = dst_port;
 276                 } else {
 277                         /* Note: persistent fwmark-based services and
 278                          * persistent port zero service are handled here.
 279                          * fwmark template:
 280                          * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
 281                          * port zero template:
 282                          * <protocol,caddr,0,vaddr,0,daddr,0>
 283                          */
 284                         if (svc->fwmark) {
 285                                 protocol = IPPROTO_IP;
 286                                 vaddr = &fwmark;
 287                         }
 288                 }
 289                 /* return *ignored = -1 so NF_DROP can be used */
 290                 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
 291                                                   vaddr, vport, &param) < 0) {
 292                         *ignored = -1;
 293                         return NULL;
 294                 }
 295         }
 296
 297         /* Check if a template already exists */
 298         ct = ip_vs_ct_in_get(&param);
 299         if (!ct || !ip_vs_check_template(ct)) {
 300                 struct ip_vs_scheduler *sched;
 301
 302                 /*
 303                  * No template found or the dest of the connection
 304                  * template is not available.
 305                  * return *ignored=0 i.e. ICMP and NF_DROP
 306                  */
 307                 sched = rcu_dereference(svc->scheduler);
 308                 dest = sched->schedule(svc, skb);
 309                 if (!dest) {
 310                         IP_VS_DBG(1, "p-schedule: no dest found.\n");
 311                         kfree(param.pe_data);
 312                         *ignored = 0;
 313                         return NULL;
 314                 }
 315
 316                 if (dst_port == svc->port && svc->port != FTPPORT)
 317                         dport = dest->port;
 318
 319                 /* Create a template
 320                  * This adds param.pe_data to the template,
 321                  * and thus param.pe_data will be destroyed
 322                  * when the template expires */
 323                 ct = ip_vs_conn_new(&param, &dest->addr, dport,
 324                                     IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
 325                 if (ct == NULL) {
 326                         kfree(param.pe_data);
 327                         *ignored = -1;
 328                         return NULL;
 329                 }
 330
 331                 ct->timeout = svc->timeout;
 332         } else {
 333                 /* set destination with the found template */
 334                 dest = ct->dest;
 335                 kfree(param.pe_data);
 336         }
 337
 338         dport = dst_port;
 339         if (dport == svc->port && dest->port)
 340                 dport = dest->port;
 341
 342         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 343                  && iph->protocol == IPPROTO_UDP) ?
 344                 IP_VS_CONN_F_ONE_PACKET : 0;
 345
 346         /*
 347          *    Create a new connection according to the template
 348          */
 349         ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol, &iph->saddr,
 350                               src_port, &iph->daddr, dst_port, &param);
 351
 352         cp = ip_vs_conn_new(&param, &dest->addr, dport, flags, dest, skb->mark);
 353         if (cp == NULL) {
 354                 ip_vs_conn_put(ct);
 355                 *ignored = -1;
 356                 return NULL;
 357         }
 358
 359         /*
 360          *    Add its control
 361          */
 362         ip_vs_control_add(cp, ct);
 363         ip_vs_conn_put(ct);
 364
 365         ip_vs_conn_stats(cp, svc);
 366         return cp;
 367 }
 368
 369
 370 /*
 371  *  IPVS main scheduling function
 372  *  It selects a server according to the virtual service, and
 373  *  creates a connection entry.
 374  *  Protocols supported: TCP, UDP
 375  *
 376  *  Usage of *ignored
 377  *
 378  * 1 :   protocol tried to schedule (eg. on SYN), found svc but the
 379  *       svc/scheduler decides that this packet should be accepted with
 380  *       NF_ACCEPT because it must not be scheduled.
 381  *
 382  * 0 :   scheduler can not find destination, so try bypass or
 383  *       return ICMP and then NF_DROP (ip_vs_leave).
 384  *
 385  * -1 :  scheduler tried to schedule but fatal error occurred, eg.
 386  *       ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
 387  *       failure such as missing Call-ID, ENOMEM on skb_linearize
 388  *       or pe_data. In this case we should return NF_DROP without
 389  *       any attempts to send ICMP with ip_vs_leave.
 390  */
 391 struct ip_vs_conn *
 392 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
 393                struct ip_vs_proto_data *pd, int *ignored,
 394                struct ip_vs_iphdr *iph)
 395 {
 396         struct ip_vs_protocol *pp = pd->pp;
 397         struct ip_vs_conn *cp = NULL;
 398         struct ip_vs_scheduler *sched;
 399         struct ip_vs_dest *dest;
 400         __be16 _ports[2], *pptr;
 401         unsigned int flags;
 402
 403         *ignored = 1;
 404         /*
 405          * IPv6 frags, only the first hit here.
 406          */
 407         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 408         if (pptr == NULL)
 409                 return NULL;
 410
 411         /*
 412          * FTPDATA needs this check when using local real server.
 413          * Never schedule Active FTPDATA connections from real server.
 414          * For LVS-NAT they must be already created. For other methods
 415          * with persistence the connection is created on SYN+ACK.
 416          */
 417         if (pptr[0] == FTPDATA) {
 418                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 419                               "Not scheduling FTPDATA");
 420                 return NULL;
 421         }
 422
 423         /*
 424          *    Do not schedule replies from local real server.
 425          */
 426         if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
 427             (cp = pp->conn_in_get(svc->af, skb, iph, 1))) {
 428                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 429                               "Not scheduling reply for existing connection");
 430                 __ip_vs_conn_put(cp);
 431                 return NULL;
 432         }
 433
 434         /*
 435          *    Persistent service
 436          */
 437         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 438                 return ip_vs_sched_persist(svc, skb, pptr[0], pptr[1], ignored,
 439                                            iph);
 440
 441         *ignored = 0;
 442
 443         /*
 444          *    Non-persistent service
 445          */
 446         if (!svc->fwmark && pptr[1] != svc->port) {
 447                 if (!svc->port)
 448                         pr_err("Schedule: port zero only supported "
 449                                "in persistent services, "
 450                                "check your ipvs configuration\n");
 451                 return NULL;
 452         }
 453
 454         sched = rcu_dereference(svc->scheduler);
 455         dest = sched->schedule(svc, skb);
 456         if (dest == NULL) {
 457                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 458                 return NULL;
 459         }
 460
 461         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 462                  && iph->protocol == IPPROTO_UDP) ?
 463                 IP_VS_CONN_F_ONE_PACKET : 0;
 464
 465         /*
 466          *    Create a connection entry.
 467          */
 468         {
 469                 struct ip_vs_conn_param p;
 470
 471                 ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol,
 472                                       &iph->saddr, pptr[0], &iph->daddr,
 473                                       pptr[1], &p);
 474                 cp = ip_vs_conn_new(&p, &dest->addr,
 475                                     dest->port ? dest->port : pptr[1],
 476                                     flags, dest, skb->mark);
 477                 if (!cp) {
 478                         *ignored = -1;
 479                         return NULL;
 480                 }
 481         }
 482
 483         IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
 484                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
 485                       ip_vs_fwd_tag(cp),
 486                       IP_VS_DBG_ADDR(svc->af, &cp->caddr), ntohs(cp->cport),
 487                       IP_VS_DBG_ADDR(svc->af, &cp->vaddr), ntohs(cp->vport),
 488                       IP_VS_DBG_ADDR(svc->af, &cp->daddr), ntohs(cp->dport),
 489                       cp->flags, atomic_read(&cp->refcnt));
 490
 491         ip_vs_conn_stats(cp, svc);
 492         return cp;
 493 }
 494
 495
 496 /*
 497  *  Pass or drop the packet.
 498  *  Called by ip_vs_in, when the virtual service is available but
 499  *  no destination is available for a new connection.
 500  */
 501 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 502                 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
 503 {
 504         __be16 _ports[2], *pptr;
 505 #ifdef CONFIG_SYSCTL
 506         struct net *net;
 507         struct netns_ipvs *ipvs;
 508         int unicast;
 509 #endif
 510
 511         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 512         if (pptr == NULL) {
 513                 return NF_DROP;
 514         }
 515
 516 #ifdef CONFIG_SYSCTL
 517         net = skb_net(skb);
 518
 519 #ifdef CONFIG_IP_VS_IPV6
 520         if (svc->af == AF_INET6)
 521                 unicast = ipv6_addr_type(&iph->daddr.in6) & IPV6_ADDR_UNICAST;
 522         else
 523 #endif
 524                 unicast = (inet_addr_type(net, iph->daddr.ip) == RTN_UNICAST);
 525
 526         /* if it is fwmark-based service, the cache_bypass sysctl is up
 527            and the destination is a non-local unicast, then create
 528            a cache_bypass connection entry */
 529         ipvs = net_ipvs(net);
 530         if (ipvs->sysctl_cache_bypass && svc->fwmark && unicast) {
 531                 int ret;
 532                 struct ip_vs_conn *cp;
 533                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
 534                                       iph->protocol == IPPROTO_UDP) ?
 535                                       IP_VS_CONN_F_ONE_PACKET : 0;
 536                 union nf_inet_addr daddr =  { .all = { 0, 0, 0, 0 } };
 537
 538                 /* create a new connection entry */
 539                 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
 540                 {
 541                         struct ip_vs_conn_param p;
 542                         ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol,
 543                                               &iph->saddr, pptr[0],
 544                                               &iph->daddr, pptr[1], &p);
 545                         cp = ip_vs_conn_new(&p, &daddr, 0,
 546                                             IP_VS_CONN_F_BYPASS | flags,
 547                                             NULL, skb->mark);
 548                         if (!cp)
 549                                 return NF_DROP;
 550                 }
 551
 552                 /* statistics */
 553                 ip_vs_in_stats(cp, skb);
 554
 555                 /* set state */
 556                 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
 557
 558                 /* transmit the first SYN packet */
 559                 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
 560                 /* do not touch skb anymore */
 561
 562                 atomic_inc(&cp->in_pkts);
 563                 ip_vs_conn_put(cp);
 564                 return ret;
 565         }
 566 #endif
 567
 568         /*
 569          * When the virtual ftp service is presented, packets destined
 570          * for other services on the VIP may get here (except services
 571          * listed in the ipvs table), pass the packets, because it is
 572          * not ipvs job to decide to drop the packets.
 573          */
 574         if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT))
 575                 return NF_ACCEPT;
 576
 577         /*
 578          * Notify the client that the destination is unreachable, and
 579          * release the socket buffer.
 580          * Since it is in IP layer, the TCP socket is not actually
 581          * created, the TCP RST packet cannot be sent, instead that
 582          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 583          */
 584 #ifdef CONFIG_IP_VS_IPV6
 585         if (svc->af == AF_INET6) {
 586                 if (!skb->dev) {
 587                         struct net *net_ = dev_net(skb_dst(skb)->dev);
 588
 589                         skb->dev = net_->loopback_dev;
 590                 }
 591                 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 592         } else
 593 #endif
 594                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 595
 596         return NF_DROP;
 597 }
 598
 599 #ifdef CONFIG_SYSCTL
 600
 601 static int sysctl_snat_reroute(struct sk_buff *skb)
 602 {
 603         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 604         return ipvs->sysctl_snat_reroute;
 605 }
 606
 607 static int sysctl_nat_icmp_send(struct net *net)
 608 {
 609         struct netns_ipvs *ipvs = net_ipvs(net);
 610         return ipvs->sysctl_nat_icmp_send;
 611 }
 612
 613 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
 614 {
 615         return ipvs->sysctl_expire_nodest_conn;
 616 }
 617
 618 #else
 619
 620 static int sysctl_snat_reroute(struct sk_buff *skb) { return 0; }
 621 static int sysctl_nat_icmp_send(struct net *net) { return 0; }
 622 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
 623
 624 #endif
 625
 626 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 627 {
 628         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 629 }
 630
 631 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
 632 {
 633         if (NF_INET_LOCAL_IN == hooknum)
 634                 return IP_DEFRAG_VS_IN;
 635         if (NF_INET_FORWARD == hooknum)
 636                 return IP_DEFRAG_VS_FWD;
 637         return IP_DEFRAG_VS_OUT;
 638 }
 639
 640 static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user)
 641 {
 642         int err;
 643
 644         local_bh_disable();
 645         err = ip_defrag(skb, user);
 646         local_bh_enable();
 647         if (!err)
 648                 ip_send_check(ip_hdr(skb));
 649
 650         return err;
 651 }
 652
 653 static int ip_vs_route_me_harder(int af, struct sk_buff *skb)
 654 {
 655 #ifdef CONFIG_IP_VS_IPV6
 656         if (af == AF_INET6) {
 657                 if (sysctl_snat_reroute(skb) && ip6_route_me_harder(skb) != 0)
 658                         return 1;
 659         } else
 660 #endif
 661                 if ((sysctl_snat_reroute(skb) ||
 662                      skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
 663                     ip_route_me_harder(skb, RTN_LOCAL) != 0)
 664                         return 1;
 665
 666         return 0;
 667 }
 668
 669 /*
 670  * Packet has been made sufficiently writable in caller
 671  * - inout: 1=in->out, 0=out->in
 672  */
 673 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 674                     struct ip_vs_conn *cp, int inout)
 675 {
 676         struct iphdr *iph        = ip_hdr(skb);
 677         unsigned int icmp_offset = iph->ihl*4;
 678         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
 679                                                       icmp_offset);
 680         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 681
 682         if (inout) {
 683                 iph->saddr = cp->vaddr.ip;
 684                 ip_send_check(iph);
 685                 ciph->daddr = cp->vaddr.ip;
 686                 ip_send_check(ciph);
 687         } else {
 688                 iph->daddr = cp->daddr.ip;
 689                 ip_send_check(iph);
 690                 ciph->saddr = cp->daddr.ip;
 691                 ip_send_check(ciph);
 692         }
 693
 694         /* the TCP/UDP/SCTP port */
 695         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
 696             IPPROTO_SCTP == ciph->protocol) {
 697                 __be16 *ports = (void *)ciph + ciph->ihl*4;
 698
 699                 if (inout)
 700                         ports[1] = cp->vport;
 701                 else
 702                         ports[0] = cp->dport;
 703         }
 704
 705         /* And finally the ICMP checksum */
 706         icmph->checksum = 0;
 707         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 708         skb->ip_summed = CHECKSUM_UNNECESSARY;
 709
 710         if (inout)
 711                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 712                         "Forwarding altered outgoing ICMP");
 713         else
 714                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 715                         "Forwarding altered incoming ICMP");
 716 }
 717
 718 #ifdef CONFIG_IP_VS_IPV6
 719 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
 720                     struct ip_vs_conn *cp, int inout)
 721 {
 722         struct ipv6hdr *iph      = ipv6_hdr(skb);
 723         unsigned int icmp_offset = 0;
 724         unsigned int offs        = 0; /* header offset*/
 725         int protocol;
 726         struct icmp6hdr *icmph;
 727         struct ipv6hdr *ciph;
 728         unsigned short fragoffs;
 729
 730         ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
 731         icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
 732         offs = icmp_offset + sizeof(struct icmp6hdr);
 733         ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
 734
 735         protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
 736
 737         if (inout) {
 738                 iph->saddr = cp->vaddr.in6;
 739                 ciph->daddr = cp->vaddr.in6;
 740         } else {
 741                 iph->daddr = cp->daddr.in6;
 742                 ciph->saddr = cp->daddr.in6;
 743         }
 744
 745         /* the TCP/UDP/SCTP port */
 746         if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 747                           IPPROTO_SCTP == protocol)) {
 748                 __be16 *ports = (void *)(skb_network_header(skb) + offs);
 749
 750                 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
 751                               ntohs(inout ? ports[1] : ports[0]),
 752                               ntohs(inout ? cp->vport : cp->dport));
 753                 if (inout)
 754                         ports[1] = cp->vport;
 755                 else
 756                         ports[0] = cp->dport;
 757         }
 758
 759         /* And finally the ICMP checksum */
 760         icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
 761                                               skb->len - icmp_offset,
 762                                               IPPROTO_ICMPV6, 0);
 763         skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
 764         skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
 765         skb->ip_summed = CHECKSUM_PARTIAL;
 766
 767         if (inout)
 768                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 769                               (void *)ciph - (void *)iph,
 770                               "Forwarding altered outgoing ICMPv6");
 771         else
 772                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 773                               (void *)ciph - (void *)iph,
 774                               "Forwarding altered incoming ICMPv6");
 775 }
 776 #endif
 777
 778 /* Handle relevant response ICMP messages - forward to the right
 779  * destination host.
 780  */
 781 static int handle_response_icmp(int af, struct sk_buff *skb,
 782                                 union nf_inet_addr *snet,
 783                                 __u8 protocol, struct ip_vs_conn *cp,
 784                                 struct ip_vs_protocol *pp,
 785                                 unsigned int offset, unsigned int ihl)
 786 {
 787         unsigned int verdict = NF_DROP;
 788
 789         if (IP_VS_FWD_METHOD(cp) != 0) {
 790                 pr_err("shouldn't reach here, because the box is on the "
 791                        "half connection in the tun/dr module.\n");
 792         }
 793
 794         /* Ensure the checksum is correct */
 795         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 796                 /* Failed checksum! */
 797                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
 798                               IP_VS_DBG_ADDR(af, snet));
 799                 goto out;
 800         }
 801
 802         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 803             IPPROTO_SCTP == protocol)
 804                 offset += 2 * sizeof(__u16);
 805         if (!skb_make_writable(skb, offset))
 806                 goto out;
 807
 808 #ifdef CONFIG_IP_VS_IPV6
 809         if (af == AF_INET6)
 810                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
 811         else
 812 #endif
 813                 ip_vs_nat_icmp(skb, pp, cp, 1);
 814
 815         if (ip_vs_route_me_harder(af, skb))
 816                 goto out;
 817
 818         /* do the statistics and put it back */
 819         ip_vs_out_stats(cp, skb);
 820
 821         skb->ipvs_property = 1;
 822         if (!(cp->flags & IP_VS_CONN_F_NFCT))
 823                 ip_vs_notrack(skb);
 824         else
 825                 ip_vs_update_conntrack(skb, cp, 0);
 826         verdict = NF_ACCEPT;
 827
 828 out:
 829         __ip_vs_conn_put(cp);
 830
 831         return verdict;
 832 }
 833
 834 /*
 835  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 836  *      Find any that might be relevant, check against existing connections.
 837  *      Currently handles error types - unreachable, quench, ttl exceeded.
 838  */
 839 static int ip_vs_out_icmp(struct sk_buff *skb, int *related,
 840                           unsigned int hooknum)
 841 {
 842         struct iphdr *iph;
 843         struct icmphdr  _icmph, *ic;
 844         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 845         struct ip_vs_iphdr ciph;
 846         struct ip_vs_conn *cp;
 847         struct ip_vs_protocol *pp;
 848         unsigned int offset, ihl;
 849         union nf_inet_addr snet;
 850
 851         *related = 1;
 852
 853         /* reassemble IP fragments */
 854         if (ip_is_fragment(ip_hdr(skb))) {
 855                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
 856                         return NF_STOLEN;
 857         }
 858
 859         iph = ip_hdr(skb);
 860         offset = ihl = iph->ihl * 4;
 861         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 862         if (ic == NULL)
 863                 return NF_DROP;
 864
 865         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
 866                   ic->type, ntohs(icmp_id(ic)),
 867                   &iph->saddr, &iph->daddr);
 868
 869         /*
 870          * Work through seeing if this is for us.
 871          * These checks are supposed to be in an order that means easy
 872          * things are checked first to speed up processing.... however
 873          * this means that some packets will manage to get a long way
 874          * down this stack and then be rejected, but that's life.
 875          */
 876         if ((ic->type != ICMP_DEST_UNREACH) &&
 877             (ic->type != ICMP_SOURCE_QUENCH) &&
 878             (ic->type != ICMP_TIME_EXCEEDED)) {
 879                 *related = 0;
 880                 return NF_ACCEPT;
 881         }
 882
 883         /* Now find the contained IP header */
 884         offset += sizeof(_icmph);
 885         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 886         if (cih == NULL)
 887                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 888
 889         pp = ip_vs_proto_get(cih->protocol);
 890         if (!pp)
 891                 return NF_ACCEPT;
 892
 893         /* Is the embedded protocol header present? */
 894         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 895                      pp->dont_defrag))
 896                 return NF_ACCEPT;
 897
 898         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
 899                       "Checking outgoing ICMP for");
 900
 901         ip_vs_fill_ip4hdr(cih, &ciph);
 902         ciph.len += offset;
 903         /* The embedded headers contain source and dest in reverse order */
 904         cp = pp->conn_out_get(AF_INET, skb, &ciph, 1);
 905         if (!cp)
 906                 return NF_ACCEPT;
 907
 908         snet.ip = iph->saddr;
 909         return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
 910                                     pp, ciph.len, ihl);
 911 }
 912
 913 #ifdef CONFIG_IP_VS_IPV6
 914 static int ip_vs_out_icmp_v6(struct sk_buff *skb, int *related,
 915                              unsigned int hooknum, struct ip_vs_iphdr *ipvsh)
 916 {
 917         struct icmp6hdr _icmph, *ic;
 918         struct ipv6hdr _ip6h, *ip6h; /* The ip header contained within ICMP */
 919         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
 920         struct ip_vs_conn *cp;
 921         struct ip_vs_protocol *pp;
 922         union nf_inet_addr snet;
 923         unsigned int writable;
 924
 925         *related = 1;
 926         ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph, ipvsh);
 927         if (ic == NULL)
 928                 return NF_DROP;
 929
 930         /*
 931          * Work through seeing if this is for us.
 932          * These checks are supposed to be in an order that means easy
 933          * things are checked first to speed up processing.... however
 934          * this means that some packets will manage to get a long way
 935          * down this stack and then be rejected, but that's life.
 936          */
 937         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
 938                 *related = 0;
 939                 return NF_ACCEPT;
 940         }
 941         /* Fragment header that is before ICMP header tells us that:
 942          * it's not an error message since they can't be fragmented.
 943          */
 944         if (ipvsh->flags & IP6_FH_F_FRAG)
 945                 return NF_DROP;
 946
 947         IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
 948                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
 949                   &ipvsh->saddr, &ipvsh->daddr);
 950
 951         /* Now find the contained IP header */
 952         ciph.len = ipvsh->len + sizeof(_icmph);
 953         ip6h = skb_header_pointer(skb, ciph.len, sizeof(_ip6h), &_ip6h);
 954         if (ip6h == NULL)
 955                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 956         ciph.saddr.in6 = ip6h->saddr; /* conn_out_get() handles reverse order */
 957         ciph.daddr.in6 = ip6h->daddr;
 958         /* skip possible IPv6 exthdrs of contained IPv6 packet */
 959         ciph.protocol = ipv6_find_hdr(skb, &ciph.len, -1, &ciph.fragoffs, NULL);
 960         if (ciph.protocol < 0)
 961                 return NF_ACCEPT; /* Contained IPv6 hdr looks wrong, ignore */
 962
 963         pp = ip_vs_proto_get(ciph.protocol);
 964         if (!pp)
 965                 return NF_ACCEPT;
 966
 967         /* The embedded headers contain source and dest in reverse order */
 968         cp = pp->conn_out_get(AF_INET6, skb, &ciph, 1);
 969         if (!cp)
 970                 return NF_ACCEPT;
 971
 972         snet.in6 = ciph.saddr.in6;
 973         writable = ciph.len;
 974         return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
 975                                     pp, writable, sizeof(struct ipv6hdr));
 976 }
 977 #endif
 978
 979 /*
 980  * Check if sctp chunc is ABORT chunk
 981  */
 982 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
 983 {
 984         sctp_chunkhdr_t *sch, schunk;
 985         sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
 986                         sizeof(schunk), &schunk);
 987         if (sch == NULL)
 988                 return 0;
 989         if (sch->type == SCTP_CID_ABORT)
 990                 return 1;
 991         return 0;
 992 }
 993
 994 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
 995 {
 996         struct tcphdr _tcph, *th;
 997
 998         th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
 999         if (th == NULL)
1000                 return 0;
1001         return th->rst;
1002 }
1003
1004 /* Handle response packets: rewrite addresses and send away...
1005  */
1006 static unsigned int
1007 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1008                 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph)
1009 {
1010         struct ip_vs_protocol *pp = pd->pp;
1011
1012         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Outgoing packet");
1013
1014         if (!skb_make_writable(skb, iph->len))
1015                 goto drop;
1016
1017         /* mangle the packet */
1018         if (pp->snat_handler && !pp->snat_handler(skb, pp, cp, iph))
1019                 goto drop;
1020
1021 #ifdef CONFIG_IP_VS_IPV6
1022         if (af == AF_INET6)
1023                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1024         else
1025 #endif
1026         {
1027                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1028                 ip_send_check(ip_hdr(skb));
1029         }
1030
1031         /*
1032          * nf_iterate does not expect change in the skb->dst->dev.
1033          * It looks like it is not fatal to enable this code for hooks
1034          * where our handlers are at the end of the chain list and
1035          * when all next handlers use skb->dst->dev and not outdev.
1036          * It will definitely route properly the inout NAT traffic
1037          * when multiple paths are used.
1038          */
1039
1040         /* For policy routing, packets originating from this
1041          * machine itself may be routed differently to packets
1042          * passing through.  We want this packet to be routed as
1043          * if it came from this machine itself.  So re-compute
1044          * the routing information.
1045          */
1046         if (ip_vs_route_me_harder(af, skb))
1047                 goto drop;
1048
1049         IP_VS_DBG_PKT(10, af, pp, skb, 0, "After SNAT");
1050
1051         ip_vs_out_stats(cp, skb);
1052         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1053         skb->ipvs_property = 1;
1054         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1055                 ip_vs_notrack(skb);
1056         else
1057                 ip_vs_update_conntrack(skb, cp, 0);
1058         ip_vs_conn_put(cp);
1059
1060         LeaveFunction(11);
1061         return NF_ACCEPT;
1062
1063 drop:
1064         ip_vs_conn_put(cp);
1065         kfree_skb(skb);
1066         LeaveFunction(11);
1067         return NF_STOLEN;
1068 }
1069
1070 /*
1071  *      Check if outgoing packet belongs to the established ip_vs_conn.
1072  */
1073 static unsigned int
1074 ip_vs_out(unsigned int hooknum, struct sk_buff *skb, int af)
1075 {
1076         struct net *net = NULL;
1077         struct ip_vs_iphdr iph;
1078         struct ip_vs_protocol *pp;
1079         struct ip_vs_proto_data *pd;
1080         struct ip_vs_conn *cp;
1081
1082         EnterFunction(11);
1083
1084         /* Already marked as IPVS request or reply? */
1085         if (skb->ipvs_property)
1086                 return NF_ACCEPT;
1087
1088         /* Bad... Do not break raw sockets */
1089         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1090                      af == AF_INET)) {
1091                 struct sock *sk = skb->sk;
1092                 struct inet_sock *inet = inet_sk(skb->sk);
1093
1094                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1095                         return NF_ACCEPT;
1096         }
1097
1098         if (unlikely(!skb_dst(skb)))
1099                 return NF_ACCEPT;
1100
1101         net = skb_net(skb);
1102         if (!net_ipvs(net)->enable)
1103                 return NF_ACCEPT;
1104
1105         ip_vs_fill_iph_skb(af, skb, &iph);
1106 #ifdef CONFIG_IP_VS_IPV6
1107         if (af == AF_INET6) {
1108                 if (!iph.fragoffs && skb_nfct_reasm(skb)) {
1109                         struct sk_buff *reasm = skb_nfct_reasm(skb);
1110                         /* Save fw mark for coming frags */
1111                         reasm->ipvs_property = 1;
1112                         reasm->mark = skb->mark;
1113                 }
1114                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1115                         int related;
1116                         int verdict = ip_vs_out_icmp_v6(skb, &related,
1117                                                         hooknum, &iph);
1118
1119                         if (related)
1120                                 return verdict;
1121                 }
1122         } else
1123 #endif
1124                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1125                         int related;
1126                         int verdict = ip_vs_out_icmp(skb, &related, hooknum);
1127
1128                         if (related)
1129                                 return verdict;
1130                 }
1131
1132         pd = ip_vs_proto_data_get(net, iph.protocol);
1133         if (unlikely(!pd))
1134                 return NF_ACCEPT;
1135         pp = pd->pp;
1136
1137         /* reassemble IP fragments */
1138 #ifdef CONFIG_IP_VS_IPV6
1139         if (af == AF_INET)
1140 #endif
1141                 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1142                         if (ip_vs_gather_frags(skb,
1143                                                ip_vs_defrag_user(hooknum)))
1144                                 return NF_STOLEN;
1145
1146                         ip_vs_fill_ip4hdr(skb_network_header(skb), &iph);
1147                 }
1148
1149         /*
1150          * Check if the packet belongs to an existing entry
1151          */
1152         cp = pp->conn_out_get(af, skb, &iph, 0);
1153
1154         if (likely(cp))
1155                 return handle_response(af, skb, pd, cp, &iph);
1156         if (sysctl_nat_icmp_send(net) &&
1157             (pp->protocol == IPPROTO_TCP ||
1158              pp->protocol == IPPROTO_UDP ||
1159              pp->protocol == IPPROTO_SCTP)) {
1160                 __be16 _ports[2], *pptr;
1161
1162                 pptr = frag_safe_skb_hp(skb, iph.len,
1163                                          sizeof(_ports), _ports, &iph);
1164                 if (pptr == NULL)
1165                         return NF_ACCEPT;       /* Not for me */
1166                 if (ip_vs_has_real_service(net, af, iph.protocol, &iph.saddr,
1167                                            pptr[0])) {
1168                         /*
1169                          * Notify the real server: there is no
1170                          * existing entry if it is not RST
1171                          * packet or not TCP packet.
1172                          */
1173                         if ((iph.protocol != IPPROTO_TCP &&
1174                              iph.protocol != IPPROTO_SCTP)
1175                              || ((iph.protocol == IPPROTO_TCP
1176                                   && !is_tcp_reset(skb, iph.len))
1177                                  || (iph.protocol == IPPROTO_SCTP
1178                                         && !is_sctp_abort(skb,
1179                                                 iph.len)))) {
1180 #ifdef CONFIG_IP_VS_IPV6
1181                                 if (af == AF_INET6) {
1182                                         if (!skb->dev)
1183                                                 skb->dev = net->loopback_dev;
1184                                         icmpv6_send(skb,
1185                                                     ICMPV6_DEST_UNREACH,
1186                                                     ICMPV6_PORT_UNREACH,
1187                                                     0);
1188                                 } else
1189 #endif
1190                                         icmp_send(skb,
1191                                                   ICMP_DEST_UNREACH,
1192                                                   ICMP_PORT_UNREACH, 0);
1193                                 return NF_DROP;
1194                         }
1195                 }
1196         }
1197         IP_VS_DBG_PKT(12, af, pp, skb, 0,
1198                       "ip_vs_out: packet continues traversal as normal");
1199         return NF_ACCEPT;
1200 }
1201
1202 /*
1203  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1204  *      used only for VS/NAT.
1205  *      Check if packet is reply for established ip_vs_conn.
1206  */
1207 static unsigned int
1208 ip_vs_reply4(unsigned int hooknum, struct sk_buff *skb,
1209              const struct net_device *in, const struct net_device *out,
1210              int (*okfn)(struct sk_buff *))
1211 {
1212         return ip_vs_out(hooknum, skb, AF_INET);
1213 }
1214
1215 /*
1216  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1217  *      Check if packet is reply for established ip_vs_conn.
1218  */
1219 static unsigned int
1220 ip_vs_local_reply4(unsigned int hooknum, struct sk_buff *skb,
1221                    const struct net_device *in, const struct net_device *out,
1222                    int (*okfn)(struct sk_buff *))
1223 {
1224         return ip_vs_out(hooknum, skb, AF_INET);
1225 }
1226
1227 #ifdef CONFIG_IP_VS_IPV6
1228
1229 /*
1230  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1231  *      used only for VS/NAT.
1232  *      Check if packet is reply for established ip_vs_conn.
1233  */
1234 static unsigned int
1235 ip_vs_reply6(unsigned int hooknum, struct sk_buff *skb,
1236              const struct net_device *in, const struct net_device *out,
1237              int (*okfn)(struct sk_buff *))
1238 {
1239         return ip_vs_out(hooknum, skb, AF_INET6);
1240 }
1241
1242 /*
1243  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1244  *      Check if packet is reply for established ip_vs_conn.
1245  */
1246 static unsigned int
1247 ip_vs_local_reply6(unsigned int hooknum, struct sk_buff *skb,
1248                    const struct net_device *in, const struct net_device *out,
1249                    int (*okfn)(struct sk_buff *))
1250 {
1251         return ip_vs_out(hooknum, skb, AF_INET6);
1252 }
1253
1254 #endif
1255
1256 /*
1257  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1258  *      Find any that might be relevant, check against existing connections,
1259  *      forward to the right destination host if relevant.
1260  *      Currently handles error types - unreachable, quench, ttl exceeded.
1261  */
1262 static int
1263 ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum)
1264 {
1265         struct net *net = NULL;
1266         struct iphdr *iph;
1267         struct icmphdr  _icmph, *ic;
1268         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1269         struct ip_vs_iphdr ciph;
1270         struct ip_vs_conn *cp;
1271         struct ip_vs_protocol *pp;
1272         struct ip_vs_proto_data *pd;
1273         unsigned int offset, offset2, ihl, verdict;
1274         bool ipip;
1275
1276         *related = 1;
1277
1278         /* reassemble IP fragments */
1279         if (ip_is_fragment(ip_hdr(skb))) {
1280                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
1281                         return NF_STOLEN;
1282         }
1283
1284         iph = ip_hdr(skb);
1285         offset = ihl = iph->ihl * 4;
1286         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1287         if (ic == NULL)
1288                 return NF_DROP;
1289
1290         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1291                   ic->type, ntohs(icmp_id(ic)),
1292                   &iph->saddr, &iph->daddr);
1293
1294         /*
1295          * Work through seeing if this is for us.
1296          * These checks are supposed to be in an order that means easy
1297          * things are checked first to speed up processing.... however
1298          * this means that some packets will manage to get a long way
1299          * down this stack and then be rejected, but that's life.
1300          */
1301         if ((ic->type != ICMP_DEST_UNREACH) &&
1302             (ic->type != ICMP_SOURCE_QUENCH) &&
1303             (ic->type != ICMP_TIME_EXCEEDED)) {
1304                 *related = 0;
1305                 return NF_ACCEPT;
1306         }
1307
1308         /* Now find the contained IP header */
1309         offset += sizeof(_icmph);
1310         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1311         if (cih == NULL)
1312                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1313
1314         net = skb_net(skb);
1315
1316         /* Special case for errors for IPIP packets */
1317         ipip = false;
1318         if (cih->protocol == IPPROTO_IPIP) {
1319                 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1320                         return NF_ACCEPT;
1321                 /* Error for our IPIP must arrive at LOCAL_IN */
1322                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1323                         return NF_ACCEPT;
1324                 offset += cih->ihl * 4;
1325                 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1326                 if (cih == NULL)
1327                         return NF_ACCEPT; /* The packet looks wrong, ignore */
1328                 ipip = true;
1329         }
1330
1331         pd = ip_vs_proto_data_get(net, cih->protocol);
1332         if (!pd)
1333                 return NF_ACCEPT;
1334         pp = pd->pp;
1335
1336         /* Is the embedded protocol header present? */
1337         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1338                      pp->dont_defrag))
1339                 return NF_ACCEPT;
1340
1341         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1342                       "Checking incoming ICMP for");
1343
1344         offset2 = offset;
1345         ip_vs_fill_ip4hdr(cih, &ciph);
1346         ciph.len += offset;
1347         offset = ciph.len;
1348         /* The embedded headers contain source and dest in reverse order.
1349          * For IPIP this is error for request, not for reply.
1350          */
1351         cp = pp->conn_in_get(AF_INET, skb, &ciph, ipip ? 0 : 1);
1352         if (!cp)
1353                 return NF_ACCEPT;
1354
1355         verdict = NF_DROP;
1356
1357         /* Ensure the checksum is correct */
1358         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1359                 /* Failed checksum! */
1360                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1361                           &iph->saddr);
1362                 goto out;
1363         }
1364
1365         if (ipip) {
1366                 __be32 info = ic->un.gateway;
1367
1368                 /* Update the MTU */
1369                 if (ic->type == ICMP_DEST_UNREACH &&
1370                     ic->code == ICMP_FRAG_NEEDED) {
1371                         struct ip_vs_dest *dest = cp->dest;
1372                         u32 mtu = ntohs(ic->un.frag.mtu);
1373
1374                         /* Strip outer IP and ICMP, go to IPIP header */
1375                         __skb_pull(skb, ihl + sizeof(_icmph));
1376                         offset2 -= ihl + sizeof(_icmph);
1377                         skb_reset_network_header(skb);
1378                         IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1379                                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1380                         ipv4_update_pmtu(skb, dev_net(skb->dev),
1381                                          mtu, 0, 0, 0, 0);
1382                         /* Client uses PMTUD? */
1383                         if (!(cih->frag_off & htons(IP_DF)))
1384                                 goto ignore_ipip;
1385                         /* Prefer the resulting PMTU */
1386                         if (dest) {
1387                                 struct ip_vs_dest_dst *dest_dst;
1388
1389                                 rcu_read_lock();
1390                                 dest_dst = rcu_dereference(dest->dest_dst);
1391                                 if (dest_dst)
1392                                         mtu = dst_mtu(dest_dst->dst_cache);
1393                                 rcu_read_unlock();
1394                         }
1395                         if (mtu > 68 + sizeof(struct iphdr))
1396                                 mtu -= sizeof(struct iphdr);
1397                         info = htonl(mtu);
1398                 }
1399                 /* Strip outer IP, ICMP and IPIP, go to IP header of
1400                  * original request.
1401                  */
1402                 __skb_pull(skb, offset2);
1403                 skb_reset_network_header(skb);
1404                 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1405                         &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1406                         ic->type, ic->code, ntohl(info));
1407                 icmp_send(skb, ic->type, ic->code, info);
1408                 /* ICMP can be shorter but anyways, account it */
1409                 ip_vs_out_stats(cp, skb);
1410
1411 ignore_ipip:
1412                 consume_skb(skb);
1413                 verdict = NF_STOLEN;
1414                 goto out;
1415         }
1416
1417         /* do the statistics and put it back */
1418         ip_vs_in_stats(cp, skb);
1419         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol)
1420                 offset += 2 * sizeof(__u16);
1421         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1422
1423 out:
1424         __ip_vs_conn_put(cp);
1425
1426         return verdict;
1427 }
1428
1429 #ifdef CONFIG_IP_VS_IPV6
1430 static int ip_vs_in_icmp_v6(struct sk_buff *skb, int *related,
1431                             unsigned int hooknum, struct ip_vs_iphdr *iph)
1432 {
1433         struct net *net = NULL;
1434         struct ipv6hdr _ip6h, *ip6h;
1435         struct icmp6hdr _icmph, *ic;
1436         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1437         struct ip_vs_conn *cp;
1438         struct ip_vs_protocol *pp;
1439         struct ip_vs_proto_data *pd;
1440         unsigned int offs_ciph, writable, verdict;
1441
1442         *related = 1;
1443
1444         ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph, iph);
1445         if (ic == NULL)
1446                 return NF_DROP;
1447
1448         /*
1449          * Work through seeing if this is for us.
1450          * These checks are supposed to be in an order that means easy
1451          * things are checked first to speed up processing.... however
1452          * this means that some packets will manage to get a long way
1453          * down this stack and then be rejected, but that's life.
1454          */
1455         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1456                 *related = 0;
1457                 return NF_ACCEPT;
1458         }
1459         /* Fragment header that is before ICMP header tells us that:
1460          * it's not an error message since they can't be fragmented.
1461          */
1462         if (iph->flags & IP6_FH_F_FRAG)
1463                 return NF_DROP;
1464
1465         IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1466                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1467                   &iph->saddr, &iph->daddr);
1468
1469         /* Now find the contained IP header */
1470         ciph.len = iph->len + sizeof(_icmph);
1471         offs_ciph = ciph.len; /* Save ip header offset */
1472         ip6h = skb_header_pointer(skb, ciph.len, sizeof(_ip6h), &_ip6h);
1473         if (ip6h == NULL)
1474                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1475         ciph.saddr.in6 = ip6h->saddr; /* conn_in_get() handles reverse order */
1476         ciph.daddr.in6 = ip6h->daddr;
1477         /* skip possible IPv6 exthdrs of contained IPv6 packet */
1478         ciph.protocol = ipv6_find_hdr(skb, &ciph.len, -1, &ciph.fragoffs, NULL);
1479         if (ciph.protocol < 0)
1480                 return NF_ACCEPT; /* Contained IPv6 hdr looks wrong, ignore */
1481
1482         net = skb_net(skb);
1483         pd = ip_vs_proto_data_get(net, ciph.protocol);
1484         if (!pd)
1485                 return NF_ACCEPT;
1486         pp = pd->pp;
1487
1488         /* Cannot handle fragmented embedded protocol */
1489         if (ciph.fragoffs)
1490                 return NF_ACCEPT;
1491
1492         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offs_ciph,
1493                       "Checking incoming ICMPv6 for");
1494
1495         /* The embedded headers contain source and dest in reverse order
1496          * if not from localhost
1497          */
1498         cp = pp->conn_in_get(AF_INET6, skb, &ciph,
1499                              (hooknum == NF_INET_LOCAL_OUT) ? 0 : 1);
1500
1501         if (!cp)
1502                 return NF_ACCEPT;
1503         /* VS/TUN, VS/DR and LOCALNODE just let it go */
1504         if ((hooknum == NF_INET_LOCAL_OUT) &&
1505             (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1506                 __ip_vs_conn_put(cp);
1507                 return NF_ACCEPT;
1508         }
1509
1510         /* do the statistics and put it back */
1511         ip_vs_in_stats(cp, skb);
1512
1513         /* Need to mangle contained IPv6 header in ICMPv6 packet */
1514         writable = ciph.len;
1515         if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1516             IPPROTO_SCTP == ciph.protocol)
1517                 writable += 2 * sizeof(__u16); /* Also mangle ports */
1518
1519         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, writable, hooknum, &ciph);
1520
1521         __ip_vs_conn_put(cp);
1522
1523         return verdict;
1524 }
1525 #endif
1526
1527
1528 /*
1529  *      Check if it's for virtual services, look it up,
1530  *      and send it on its way...
1531  */
1532 static unsigned int
1533 ip_vs_in(unsigned int hooknum, struct sk_buff *skb, int af)
1534 {
1535         struct net *net;
1536         struct ip_vs_iphdr iph;
1537         struct ip_vs_protocol *pp;
1538         struct ip_vs_proto_data *pd;
1539         struct ip_vs_conn *cp;
1540         int ret, pkts;
1541         struct netns_ipvs *ipvs;
1542
1543         /* Already marked as IPVS request or reply? */
1544         if (skb->ipvs_property)
1545                 return NF_ACCEPT;
1546
1547         /*
1548          *      Big tappo:
1549          *      - remote client: only PACKET_HOST
1550          *      - route: used for struct net when skb->dev is unset
1551          */
1552         if (unlikely((skb->pkt_type != PACKET_HOST &&
1553                       hooknum != NF_INET_LOCAL_OUT) ||
1554                      !skb_dst(skb))) {
1555                 ip_vs_fill_iph_skb(af, skb, &iph);
1556                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1557                               " ignored in hook %u\n",
1558                               skb->pkt_type, iph.protocol,
1559                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1560                 return NF_ACCEPT;
1561         }
1562         /* ipvs enabled in this netns ? */
1563         net = skb_net(skb);
1564         ipvs = net_ipvs(net);
1565         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1566                 return NF_ACCEPT;
1567
1568         ip_vs_fill_iph_skb(af, skb, &iph);
1569
1570         /* Bad... Do not break raw sockets */
1571         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1572                      af == AF_INET)) {
1573                 struct sock *sk = skb->sk;
1574                 struct inet_sock *inet = inet_sk(skb->sk);
1575
1576                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1577                         return NF_ACCEPT;
1578         }
1579
1580 #ifdef CONFIG_IP_VS_IPV6
1581         if (af == AF_INET6) {
1582                 if (!iph.fragoffs && skb_nfct_reasm(skb)) {
1583                         struct sk_buff *reasm = skb_nfct_reasm(skb);
1584                         /* Save fw mark for coming frags. */
1585                         reasm->ipvs_property = 1;
1586                         reasm->mark = skb->mark;
1587                 }
1588                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1589                         int related;
1590                         int verdict = ip_vs_in_icmp_v6(skb, &related, hooknum,
1591                                                        &iph);
1592
1593                         if (related)
1594                                 return verdict;
1595                 }
1596         } else
1597 #endif
1598                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1599                         int related;
1600                         int verdict = ip_vs_in_icmp(skb, &related, hooknum);
1601
1602                         if (related)
1603                                 return verdict;
1604                 }
1605
1606         /* Protocol supported? */
1607         pd = ip_vs_proto_data_get(net, iph.protocol);
1608         if (unlikely(!pd))
1609                 return NF_ACCEPT;
1610         pp = pd->pp;
1611         /*
1612          * Check if the packet belongs to an existing connection entry
1613          */
1614         cp = pp->conn_in_get(af, skb, &iph, 0);
1615         if (unlikely(!cp) && !iph.fragoffs) {
1616                 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1617                  * replayed fragment zero will already have created the cp
1618                  */
1619                 int v;
1620
1621                 /* Schedule and create new connection entry into &cp */
1622                 if (!pp->conn_schedule(af, skb, pd, &v, &cp, &iph))
1623                         return v;
1624         }
1625
1626         if (unlikely(!cp)) {
1627                 /* sorry, all this trouble for a no-hit :) */
1628                 IP_VS_DBG_PKT(12, af, pp, skb, 0,
1629                               "ip_vs_in: packet continues traversal as normal");
1630                 if (iph.fragoffs && !skb_nfct_reasm(skb)) {
1631                         /* Fragment that couldn't be mapped to a conn entry
1632                          * and don't have any pointer to a reasm skb
1633                          * is missing module nf_defrag_ipv6
1634                          */
1635                         IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1636                         IP_VS_DBG_PKT(7, af, pp, skb, 0, "unhandled fragment");
1637                 }
1638                 return NF_ACCEPT;
1639         }
1640
1641         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Incoming packet");
1642         /* Check the server status */
1643         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1644                 /* the destination server is not available */
1645
1646                 if (sysctl_expire_nodest_conn(ipvs)) {
1647                         /* try to expire the connection immediately */
1648                         ip_vs_conn_expire_now(cp);
1649                 }
1650                 /* don't restart its timer, and silently
1651                    drop the packet. */
1652                 __ip_vs_conn_put(cp);
1653                 return NF_DROP;
1654         }
1655
1656         ip_vs_in_stats(cp, skb);
1657         ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1658         if (cp->packet_xmit)
1659                 ret = cp->packet_xmit(skb, cp, pp, &iph);
1660                 /* do not touch skb anymore */
1661         else {
1662                 IP_VS_DBG_RL("warning: packet_xmit is null");
1663                 ret = NF_ACCEPT;
1664         }
1665
1666         /* Increase its packet counter and check if it is needed
1667          * to be synchronized
1668          *
1669          * Sync connection if it is about to close to
1670          * encorage the standby servers to update the connections timeout
1671          *
1672          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1673          */
1674
1675         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
1676                 pkts = sysctl_sync_threshold(ipvs);
1677         else
1678                 pkts = atomic_add_return(1, &cp->in_pkts);
1679
1680         if (ipvs->sync_state & IP_VS_STATE_MASTER)
1681                 ip_vs_sync_conn(net, cp, pkts);
1682
1683         ip_vs_conn_put(cp);
1684         return ret;
1685 }
1686
1687 /*
1688  *      AF_INET handler in NF_INET_LOCAL_IN chain
1689  *      Schedule and forward packets from remote clients
1690  */
1691 static unsigned int
1692 ip_vs_remote_request4(unsigned int hooknum, struct sk_buff *skb,
1693                       const struct net_device *in,
1694                       const struct net_device *out,
1695                       int (*okfn)(struct sk_buff *))
1696 {
1697         return ip_vs_in(hooknum, skb, AF_INET);
1698 }
1699
1700 /*
1701  *      AF_INET handler in NF_INET_LOCAL_OUT chain
1702  *      Schedule and forward packets from local clients
1703  */
1704 static unsigned int
1705 ip_vs_local_request4(unsigned int hooknum, struct sk_buff *skb,
1706                      const struct net_device *in, const struct net_device *out,
1707                      int (*okfn)(struct sk_buff *))
1708 {
1709         return ip_vs_in(hooknum, skb, AF_INET);
1710 }
1711
1712 #ifdef CONFIG_IP_VS_IPV6
1713
1714 /*
1715  * AF_INET6 fragment handling
1716  * Copy info from first fragment, to the rest of them.
1717  */
1718 static unsigned int
1719 ip_vs_preroute_frag6(unsigned int hooknum, struct sk_buff *skb,
1720                      const struct net_device *in,
1721                      const struct net_device *out,
1722                      int (*okfn)(struct sk_buff *))
1723 {
1724         struct sk_buff *reasm = skb_nfct_reasm(skb);
1725         struct net *net;
1726
1727         /* Skip if not a "replay" from nf_ct_frag6_output or first fragment.
1728          * ipvs_property is set when checking first fragment
1729          * in ip_vs_in() and ip_vs_out().
1730          */
1731         if (reasm)
1732                 IP_VS_DBG(2, "Fragment recv prop:%d\n", reasm->ipvs_property);
1733         if (!reasm || !reasm->ipvs_property)
1734                 return NF_ACCEPT;
1735
1736         net = skb_net(skb);
1737         if (!net_ipvs(net)->enable)
1738                 return NF_ACCEPT;
1739
1740         /* Copy stored fw mark, saved in ip_vs_{in,out} */
1741         skb->mark = reasm->mark;
1742
1743         return NF_ACCEPT;
1744 }
1745
1746 /*
1747  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
1748  *      Schedule and forward packets from remote clients
1749  */
1750 static unsigned int
1751 ip_vs_remote_request6(unsigned int hooknum, struct sk_buff *skb,
1752                       const struct net_device *in,
1753                       const struct net_device *out,
1754                       int (*okfn)(struct sk_buff *))
1755 {
1756         return ip_vs_in(hooknum, skb, AF_INET6);
1757 }
1758
1759 /*
1760  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
1761  *      Schedule and forward packets from local clients
1762  */
1763 static unsigned int
1764 ip_vs_local_request6(unsigned int hooknum, struct sk_buff *skb,
1765                      const struct net_device *in, const struct net_device *out,
1766                      int (*okfn)(struct sk_buff *))
1767 {
1768         return ip_vs_in(hooknum, skb, AF_INET6);
1769 }
1770
1771 #endif
1772
1773
1774 /*
1775  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1776  *      related packets destined for 0.0.0.0/0.
1777  *      When fwmark-based virtual service is used, such as transparent
1778  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
1779  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1780  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1781  *      and send them to ip_vs_in_icmp.
1782  */
1783 static unsigned int
1784 ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff *skb,
1785                    const struct net_device *in, const struct net_device *out,
1786                    int (*okfn)(struct sk_buff *))
1787 {
1788         int r;
1789         struct net *net;
1790         struct netns_ipvs *ipvs;
1791
1792         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1793                 return NF_ACCEPT;
1794
1795         /* ipvs enabled in this netns ? */
1796         net = skb_net(skb);
1797         ipvs = net_ipvs(net);
1798         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1799                 return NF_ACCEPT;
1800
1801         return ip_vs_in_icmp(skb, &r, hooknum);
1802 }
1803
1804 #ifdef CONFIG_IP_VS_IPV6
1805 static unsigned int
1806 ip_vs_forward_icmp_v6(unsigned int hooknum, struct sk_buff *skb,
1807                       const struct net_device *in, const struct net_device *out,
1808                       int (*okfn)(struct sk_buff *))
1809 {
1810         int r;
1811         struct net *net;
1812         struct netns_ipvs *ipvs;
1813         struct ip_vs_iphdr iphdr;
1814
1815         ip_vs_fill_iph_skb(AF_INET6, skb, &iphdr);
1816         if (iphdr.protocol != IPPROTO_ICMPV6)
1817                 return NF_ACCEPT;
1818
1819         /* ipvs enabled in this netns ? */
1820         net = skb_net(skb);
1821         ipvs = net_ipvs(net);
1822         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1823                 return NF_ACCEPT;
1824
1825         return ip_vs_in_icmp_v6(skb, &r, hooknum, &iphdr);
1826 }
1827 #endif
1828
1829
1830 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
1831         /* After packet filtering, change source only for VS/NAT */
1832         {
1833                 .hook           = ip_vs_reply4,
1834                 .owner          = THIS_MODULE,
1835                 .pf             = NFPROTO_IPV4,
1836                 .hooknum        = NF_INET_LOCAL_IN,
1837                 .priority       = NF_IP_PRI_NAT_SRC - 2,
1838         },
1839         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1840          * or VS/NAT(change destination), so that filtering rules can be
1841          * applied to IPVS. */
1842         {
1843                 .hook           = ip_vs_remote_request4,
1844                 .owner          = THIS_MODULE,
1845                 .pf             = NFPROTO_IPV4,
1846                 .hooknum        = NF_INET_LOCAL_IN,
1847                 .priority       = NF_IP_PRI_NAT_SRC - 1,
1848         },
1849         /* Before ip_vs_in, change source only for VS/NAT */
1850         {
1851                 .hook           = ip_vs_local_reply4,
1852                 .owner          = THIS_MODULE,
1853                 .pf             = NFPROTO_IPV4,
1854                 .hooknum        = NF_INET_LOCAL_OUT,
1855                 .priority       = NF_IP_PRI_NAT_DST + 1,
1856         },
1857         /* After mangle, schedule and forward local requests */
1858         {
1859                 .hook           = ip_vs_local_request4,
1860                 .owner          = THIS_MODULE,
1861                 .pf             = NFPROTO_IPV4,
1862                 .hooknum        = NF_INET_LOCAL_OUT,
1863                 .priority       = NF_IP_PRI_NAT_DST + 2,
1864         },
1865         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1866          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1867         {
1868                 .hook           = ip_vs_forward_icmp,
1869                 .owner          = THIS_MODULE,
1870                 .pf             = NFPROTO_IPV4,
1871                 .hooknum        = NF_INET_FORWARD,
1872                 .priority       = 99,
1873         },
1874         /* After packet filtering, change source only for VS/NAT */
1875         {
1876                 .hook           = ip_vs_reply4,
1877                 .owner          = THIS_MODULE,
1878                 .pf             = NFPROTO_IPV4,
1879                 .hooknum        = NF_INET_FORWARD,
1880                 .priority       = 100,
1881         },
1882 #ifdef CONFIG_IP_VS_IPV6
1883         /* After mangle & nat fetch 2:nd fragment and following */
1884         {
1885                 .hook           = ip_vs_preroute_frag6,
1886                 .owner          = THIS_MODULE,
1887                 .pf             = NFPROTO_IPV6,
1888                 .hooknum        = NF_INET_PRE_ROUTING,
1889                 .priority       = NF_IP6_PRI_NAT_DST + 1,
1890         },
1891         /* After packet filtering, change source only for VS/NAT */
1892         {
1893                 .hook           = ip_vs_reply6,
1894                 .owner          = THIS_MODULE,
1895                 .pf             = NFPROTO_IPV6,
1896                 .hooknum        = NF_INET_LOCAL_IN,
1897                 .priority       = NF_IP6_PRI_NAT_SRC - 2,
1898         },
1899         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1900          * or VS/NAT(change destination), so that filtering rules can be
1901          * applied to IPVS. */
1902         {
1903                 .hook           = ip_vs_remote_request6,
1904                 .owner          = THIS_MODULE,
1905                 .pf             = NFPROTO_IPV6,
1906                 .hooknum        = NF_INET_LOCAL_IN,
1907                 .priority       = NF_IP6_PRI_NAT_SRC - 1,
1908         },
1909         /* Before ip_vs_in, change source only for VS/NAT */
1910         {
1911                 .hook           = ip_vs_local_reply6,
1912                 .owner          = THIS_MODULE,
1913                 .pf             = NFPROTO_IPV4,
1914                 .hooknum        = NF_INET_LOCAL_OUT,
1915                 .priority       = NF_IP6_PRI_NAT_DST + 1,
1916         },
1917         /* After mangle, schedule and forward local requests */
1918         {
1919                 .hook           = ip_vs_local_request6,
1920                 .owner          = THIS_MODULE,
1921                 .pf             = NFPROTO_IPV6,
1922                 .hooknum        = NF_INET_LOCAL_OUT,
1923                 .priority       = NF_IP6_PRI_NAT_DST + 2,
1924         },
1925         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1926          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1927         {
1928                 .hook           = ip_vs_forward_icmp_v6,
1929                 .owner          = THIS_MODULE,
1930                 .pf             = NFPROTO_IPV6,
1931                 .hooknum        = NF_INET_FORWARD,
1932                 .priority       = 99,
1933         },
1934         /* After packet filtering, change source only for VS/NAT */
1935         {
1936                 .hook           = ip_vs_reply6,
1937                 .owner          = THIS_MODULE,
1938                 .pf             = NFPROTO_IPV6,
1939                 .hooknum        = NF_INET_FORWARD,
1940                 .priority       = 100,
1941         },
1942 #endif
1943 };
1944 /*
1945  *      Initialize IP Virtual Server netns mem.
1946  */
1947 static int __net_init __ip_vs_init(struct net *net)
1948 {
1949         struct netns_ipvs *ipvs;
1950
1951         ipvs = net_generic(net, ip_vs_net_id);
1952         if (ipvs == NULL)
1953                 return -ENOMEM;
1954
1955         /* Hold the beast until a service is registerd */
1956         ipvs->enable = 0;
1957         ipvs->net = net;
1958         /* Counters used for creating unique names */
1959         ipvs->gen = atomic_read(&ipvs_netns_cnt);
1960         atomic_inc(&ipvs_netns_cnt);
1961         net->ipvs = ipvs;
1962
1963         if (ip_vs_estimator_net_init(net) < 0)
1964                 goto estimator_fail;
1965
1966         if (ip_vs_control_net_init(net) < 0)
1967                 goto control_fail;
1968
1969         if (ip_vs_protocol_net_init(net) < 0)
1970                 goto protocol_fail;
1971
1972         if (ip_vs_app_net_init(net) < 0)
1973                 goto app_fail;
1974
1975         if (ip_vs_conn_net_init(net) < 0)
1976                 goto conn_fail;
1977
1978         if (ip_vs_sync_net_init(net) < 0)
1979                 goto sync_fail;
1980
1981         printk(KERN_INFO "IPVS: Creating netns size=%zu id=%d\n",
1982                          sizeof(struct netns_ipvs), ipvs->gen);
1983         return 0;
1984 /*
1985  * Error handling
1986  */
1987
1988 sync_fail:
1989         ip_vs_conn_net_cleanup(net);
1990 conn_fail:
1991         ip_vs_app_net_cleanup(net);
1992 app_fail:
1993         ip_vs_protocol_net_cleanup(net);
1994 protocol_fail:
1995         ip_vs_control_net_cleanup(net);
1996 control_fail:
1997         ip_vs_estimator_net_cleanup(net);
1998 estimator_fail:
1999         net->ipvs = NULL;
2000         return -ENOMEM;
2001 }
2002
2003 static void __net_exit __ip_vs_cleanup(struct net *net)
2004 {
2005         ip_vs_service_net_cleanup(net); /* ip_vs_flush() with locks */
2006         ip_vs_conn_net_cleanup(net);
2007         ip_vs_app_net_cleanup(net);
2008         ip_vs_protocol_net_cleanup(net);
2009         ip_vs_control_net_cleanup(net);
2010         ip_vs_estimator_net_cleanup(net);
2011         IP_VS_DBG(2, "ipvs netns %d released\n", net_ipvs(net)->gen);
2012         net->ipvs = NULL;
2013 }
2014
2015 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
2016 {
2017         EnterFunction(2);
2018         net_ipvs(net)->enable = 0;      /* Disable packet reception */
2019         smp_wmb();
2020         ip_vs_sync_net_cleanup(net);
2021         LeaveFunction(2);
2022 }
2023
2024 static struct pernet_operations ipvs_core_ops = {
2025         .init = __ip_vs_init,
2026         .exit = __ip_vs_cleanup,
2027         .id   = &ip_vs_net_id,
2028         .size = sizeof(struct netns_ipvs),
2029 };
2030
2031 static struct pernet_operations ipvs_core_dev_ops = {
2032         .exit = __ip_vs_dev_cleanup,
2033 };
2034
2035 /*
2036  *      Initialize IP Virtual Server
2037  */
2038 static int __init ip_vs_init(void)
2039 {
2040         int ret;
2041
2042         ret = ip_vs_control_init();
2043         if (ret < 0) {
2044                 pr_err("can't setup control.\n");
2045                 goto exit;
2046         }
2047
2048         ip_vs_protocol_init();
2049
2050         ret = ip_vs_conn_init();
2051         if (ret < 0) {
2052                 pr_err("can't setup connection table.\n");
2053                 goto cleanup_protocol;
2054         }
2055
2056         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
2057         if (ret < 0)
2058                 goto cleanup_conn;
2059
2060         ret = register_pernet_device(&ipvs_core_dev_ops);
2061         if (ret < 0)
2062                 goto cleanup_sub;
2063
2064         ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2065         if (ret < 0) {
2066                 pr_err("can't register hooks.\n");
2067                 goto cleanup_dev;
2068         }
2069
2070         ret = ip_vs_register_nl_ioctl();
2071         if (ret < 0) {
2072                 pr_err("can't register netlink/ioctl.\n");
2073                 goto cleanup_hooks;
2074         }
2075
2076         pr_info("ipvs loaded.\n");
2077
2078         return ret;
2079
2080 cleanup_hooks:
2081         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2082 cleanup_dev:
2083         unregister_pernet_device(&ipvs_core_dev_ops);
2084 cleanup_sub:
2085         unregister_pernet_subsys(&ipvs_core_ops);
2086 cleanup_conn:
2087         ip_vs_conn_cleanup();
2088 cleanup_protocol:
2089         ip_vs_protocol_cleanup();
2090         ip_vs_control_cleanup();
2091 exit:
2092         return ret;
2093 }
2094
2095 static void __exit ip_vs_cleanup(void)
2096 {
2097         ip_vs_unregister_nl_ioctl();
2098         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2099         unregister_pernet_device(&ipvs_core_dev_ops);
2100         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
2101         ip_vs_conn_cleanup();
2102         ip_vs_protocol_cleanup();
2103         ip_vs_control_cleanup();
2104         pr_info("ipvs unloaded.\n");
2105 }
2106
2107 module_init(ip_vs_init);
2108 module_exit(ip_vs_cleanup);
2109 MODULE_LICENSE("GPL");