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