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