kernel/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                 u64_stats_update_begin(&s->syncp);
 123                 s->cnt.inpkts++;
 124                 s->cnt.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                 u64_stats_update_begin(&s->syncp);
 131                 s->cnt.inpkts++;
 132                 s->cnt.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                 u64_stats_update_begin(&s->syncp);
 138                 s->cnt.inpkts++;
 139                 s->cnt.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                 u64_stats_update_begin(&s->syncp);
 157                 s->cnt.outpkts++;
 158                 s->cnt.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                 u64_stats_update_begin(&s->syncp);
 165                 s->cnt.outpkts++;
 166                 s->cnt.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                 u64_stats_update_begin(&s->syncp);
 172                 s->cnt.outpkts++;
 173                 s->cnt.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         u64_stats_update_begin(&s->syncp);
 187         s->cnt.conns++;
 188         u64_stats_update_end(&s->syncp);
 189
 190         s = this_cpu_ptr(svc->stats.cpustats);
 191         u64_stats_update_begin(&s->syncp);
 192         s->cnt.conns++;
 193         u64_stats_update_end(&s->syncp);
 194
 195         s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 196         u64_stats_update_begin(&s->syncp);
 197         s->cnt.conns++;
 198         u64_stats_update_end(&s->syncp);
 199 }
 200
 201
 202 static inline void
 203 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
 204                 const struct sk_buff *skb,
 205                 struct ip_vs_proto_data *pd)
 206 {
 207         if (likely(pd->pp->state_transition))
 208                 pd->pp->state_transition(cp, direction, skb, pd);
 209 }
 210
 211 static inline int
 212 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
 213                               struct sk_buff *skb, int protocol,
 214                               const union nf_inet_addr *caddr, __be16 cport,
 215                               const union nf_inet_addr *vaddr, __be16 vport,
 216                               struct ip_vs_conn_param *p)
 217 {
 218         ip_vs_conn_fill_param(svc->net, svc->af, protocol, caddr, cport, vaddr,
 219                               vport, p);
 220         p->pe = rcu_dereference(svc->pe);
 221         if (p->pe && p->pe->fill_param)
 222                 return p->pe->fill_param(p, skb);
 223
 224         return 0;
 225 }
 226
 227 /*
 228  *  IPVS persistent scheduling function
 229  *  It creates a connection entry according to its template if exists,
 230  *  or selects a server and creates a connection entry plus a template.
 231  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
 232  *  Protocols supported: TCP, UDP
 233  */
 234 static struct ip_vs_conn *
 235 ip_vs_sched_persist(struct ip_vs_service *svc,
 236                     struct sk_buff *skb, __be16 src_port, __be16 dst_port,
 237                     int *ignored, struct ip_vs_iphdr *iph)
 238 {
 239         struct ip_vs_conn *cp = NULL;
 240         struct ip_vs_dest *dest;
 241         struct ip_vs_conn *ct;
 242         __be16 dport = 0;               /* destination port to forward */
 243         unsigned int flags;
 244         struct ip_vs_conn_param param;
 245         const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
 246         union nf_inet_addr snet;        /* source network of the client,
 247                                            after masking */
 248
 249         /* Mask saddr with the netmask to adjust template granularity */
 250 #ifdef CONFIG_IP_VS_IPV6
 251         if (svc->af == AF_INET6)
 252                 ipv6_addr_prefix(&snet.in6, &iph->saddr.in6,
 253                                  (__force __u32) svc->netmask);
 254         else
 255 #endif
 256                 snet.ip = iph->saddr.ip & svc->netmask;
 257
 258         IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
 259                       "mnet %s\n",
 260                       IP_VS_DBG_ADDR(svc->af, &iph->saddr), ntohs(src_port),
 261                       IP_VS_DBG_ADDR(svc->af, &iph->daddr), ntohs(dst_port),
 262                       IP_VS_DBG_ADDR(svc->af, &snet));
 263
 264         /*
 265          * As far as we know, FTP is a very complicated network protocol, and
 266          * it uses control connection and data connections. For active FTP,
 267          * FTP server initialize data connection to the client, its source port
 268          * is often 20. For passive FTP, FTP server tells the clients the port
 269          * that it passively listens to,  and the client issues the data
 270          * connection. In the tunneling or direct routing mode, the load
 271          * balancer is on the client-to-server half of connection, the port
 272          * number is unknown to the load balancer. So, a conn template like
 273          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
 274          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
 275          * is created for other persistent services.
 276          */
 277         {
 278                 int protocol = iph->protocol;
 279                 const union nf_inet_addr *vaddr = &iph->daddr;
 280                 __be16 vport = 0;
 281
 282                 if (dst_port == svc->port) {
 283                         /* non-FTP template:
 284                          * <protocol, caddr, 0, vaddr, vport, daddr, dport>
 285                          * FTP template:
 286                          * <protocol, caddr, 0, vaddr, 0, daddr, 0>
 287                          */
 288                         if (svc->port != FTPPORT)
 289                                 vport = dst_port;
 290                 } else {
 291                         /* Note: persistent fwmark-based services and
 292                          * persistent port zero service are handled here.
 293                          * fwmark template:
 294                          * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
 295                          * port zero template:
 296                          * <protocol,caddr,0,vaddr,0,daddr,0>
 297                          */
 298                         if (svc->fwmark) {
 299                                 protocol = IPPROTO_IP;
 300                                 vaddr = &fwmark;
 301                         }
 302                 }
 303                 /* return *ignored = -1 so NF_DROP can be used */
 304                 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
 305                                                   vaddr, vport, &param) < 0) {
 306                         *ignored = -1;
 307                         return NULL;
 308                 }
 309         }
 310
 311         /* Check if a template already exists */
 312         ct = ip_vs_ct_in_get(&param);
 313         if (!ct || !ip_vs_check_template(ct)) {
 314                 struct ip_vs_scheduler *sched;
 315
 316                 /*
 317                  * No template found or the dest of the connection
 318                  * template is not available.
 319                  * return *ignored=0 i.e. ICMP and NF_DROP
 320                  */
 321                 sched = rcu_dereference(svc->scheduler);
 322                 dest = sched->schedule(svc, skb, iph);
 323                 if (!dest) {
 324                         IP_VS_DBG(1, "p-schedule: no dest found.\n");
 325                         kfree(param.pe_data);
 326                         *ignored = 0;
 327                         return NULL;
 328                 }
 329
 330                 if (dst_port == svc->port && svc->port != FTPPORT)
 331                         dport = dest->port;
 332
 333                 /* Create a template
 334                  * This adds param.pe_data to the template,
 335                  * and thus param.pe_data will be destroyed
 336                  * when the template expires */
 337                 ct = ip_vs_conn_new(&param, dest->af, &dest->addr, dport,
 338                                     IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
 339                 if (ct == NULL) {
 340                         kfree(param.pe_data);
 341                         *ignored = -1;
 342                         return NULL;
 343                 }
 344
 345                 ct->timeout = svc->timeout;
 346         } else {
 347                 /* set destination with the found template */
 348                 dest = ct->dest;
 349                 kfree(param.pe_data);
 350         }
 351
 352         dport = dst_port;
 353         if (dport == svc->port && dest->port)
 354                 dport = dest->port;
 355
 356         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 357                  && iph->protocol == IPPROTO_UDP) ?
 358                 IP_VS_CONN_F_ONE_PACKET : 0;
 359
 360         /*
 361          *    Create a new connection according to the template
 362          */
 363         ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol, &iph->saddr,
 364                               src_port, &iph->daddr, dst_port, &param);
 365
 366         cp = ip_vs_conn_new(&param, dest->af, &dest->addr, dport, flags, dest,
 367                             skb->mark);
 368         if (cp == NULL) {
 369                 ip_vs_conn_put(ct);
 370                 *ignored = -1;
 371                 return NULL;
 372         }
 373
 374         /*
 375          *    Add its control
 376          */
 377         ip_vs_control_add(cp, ct);
 378         ip_vs_conn_put(ct);
 379
 380         ip_vs_conn_stats(cp, svc);
 381         return cp;
 382 }
 383
 384
 385 /*
 386  *  IPVS main scheduling function
 387  *  It selects a server according to the virtual service, and
 388  *  creates a connection entry.
 389  *  Protocols supported: TCP, UDP
 390  *
 391  *  Usage of *ignored
 392  *
 393  * 1 :   protocol tried to schedule (eg. on SYN), found svc but the
 394  *       svc/scheduler decides that this packet should be accepted with
 395  *       NF_ACCEPT because it must not be scheduled.
 396  *
 397  * 0 :   scheduler can not find destination, so try bypass or
 398  *       return ICMP and then NF_DROP (ip_vs_leave).
 399  *
 400  * -1 :  scheduler tried to schedule but fatal error occurred, eg.
 401  *       ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
 402  *       failure such as missing Call-ID, ENOMEM on skb_linearize
 403  *       or pe_data. In this case we should return NF_DROP without
 404  *       any attempts to send ICMP with ip_vs_leave.
 405  */
 406 struct ip_vs_conn *
 407 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
 408                struct ip_vs_proto_data *pd, int *ignored,
 409                struct ip_vs_iphdr *iph)
 410 {
 411         struct ip_vs_protocol *pp = pd->pp;
 412         struct ip_vs_conn *cp = NULL;
 413         struct ip_vs_scheduler *sched;
 414         struct ip_vs_dest *dest;
 415         __be16 _ports[2], *pptr;
 416         unsigned int flags;
 417
 418         *ignored = 1;
 419         /*
 420          * IPv6 frags, only the first hit here.
 421          */
 422         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 423         if (pptr == NULL)
 424                 return NULL;
 425
 426         /*
 427          * FTPDATA needs this check when using local real server.
 428          * Never schedule Active FTPDATA connections from real server.
 429          * For LVS-NAT they must be already created. For other methods
 430          * with persistence the connection is created on SYN+ACK.
 431          */
 432         if (pptr[0] == FTPDATA) {
 433                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 434                               "Not scheduling FTPDATA");
 435                 return NULL;
 436         }
 437
 438         /*
 439          *    Do not schedule replies from local real server.
 440          */
 441         if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
 442             (cp = pp->conn_in_get(svc->af, skb, iph, 1))) {
 443                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 444                               "Not scheduling reply for existing connection");
 445                 __ip_vs_conn_put(cp);
 446                 return NULL;
 447         }
 448
 449         /*
 450          *    Persistent service
 451          */
 452         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 453                 return ip_vs_sched_persist(svc, skb, pptr[0], pptr[1], ignored,
 454                                            iph);
 455
 456         *ignored = 0;
 457
 458         /*
 459          *    Non-persistent service
 460          */
 461         if (!svc->fwmark && pptr[1] != svc->port) {
 462                 if (!svc->port)
 463                         pr_err("Schedule: port zero only supported "
 464                                "in persistent services, "
 465                                "check your ipvs configuration\n");
 466                 return NULL;
 467         }
 468
 469         sched = rcu_dereference(svc->scheduler);
 470         dest = sched->schedule(svc, skb, iph);
 471         if (dest == NULL) {
 472                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 473                 return NULL;
 474         }
 475
 476         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 477                  && iph->protocol == IPPROTO_UDP) ?
 478                 IP_VS_CONN_F_ONE_PACKET : 0;
 479
 480         /*
 481          *    Create a connection entry.
 482          */
 483         {
 484                 struct ip_vs_conn_param p;
 485
 486                 ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol,
 487                                       &iph->saddr, pptr[0], &iph->daddr,
 488                                       pptr[1], &p);
 489                 cp = ip_vs_conn_new(&p, dest->af, &dest->addr,
 490                                     dest->port ? dest->port : pptr[1],
 491                                     flags, dest, skb->mark);
 492                 if (!cp) {
 493                         *ignored = -1;
 494                         return NULL;
 495                 }
 496         }
 497
 498         IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
 499                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
 500                       ip_vs_fwd_tag(cp),
 501                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 502                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 503                       IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
 504                       cp->flags, atomic_read(&cp->refcnt));
 505
 506         ip_vs_conn_stats(cp, svc);
 507         return cp;
 508 }
 509
 510
 511 /*
 512  *  Pass or drop the packet.
 513  *  Called by ip_vs_in, when the virtual service is available but
 514  *  no destination is available for a new connection.
 515  */
 516 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 517                 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
 518 {
 519         __be16 _ports[2], *pptr;
 520 #ifdef CONFIG_SYSCTL
 521         struct net *net;
 522         struct netns_ipvs *ipvs;
 523         int unicast;
 524 #endif
 525
 526         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 527         if (pptr == NULL) {
 528                 return NF_DROP;
 529         }
 530
 531 #ifdef CONFIG_SYSCTL
 532         net = skb_net(skb);
 533
 534 #ifdef CONFIG_IP_VS_IPV6
 535         if (svc->af == AF_INET6)
 536                 unicast = ipv6_addr_type(&iph->daddr.in6) & IPV6_ADDR_UNICAST;
 537         else
 538 #endif
 539                 unicast = (inet_addr_type(net, iph->daddr.ip) == RTN_UNICAST);
 540
 541         /* if it is fwmark-based service, the cache_bypass sysctl is up
 542            and the destination is a non-local unicast, then create
 543            a cache_bypass connection entry */
 544         ipvs = net_ipvs(net);
 545         if (ipvs->sysctl_cache_bypass && svc->fwmark && unicast) {
 546                 int ret;
 547                 struct ip_vs_conn *cp;
 548                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
 549                                       iph->protocol == IPPROTO_UDP) ?
 550                                       IP_VS_CONN_F_ONE_PACKET : 0;
 551                 union nf_inet_addr daddr =  { .all = { 0, 0, 0, 0 } };
 552
 553                 /* create a new connection entry */
 554                 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
 555                 {
 556                         struct ip_vs_conn_param p;
 557                         ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol,
 558                                               &iph->saddr, pptr[0],
 559                                               &iph->daddr, pptr[1], &p);
 560                         cp = ip_vs_conn_new(&p, svc->af, &daddr, 0,
 561                                             IP_VS_CONN_F_BYPASS | flags,
 562                                             NULL, skb->mark);
 563                         if (!cp)
 564                                 return NF_DROP;
 565                 }
 566
 567                 /* statistics */
 568                 ip_vs_in_stats(cp, skb);
 569
 570                 /* set state */
 571                 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
 572
 573                 /* transmit the first SYN packet */
 574                 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
 575                 /* do not touch skb anymore */
 576
 577                 atomic_inc(&cp->in_pkts);
 578                 ip_vs_conn_put(cp);
 579                 return ret;
 580         }
 581 #endif
 582
 583         /*
 584          * When the virtual ftp service is presented, packets destined
 585          * for other services on the VIP may get here (except services
 586          * listed in the ipvs table), pass the packets, because it is
 587          * not ipvs job to decide to drop the packets.
 588          */
 589         if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT))
 590                 return NF_ACCEPT;
 591
 592         /*
 593          * Notify the client that the destination is unreachable, and
 594          * release the socket buffer.
 595          * Since it is in IP layer, the TCP socket is not actually
 596          * created, the TCP RST packet cannot be sent, instead that
 597          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 598          */
 599 #ifdef CONFIG_IP_VS_IPV6
 600         if (svc->af == AF_INET6) {
 601                 if (!skb->dev) {
 602                         struct net *net_ = dev_net(skb_dst(skb)->dev);
 603
 604                         skb->dev = net_->loopback_dev;
 605                 }
 606                 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 607         } else
 608 #endif
 609                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 610
 611         return NF_DROP;
 612 }
 613
 614 #ifdef CONFIG_SYSCTL
 615
 616 static int sysctl_snat_reroute(struct sk_buff *skb)
 617 {
 618         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 619         return ipvs->sysctl_snat_reroute;
 620 }
 621
 622 static int sysctl_nat_icmp_send(struct net *net)
 623 {
 624         struct netns_ipvs *ipvs = net_ipvs(net);
 625         return ipvs->sysctl_nat_icmp_send;
 626 }
 627
 628 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
 629 {
 630         return ipvs->sysctl_expire_nodest_conn;
 631 }
 632
 633 #else
 634
 635 static int sysctl_snat_reroute(struct sk_buff *skb) { return 0; }
 636 static int sysctl_nat_icmp_send(struct net *net) { return 0; }
 637 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
 638
 639 #endif
 640
 641 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 642 {
 643         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 644 }
 645
 646 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
 647 {
 648         if (NF_INET_LOCAL_IN == hooknum)
 649                 return IP_DEFRAG_VS_IN;
 650         if (NF_INET_FORWARD == hooknum)
 651                 return IP_DEFRAG_VS_FWD;
 652         return IP_DEFRAG_VS_OUT;
 653 }
 654
 655 static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user)
 656 {
 657         int err;
 658
 659         local_bh_disable();
 660         err = ip_defrag(skb, user);
 661         local_bh_enable();
 662         if (!err)
 663                 ip_send_check(ip_hdr(skb));
 664
 665         return err;
 666 }
 667
 668 static int ip_vs_route_me_harder(int af, struct sk_buff *skb,
 669                                  unsigned int hooknum)
 670 {
 671         if (!sysctl_snat_reroute(skb))
 672                 return 0;
 673         /* Reroute replies only to remote clients (FORWARD and LOCAL_OUT) */
 674         if (NF_INET_LOCAL_IN == hooknum)
 675                 return 0;
 676 #ifdef CONFIG_IP_VS_IPV6
 677         if (af == AF_INET6) {
 678                 struct dst_entry *dst = skb_dst(skb);
 679
 680                 if (dst->dev && !(dst->dev->flags & IFF_LOOPBACK) &&
 681                     ip6_route_me_harder(skb) != 0)
 682                         return 1;
 683         } else
 684 #endif
 685                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
 686                     ip_route_me_harder(skb, RTN_LOCAL) != 0)
 687                         return 1;
 688
 689         return 0;
 690 }
 691
 692 /*
 693  * Packet has been made sufficiently writable in caller
 694  * - inout: 1=in->out, 0=out->in
 695  */
 696 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 697                     struct ip_vs_conn *cp, int inout)
 698 {
 699         struct iphdr *iph        = ip_hdr(skb);
 700         unsigned int icmp_offset = iph->ihl*4;
 701         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
 702                                                       icmp_offset);
 703         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 704
 705         if (inout) {
 706                 iph->saddr = cp->vaddr.ip;
 707                 ip_send_check(iph);
 708                 ciph->daddr = cp->vaddr.ip;
 709                 ip_send_check(ciph);
 710         } else {
 711                 iph->daddr = cp->daddr.ip;
 712                 ip_send_check(iph);
 713                 ciph->saddr = cp->daddr.ip;
 714                 ip_send_check(ciph);
 715         }
 716
 717         /* the TCP/UDP/SCTP port */
 718         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
 719             IPPROTO_SCTP == ciph->protocol) {
 720                 __be16 *ports = (void *)ciph + ciph->ihl*4;
 721
 722                 if (inout)
 723                         ports[1] = cp->vport;
 724                 else
 725                         ports[0] = cp->dport;
 726         }
 727
 728         /* And finally the ICMP checksum */
 729         icmph->checksum = 0;
 730         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 731         skb->ip_summed = CHECKSUM_UNNECESSARY;
 732
 733         if (inout)
 734                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 735                         "Forwarding altered outgoing ICMP");
 736         else
 737                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 738                         "Forwarding altered incoming ICMP");
 739 }
 740
 741 #ifdef CONFIG_IP_VS_IPV6
 742 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
 743                     struct ip_vs_conn *cp, int inout)
 744 {
 745         struct ipv6hdr *iph      = ipv6_hdr(skb);
 746         unsigned int icmp_offset = 0;
 747         unsigned int offs        = 0; /* header offset*/
 748         int protocol;
 749         struct icmp6hdr *icmph;
 750         struct ipv6hdr *ciph;
 751         unsigned short fragoffs;
 752
 753         ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
 754         icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
 755         offs = icmp_offset + sizeof(struct icmp6hdr);
 756         ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
 757
 758         protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
 759
 760         if (inout) {
 761                 iph->saddr = cp->vaddr.in6;
 762                 ciph->daddr = cp->vaddr.in6;
 763         } else {
 764                 iph->daddr = cp->daddr.in6;
 765                 ciph->saddr = cp->daddr.in6;
 766         }
 767
 768         /* the TCP/UDP/SCTP port */
 769         if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 770                           IPPROTO_SCTP == protocol)) {
 771                 __be16 *ports = (void *)(skb_network_header(skb) + offs);
 772
 773                 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
 774                               ntohs(inout ? ports[1] : ports[0]),
 775                               ntohs(inout ? cp->vport : cp->dport));
 776                 if (inout)
 777                         ports[1] = cp->vport;
 778                 else
 779                         ports[0] = cp->dport;
 780         }
 781
 782         /* And finally the ICMP checksum */
 783         icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
 784                                               skb->len - icmp_offset,
 785                                               IPPROTO_ICMPV6, 0);
 786         skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
 787         skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
 788         skb->ip_summed = CHECKSUM_PARTIAL;
 789
 790         if (inout)
 791                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 792                               (void *)ciph - (void *)iph,
 793                               "Forwarding altered outgoing ICMPv6");
 794         else
 795                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 796                               (void *)ciph - (void *)iph,
 797                               "Forwarding altered incoming ICMPv6");
 798 }
 799 #endif
 800
 801 /* Handle relevant response ICMP messages - forward to the right
 802  * destination host.
 803  */
 804 static int handle_response_icmp(int af, struct sk_buff *skb,
 805                                 union nf_inet_addr *snet,
 806                                 __u8 protocol, struct ip_vs_conn *cp,
 807                                 struct ip_vs_protocol *pp,
 808                                 unsigned int offset, unsigned int ihl,
 809                                 unsigned int hooknum)
 810 {
 811         unsigned int verdict = NF_DROP;
 812
 813         if (IP_VS_FWD_METHOD(cp) != 0) {
 814                 pr_err("shouldn't reach here, because the box is on the "
 815                        "half connection in the tun/dr module.\n");
 816         }
 817
 818         /* Ensure the checksum is correct */
 819         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 820                 /* Failed checksum! */
 821                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
 822                               IP_VS_DBG_ADDR(af, snet));
 823                 goto out;
 824         }
 825
 826         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 827             IPPROTO_SCTP == protocol)
 828                 offset += 2 * sizeof(__u16);
 829         if (!skb_make_writable(skb, offset))
 830                 goto out;
 831
 832 #ifdef CONFIG_IP_VS_IPV6
 833         if (af == AF_INET6)
 834                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
 835         else
 836 #endif
 837                 ip_vs_nat_icmp(skb, pp, cp, 1);
 838
 839         if (ip_vs_route_me_harder(af, skb, hooknum))
 840                 goto out;
 841
 842         /* do the statistics and put it back */
 843         ip_vs_out_stats(cp, skb);
 844
 845         skb->ipvs_property = 1;
 846         if (!(cp->flags & IP_VS_CONN_F_NFCT))
 847                 ip_vs_notrack(skb);
 848         else
 849                 ip_vs_update_conntrack(skb, cp, 0);
 850         verdict = NF_ACCEPT;
 851
 852 out:
 853         __ip_vs_conn_put(cp);
 854
 855         return verdict;
 856 }
 857
 858 /*
 859  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 860  *      Find any that might be relevant, check against existing connections.
 861  *      Currently handles error types - unreachable, quench, ttl exceeded.
 862  */
 863 static int ip_vs_out_icmp(struct sk_buff *skb, int *related,
 864                           unsigned int hooknum)
 865 {
 866         struct iphdr *iph;
 867         struct icmphdr  _icmph, *ic;
 868         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 869         struct ip_vs_iphdr ciph;
 870         struct ip_vs_conn *cp;
 871         struct ip_vs_protocol *pp;
 872         unsigned int offset, ihl;
 873         union nf_inet_addr snet;
 874
 875         *related = 1;
 876
 877         /* reassemble IP fragments */
 878         if (ip_is_fragment(ip_hdr(skb))) {
 879                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
 880                         return NF_STOLEN;
 881         }
 882
 883         iph = ip_hdr(skb);
 884         offset = ihl = iph->ihl * 4;
 885         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 886         if (ic == NULL)
 887                 return NF_DROP;
 888
 889         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
 890                   ic->type, ntohs(icmp_id(ic)),
 891                   &iph->saddr, &iph->daddr);
 892
 893         /*
 894          * Work through seeing if this is for us.
 895          * These checks are supposed to be in an order that means easy
 896          * things are checked first to speed up processing.... however
 897          * this means that some packets will manage to get a long way
 898          * down this stack and then be rejected, but that's life.
 899          */
 900         if ((ic->type != ICMP_DEST_UNREACH) &&
 901             (ic->type != ICMP_SOURCE_QUENCH) &&
 902             (ic->type != ICMP_TIME_EXCEEDED)) {
 903                 *related = 0;
 904                 return NF_ACCEPT;
 905         }
 906
 907         /* Now find the contained IP header */
 908         offset += sizeof(_icmph);
 909         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 910         if (cih == NULL)
 911                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 912
 913         pp = ip_vs_proto_get(cih->protocol);
 914         if (!pp)
 915                 return NF_ACCEPT;
 916
 917         /* Is the embedded protocol header present? */
 918         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 919                      pp->dont_defrag))
 920                 return NF_ACCEPT;
 921
 922         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
 923                       "Checking outgoing ICMP for");
 924
 925         ip_vs_fill_ip4hdr(cih, &ciph);
 926         ciph.len += offset;
 927         /* The embedded headers contain source and dest in reverse order */
 928         cp = pp->conn_out_get(AF_INET, skb, &ciph, 1);
 929         if (!cp)
 930                 return NF_ACCEPT;
 931
 932         snet.ip = iph->saddr;
 933         return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
 934                                     pp, ciph.len, ihl, hooknum);
 935 }
 936
 937 #ifdef CONFIG_IP_VS_IPV6
 938 static int ip_vs_out_icmp_v6(struct sk_buff *skb, int *related,
 939                              unsigned int hooknum, struct ip_vs_iphdr *ipvsh)
 940 {
 941         struct icmp6hdr _icmph, *ic;
 942         struct ipv6hdr _ip6h, *ip6h; /* The ip header contained within ICMP */
 943         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
 944         struct ip_vs_conn *cp;
 945         struct ip_vs_protocol *pp;
 946         union nf_inet_addr snet;
 947         unsigned int writable;
 948
 949         *related = 1;
 950         ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph, ipvsh);
 951         if (ic == NULL)
 952                 return NF_DROP;
 953
 954         /*
 955          * Work through seeing if this is for us.
 956          * These checks are supposed to be in an order that means easy
 957          * things are checked first to speed up processing.... however
 958          * this means that some packets will manage to get a long way
 959          * down this stack and then be rejected, but that's life.
 960          */
 961         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
 962                 *related = 0;
 963                 return NF_ACCEPT;
 964         }
 965         /* Fragment header that is before ICMP header tells us that:
 966          * it's not an error message since they can't be fragmented.
 967          */
 968         if (ipvsh->flags & IP6_FH_F_FRAG)
 969                 return NF_DROP;
 970
 971         IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
 972                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
 973                   &ipvsh->saddr, &ipvsh->daddr);
 974
 975         /* Now find the contained IP header */
 976         ciph.len = ipvsh->len + sizeof(_icmph);
 977         ip6h = skb_header_pointer(skb, ciph.len, sizeof(_ip6h), &_ip6h);
 978         if (ip6h == NULL)
 979                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 980         ciph.saddr.in6 = ip6h->saddr; /* conn_out_get() handles reverse order */
 981         ciph.daddr.in6 = ip6h->daddr;
 982         /* skip possible IPv6 exthdrs of contained IPv6 packet */
 983         ciph.protocol = ipv6_find_hdr(skb, &ciph.len, -1, &ciph.fragoffs, NULL);
 984         if (ciph.protocol < 0)
 985                 return NF_ACCEPT; /* Contained IPv6 hdr looks wrong, ignore */
 986
 987         pp = ip_vs_proto_get(ciph.protocol);
 988         if (!pp)
 989                 return NF_ACCEPT;
 990
 991         /* The embedded headers contain source and dest in reverse order */
 992         cp = pp->conn_out_get(AF_INET6, skb, &ciph, 1);
 993         if (!cp)
 994                 return NF_ACCEPT;
 995
 996         snet.in6 = ciph.saddr.in6;
 997         writable = ciph.len;
 998         return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
 999                                     pp, writable, sizeof(struct ipv6hdr),
1000                                     hooknum);
1001 }
1002 #endif
1003
1004 /*
1005  * Check if sctp chunc is ABORT chunk
1006  */
1007 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
1008 {
1009         sctp_chunkhdr_t *sch, schunk;
1010         sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
1011                         sizeof(schunk), &schunk);
1012         if (sch == NULL)
1013                 return 0;
1014         if (sch->type == SCTP_CID_ABORT)
1015                 return 1;
1016         return 0;
1017 }
1018
1019 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
1020 {
1021         struct tcphdr _tcph, *th;
1022
1023         th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
1024         if (th == NULL)
1025                 return 0;
1026         return th->rst;
1027 }
1028
1029 static inline bool is_new_conn(const struct sk_buff *skb,
1030                                struct ip_vs_iphdr *iph)
1031 {
1032         switch (iph->protocol) {
1033         case IPPROTO_TCP: {
1034                 struct tcphdr _tcph, *th;
1035
1036                 th = skb_header_pointer(skb, iph->len, sizeof(_tcph), &_tcph);
1037                 if (th == NULL)
1038                         return false;
1039                 return th->syn;
1040         }
1041         case IPPROTO_SCTP: {
1042                 sctp_chunkhdr_t *sch, schunk;
1043
1044                 sch = skb_header_pointer(skb, iph->len + sizeof(sctp_sctphdr_t),
1045                                          sizeof(schunk), &schunk);
1046                 if (sch == NULL)
1047                         return false;
1048                 return sch->type == SCTP_CID_INIT;
1049         }
1050         default:
1051                 return false;
1052         }
1053 }
1054
1055 static inline bool is_new_conn_expected(const struct ip_vs_conn *cp,
1056                                         int conn_reuse_mode)
1057 {
1058         /* Controlled (FTP DATA or persistence)? */
1059         if (cp->control)
1060                 return false;
1061
1062         switch (cp->protocol) {
1063         case IPPROTO_TCP:
1064                 return (cp->state == IP_VS_TCP_S_TIME_WAIT) ||
1065                         ((conn_reuse_mode & 2) &&
1066                          (cp->state == IP_VS_TCP_S_FIN_WAIT) &&
1067                          (cp->flags & IP_VS_CONN_F_NOOUTPUT));
1068         case IPPROTO_SCTP:
1069                 return cp->state == IP_VS_SCTP_S_CLOSED;
1070         default:
1071                 return false;
1072         }
1073 }
1074
1075 /* Handle response packets: rewrite addresses and send away...
1076  */
1077 static unsigned int
1078 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1079                 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph,
1080                 unsigned int hooknum)
1081 {
1082         struct ip_vs_protocol *pp = pd->pp;
1083
1084         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Outgoing packet");
1085
1086         if (!skb_make_writable(skb, iph->len))
1087                 goto drop;
1088
1089         /* mangle the packet */
1090         if (pp->snat_handler && !pp->snat_handler(skb, pp, cp, iph))
1091                 goto drop;
1092
1093 #ifdef CONFIG_IP_VS_IPV6
1094         if (af == AF_INET6)
1095                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1096         else
1097 #endif
1098         {
1099                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1100                 ip_send_check(ip_hdr(skb));
1101         }
1102
1103         /*
1104          * nf_iterate does not expect change in the skb->dst->dev.
1105          * It looks like it is not fatal to enable this code for hooks
1106          * where our handlers are at the end of the chain list and
1107          * when all next handlers use skb->dst->dev and not outdev.
1108          * It will definitely route properly the inout NAT traffic
1109          * when multiple paths are used.
1110          */
1111
1112         /* For policy routing, packets originating from this
1113          * machine itself may be routed differently to packets
1114          * passing through.  We want this packet to be routed as
1115          * if it came from this machine itself.  So re-compute
1116          * the routing information.
1117          */
1118         if (ip_vs_route_me_harder(af, skb, hooknum))
1119                 goto drop;
1120
1121         IP_VS_DBG_PKT(10, af, pp, skb, 0, "After SNAT");
1122
1123         ip_vs_out_stats(cp, skb);
1124         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1125         skb->ipvs_property = 1;
1126         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1127                 ip_vs_notrack(skb);
1128         else
1129                 ip_vs_update_conntrack(skb, cp, 0);
1130         ip_vs_conn_put(cp);
1131
1132         LeaveFunction(11);
1133         return NF_ACCEPT;
1134
1135 drop:
1136         ip_vs_conn_put(cp);
1137         kfree_skb(skb);
1138         LeaveFunction(11);
1139         return NF_STOLEN;
1140 }
1141
1142 /*
1143  *      Check if outgoing packet belongs to the established ip_vs_conn.
1144  */
1145 static unsigned int
1146 ip_vs_out(unsigned int hooknum, struct sk_buff *skb, int af)
1147 {
1148         struct net *net = NULL;
1149         struct ip_vs_iphdr iph;
1150         struct ip_vs_protocol *pp;
1151         struct ip_vs_proto_data *pd;
1152         struct ip_vs_conn *cp;
1153
1154         EnterFunction(11);
1155
1156         /* Already marked as IPVS request or reply? */
1157         if (skb->ipvs_property)
1158                 return NF_ACCEPT;
1159
1160         /* Bad... Do not break raw sockets */
1161         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1162                      af == AF_INET)) {
1163                 struct sock *sk = skb->sk;
1164                 struct inet_sock *inet = inet_sk(skb->sk);
1165
1166                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1167                         return NF_ACCEPT;
1168         }
1169
1170         if (unlikely(!skb_dst(skb)))
1171                 return NF_ACCEPT;
1172
1173         net = skb_net(skb);
1174         if (!net_ipvs(net)->enable)
1175                 return NF_ACCEPT;
1176
1177         ip_vs_fill_iph_skb(af, skb, &iph);
1178 #ifdef CONFIG_IP_VS_IPV6
1179         if (af == AF_INET6) {
1180                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1181                         int related;
1182                         int verdict = ip_vs_out_icmp_v6(skb, &related,
1183                                                         hooknum, &iph);
1184
1185                         if (related)
1186                                 return verdict;
1187                 }
1188         } else
1189 #endif
1190                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1191                         int related;
1192                         int verdict = ip_vs_out_icmp(skb, &related, hooknum);
1193
1194                         if (related)
1195                                 return verdict;
1196                 }
1197
1198         pd = ip_vs_proto_data_get(net, iph.protocol);
1199         if (unlikely(!pd))
1200                 return NF_ACCEPT;
1201         pp = pd->pp;
1202
1203         /* reassemble IP fragments */
1204 #ifdef CONFIG_IP_VS_IPV6
1205         if (af == AF_INET)
1206 #endif
1207                 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1208                         if (ip_vs_gather_frags(skb,
1209                                                ip_vs_defrag_user(hooknum)))
1210                                 return NF_STOLEN;
1211
1212                         ip_vs_fill_ip4hdr(skb_network_header(skb), &iph);
1213                 }
1214
1215         /*
1216          * Check if the packet belongs to an existing entry
1217          */
1218         cp = pp->conn_out_get(af, skb, &iph, 0);
1219
1220         if (likely(cp))
1221                 return handle_response(af, skb, pd, cp, &iph, hooknum);
1222         if (sysctl_nat_icmp_send(net) &&
1223             (pp->protocol == IPPROTO_TCP ||
1224              pp->protocol == IPPROTO_UDP ||
1225              pp->protocol == IPPROTO_SCTP)) {
1226                 __be16 _ports[2], *pptr;
1227
1228                 pptr = frag_safe_skb_hp(skb, iph.len,
1229                                          sizeof(_ports), _ports, &iph);
1230                 if (pptr == NULL)
1231                         return NF_ACCEPT;       /* Not for me */
1232                 if (ip_vs_has_real_service(net, af, iph.protocol, &iph.saddr,
1233                                            pptr[0])) {
1234                         /*
1235                          * Notify the real server: there is no
1236                          * existing entry if it is not RST
1237                          * packet or not TCP packet.
1238                          */
1239                         if ((iph.protocol != IPPROTO_TCP &&
1240                              iph.protocol != IPPROTO_SCTP)
1241                              || ((iph.protocol == IPPROTO_TCP
1242                                   && !is_tcp_reset(skb, iph.len))
1243                                  || (iph.protocol == IPPROTO_SCTP
1244                                         && !is_sctp_abort(skb,
1245                                                 iph.len)))) {
1246 #ifdef CONFIG_IP_VS_IPV6
1247                                 if (af == AF_INET6) {
1248                                         if (!skb->dev)
1249                                                 skb->dev = net->loopback_dev;
1250                                         icmpv6_send(skb,
1251                                                     ICMPV6_DEST_UNREACH,
1252                                                     ICMPV6_PORT_UNREACH,
1253                                                     0);
1254                                 } else
1255 #endif
1256                                         icmp_send(skb,
1257                                                   ICMP_DEST_UNREACH,
1258                                                   ICMP_PORT_UNREACH, 0);
1259                                 return NF_DROP;
1260                         }
1261                 }
1262         }
1263         IP_VS_DBG_PKT(12, af, pp, skb, 0,
1264                       "ip_vs_out: packet continues traversal as normal");
1265         return NF_ACCEPT;
1266 }
1267
1268 /*
1269  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1270  *      used only for VS/NAT.
1271  *      Check if packet is reply for established ip_vs_conn.
1272  */
1273 static unsigned int
1274 ip_vs_reply4(const struct nf_hook_ops *ops, struct sk_buff *skb,
1275              const struct nf_hook_state *state)
1276 {
1277         return ip_vs_out(ops->hooknum, skb, AF_INET);
1278 }
1279
1280 /*
1281  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1282  *      Check if packet is reply for established ip_vs_conn.
1283  */
1284 static unsigned int
1285 ip_vs_local_reply4(const struct nf_hook_ops *ops, struct sk_buff *skb,
1286                    const struct nf_hook_state *state)
1287 {
1288         return ip_vs_out(ops->hooknum, skb, AF_INET);
1289 }
1290
1291 #ifdef CONFIG_IP_VS_IPV6
1292
1293 /*
1294  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1295  *      used only for VS/NAT.
1296  *      Check if packet is reply for established ip_vs_conn.
1297  */
1298 static unsigned int
1299 ip_vs_reply6(const struct nf_hook_ops *ops, struct sk_buff *skb,
1300              const struct nf_hook_state *state)
1301 {
1302         return ip_vs_out(ops->hooknum, skb, AF_INET6);
1303 }
1304
1305 /*
1306  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1307  *      Check if packet is reply for established ip_vs_conn.
1308  */
1309 static unsigned int
1310 ip_vs_local_reply6(const struct nf_hook_ops *ops, struct sk_buff *skb,
1311                    const struct nf_hook_state *state)
1312 {
1313         return ip_vs_out(ops->hooknum, skb, AF_INET6);
1314 }
1315
1316 #endif
1317
1318 /*
1319  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1320  *      Find any that might be relevant, check against existing connections,
1321  *      forward to the right destination host if relevant.
1322  *      Currently handles error types - unreachable, quench, ttl exceeded.
1323  */
1324 static int
1325 ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum)
1326 {
1327         struct net *net = NULL;
1328         struct iphdr *iph;
1329         struct icmphdr  _icmph, *ic;
1330         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1331         struct ip_vs_iphdr ciph;
1332         struct ip_vs_conn *cp;
1333         struct ip_vs_protocol *pp;
1334         struct ip_vs_proto_data *pd;
1335         unsigned int offset, offset2, ihl, verdict;
1336         bool ipip;
1337
1338         *related = 1;
1339
1340         /* reassemble IP fragments */
1341         if (ip_is_fragment(ip_hdr(skb))) {
1342                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
1343                         return NF_STOLEN;
1344         }
1345
1346         iph = ip_hdr(skb);
1347         offset = ihl = iph->ihl * 4;
1348         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1349         if (ic == NULL)
1350                 return NF_DROP;
1351
1352         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1353                   ic->type, ntohs(icmp_id(ic)),
1354                   &iph->saddr, &iph->daddr);
1355
1356         /*
1357          * Work through seeing if this is for us.
1358          * These checks are supposed to be in an order that means easy
1359          * things are checked first to speed up processing.... however
1360          * this means that some packets will manage to get a long way
1361          * down this stack and then be rejected, but that's life.
1362          */
1363         if ((ic->type != ICMP_DEST_UNREACH) &&
1364             (ic->type != ICMP_SOURCE_QUENCH) &&
1365             (ic->type != ICMP_TIME_EXCEEDED)) {
1366                 *related = 0;
1367                 return NF_ACCEPT;
1368         }
1369
1370         /* Now find the contained IP header */
1371         offset += sizeof(_icmph);
1372         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1373         if (cih == NULL)
1374                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1375
1376         net = skb_net(skb);
1377
1378         /* Special case for errors for IPIP packets */
1379         ipip = false;
1380         if (cih->protocol == IPPROTO_IPIP) {
1381                 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1382                         return NF_ACCEPT;
1383                 /* Error for our IPIP must arrive at LOCAL_IN */
1384                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1385                         return NF_ACCEPT;
1386                 offset += cih->ihl * 4;
1387                 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1388                 if (cih == NULL)
1389                         return NF_ACCEPT; /* The packet looks wrong, ignore */
1390                 ipip = true;
1391         }
1392
1393         pd = ip_vs_proto_data_get(net, cih->protocol);
1394         if (!pd)
1395                 return NF_ACCEPT;
1396         pp = pd->pp;
1397
1398         /* Is the embedded protocol header present? */
1399         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1400                      pp->dont_defrag))
1401                 return NF_ACCEPT;
1402
1403         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1404                       "Checking incoming ICMP for");
1405
1406         offset2 = offset;
1407         ip_vs_fill_ip4hdr(cih, &ciph);
1408         ciph.len += offset;
1409         offset = ciph.len;
1410         /* The embedded headers contain source and dest in reverse order.
1411          * For IPIP this is error for request, not for reply.
1412          */
1413         cp = pp->conn_in_get(AF_INET, skb, &ciph, ipip ? 0 : 1);
1414         if (!cp)
1415                 return NF_ACCEPT;
1416
1417         verdict = NF_DROP;
1418
1419         /* Ensure the checksum is correct */
1420         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1421                 /* Failed checksum! */
1422                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1423                           &iph->saddr);
1424                 goto out;
1425         }
1426
1427         if (ipip) {
1428                 __be32 info = ic->un.gateway;
1429                 __u8 type = ic->type;
1430                 __u8 code = ic->code;
1431
1432                 /* Update the MTU */
1433                 if (ic->type == ICMP_DEST_UNREACH &&
1434                     ic->code == ICMP_FRAG_NEEDED) {
1435                         struct ip_vs_dest *dest = cp->dest;
1436                         u32 mtu = ntohs(ic->un.frag.mtu);
1437                         __be16 frag_off = cih->frag_off;
1438
1439                         /* Strip outer IP and ICMP, go to IPIP header */
1440                         if (pskb_pull(skb, ihl + sizeof(_icmph)) == NULL)
1441                                 goto ignore_ipip;
1442                         offset2 -= ihl + sizeof(_icmph);
1443                         skb_reset_network_header(skb);
1444                         IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1445                                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1446                         ipv4_update_pmtu(skb, dev_net(skb->dev),
1447                                          mtu, 0, 0, 0, 0);
1448                         /* Client uses PMTUD? */
1449                         if (!(frag_off & htons(IP_DF)))
1450                                 goto ignore_ipip;
1451                         /* Prefer the resulting PMTU */
1452                         if (dest) {
1453                                 struct ip_vs_dest_dst *dest_dst;
1454
1455                                 rcu_read_lock();
1456                                 dest_dst = rcu_dereference(dest->dest_dst);
1457                                 if (dest_dst)
1458                                         mtu = dst_mtu(dest_dst->dst_cache);
1459                                 rcu_read_unlock();
1460                         }
1461                         if (mtu > 68 + sizeof(struct iphdr))
1462                                 mtu -= sizeof(struct iphdr);
1463                         info = htonl(mtu);
1464                 }
1465                 /* Strip outer IP, ICMP and IPIP, go to IP header of
1466                  * original request.
1467                  */
1468                 if (pskb_pull(skb, offset2) == NULL)
1469                         goto ignore_ipip;
1470                 skb_reset_network_header(skb);
1471                 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1472                         &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1473                         type, code, ntohl(info));
1474                 icmp_send(skb, type, code, info);
1475                 /* ICMP can be shorter but anyways, account it */
1476                 ip_vs_out_stats(cp, skb);
1477
1478 ignore_ipip:
1479                 consume_skb(skb);
1480                 verdict = NF_STOLEN;
1481                 goto out;
1482         }
1483
1484         /* do the statistics and put it back */
1485         ip_vs_in_stats(cp, skb);
1486         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol ||
1487             IPPROTO_SCTP == cih->protocol)
1488                 offset += 2 * sizeof(__u16);
1489         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1490
1491 out:
1492         __ip_vs_conn_put(cp);
1493
1494         return verdict;
1495 }
1496
1497 #ifdef CONFIG_IP_VS_IPV6
1498 static int ip_vs_in_icmp_v6(struct sk_buff *skb, int *related,
1499                             unsigned int hooknum, struct ip_vs_iphdr *iph)
1500 {
1501         struct net *net = NULL;
1502         struct ipv6hdr _ip6h, *ip6h;
1503         struct icmp6hdr _icmph, *ic;
1504         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1505         struct ip_vs_conn *cp;
1506         struct ip_vs_protocol *pp;
1507         struct ip_vs_proto_data *pd;
1508         unsigned int offs_ciph, writable, verdict;
1509
1510         *related = 1;
1511
1512         ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph, iph);
1513         if (ic == NULL)
1514                 return NF_DROP;
1515
1516         /*
1517          * Work through seeing if this is for us.
1518          * These checks are supposed to be in an order that means easy
1519          * things are checked first to speed up processing.... however
1520          * this means that some packets will manage to get a long way
1521          * down this stack and then be rejected, but that's life.
1522          */
1523         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1524                 *related = 0;
1525                 return NF_ACCEPT;
1526         }
1527         /* Fragment header that is before ICMP header tells us that:
1528          * it's not an error message since they can't be fragmented.
1529          */
1530         if (iph->flags & IP6_FH_F_FRAG)
1531                 return NF_DROP;
1532
1533         IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1534                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1535                   &iph->saddr, &iph->daddr);
1536
1537         /* Now find the contained IP header */
1538         ciph.len = iph->len + sizeof(_icmph);
1539         offs_ciph = ciph.len; /* Save ip header offset */
1540         ip6h = skb_header_pointer(skb, ciph.len, sizeof(_ip6h), &_ip6h);
1541         if (ip6h == NULL)
1542                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1543         ciph.saddr.in6 = ip6h->saddr; /* conn_in_get() handles reverse order */
1544         ciph.daddr.in6 = ip6h->daddr;
1545         /* skip possible IPv6 exthdrs of contained IPv6 packet */
1546         ciph.protocol = ipv6_find_hdr(skb, &ciph.len, -1, &ciph.fragoffs, NULL);
1547         if (ciph.protocol < 0)
1548                 return NF_ACCEPT; /* Contained IPv6 hdr looks wrong, ignore */
1549
1550         net = skb_net(skb);
1551         pd = ip_vs_proto_data_get(net, ciph.protocol);
1552         if (!pd)
1553                 return NF_ACCEPT;
1554         pp = pd->pp;
1555
1556         /* Cannot handle fragmented embedded protocol */
1557         if (ciph.fragoffs)
1558                 return NF_ACCEPT;
1559
1560         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offs_ciph,
1561                       "Checking incoming ICMPv6 for");
1562
1563         /* The embedded headers contain source and dest in reverse order
1564          * if not from localhost
1565          */
1566         cp = pp->conn_in_get(AF_INET6, skb, &ciph,
1567                              (hooknum == NF_INET_LOCAL_OUT) ? 0 : 1);
1568
1569         if (!cp)
1570                 return NF_ACCEPT;
1571         /* VS/TUN, VS/DR and LOCALNODE just let it go */
1572         if ((hooknum == NF_INET_LOCAL_OUT) &&
1573             (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1574                 __ip_vs_conn_put(cp);
1575                 return NF_ACCEPT;
1576         }
1577
1578         /* do the statistics and put it back */
1579         ip_vs_in_stats(cp, skb);
1580
1581         /* Need to mangle contained IPv6 header in ICMPv6 packet */
1582         writable = ciph.len;
1583         if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1584             IPPROTO_SCTP == ciph.protocol)
1585                 writable += 2 * sizeof(__u16); /* Also mangle ports */
1586
1587         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, writable, hooknum, &ciph);
1588
1589         __ip_vs_conn_put(cp);
1590
1591         return verdict;
1592 }
1593 #endif
1594
1595
1596 /*
1597  *      Check if it's for virtual services, look it up,
1598  *      and send it on its way...
1599  */
1600 static unsigned int
1601 ip_vs_in(unsigned int hooknum, struct sk_buff *skb, int af)
1602 {
1603         struct net *net;
1604         struct ip_vs_iphdr iph;
1605         struct ip_vs_protocol *pp;
1606         struct ip_vs_proto_data *pd;
1607         struct ip_vs_conn *cp;
1608         int ret, pkts;
1609         struct netns_ipvs *ipvs;
1610         int conn_reuse_mode;
1611
1612         /* Already marked as IPVS request or reply? */
1613         if (skb->ipvs_property)
1614                 return NF_ACCEPT;
1615
1616         /*
1617          *      Big tappo:
1618          *      - remote client: only PACKET_HOST
1619          *      - route: used for struct net when skb->dev is unset
1620          */
1621         if (unlikely((skb->pkt_type != PACKET_HOST &&
1622                       hooknum != NF_INET_LOCAL_OUT) ||
1623                      !skb_dst(skb))) {
1624                 ip_vs_fill_iph_skb(af, skb, &iph);
1625                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1626                               " ignored in hook %u\n",
1627                               skb->pkt_type, iph.protocol,
1628                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1629                 return NF_ACCEPT;
1630         }
1631         /* ipvs enabled in this netns ? */
1632         net = skb_net(skb);
1633         ipvs = net_ipvs(net);
1634         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1635                 return NF_ACCEPT;
1636
1637         ip_vs_fill_iph_skb(af, skb, &iph);
1638
1639         /* Bad... Do not break raw sockets */
1640         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1641                      af == AF_INET)) {
1642                 struct sock *sk = skb->sk;
1643                 struct inet_sock *inet = inet_sk(skb->sk);
1644
1645                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1646                         return NF_ACCEPT;
1647         }
1648
1649 #ifdef CONFIG_IP_VS_IPV6
1650         if (af == AF_INET6) {
1651                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1652                         int related;
1653                         int verdict = ip_vs_in_icmp_v6(skb, &related, hooknum,
1654                                                        &iph);
1655
1656                         if (related)
1657                                 return verdict;
1658                 }
1659         } else
1660 #endif
1661                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1662                         int related;
1663                         int verdict = ip_vs_in_icmp(skb, &related, hooknum);
1664
1665                         if (related)
1666                                 return verdict;
1667                 }
1668
1669         /* Protocol supported? */
1670         pd = ip_vs_proto_data_get(net, iph.protocol);
1671         if (unlikely(!pd))
1672                 return NF_ACCEPT;
1673         pp = pd->pp;
1674         /*
1675          * Check if the packet belongs to an existing connection entry
1676          */
1677         cp = pp->conn_in_get(af, skb, &iph, 0);
1678
1679         conn_reuse_mode = sysctl_conn_reuse_mode(ipvs);
1680         if (conn_reuse_mode && !iph.fragoffs &&
1681             is_new_conn(skb, &iph) && cp &&
1682             ((unlikely(sysctl_expire_nodest_conn(ipvs)) && cp->dest &&
1683               unlikely(!atomic_read(&cp->dest->weight))) ||
1684              unlikely(is_new_conn_expected(cp, conn_reuse_mode)))) {
1685                 if (!atomic_read(&cp->n_control))
1686                         ip_vs_conn_expire_now(cp);
1687                 __ip_vs_conn_put(cp);
1688                 cp = NULL;
1689         }
1690
1691         if (unlikely(!cp) && !iph.fragoffs) {
1692                 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1693                  * replayed fragment zero will already have created the cp
1694                  */
1695                 int v;
1696
1697                 /* Schedule and create new connection entry into &cp */
1698                 if (!pp->conn_schedule(af, skb, pd, &v, &cp, &iph))
1699                         return v;
1700         }
1701
1702         if (unlikely(!cp)) {
1703                 /* sorry, all this trouble for a no-hit :) */
1704                 IP_VS_DBG_PKT(12, af, pp, skb, 0,
1705                               "ip_vs_in: packet continues traversal as normal");
1706                 if (iph.fragoffs) {
1707                         /* Fragment that couldn't be mapped to a conn entry
1708                          * is missing module nf_defrag_ipv6
1709                          */
1710                         IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1711                         IP_VS_DBG_PKT(7, af, pp, skb, 0, "unhandled fragment");
1712                 }
1713                 return NF_ACCEPT;
1714         }
1715
1716         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Incoming packet");
1717         /* Check the server status */
1718         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1719                 /* the destination server is not available */
1720
1721                 if (sysctl_expire_nodest_conn(ipvs)) {
1722                         /* try to expire the connection immediately */
1723                         ip_vs_conn_expire_now(cp);
1724                 }
1725                 /* don't restart its timer, and silently
1726                    drop the packet. */
1727                 __ip_vs_conn_put(cp);
1728                 return NF_DROP;
1729         }
1730
1731         ip_vs_in_stats(cp, skb);
1732         ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1733         if (cp->packet_xmit)
1734                 ret = cp->packet_xmit(skb, cp, pp, &iph);
1735                 /* do not touch skb anymore */
1736         else {
1737                 IP_VS_DBG_RL("warning: packet_xmit is null");
1738                 ret = NF_ACCEPT;
1739         }
1740
1741         /* Increase its packet counter and check if it is needed
1742          * to be synchronized
1743          *
1744          * Sync connection if it is about to close to
1745          * encorage the standby servers to update the connections timeout
1746          *
1747          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1748          */
1749
1750         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
1751                 pkts = sysctl_sync_threshold(ipvs);
1752         else
1753                 pkts = atomic_add_return(1, &cp->in_pkts);
1754
1755         if (ipvs->sync_state & IP_VS_STATE_MASTER)
1756                 ip_vs_sync_conn(net, cp, pkts);
1757
1758         ip_vs_conn_put(cp);
1759         return ret;
1760 }
1761
1762 /*
1763  *      AF_INET handler in NF_INET_LOCAL_IN chain
1764  *      Schedule and forward packets from remote clients
1765  */
1766 static unsigned int
1767 ip_vs_remote_request4(const struct nf_hook_ops *ops, struct sk_buff *skb,
1768                       const struct nf_hook_state *state)
1769 {
1770         return ip_vs_in(ops->hooknum, skb, AF_INET);
1771 }
1772
1773 /*
1774  *      AF_INET handler in NF_INET_LOCAL_OUT chain
1775  *      Schedule and forward packets from local clients
1776  */
1777 static unsigned int
1778 ip_vs_local_request4(const struct nf_hook_ops *ops, struct sk_buff *skb,
1779                      const struct nf_hook_state *state)
1780 {
1781         return ip_vs_in(ops->hooknum, skb, AF_INET);
1782 }
1783
1784 #ifdef CONFIG_IP_VS_IPV6
1785
1786 /*
1787  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
1788  *      Schedule and forward packets from remote clients
1789  */
1790 static unsigned int
1791 ip_vs_remote_request6(const struct nf_hook_ops *ops, struct sk_buff *skb,
1792                       const struct nf_hook_state *state)
1793 {
1794         return ip_vs_in(ops->hooknum, skb, AF_INET6);
1795 }
1796
1797 /*
1798  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
1799  *      Schedule and forward packets from local clients
1800  */
1801 static unsigned int
1802 ip_vs_local_request6(const struct nf_hook_ops *ops, struct sk_buff *skb,
1803                      const struct nf_hook_state *state)
1804 {
1805         return ip_vs_in(ops->hooknum, skb, AF_INET6);
1806 }
1807
1808 #endif
1809
1810
1811 /*
1812  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1813  *      related packets destined for 0.0.0.0/0.
1814  *      When fwmark-based virtual service is used, such as transparent
1815  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
1816  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1817  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1818  *      and send them to ip_vs_in_icmp.
1819  */
1820 static unsigned int
1821 ip_vs_forward_icmp(const struct nf_hook_ops *ops, struct sk_buff *skb,
1822                    const struct nf_hook_state *state)
1823 {
1824         int r;
1825         struct net *net;
1826         struct netns_ipvs *ipvs;
1827
1828         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1829                 return NF_ACCEPT;
1830
1831         /* ipvs enabled in this netns ? */
1832         net = skb_net(skb);
1833         ipvs = net_ipvs(net);
1834         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1835                 return NF_ACCEPT;
1836
1837         return ip_vs_in_icmp(skb, &r, ops->hooknum);
1838 }
1839
1840 #ifdef CONFIG_IP_VS_IPV6
1841 static unsigned int
1842 ip_vs_forward_icmp_v6(const struct nf_hook_ops *ops, struct sk_buff *skb,
1843                       const struct nf_hook_state *state)
1844 {
1845         int r;
1846         struct net *net;
1847         struct netns_ipvs *ipvs;
1848         struct ip_vs_iphdr iphdr;
1849
1850         ip_vs_fill_iph_skb(AF_INET6, skb, &iphdr);
1851         if (iphdr.protocol != IPPROTO_ICMPV6)
1852                 return NF_ACCEPT;
1853
1854         /* ipvs enabled in this netns ? */
1855         net = skb_net(skb);
1856         ipvs = net_ipvs(net);
1857         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1858                 return NF_ACCEPT;
1859
1860         return ip_vs_in_icmp_v6(skb, &r, ops->hooknum, &iphdr);
1861 }
1862 #endif
1863
1864
1865 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
1866         /* After packet filtering, change source only for VS/NAT */
1867         {
1868                 .hook           = ip_vs_reply4,
1869                 .owner          = THIS_MODULE,
1870                 .pf             = NFPROTO_IPV4,
1871                 .hooknum        = NF_INET_LOCAL_IN,
1872                 .priority       = NF_IP_PRI_NAT_SRC - 2,
1873         },
1874         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1875          * or VS/NAT(change destination), so that filtering rules can be
1876          * applied to IPVS. */
1877         {
1878                 .hook           = ip_vs_remote_request4,
1879                 .owner          = THIS_MODULE,
1880                 .pf             = NFPROTO_IPV4,
1881                 .hooknum        = NF_INET_LOCAL_IN,
1882                 .priority       = NF_IP_PRI_NAT_SRC - 1,
1883         },
1884         /* Before ip_vs_in, change source only for VS/NAT */
1885         {
1886                 .hook           = ip_vs_local_reply4,
1887                 .owner          = THIS_MODULE,
1888                 .pf             = NFPROTO_IPV4,
1889                 .hooknum        = NF_INET_LOCAL_OUT,
1890                 .priority       = NF_IP_PRI_NAT_DST + 1,
1891         },
1892         /* After mangle, schedule and forward local requests */
1893         {
1894                 .hook           = ip_vs_local_request4,
1895                 .owner          = THIS_MODULE,
1896                 .pf             = NFPROTO_IPV4,
1897                 .hooknum        = NF_INET_LOCAL_OUT,
1898                 .priority       = NF_IP_PRI_NAT_DST + 2,
1899         },
1900         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1901          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1902         {
1903                 .hook           = ip_vs_forward_icmp,
1904                 .owner          = THIS_MODULE,
1905                 .pf             = NFPROTO_IPV4,
1906                 .hooknum        = NF_INET_FORWARD,
1907                 .priority       = 99,
1908         },
1909         /* After packet filtering, change source only for VS/NAT */
1910         {
1911                 .hook           = ip_vs_reply4,
1912                 .owner          = THIS_MODULE,
1913                 .pf             = NFPROTO_IPV4,
1914                 .hooknum        = NF_INET_FORWARD,
1915                 .priority       = 100,
1916         },
1917 #ifdef CONFIG_IP_VS_IPV6
1918         /* After packet filtering, change source only for VS/NAT */
1919         {
1920                 .hook           = ip_vs_reply6,
1921                 .owner          = THIS_MODULE,
1922                 .pf             = NFPROTO_IPV6,
1923                 .hooknum        = NF_INET_LOCAL_IN,
1924                 .priority       = NF_IP6_PRI_NAT_SRC - 2,
1925         },
1926         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1927          * or VS/NAT(change destination), so that filtering rules can be
1928          * applied to IPVS. */
1929         {
1930                 .hook           = ip_vs_remote_request6,
1931                 .owner          = THIS_MODULE,
1932                 .pf             = NFPROTO_IPV6,
1933                 .hooknum        = NF_INET_LOCAL_IN,
1934                 .priority       = NF_IP6_PRI_NAT_SRC - 1,
1935         },
1936         /* Before ip_vs_in, change source only for VS/NAT */
1937         {
1938                 .hook           = ip_vs_local_reply6,
1939                 .owner          = THIS_MODULE,
1940                 .pf             = NFPROTO_IPV6,
1941                 .hooknum        = NF_INET_LOCAL_OUT,
1942                 .priority       = NF_IP6_PRI_NAT_DST + 1,
1943         },
1944         /* After mangle, schedule and forward local requests */
1945         {
1946                 .hook           = ip_vs_local_request6,
1947                 .owner          = THIS_MODULE,
1948                 .pf             = NFPROTO_IPV6,
1949                 .hooknum        = NF_INET_LOCAL_OUT,
1950                 .priority       = NF_IP6_PRI_NAT_DST + 2,
1951         },
1952         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1953          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1954         {
1955                 .hook           = ip_vs_forward_icmp_v6,
1956                 .owner          = THIS_MODULE,
1957                 .pf             = NFPROTO_IPV6,
1958                 .hooknum        = NF_INET_FORWARD,
1959                 .priority       = 99,
1960         },
1961         /* After packet filtering, change source only for VS/NAT */
1962         {
1963                 .hook           = ip_vs_reply6,
1964                 .owner          = THIS_MODULE,
1965                 .pf             = NFPROTO_IPV6,
1966                 .hooknum        = NF_INET_FORWARD,
1967                 .priority       = 100,
1968         },
1969 #endif
1970 };
1971 /*
1972  *      Initialize IP Virtual Server netns mem.
1973  */
1974 static int __net_init __ip_vs_init(struct net *net)
1975 {
1976         struct netns_ipvs *ipvs;
1977
1978         ipvs = net_generic(net, ip_vs_net_id);
1979         if (ipvs == NULL)
1980                 return -ENOMEM;
1981
1982         /* Hold the beast until a service is registerd */
1983         ipvs->enable = 0;
1984         ipvs->net = net;
1985         /* Counters used for creating unique names */
1986         ipvs->gen = atomic_read(&ipvs_netns_cnt);
1987         atomic_inc(&ipvs_netns_cnt);
1988         net->ipvs = ipvs;
1989
1990         if (ip_vs_estimator_net_init(net) < 0)
1991                 goto estimator_fail;
1992
1993         if (ip_vs_control_net_init(net) < 0)
1994                 goto control_fail;
1995
1996         if (ip_vs_protocol_net_init(net) < 0)
1997                 goto protocol_fail;
1998
1999         if (ip_vs_app_net_init(net) < 0)
2000                 goto app_fail;
2001
2002         if (ip_vs_conn_net_init(net) < 0)
2003                 goto conn_fail;
2004
2005         if (ip_vs_sync_net_init(net) < 0)
2006                 goto sync_fail;
2007
2008         printk(KERN_INFO "IPVS: Creating netns size=%zu id=%d\n",
2009                          sizeof(struct netns_ipvs), ipvs->gen);
2010         return 0;
2011 /*
2012  * Error handling
2013  */
2014
2015 sync_fail:
2016         ip_vs_conn_net_cleanup(net);
2017 conn_fail:
2018         ip_vs_app_net_cleanup(net);
2019 app_fail:
2020         ip_vs_protocol_net_cleanup(net);
2021 protocol_fail:
2022         ip_vs_control_net_cleanup(net);
2023 control_fail:
2024         ip_vs_estimator_net_cleanup(net);
2025 estimator_fail:
2026         net->ipvs = NULL;
2027         return -ENOMEM;
2028 }
2029
2030 static void __net_exit __ip_vs_cleanup(struct net *net)
2031 {
2032         ip_vs_service_net_cleanup(net); /* ip_vs_flush() with locks */
2033         ip_vs_conn_net_cleanup(net);
2034         ip_vs_app_net_cleanup(net);
2035         ip_vs_protocol_net_cleanup(net);
2036         ip_vs_control_net_cleanup(net);
2037         ip_vs_estimator_net_cleanup(net);
2038         IP_VS_DBG(2, "ipvs netns %d released\n", net_ipvs(net)->gen);
2039         net->ipvs = NULL;
2040 }
2041
2042 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
2043 {
2044         EnterFunction(2);
2045         net_ipvs(net)->enable = 0;      /* Disable packet reception */
2046         smp_wmb();
2047         ip_vs_sync_net_cleanup(net);
2048         LeaveFunction(2);
2049 }
2050
2051 static struct pernet_operations ipvs_core_ops = {
2052         .init = __ip_vs_init,
2053         .exit = __ip_vs_cleanup,
2054         .id   = &ip_vs_net_id,
2055         .size = sizeof(struct netns_ipvs),
2056 };
2057
2058 static struct pernet_operations ipvs_core_dev_ops = {
2059         .exit = __ip_vs_dev_cleanup,
2060 };
2061
2062 /*
2063  *      Initialize IP Virtual Server
2064  */
2065 static int __init ip_vs_init(void)
2066 {
2067         int ret;
2068
2069         ret = ip_vs_control_init();
2070         if (ret < 0) {
2071                 pr_err("can't setup control.\n");
2072                 goto exit;
2073         }
2074
2075         ip_vs_protocol_init();
2076
2077         ret = ip_vs_conn_init();
2078         if (ret < 0) {
2079                 pr_err("can't setup connection table.\n");
2080                 goto cleanup_protocol;
2081         }
2082
2083         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
2084         if (ret < 0)
2085                 goto cleanup_conn;
2086
2087         ret = register_pernet_device(&ipvs_core_dev_ops);
2088         if (ret < 0)
2089                 goto cleanup_sub;
2090
2091         ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2092         if (ret < 0) {
2093                 pr_err("can't register hooks.\n");
2094                 goto cleanup_dev;
2095         }
2096
2097         ret = ip_vs_register_nl_ioctl();
2098         if (ret < 0) {
2099                 pr_err("can't register netlink/ioctl.\n");
2100                 goto cleanup_hooks;
2101         }
2102
2103         pr_info("ipvs loaded.\n");
2104
2105         return ret;
2106
2107 cleanup_hooks:
2108         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2109 cleanup_dev:
2110         unregister_pernet_device(&ipvs_core_dev_ops);
2111 cleanup_sub:
2112         unregister_pernet_subsys(&ipvs_core_ops);
2113 cleanup_conn:
2114         ip_vs_conn_cleanup();
2115 cleanup_protocol:
2116         ip_vs_protocol_cleanup();
2117         ip_vs_control_cleanup();
2118 exit:
2119         return ret;
2120 }
2121
2122 static void __exit ip_vs_cleanup(void)
2123 {
2124         ip_vs_unregister_nl_ioctl();
2125         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2126         unregister_pernet_device(&ipvs_core_dev_ops);
2127         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
2128         ip_vs_conn_cleanup();
2129         ip_vs_protocol_cleanup();
2130         ip_vs_control_cleanup();
2131         pr_info("ipvs unloaded.\n");
2132 }
2133
2134 module_init(ip_vs_init);
2135 module_exit(ip_vs_cleanup);
2136 MODULE_LICENSE("GPL");