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 = svc->ipvs;
 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->ipvs, 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         const union nf_inet_addr *src_addr, *dst_addr;
 249
 250         if (likely(!ip_vs_iph_inverse(iph))) {
 251                 src_addr = &iph->saddr;
 252                 dst_addr = &iph->daddr;
 253         } else {
 254                 src_addr = &iph->daddr;
 255                 dst_addr = &iph->saddr;
 256         }
 257
 258
 259         /* Mask saddr with the netmask to adjust template granularity */
 260 #ifdef CONFIG_IP_VS_IPV6
 261         if (svc->af == AF_INET6)
 262                 ipv6_addr_prefix(&snet.in6, &src_addr->in6,
 263                                  (__force __u32) svc->netmask);
 264         else
 265 #endif
 266                 snet.ip = src_addr->ip & svc->netmask;
 267
 268         IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
 269                       "mnet %s\n",
 270                       IP_VS_DBG_ADDR(svc->af, src_addr), ntohs(src_port),
 271                       IP_VS_DBG_ADDR(svc->af, dst_addr), ntohs(dst_port),
 272                       IP_VS_DBG_ADDR(svc->af, &snet));
 273
 274         /*
 275          * As far as we know, FTP is a very complicated network protocol, and
 276          * it uses control connection and data connections. For active FTP,
 277          * FTP server initialize data connection to the client, its source port
 278          * is often 20. For passive FTP, FTP server tells the clients the port
 279          * that it passively listens to,  and the client issues the data
 280          * connection. In the tunneling or direct routing mode, the load
 281          * balancer is on the client-to-server half of connection, the port
 282          * number is unknown to the load balancer. So, a conn template like
 283          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
 284          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
 285          * is created for other persistent services.
 286          */
 287         {
 288                 int protocol = iph->protocol;
 289                 const union nf_inet_addr *vaddr = dst_addr;
 290                 __be16 vport = 0;
 291
 292                 if (dst_port == svc->port) {
 293                         /* non-FTP template:
 294                          * <protocol, caddr, 0, vaddr, vport, daddr, dport>
 295                          * FTP template:
 296                          * <protocol, caddr, 0, vaddr, 0, daddr, 0>
 297                          */
 298                         if (svc->port != FTPPORT)
 299                                 vport = dst_port;
 300                 } else {
 301                         /* Note: persistent fwmark-based services and
 302                          * persistent port zero service are handled here.
 303                          * fwmark template:
 304                          * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
 305                          * port zero template:
 306                          * <protocol,caddr,0,vaddr,0,daddr,0>
 307                          */
 308                         if (svc->fwmark) {
 309                                 protocol = IPPROTO_IP;
 310                                 vaddr = &fwmark;
 311                         }
 312                 }
 313                 /* return *ignored = -1 so NF_DROP can be used */
 314                 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
 315                                                   vaddr, vport, &param) < 0) {
 316                         *ignored = -1;
 317                         return NULL;
 318                 }
 319         }
 320
 321         /* Check if a template already exists */
 322         ct = ip_vs_ct_in_get(&param);
 323         if (!ct || !ip_vs_check_template(ct)) {
 324                 struct ip_vs_scheduler *sched;
 325
 326                 /*
 327                  * No template found or the dest of the connection
 328                  * template is not available.
 329                  * return *ignored=0 i.e. ICMP and NF_DROP
 330                  */
 331                 sched = rcu_dereference(svc->scheduler);
 332                 if (sched) {
 333                         /* read svc->sched_data after svc->scheduler */
 334                         smp_rmb();
 335                         dest = sched->schedule(svc, skb, iph);
 336                 } else {
 337                         dest = NULL;
 338                 }
 339                 if (!dest) {
 340                         IP_VS_DBG(1, "p-schedule: no dest found.\n");
 341                         kfree(param.pe_data);
 342                         *ignored = 0;
 343                         return NULL;
 344                 }
 345
 346                 if (dst_port == svc->port && svc->port != FTPPORT)
 347                         dport = dest->port;
 348
 349                 /* Create a template
 350                  * This adds param.pe_data to the template,
 351                  * and thus param.pe_data will be destroyed
 352                  * when the template expires */
 353                 ct = ip_vs_conn_new(&param, dest->af, &dest->addr, dport,
 354                                     IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
 355                 if (ct == NULL) {
 356                         kfree(param.pe_data);
 357                         *ignored = -1;
 358                         return NULL;
 359                 }
 360
 361                 ct->timeout = svc->timeout;
 362         } else {
 363                 /* set destination with the found template */
 364                 dest = ct->dest;
 365                 kfree(param.pe_data);
 366         }
 367
 368         dport = dst_port;
 369         if (dport == svc->port && dest->port)
 370                 dport = dest->port;
 371
 372         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 373                  && iph->protocol == IPPROTO_UDP) ?
 374                 IP_VS_CONN_F_ONE_PACKET : 0;
 375
 376         /*
 377          *    Create a new connection according to the template
 378          */
 379         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol, src_addr,
 380                               src_port, dst_addr, dst_port, &param);
 381
 382         cp = ip_vs_conn_new(&param, dest->af, &dest->addr, dport, flags, dest,
 383                             skb->mark);
 384         if (cp == NULL) {
 385                 ip_vs_conn_put(ct);
 386                 *ignored = -1;
 387                 return NULL;
 388         }
 389
 390         /*
 391          *    Add its control
 392          */
 393         ip_vs_control_add(cp, ct);
 394         ip_vs_conn_put(ct);
 395
 396         ip_vs_conn_stats(cp, svc);
 397         return cp;
 398 }
 399
 400
 401 /*
 402  *  IPVS main scheduling function
 403  *  It selects a server according to the virtual service, and
 404  *  creates a connection entry.
 405  *  Protocols supported: TCP, UDP
 406  *
 407  *  Usage of *ignored
 408  *
 409  * 1 :   protocol tried to schedule (eg. on SYN), found svc but the
 410  *       svc/scheduler decides that this packet should be accepted with
 411  *       NF_ACCEPT because it must not be scheduled.
 412  *
 413  * 0 :   scheduler can not find destination, so try bypass or
 414  *       return ICMP and then NF_DROP (ip_vs_leave).
 415  *
 416  * -1 :  scheduler tried to schedule but fatal error occurred, eg.
 417  *       ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
 418  *       failure such as missing Call-ID, ENOMEM on skb_linearize
 419  *       or pe_data. In this case we should return NF_DROP without
 420  *       any attempts to send ICMP with ip_vs_leave.
 421  */
 422 struct ip_vs_conn *
 423 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
 424                struct ip_vs_proto_data *pd, int *ignored,
 425                struct ip_vs_iphdr *iph)
 426 {
 427         struct ip_vs_protocol *pp = pd->pp;
 428         struct ip_vs_conn *cp = NULL;
 429         struct ip_vs_scheduler *sched;
 430         struct ip_vs_dest *dest;
 431         __be16 _ports[2], *pptr, cport, vport;
 432         const void *caddr, *vaddr;
 433         unsigned int flags;
 434
 435         *ignored = 1;
 436         /*
 437          * IPv6 frags, only the first hit here.
 438          */
 439         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 440         if (pptr == NULL)
 441                 return NULL;
 442
 443         if (likely(!ip_vs_iph_inverse(iph))) {
 444                 cport = pptr[0];
 445                 caddr = &iph->saddr;
 446                 vport = pptr[1];
 447                 vaddr = &iph->daddr;
 448         } else {
 449                 cport = pptr[1];
 450                 caddr = &iph->daddr;
 451                 vport = pptr[0];
 452                 vaddr = &iph->saddr;
 453         }
 454
 455         /*
 456          * FTPDATA needs this check when using local real server.
 457          * Never schedule Active FTPDATA connections from real server.
 458          * For LVS-NAT they must be already created. For other methods
 459          * with persistence the connection is created on SYN+ACK.
 460          */
 461         if (cport == FTPDATA) {
 462                 IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
 463                               "Not scheduling FTPDATA");
 464                 return NULL;
 465         }
 466
 467         /*
 468          *    Do not schedule replies from local real server.
 469          */
 470         if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK)) {
 471                 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
 472                 cp = pp->conn_in_get(svc->af, skb, iph);
 473                 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
 474
 475                 if (cp) {
 476                         IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
 477                                       "Not scheduling reply for existing"
 478                                       " connection");
 479                         __ip_vs_conn_put(cp);
 480                         return NULL;
 481                 }
 482         }
 483
 484         /*
 485          *    Persistent service
 486          */
 487         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 488                 return ip_vs_sched_persist(svc, skb, cport, vport, ignored,
 489                                            iph);
 490
 491         *ignored = 0;
 492
 493         /*
 494          *    Non-persistent service
 495          */
 496         if (!svc->fwmark && vport != svc->port) {
 497                 if (!svc->port)
 498                         pr_err("Schedule: port zero only supported "
 499                                "in persistent services, "
 500                                "check your ipvs configuration\n");
 501                 return NULL;
 502         }
 503
 504         sched = rcu_dereference(svc->scheduler);
 505         if (sched) {
 506                 /* read svc->sched_data after svc->scheduler */
 507                 smp_rmb();
 508                 dest = sched->schedule(svc, skb, iph);
 509         } else {
 510                 dest = NULL;
 511         }
 512         if (dest == NULL) {
 513                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 514                 return NULL;
 515         }
 516
 517         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 518                  && iph->protocol == IPPROTO_UDP) ?
 519                 IP_VS_CONN_F_ONE_PACKET : 0;
 520
 521         /*
 522          *    Create a connection entry.
 523          */
 524         {
 525                 struct ip_vs_conn_param p;
 526
 527                 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
 528                                       caddr, cport, vaddr, vport, &p);
 529                 cp = ip_vs_conn_new(&p, dest->af, &dest->addr,
 530                                     dest->port ? dest->port : vport,
 531                                     flags, dest, skb->mark);
 532                 if (!cp) {
 533                         *ignored = -1;
 534                         return NULL;
 535                 }
 536         }
 537
 538         IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
 539                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
 540                       ip_vs_fwd_tag(cp),
 541                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 542                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 543                       IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
 544                       cp->flags, atomic_read(&cp->refcnt));
 545
 546         ip_vs_conn_stats(cp, svc);
 547         return cp;
 548 }
 549
 550 #ifdef CONFIG_SYSCTL
 551 static inline int ip_vs_addr_is_unicast(struct net *net, int af,
 552                                         union nf_inet_addr *addr)
 553 {
 554 #ifdef CONFIG_IP_VS_IPV6
 555         if (af == AF_INET6)
 556                 return ipv6_addr_type(&addr->in6) & IPV6_ADDR_UNICAST;
 557 #endif
 558         return (inet_addr_type(net, addr->ip) == RTN_UNICAST);
 559 }
 560 #endif
 561
 562 /*
 563  *  Pass or drop the packet.
 564  *  Called by ip_vs_in, when the virtual service is available but
 565  *  no destination is available for a new connection.
 566  */
 567 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 568                 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
 569 {
 570         __be16 _ports[2], *pptr, dport;
 571 #ifdef CONFIG_SYSCTL
 572         struct net *net;
 573         struct netns_ipvs *ipvs;
 574 #endif
 575
 576         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 577         if (!pptr)
 578                 return NF_DROP;
 579         dport = likely(!ip_vs_iph_inverse(iph)) ? pptr[1] : pptr[0];
 580
 581 #ifdef CONFIG_SYSCTL
 582         net = skb_net(skb);
 583
 584
 585         /* if it is fwmark-based service, the cache_bypass sysctl is up
 586            and the destination is a non-local unicast, then create
 587            a cache_bypass connection entry */
 588         ipvs = net_ipvs(net);
 589         if (ipvs->sysctl_cache_bypass && svc->fwmark &&
 590             !(iph->hdr_flags & (IP_VS_HDR_INVERSE | IP_VS_HDR_ICMP)) &&
 591             ip_vs_addr_is_unicast(net, svc->af, &iph->daddr)) {
 592                 int ret;
 593                 struct ip_vs_conn *cp;
 594                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
 595                                       iph->protocol == IPPROTO_UDP) ?
 596                                       IP_VS_CONN_F_ONE_PACKET : 0;
 597                 union nf_inet_addr daddr =  { .all = { 0, 0, 0, 0 } };
 598
 599                 /* create a new connection entry */
 600                 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
 601                 {
 602                         struct ip_vs_conn_param p;
 603                         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
 604                                               &iph->saddr, pptr[0],
 605                                               &iph->daddr, pptr[1], &p);
 606                         cp = ip_vs_conn_new(&p, svc->af, &daddr, 0,
 607                                             IP_VS_CONN_F_BYPASS | flags,
 608                                             NULL, skb->mark);
 609                         if (!cp)
 610                                 return NF_DROP;
 611                 }
 612
 613                 /* statistics */
 614                 ip_vs_in_stats(cp, skb);
 615
 616                 /* set state */
 617                 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
 618
 619                 /* transmit the first SYN packet */
 620                 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
 621                 /* do not touch skb anymore */
 622
 623                 atomic_inc(&cp->in_pkts);
 624                 ip_vs_conn_put(cp);
 625                 return ret;
 626         }
 627 #endif
 628
 629         /*
 630          * When the virtual ftp service is presented, packets destined
 631          * for other services on the VIP may get here (except services
 632          * listed in the ipvs table), pass the packets, because it is
 633          * not ipvs job to decide to drop the packets.
 634          */
 635         if (svc->port == FTPPORT && dport != FTPPORT)
 636                 return NF_ACCEPT;
 637
 638         if (unlikely(ip_vs_iph_icmp(iph)))
 639                 return NF_DROP;
 640
 641         /*
 642          * Notify the client that the destination is unreachable, and
 643          * release the socket buffer.
 644          * Since it is in IP layer, the TCP socket is not actually
 645          * created, the TCP RST packet cannot be sent, instead that
 646          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 647          */
 648 #ifdef CONFIG_IP_VS_IPV6
 649         if (svc->af == AF_INET6) {
 650                 if (!skb->dev) {
 651                         struct net *net_ = dev_net(skb_dst(skb)->dev);
 652
 653                         skb->dev = net_->loopback_dev;
 654                 }
 655                 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 656         } else
 657 #endif
 658                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 659
 660         return NF_DROP;
 661 }
 662
 663 #ifdef CONFIG_SYSCTL
 664
 665 static int sysctl_snat_reroute(struct sk_buff *skb)
 666 {
 667         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 668         return ipvs->sysctl_snat_reroute;
 669 }
 670
 671 static int sysctl_nat_icmp_send(struct net *net)
 672 {
 673         struct netns_ipvs *ipvs = net_ipvs(net);
 674         return ipvs->sysctl_nat_icmp_send;
 675 }
 676
 677 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
 678 {
 679         return ipvs->sysctl_expire_nodest_conn;
 680 }
 681
 682 #else
 683
 684 static int sysctl_snat_reroute(struct sk_buff *skb) { return 0; }
 685 static int sysctl_nat_icmp_send(struct net *net) { return 0; }
 686 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
 687
 688 #endif
 689
 690 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 691 {
 692         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 693 }
 694
 695 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
 696 {
 697         if (NF_INET_LOCAL_IN == hooknum)
 698                 return IP_DEFRAG_VS_IN;
 699         if (NF_INET_FORWARD == hooknum)
 700                 return IP_DEFRAG_VS_FWD;
 701         return IP_DEFRAG_VS_OUT;
 702 }
 703
 704 static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user)
 705 {
 706         int err;
 707
 708         local_bh_disable();
 709         err = ip_defrag(skb, user);
 710         local_bh_enable();
 711         if (!err)
 712                 ip_send_check(ip_hdr(skb));
 713
 714         return err;
 715 }
 716
 717 static int ip_vs_route_me_harder(int af, struct sk_buff *skb,
 718                                  unsigned int hooknum)
 719 {
 720         if (!sysctl_snat_reroute(skb))
 721                 return 0;
 722         /* Reroute replies only to remote clients (FORWARD and LOCAL_OUT) */
 723         if (NF_INET_LOCAL_IN == hooknum)
 724                 return 0;
 725 #ifdef CONFIG_IP_VS_IPV6
 726         if (af == AF_INET6) {
 727                 struct dst_entry *dst = skb_dst(skb);
 728
 729                 if (dst->dev && !(dst->dev->flags & IFF_LOOPBACK) &&
 730                     ip6_route_me_harder(skb) != 0)
 731                         return 1;
 732         } else
 733 #endif
 734                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
 735                     ip_route_me_harder(skb, RTN_LOCAL) != 0)
 736                         return 1;
 737
 738         return 0;
 739 }
 740
 741 /*
 742  * Packet has been made sufficiently writable in caller
 743  * - inout: 1=in->out, 0=out->in
 744  */
 745 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 746                     struct ip_vs_conn *cp, int inout)
 747 {
 748         struct iphdr *iph        = ip_hdr(skb);
 749         unsigned int icmp_offset = iph->ihl*4;
 750         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
 751                                                       icmp_offset);
 752         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 753
 754         if (inout) {
 755                 iph->saddr = cp->vaddr.ip;
 756                 ip_send_check(iph);
 757                 ciph->daddr = cp->vaddr.ip;
 758                 ip_send_check(ciph);
 759         } else {
 760                 iph->daddr = cp->daddr.ip;
 761                 ip_send_check(iph);
 762                 ciph->saddr = cp->daddr.ip;
 763                 ip_send_check(ciph);
 764         }
 765
 766         /* the TCP/UDP/SCTP port */
 767         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
 768             IPPROTO_SCTP == ciph->protocol) {
 769                 __be16 *ports = (void *)ciph + ciph->ihl*4;
 770
 771                 if (inout)
 772                         ports[1] = cp->vport;
 773                 else
 774                         ports[0] = cp->dport;
 775         }
 776
 777         /* And finally the ICMP checksum */
 778         icmph->checksum = 0;
 779         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 780         skb->ip_summed = CHECKSUM_UNNECESSARY;
 781
 782         if (inout)
 783                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 784                         "Forwarding altered outgoing ICMP");
 785         else
 786                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 787                         "Forwarding altered incoming ICMP");
 788 }
 789
 790 #ifdef CONFIG_IP_VS_IPV6
 791 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
 792                     struct ip_vs_conn *cp, int inout)
 793 {
 794         struct ipv6hdr *iph      = ipv6_hdr(skb);
 795         unsigned int icmp_offset = 0;
 796         unsigned int offs        = 0; /* header offset*/
 797         int protocol;
 798         struct icmp6hdr *icmph;
 799         struct ipv6hdr *ciph;
 800         unsigned short fragoffs;
 801
 802         ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
 803         icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
 804         offs = icmp_offset + sizeof(struct icmp6hdr);
 805         ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
 806
 807         protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
 808
 809         if (inout) {
 810                 iph->saddr = cp->vaddr.in6;
 811                 ciph->daddr = cp->vaddr.in6;
 812         } else {
 813                 iph->daddr = cp->daddr.in6;
 814                 ciph->saddr = cp->daddr.in6;
 815         }
 816
 817         /* the TCP/UDP/SCTP port */
 818         if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 819                           IPPROTO_SCTP == protocol)) {
 820                 __be16 *ports = (void *)(skb_network_header(skb) + offs);
 821
 822                 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
 823                               ntohs(inout ? ports[1] : ports[0]),
 824                               ntohs(inout ? cp->vport : cp->dport));
 825                 if (inout)
 826                         ports[1] = cp->vport;
 827                 else
 828                         ports[0] = cp->dport;
 829         }
 830
 831         /* And finally the ICMP checksum */
 832         icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
 833                                               skb->len - icmp_offset,
 834                                               IPPROTO_ICMPV6, 0);
 835         skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
 836         skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
 837         skb->ip_summed = CHECKSUM_PARTIAL;
 838
 839         if (inout)
 840                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 841                               (void *)ciph - (void *)iph,
 842                               "Forwarding altered outgoing ICMPv6");
 843         else
 844                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 845                               (void *)ciph - (void *)iph,
 846                               "Forwarding altered incoming ICMPv6");
 847 }
 848 #endif
 849
 850 /* Handle relevant response ICMP messages - forward to the right
 851  * destination host.
 852  */
 853 static int handle_response_icmp(int af, struct sk_buff *skb,
 854                                 union nf_inet_addr *snet,
 855                                 __u8 protocol, struct ip_vs_conn *cp,
 856                                 struct ip_vs_protocol *pp,
 857                                 unsigned int offset, unsigned int ihl,
 858                                 unsigned int hooknum)
 859 {
 860         unsigned int verdict = NF_DROP;
 861
 862         if (IP_VS_FWD_METHOD(cp) != 0) {
 863                 pr_err("shouldn't reach here, because the box is on the "
 864                        "half connection in the tun/dr module.\n");
 865         }
 866
 867         /* Ensure the checksum is correct */
 868         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 869                 /* Failed checksum! */
 870                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
 871                               IP_VS_DBG_ADDR(af, snet));
 872                 goto out;
 873         }
 874
 875         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 876             IPPROTO_SCTP == protocol)
 877                 offset += 2 * sizeof(__u16);
 878         if (!skb_make_writable(skb, offset))
 879                 goto out;
 880
 881 #ifdef CONFIG_IP_VS_IPV6
 882         if (af == AF_INET6)
 883                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
 884         else
 885 #endif
 886                 ip_vs_nat_icmp(skb, pp, cp, 1);
 887
 888         if (ip_vs_route_me_harder(af, skb, hooknum))
 889                 goto out;
 890
 891         /* do the statistics and put it back */
 892         ip_vs_out_stats(cp, skb);
 893
 894         skb->ipvs_property = 1;
 895         if (!(cp->flags & IP_VS_CONN_F_NFCT))
 896                 ip_vs_notrack(skb);
 897         else
 898                 ip_vs_update_conntrack(skb, cp, 0);
 899         verdict = NF_ACCEPT;
 900
 901 out:
 902         __ip_vs_conn_put(cp);
 903
 904         return verdict;
 905 }
 906
 907 /*
 908  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 909  *      Find any that might be relevant, check against existing connections.
 910  *      Currently handles error types - unreachable, quench, ttl exceeded.
 911  */
 912 static int ip_vs_out_icmp(struct sk_buff *skb, int *related,
 913                           unsigned int hooknum)
 914 {
 915         struct iphdr *iph;
 916         struct icmphdr  _icmph, *ic;
 917         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 918         struct ip_vs_iphdr ciph;
 919         struct ip_vs_conn *cp;
 920         struct ip_vs_protocol *pp;
 921         unsigned int offset, ihl;
 922         union nf_inet_addr snet;
 923
 924         *related = 1;
 925
 926         /* reassemble IP fragments */
 927         if (ip_is_fragment(ip_hdr(skb))) {
 928                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
 929                         return NF_STOLEN;
 930         }
 931
 932         iph = ip_hdr(skb);
 933         offset = ihl = iph->ihl * 4;
 934         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 935         if (ic == NULL)
 936                 return NF_DROP;
 937
 938         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
 939                   ic->type, ntohs(icmp_id(ic)),
 940                   &iph->saddr, &iph->daddr);
 941
 942         /*
 943          * Work through seeing if this is for us.
 944          * These checks are supposed to be in an order that means easy
 945          * things are checked first to speed up processing.... however
 946          * this means that some packets will manage to get a long way
 947          * down this stack and then be rejected, but that's life.
 948          */
 949         if ((ic->type != ICMP_DEST_UNREACH) &&
 950             (ic->type != ICMP_SOURCE_QUENCH) &&
 951             (ic->type != ICMP_TIME_EXCEEDED)) {
 952                 *related = 0;
 953                 return NF_ACCEPT;
 954         }
 955
 956         /* Now find the contained IP header */
 957         offset += sizeof(_icmph);
 958         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 959         if (cih == NULL)
 960                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 961
 962         pp = ip_vs_proto_get(cih->protocol);
 963         if (!pp)
 964                 return NF_ACCEPT;
 965
 966         /* Is the embedded protocol header present? */
 967         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 968                      pp->dont_defrag))
 969                 return NF_ACCEPT;
 970
 971         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
 972                       "Checking outgoing ICMP for");
 973
 974         ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, true, &ciph);
 975
 976         /* The embedded headers contain source and dest in reverse order */
 977         cp = pp->conn_out_get(AF_INET, skb, &ciph);
 978         if (!cp)
 979                 return NF_ACCEPT;
 980
 981         snet.ip = iph->saddr;
 982         return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
 983                                     pp, ciph.len, ihl, hooknum);
 984 }
 985
 986 #ifdef CONFIG_IP_VS_IPV6
 987 static int ip_vs_out_icmp_v6(struct sk_buff *skb, int *related,
 988                              unsigned int hooknum, struct ip_vs_iphdr *ipvsh)
 989 {
 990         struct icmp6hdr _icmph, *ic;
 991         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
 992         struct ip_vs_conn *cp;
 993         struct ip_vs_protocol *pp;
 994         union nf_inet_addr snet;
 995         unsigned int offset;
 996
 997         *related = 1;
 998         ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph, ipvsh);
 999         if (ic == NULL)
1000                 return NF_DROP;
1001
1002         /*
1003          * Work through seeing if this is for us.
1004          * These checks are supposed to be in an order that means easy
1005          * things are checked first to speed up processing.... however
1006          * this means that some packets will manage to get a long way
1007          * down this stack and then be rejected, but that's life.
1008          */
1009         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1010                 *related = 0;
1011                 return NF_ACCEPT;
1012         }
1013         /* Fragment header that is before ICMP header tells us that:
1014          * it's not an error message since they can't be fragmented.
1015          */
1016         if (ipvsh->flags & IP6_FH_F_FRAG)
1017                 return NF_DROP;
1018
1019         IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1020                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1021                   &ipvsh->saddr, &ipvsh->daddr);
1022
1023         if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, ipvsh->len + sizeof(_icmph),
1024                                      true, &ciph))
1025                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1026
1027         pp = ip_vs_proto_get(ciph.protocol);
1028         if (!pp)
1029                 return NF_ACCEPT;
1030
1031         /* The embedded headers contain source and dest in reverse order */
1032         cp = pp->conn_out_get(AF_INET6, skb, &ciph);
1033         if (!cp)
1034                 return NF_ACCEPT;
1035
1036         snet.in6 = ciph.saddr.in6;
1037         offset = ciph.len;
1038         return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
1039                                     pp, offset, sizeof(struct ipv6hdr),
1040                                     hooknum);
1041 }
1042 #endif
1043
1044 /*
1045  * Check if sctp chunc is ABORT chunk
1046  */
1047 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
1048 {
1049         sctp_chunkhdr_t *sch, schunk;
1050         sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
1051                         sizeof(schunk), &schunk);
1052         if (sch == NULL)
1053                 return 0;
1054         if (sch->type == SCTP_CID_ABORT)
1055                 return 1;
1056         return 0;
1057 }
1058
1059 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
1060 {
1061         struct tcphdr _tcph, *th;
1062
1063         th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
1064         if (th == NULL)
1065                 return 0;
1066         return th->rst;
1067 }
1068
1069 static inline bool is_new_conn(const struct sk_buff *skb,
1070                                struct ip_vs_iphdr *iph)
1071 {
1072         switch (iph->protocol) {
1073         case IPPROTO_TCP: {
1074                 struct tcphdr _tcph, *th;
1075
1076                 th = skb_header_pointer(skb, iph->len, sizeof(_tcph), &_tcph);
1077                 if (th == NULL)
1078                         return false;
1079                 return th->syn;
1080         }
1081         case IPPROTO_SCTP: {
1082                 sctp_chunkhdr_t *sch, schunk;
1083
1084                 sch = skb_header_pointer(skb, iph->len + sizeof(sctp_sctphdr_t),
1085                                          sizeof(schunk), &schunk);
1086                 if (sch == NULL)
1087                         return false;
1088                 return sch->type == SCTP_CID_INIT;
1089         }
1090         default:
1091                 return false;
1092         }
1093 }
1094
1095 static inline bool is_new_conn_expected(const struct ip_vs_conn *cp,
1096                                         int conn_reuse_mode)
1097 {
1098         /* Controlled (FTP DATA or persistence)? */
1099         if (cp->control)
1100                 return false;
1101
1102         switch (cp->protocol) {
1103         case IPPROTO_TCP:
1104                 return (cp->state == IP_VS_TCP_S_TIME_WAIT) ||
1105                         ((conn_reuse_mode & 2) &&
1106                          (cp->state == IP_VS_TCP_S_FIN_WAIT) &&
1107                          (cp->flags & IP_VS_CONN_F_NOOUTPUT));
1108         case IPPROTO_SCTP:
1109                 return cp->state == IP_VS_SCTP_S_CLOSED;
1110         default:
1111                 return false;
1112         }
1113 }
1114
1115 /* Handle response packets: rewrite addresses and send away...
1116  */
1117 static unsigned int
1118 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1119                 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph,
1120                 unsigned int hooknum)
1121 {
1122         struct ip_vs_protocol *pp = pd->pp;
1123
1124         IP_VS_DBG_PKT(11, af, pp, skb, iph->off, "Outgoing packet");
1125
1126         if (!skb_make_writable(skb, iph->len))
1127                 goto drop;
1128
1129         /* mangle the packet */
1130         if (pp->snat_handler && !pp->snat_handler(skb, pp, cp, iph))
1131                 goto drop;
1132
1133 #ifdef CONFIG_IP_VS_IPV6
1134         if (af == AF_INET6)
1135                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1136         else
1137 #endif
1138         {
1139                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1140                 ip_send_check(ip_hdr(skb));
1141         }
1142
1143         /*
1144          * nf_iterate does not expect change in the skb->dst->dev.
1145          * It looks like it is not fatal to enable this code for hooks
1146          * where our handlers are at the end of the chain list and
1147          * when all next handlers use skb->dst->dev and not outdev.
1148          * It will definitely route properly the inout NAT traffic
1149          * when multiple paths are used.
1150          */
1151
1152         /* For policy routing, packets originating from this
1153          * machine itself may be routed differently to packets
1154          * passing through.  We want this packet to be routed as
1155          * if it came from this machine itself.  So re-compute
1156          * the routing information.
1157          */
1158         if (ip_vs_route_me_harder(af, skb, hooknum))
1159                 goto drop;
1160
1161         IP_VS_DBG_PKT(10, af, pp, skb, iph->off, "After SNAT");
1162
1163         ip_vs_out_stats(cp, skb);
1164         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1165         skb->ipvs_property = 1;
1166         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1167                 ip_vs_notrack(skb);
1168         else
1169                 ip_vs_update_conntrack(skb, cp, 0);
1170         ip_vs_conn_put(cp);
1171
1172         LeaveFunction(11);
1173         return NF_ACCEPT;
1174
1175 drop:
1176         ip_vs_conn_put(cp);
1177         kfree_skb(skb);
1178         LeaveFunction(11);
1179         return NF_STOLEN;
1180 }
1181
1182 /*
1183  *      Check if outgoing packet belongs to the established ip_vs_conn.
1184  */
1185 static unsigned int
1186 ip_vs_out(unsigned int hooknum, struct sk_buff *skb, int af)
1187 {
1188         struct net *net = NULL;
1189         struct ip_vs_iphdr iph;
1190         struct ip_vs_protocol *pp;
1191         struct ip_vs_proto_data *pd;
1192         struct ip_vs_conn *cp;
1193
1194         EnterFunction(11);
1195
1196         /* Already marked as IPVS request or reply? */
1197         if (skb->ipvs_property)
1198                 return NF_ACCEPT;
1199
1200         /* Bad... Do not break raw sockets */
1201         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1202                      af == AF_INET)) {
1203                 struct sock *sk = skb->sk;
1204                 struct inet_sock *inet = inet_sk(skb->sk);
1205
1206                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1207                         return NF_ACCEPT;
1208         }
1209
1210         if (unlikely(!skb_dst(skb)))
1211                 return NF_ACCEPT;
1212
1213         net = skb_net(skb);
1214         if (!net_ipvs(net)->enable)
1215                 return NF_ACCEPT;
1216
1217         ip_vs_fill_iph_skb(af, skb, false, &iph);
1218 #ifdef CONFIG_IP_VS_IPV6
1219         if (af == AF_INET6) {
1220                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1221                         int related;
1222                         int verdict = ip_vs_out_icmp_v6(skb, &related,
1223                                                         hooknum, &iph);
1224
1225                         if (related)
1226                                 return verdict;
1227                 }
1228         } else
1229 #endif
1230                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1231                         int related;
1232                         int verdict = ip_vs_out_icmp(skb, &related, hooknum);
1233
1234                         if (related)
1235                                 return verdict;
1236                 }
1237
1238         pd = ip_vs_proto_data_get(net, iph.protocol);
1239         if (unlikely(!pd))
1240                 return NF_ACCEPT;
1241         pp = pd->pp;
1242
1243         /* reassemble IP fragments */
1244 #ifdef CONFIG_IP_VS_IPV6
1245         if (af == AF_INET)
1246 #endif
1247                 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1248                         if (ip_vs_gather_frags(skb,
1249                                                ip_vs_defrag_user(hooknum)))
1250                                 return NF_STOLEN;
1251
1252                         ip_vs_fill_iph_skb(AF_INET, skb, false, &iph);
1253                 }
1254
1255         /*
1256          * Check if the packet belongs to an existing entry
1257          */
1258         cp = pp->conn_out_get(af, skb, &iph);
1259
1260         if (likely(cp))
1261                 return handle_response(af, skb, pd, cp, &iph, hooknum);
1262         if (sysctl_nat_icmp_send(net) &&
1263             (pp->protocol == IPPROTO_TCP ||
1264              pp->protocol == IPPROTO_UDP ||
1265              pp->protocol == IPPROTO_SCTP)) {
1266                 __be16 _ports[2], *pptr;
1267
1268                 pptr = frag_safe_skb_hp(skb, iph.len,
1269                                          sizeof(_ports), _ports, &iph);
1270                 if (pptr == NULL)
1271                         return NF_ACCEPT;       /* Not for me */
1272                 if (ip_vs_has_real_service(net, af, iph.protocol, &iph.saddr,
1273                                            pptr[0])) {
1274                         /*
1275                          * Notify the real server: there is no
1276                          * existing entry if it is not RST
1277                          * packet or not TCP packet.
1278                          */
1279                         if ((iph.protocol != IPPROTO_TCP &&
1280                              iph.protocol != IPPROTO_SCTP)
1281                              || ((iph.protocol == IPPROTO_TCP
1282                                   && !is_tcp_reset(skb, iph.len))
1283                                  || (iph.protocol == IPPROTO_SCTP
1284                                         && !is_sctp_abort(skb,
1285                                                 iph.len)))) {
1286 #ifdef CONFIG_IP_VS_IPV6
1287                                 if (af == AF_INET6) {
1288                                         if (!skb->dev)
1289                                                 skb->dev = net->loopback_dev;
1290                                         icmpv6_send(skb,
1291                                                     ICMPV6_DEST_UNREACH,
1292                                                     ICMPV6_PORT_UNREACH,
1293                                                     0);
1294                                 } else
1295 #endif
1296                                         icmp_send(skb,
1297                                                   ICMP_DEST_UNREACH,
1298                                                   ICMP_PORT_UNREACH, 0);
1299                                 return NF_DROP;
1300                         }
1301                 }
1302         }
1303         IP_VS_DBG_PKT(12, af, pp, skb, iph.off,
1304                       "ip_vs_out: packet continues traversal as normal");
1305         return NF_ACCEPT;
1306 }
1307
1308 /*
1309  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1310  *      used only for VS/NAT.
1311  *      Check if packet is reply for established ip_vs_conn.
1312  */
1313 static unsigned int
1314 ip_vs_reply4(void *priv, struct sk_buff *skb,
1315              const struct nf_hook_state *state)
1316 {
1317         return ip_vs_out(state->hook, skb, AF_INET);
1318 }
1319
1320 /*
1321  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1322  *      Check if packet is reply for established ip_vs_conn.
1323  */
1324 static unsigned int
1325 ip_vs_local_reply4(void *priv, struct sk_buff *skb,
1326                    const struct nf_hook_state *state)
1327 {
1328         return ip_vs_out(state->hook, skb, AF_INET);
1329 }
1330
1331 #ifdef CONFIG_IP_VS_IPV6
1332
1333 /*
1334  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1335  *      used only for VS/NAT.
1336  *      Check if packet is reply for established ip_vs_conn.
1337  */
1338 static unsigned int
1339 ip_vs_reply6(void *priv, struct sk_buff *skb,
1340              const struct nf_hook_state *state)
1341 {
1342         return ip_vs_out(state->hook, skb, AF_INET6);
1343 }
1344
1345 /*
1346  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1347  *      Check if packet is reply for established ip_vs_conn.
1348  */
1349 static unsigned int
1350 ip_vs_local_reply6(void *priv, struct sk_buff *skb,
1351                    const struct nf_hook_state *state)
1352 {
1353         return ip_vs_out(state->hook, skb, AF_INET6);
1354 }
1355
1356 #endif
1357
1358 static unsigned int
1359 ip_vs_try_to_schedule(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1360                       int *verdict, struct ip_vs_conn **cpp,
1361                       struct ip_vs_iphdr *iph)
1362 {
1363         struct ip_vs_protocol *pp = pd->pp;
1364
1365         if (!iph->fragoffs) {
1366                 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1367                  * replayed fragment zero will already have created the cp
1368                  */
1369
1370                 /* Schedule and create new connection entry into cpp */
1371                 if (!pp->conn_schedule(af, skb, pd, verdict, cpp, iph))
1372                         return 0;
1373         }
1374
1375         if (unlikely(!*cpp)) {
1376                 /* sorry, all this trouble for a no-hit :) */
1377                 IP_VS_DBG_PKT(12, af, pp, skb, iph->off,
1378                               "ip_vs_in: packet continues traversal as normal");
1379                 if (iph->fragoffs) {
1380                         /* Fragment that couldn't be mapped to a conn entry
1381                          * is missing module nf_defrag_ipv6
1382                          */
1383                         IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1384                         IP_VS_DBG_PKT(7, af, pp, skb, iph->off,
1385                                       "unhandled fragment");
1386                 }
1387                 *verdict = NF_ACCEPT;
1388                 return 0;
1389         }
1390
1391         return 1;
1392 }
1393
1394 /*
1395  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1396  *      Find any that might be relevant, check against existing connections,
1397  *      forward to the right destination host if relevant.
1398  *      Currently handles error types - unreachable, quench, ttl exceeded.
1399  */
1400 static int
1401 ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum)
1402 {
1403         struct net *net = NULL;
1404         struct iphdr *iph;
1405         struct icmphdr  _icmph, *ic;
1406         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1407         struct ip_vs_iphdr ciph;
1408         struct ip_vs_conn *cp;
1409         struct ip_vs_protocol *pp;
1410         struct ip_vs_proto_data *pd;
1411         unsigned int offset, offset2, ihl, verdict;
1412         bool ipip, new_cp = false;
1413
1414         *related = 1;
1415
1416         /* reassemble IP fragments */
1417         if (ip_is_fragment(ip_hdr(skb))) {
1418                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
1419                         return NF_STOLEN;
1420         }
1421
1422         iph = ip_hdr(skb);
1423         offset = ihl = iph->ihl * 4;
1424         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1425         if (ic == NULL)
1426                 return NF_DROP;
1427
1428         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1429                   ic->type, ntohs(icmp_id(ic)),
1430                   &iph->saddr, &iph->daddr);
1431
1432         /*
1433          * Work through seeing if this is for us.
1434          * These checks are supposed to be in an order that means easy
1435          * things are checked first to speed up processing.... however
1436          * this means that some packets will manage to get a long way
1437          * down this stack and then be rejected, but that's life.
1438          */
1439         if ((ic->type != ICMP_DEST_UNREACH) &&
1440             (ic->type != ICMP_SOURCE_QUENCH) &&
1441             (ic->type != ICMP_TIME_EXCEEDED)) {
1442                 *related = 0;
1443                 return NF_ACCEPT;
1444         }
1445
1446         /* Now find the contained IP header */
1447         offset += sizeof(_icmph);
1448         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1449         if (cih == NULL)
1450                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1451
1452         net = skb_net(skb);
1453
1454         /* Special case for errors for IPIP packets */
1455         ipip = false;
1456         if (cih->protocol == IPPROTO_IPIP) {
1457                 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1458                         return NF_ACCEPT;
1459                 /* Error for our IPIP must arrive at LOCAL_IN */
1460                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1461                         return NF_ACCEPT;
1462                 offset += cih->ihl * 4;
1463                 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1464                 if (cih == NULL)
1465                         return NF_ACCEPT; /* The packet looks wrong, ignore */
1466                 ipip = true;
1467         }
1468
1469         pd = ip_vs_proto_data_get(net, cih->protocol);
1470         if (!pd)
1471                 return NF_ACCEPT;
1472         pp = pd->pp;
1473
1474         /* Is the embedded protocol header present? */
1475         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1476                      pp->dont_defrag))
1477                 return NF_ACCEPT;
1478
1479         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1480                       "Checking incoming ICMP for");
1481
1482         offset2 = offset;
1483         ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, !ipip, &ciph);
1484         offset = ciph.len;
1485
1486         /* The embedded headers contain source and dest in reverse order.
1487          * For IPIP this is error for request, not for reply.
1488          */
1489         cp = pp->conn_in_get(AF_INET, skb, &ciph);
1490
1491         if (!cp) {
1492                 int v;
1493
1494                 if (!sysctl_schedule_icmp(net_ipvs(net)))
1495                         return NF_ACCEPT;
1496
1497                 if (!ip_vs_try_to_schedule(AF_INET, skb, pd, &v, &cp, &ciph))
1498                         return v;
1499                 new_cp = true;
1500         }
1501
1502         verdict = NF_DROP;
1503
1504         /* Ensure the checksum is correct */
1505         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1506                 /* Failed checksum! */
1507                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1508                           &iph->saddr);
1509                 goto out;
1510         }
1511
1512         if (ipip) {
1513                 __be32 info = ic->un.gateway;
1514                 __u8 type = ic->type;
1515                 __u8 code = ic->code;
1516
1517                 /* Update the MTU */
1518                 if (ic->type == ICMP_DEST_UNREACH &&
1519                     ic->code == ICMP_FRAG_NEEDED) {
1520                         struct ip_vs_dest *dest = cp->dest;
1521                         u32 mtu = ntohs(ic->un.frag.mtu);
1522                         __be16 frag_off = cih->frag_off;
1523
1524                         /* Strip outer IP and ICMP, go to IPIP header */
1525                         if (pskb_pull(skb, ihl + sizeof(_icmph)) == NULL)
1526                                 goto ignore_ipip;
1527                         offset2 -= ihl + sizeof(_icmph);
1528                         skb_reset_network_header(skb);
1529                         IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1530                                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1531                         ipv4_update_pmtu(skb, dev_net(skb->dev),
1532                                          mtu, 0, 0, 0, 0);
1533                         /* Client uses PMTUD? */
1534                         if (!(frag_off & htons(IP_DF)))
1535                                 goto ignore_ipip;
1536                         /* Prefer the resulting PMTU */
1537                         if (dest) {
1538                                 struct ip_vs_dest_dst *dest_dst;
1539
1540                                 rcu_read_lock();
1541                                 dest_dst = rcu_dereference(dest->dest_dst);
1542                                 if (dest_dst)
1543                                         mtu = dst_mtu(dest_dst->dst_cache);
1544                                 rcu_read_unlock();
1545                         }
1546                         if (mtu > 68 + sizeof(struct iphdr))
1547                                 mtu -= sizeof(struct iphdr);
1548                         info = htonl(mtu);
1549                 }
1550                 /* Strip outer IP, ICMP and IPIP, go to IP header of
1551                  * original request.
1552                  */
1553                 if (pskb_pull(skb, offset2) == NULL)
1554                         goto ignore_ipip;
1555                 skb_reset_network_header(skb);
1556                 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1557                         &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1558                         type, code, ntohl(info));
1559                 icmp_send(skb, type, code, info);
1560                 /* ICMP can be shorter but anyways, account it */
1561                 ip_vs_out_stats(cp, skb);
1562
1563 ignore_ipip:
1564                 consume_skb(skb);
1565                 verdict = NF_STOLEN;
1566                 goto out;
1567         }
1568
1569         /* do the statistics and put it back */
1570         ip_vs_in_stats(cp, skb);
1571         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol ||
1572             IPPROTO_SCTP == cih->protocol)
1573                 offset += 2 * sizeof(__u16);
1574         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1575
1576 out:
1577         if (likely(!new_cp))
1578                 __ip_vs_conn_put(cp);
1579         else
1580                 ip_vs_conn_put(cp);
1581
1582         return verdict;
1583 }
1584
1585 #ifdef CONFIG_IP_VS_IPV6
1586 static int ip_vs_in_icmp_v6(struct sk_buff *skb, int *related,
1587                             unsigned int hooknum, struct ip_vs_iphdr *iph)
1588 {
1589         struct net *net = NULL;
1590         struct icmp6hdr _icmph, *ic;
1591         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1592         struct ip_vs_conn *cp;
1593         struct ip_vs_protocol *pp;
1594         struct ip_vs_proto_data *pd;
1595         unsigned int offset, verdict;
1596         bool new_cp = false;
1597
1598         *related = 1;
1599
1600         ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph, iph);
1601         if (ic == NULL)
1602                 return NF_DROP;
1603
1604         /*
1605          * Work through seeing if this is for us.
1606          * These checks are supposed to be in an order that means easy
1607          * things are checked first to speed up processing.... however
1608          * this means that some packets will manage to get a long way
1609          * down this stack and then be rejected, but that's life.
1610          */
1611         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1612                 *related = 0;
1613                 return NF_ACCEPT;
1614         }
1615         /* Fragment header that is before ICMP header tells us that:
1616          * it's not an error message since they can't be fragmented.
1617          */
1618         if (iph->flags & IP6_FH_F_FRAG)
1619                 return NF_DROP;
1620
1621         IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1622                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1623                   &iph->saddr, &iph->daddr);
1624
1625         offset = iph->len + sizeof(_icmph);
1626         if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, offset, true, &ciph))
1627                 return NF_ACCEPT;
1628
1629         net = skb_net(skb);
1630         pd = ip_vs_proto_data_get(net, ciph.protocol);
1631         if (!pd)
1632                 return NF_ACCEPT;
1633         pp = pd->pp;
1634
1635         /* Cannot handle fragmented embedded protocol */
1636         if (ciph.fragoffs)
1637                 return NF_ACCEPT;
1638
1639         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
1640                       "Checking incoming ICMPv6 for");
1641
1642         /* The embedded headers contain source and dest in reverse order
1643          * if not from localhost
1644          */
1645         cp = pp->conn_in_get(AF_INET6, skb, &ciph);
1646
1647         if (!cp) {
1648                 int v;
1649
1650                 if (!sysctl_schedule_icmp(net_ipvs(net)))
1651                         return NF_ACCEPT;
1652
1653                 if (!ip_vs_try_to_schedule(AF_INET6, skb, pd, &v, &cp, &ciph))
1654                         return v;
1655
1656                 new_cp = true;
1657         }
1658
1659         /* VS/TUN, VS/DR and LOCALNODE just let it go */
1660         if ((hooknum == NF_INET_LOCAL_OUT) &&
1661             (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1662                 verdict = NF_ACCEPT;
1663                 goto out;
1664         }
1665
1666         /* do the statistics and put it back */
1667         ip_vs_in_stats(cp, skb);
1668
1669         /* Need to mangle contained IPv6 header in ICMPv6 packet */
1670         offset = ciph.len;
1671         if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1672             IPPROTO_SCTP == ciph.protocol)
1673                 offset += 2 * sizeof(__u16); /* Also mangle ports */
1674
1675         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset, hooknum, &ciph);
1676
1677 out:
1678         if (likely(!new_cp))
1679                 __ip_vs_conn_put(cp);
1680         else
1681                 ip_vs_conn_put(cp);
1682
1683         return verdict;
1684 }
1685 #endif
1686
1687
1688 /*
1689  *      Check if it's for virtual services, look it up,
1690  *      and send it on its way...
1691  */
1692 static unsigned int
1693 ip_vs_in(unsigned int hooknum, struct sk_buff *skb, int af)
1694 {
1695         struct net *net;
1696         struct ip_vs_iphdr iph;
1697         struct ip_vs_protocol *pp;
1698         struct ip_vs_proto_data *pd;
1699         struct ip_vs_conn *cp;
1700         int ret, pkts;
1701         struct netns_ipvs *ipvs;
1702         int conn_reuse_mode;
1703
1704         /* Already marked as IPVS request or reply? */
1705         if (skb->ipvs_property)
1706                 return NF_ACCEPT;
1707
1708         /*
1709          *      Big tappo:
1710          *      - remote client: only PACKET_HOST
1711          *      - route: used for struct net when skb->dev is unset
1712          */
1713         if (unlikely((skb->pkt_type != PACKET_HOST &&
1714                       hooknum != NF_INET_LOCAL_OUT) ||
1715                      !skb_dst(skb))) {
1716                 ip_vs_fill_iph_skb(af, skb, false, &iph);
1717                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1718                               " ignored in hook %u\n",
1719                               skb->pkt_type, iph.protocol,
1720                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1721                 return NF_ACCEPT;
1722         }
1723         /* ipvs enabled in this netns ? */
1724         net = skb_net(skb);
1725         ipvs = net_ipvs(net);
1726         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1727                 return NF_ACCEPT;
1728
1729         ip_vs_fill_iph_skb(af, skb, false, &iph);
1730
1731         /* Bad... Do not break raw sockets */
1732         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1733                      af == AF_INET)) {
1734                 struct sock *sk = skb->sk;
1735                 struct inet_sock *inet = inet_sk(skb->sk);
1736
1737                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1738                         return NF_ACCEPT;
1739         }
1740
1741 #ifdef CONFIG_IP_VS_IPV6
1742         if (af == AF_INET6) {
1743                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1744                         int related;
1745                         int verdict = ip_vs_in_icmp_v6(skb, &related, hooknum,
1746                                                        &iph);
1747
1748                         if (related)
1749                                 return verdict;
1750                 }
1751         } else
1752 #endif
1753                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1754                         int related;
1755                         int verdict = ip_vs_in_icmp(skb, &related, hooknum);
1756
1757                         if (related)
1758                                 return verdict;
1759                 }
1760
1761         /* Protocol supported? */
1762         pd = ip_vs_proto_data_get(net, iph.protocol);
1763         if (unlikely(!pd)) {
1764                 /* The only way we'll see this packet again is if it's
1765                  * encapsulated, so mark it with ipvs_property=1 so we
1766                  * skip it if we're ignoring tunneled packets
1767                  */
1768                 if (sysctl_ignore_tunneled(ipvs))
1769                         skb->ipvs_property = 1;
1770
1771                 return NF_ACCEPT;
1772         }
1773         pp = pd->pp;
1774         /*
1775          * Check if the packet belongs to an existing connection entry
1776          */
1777         cp = pp->conn_in_get(af, skb, &iph);
1778
1779         conn_reuse_mode = sysctl_conn_reuse_mode(ipvs);
1780         if (conn_reuse_mode && !iph.fragoffs &&
1781             is_new_conn(skb, &iph) && cp &&
1782             ((unlikely(sysctl_expire_nodest_conn(ipvs)) && cp->dest &&
1783               unlikely(!atomic_read(&cp->dest->weight))) ||
1784              unlikely(is_new_conn_expected(cp, conn_reuse_mode)))) {
1785                 if (!atomic_read(&cp->n_control))
1786                         ip_vs_conn_expire_now(cp);
1787                 __ip_vs_conn_put(cp);
1788                 cp = NULL;
1789         }
1790
1791         if (unlikely(!cp)) {
1792                 int v;
1793
1794                 if (!ip_vs_try_to_schedule(af, skb, pd, &v, &cp, &iph))
1795                         return v;
1796         }
1797
1798         IP_VS_DBG_PKT(11, af, pp, skb, iph.off, "Incoming packet");
1799
1800         /* Check the server status */
1801         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1802                 /* the destination server is not available */
1803
1804                 if (sysctl_expire_nodest_conn(ipvs)) {
1805                         /* try to expire the connection immediately */
1806                         ip_vs_conn_expire_now(cp);
1807                 }
1808                 /* don't restart its timer, and silently
1809                    drop the packet. */
1810                 __ip_vs_conn_put(cp);
1811                 return NF_DROP;
1812         }
1813
1814         ip_vs_in_stats(cp, skb);
1815         ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1816         if (cp->packet_xmit)
1817                 ret = cp->packet_xmit(skb, cp, pp, &iph);
1818                 /* do not touch skb anymore */
1819         else {
1820                 IP_VS_DBG_RL("warning: packet_xmit is null");
1821                 ret = NF_ACCEPT;
1822         }
1823
1824         /* Increase its packet counter and check if it is needed
1825          * to be synchronized
1826          *
1827          * Sync connection if it is about to close to
1828          * encorage the standby servers to update the connections timeout
1829          *
1830          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1831          */
1832
1833         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
1834                 pkts = sysctl_sync_threshold(ipvs);
1835         else
1836                 pkts = atomic_add_return(1, &cp->in_pkts);
1837
1838         if (ipvs->sync_state & IP_VS_STATE_MASTER)
1839                 ip_vs_sync_conn(net, cp, pkts);
1840
1841         ip_vs_conn_put(cp);
1842         return ret;
1843 }
1844
1845 /*
1846  *      AF_INET handler in NF_INET_LOCAL_IN chain
1847  *      Schedule and forward packets from remote clients
1848  */
1849 static unsigned int
1850 ip_vs_remote_request4(void *priv, struct sk_buff *skb,
1851                       const struct nf_hook_state *state)
1852 {
1853         return ip_vs_in(state->hook, skb, AF_INET);
1854 }
1855
1856 /*
1857  *      AF_INET handler in NF_INET_LOCAL_OUT chain
1858  *      Schedule and forward packets from local clients
1859  */
1860 static unsigned int
1861 ip_vs_local_request4(void *priv, struct sk_buff *skb,
1862                      const struct nf_hook_state *state)
1863 {
1864         return ip_vs_in(state->hook, skb, AF_INET);
1865 }
1866
1867 #ifdef CONFIG_IP_VS_IPV6
1868
1869 /*
1870  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
1871  *      Schedule and forward packets from remote clients
1872  */
1873 static unsigned int
1874 ip_vs_remote_request6(void *priv, struct sk_buff *skb,
1875                       const struct nf_hook_state *state)
1876 {
1877         return ip_vs_in(state->hook, skb, AF_INET6);
1878 }
1879
1880 /*
1881  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
1882  *      Schedule and forward packets from local clients
1883  */
1884 static unsigned int
1885 ip_vs_local_request6(void *priv, struct sk_buff *skb,
1886                      const struct nf_hook_state *state)
1887 {
1888         return ip_vs_in(state->hook, skb, AF_INET6);
1889 }
1890
1891 #endif
1892
1893
1894 /*
1895  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1896  *      related packets destined for 0.0.0.0/0.
1897  *      When fwmark-based virtual service is used, such as transparent
1898  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
1899  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1900  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1901  *      and send them to ip_vs_in_icmp.
1902  */
1903 static unsigned int
1904 ip_vs_forward_icmp(void *priv, struct sk_buff *skb,
1905                    const struct nf_hook_state *state)
1906 {
1907         int r;
1908         struct netns_ipvs *ipvs;
1909
1910         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1911                 return NF_ACCEPT;
1912
1913         /* ipvs enabled in this netns ? */
1914         ipvs = net_ipvs(state->net);
1915         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1916                 return NF_ACCEPT;
1917
1918         return ip_vs_in_icmp(skb, &r, state->hook);
1919 }
1920
1921 #ifdef CONFIG_IP_VS_IPV6
1922 static unsigned int
1923 ip_vs_forward_icmp_v6(void *priv, struct sk_buff *skb,
1924                       const struct nf_hook_state *state)
1925 {
1926         int r;
1927         struct netns_ipvs *ipvs;
1928         struct ip_vs_iphdr iphdr;
1929
1930         ip_vs_fill_iph_skb(AF_INET6, skb, false, &iphdr);
1931         if (iphdr.protocol != IPPROTO_ICMPV6)
1932                 return NF_ACCEPT;
1933
1934         /* ipvs enabled in this netns ? */
1935         ipvs = net_ipvs(state->net);
1936         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1937                 return NF_ACCEPT;
1938
1939         return ip_vs_in_icmp_v6(skb, &r, state->hook, &iphdr);
1940 }
1941 #endif
1942
1943
1944 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
1945         /* After packet filtering, change source only for VS/NAT */
1946         {
1947                 .hook           = ip_vs_reply4,
1948                 .owner          = THIS_MODULE,
1949                 .pf             = NFPROTO_IPV4,
1950                 .hooknum        = NF_INET_LOCAL_IN,
1951                 .priority       = NF_IP_PRI_NAT_SRC - 2,
1952         },
1953         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1954          * or VS/NAT(change destination), so that filtering rules can be
1955          * applied to IPVS. */
1956         {
1957                 .hook           = ip_vs_remote_request4,
1958                 .owner          = THIS_MODULE,
1959                 .pf             = NFPROTO_IPV4,
1960                 .hooknum        = NF_INET_LOCAL_IN,
1961                 .priority       = NF_IP_PRI_NAT_SRC - 1,
1962         },
1963         /* Before ip_vs_in, change source only for VS/NAT */
1964         {
1965                 .hook           = ip_vs_local_reply4,
1966                 .owner          = THIS_MODULE,
1967                 .pf             = NFPROTO_IPV4,
1968                 .hooknum        = NF_INET_LOCAL_OUT,
1969                 .priority       = NF_IP_PRI_NAT_DST + 1,
1970         },
1971         /* After mangle, schedule and forward local requests */
1972         {
1973                 .hook           = ip_vs_local_request4,
1974                 .owner          = THIS_MODULE,
1975                 .pf             = NFPROTO_IPV4,
1976                 .hooknum        = NF_INET_LOCAL_OUT,
1977                 .priority       = NF_IP_PRI_NAT_DST + 2,
1978         },
1979         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1980          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1981         {
1982                 .hook           = ip_vs_forward_icmp,
1983                 .owner          = THIS_MODULE,
1984                 .pf             = NFPROTO_IPV4,
1985                 .hooknum        = NF_INET_FORWARD,
1986                 .priority       = 99,
1987         },
1988         /* After packet filtering, change source only for VS/NAT */
1989         {
1990                 .hook           = ip_vs_reply4,
1991                 .owner          = THIS_MODULE,
1992                 .pf             = NFPROTO_IPV4,
1993                 .hooknum        = NF_INET_FORWARD,
1994                 .priority       = 100,
1995         },
1996 #ifdef CONFIG_IP_VS_IPV6
1997         /* After packet filtering, change source only for VS/NAT */
1998         {
1999                 .hook           = ip_vs_reply6,
2000                 .owner          = THIS_MODULE,
2001                 .pf             = NFPROTO_IPV6,
2002                 .hooknum        = NF_INET_LOCAL_IN,
2003                 .priority       = NF_IP6_PRI_NAT_SRC - 2,
2004         },
2005         /* After packet filtering, forward packet through VS/DR, VS/TUN,
2006          * or VS/NAT(change destination), so that filtering rules can be
2007          * applied to IPVS. */
2008         {
2009                 .hook           = ip_vs_remote_request6,
2010                 .owner          = THIS_MODULE,
2011                 .pf             = NFPROTO_IPV6,
2012                 .hooknum        = NF_INET_LOCAL_IN,
2013                 .priority       = NF_IP6_PRI_NAT_SRC - 1,
2014         },
2015         /* Before ip_vs_in, change source only for VS/NAT */
2016         {
2017                 .hook           = ip_vs_local_reply6,
2018                 .owner          = THIS_MODULE,
2019                 .pf             = NFPROTO_IPV6,
2020                 .hooknum        = NF_INET_LOCAL_OUT,
2021                 .priority       = NF_IP6_PRI_NAT_DST + 1,
2022         },
2023         /* After mangle, schedule and forward local requests */
2024         {
2025                 .hook           = ip_vs_local_request6,
2026                 .owner          = THIS_MODULE,
2027                 .pf             = NFPROTO_IPV6,
2028                 .hooknum        = NF_INET_LOCAL_OUT,
2029                 .priority       = NF_IP6_PRI_NAT_DST + 2,
2030         },
2031         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2032          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2033         {
2034                 .hook           = ip_vs_forward_icmp_v6,
2035                 .owner          = THIS_MODULE,
2036                 .pf             = NFPROTO_IPV6,
2037                 .hooknum        = NF_INET_FORWARD,
2038                 .priority       = 99,
2039         },
2040         /* After packet filtering, change source only for VS/NAT */
2041         {
2042                 .hook           = ip_vs_reply6,
2043                 .owner          = THIS_MODULE,
2044                 .pf             = NFPROTO_IPV6,
2045                 .hooknum        = NF_INET_FORWARD,
2046                 .priority       = 100,
2047         },
2048 #endif
2049 };
2050 /*
2051  *      Initialize IP Virtual Server netns mem.
2052  */
2053 static int __net_init __ip_vs_init(struct net *net)
2054 {
2055         struct netns_ipvs *ipvs;
2056
2057         ipvs = net_generic(net, ip_vs_net_id);
2058         if (ipvs == NULL)
2059                 return -ENOMEM;
2060
2061         /* Hold the beast until a service is registerd */
2062         ipvs->enable = 0;
2063         ipvs->net = net;
2064         /* Counters used for creating unique names */
2065         ipvs->gen = atomic_read(&ipvs_netns_cnt);
2066         atomic_inc(&ipvs_netns_cnt);
2067         net->ipvs = ipvs;
2068
2069         if (ip_vs_estimator_net_init(net) < 0)
2070                 goto estimator_fail;
2071
2072         if (ip_vs_control_net_init(net) < 0)
2073                 goto control_fail;
2074
2075         if (ip_vs_protocol_net_init(net) < 0)
2076                 goto protocol_fail;
2077
2078         if (ip_vs_app_net_init(net) < 0)
2079                 goto app_fail;
2080
2081         if (ip_vs_conn_net_init(net) < 0)
2082                 goto conn_fail;
2083
2084         if (ip_vs_sync_net_init(net) < 0)
2085                 goto sync_fail;
2086
2087         printk(KERN_INFO "IPVS: Creating netns size=%zu id=%d\n",
2088                          sizeof(struct netns_ipvs), ipvs->gen);
2089         return 0;
2090 /*
2091  * Error handling
2092  */
2093
2094 sync_fail:
2095         ip_vs_conn_net_cleanup(net);
2096 conn_fail:
2097         ip_vs_app_net_cleanup(net);
2098 app_fail:
2099         ip_vs_protocol_net_cleanup(net);
2100 protocol_fail:
2101         ip_vs_control_net_cleanup(net);
2102 control_fail:
2103         ip_vs_estimator_net_cleanup(net);
2104 estimator_fail:
2105         net->ipvs = NULL;
2106         return -ENOMEM;
2107 }
2108
2109 static void __net_exit __ip_vs_cleanup(struct net *net)
2110 {
2111         ip_vs_service_net_cleanup(net); /* ip_vs_flush() with locks */
2112         ip_vs_conn_net_cleanup(net);
2113         ip_vs_app_net_cleanup(net);
2114         ip_vs_protocol_net_cleanup(net);
2115         ip_vs_control_net_cleanup(net);
2116         ip_vs_estimator_net_cleanup(net);
2117         IP_VS_DBG(2, "ipvs netns %d released\n", net_ipvs(net)->gen);
2118         net->ipvs = NULL;
2119 }
2120
2121 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
2122 {
2123         EnterFunction(2);
2124         net_ipvs(net)->enable = 0;      /* Disable packet reception */
2125         smp_wmb();
2126         ip_vs_sync_net_cleanup(net);
2127         LeaveFunction(2);
2128 }
2129
2130 static struct pernet_operations ipvs_core_ops = {
2131         .init = __ip_vs_init,
2132         .exit = __ip_vs_cleanup,
2133         .id   = &ip_vs_net_id,
2134         .size = sizeof(struct netns_ipvs),
2135 };
2136
2137 static struct pernet_operations ipvs_core_dev_ops = {
2138         .exit = __ip_vs_dev_cleanup,
2139 };
2140
2141 /*
2142  *      Initialize IP Virtual Server
2143  */
2144 static int __init ip_vs_init(void)
2145 {
2146         int ret;
2147
2148         ret = ip_vs_control_init();
2149         if (ret < 0) {
2150                 pr_err("can't setup control.\n");
2151                 goto exit;
2152         }
2153
2154         ip_vs_protocol_init();
2155
2156         ret = ip_vs_conn_init();
2157         if (ret < 0) {
2158                 pr_err("can't setup connection table.\n");
2159                 goto cleanup_protocol;
2160         }
2161
2162         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
2163         if (ret < 0)
2164                 goto cleanup_conn;
2165
2166         ret = register_pernet_device(&ipvs_core_dev_ops);
2167         if (ret < 0)
2168                 goto cleanup_sub;
2169
2170         ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2171         if (ret < 0) {
2172                 pr_err("can't register hooks.\n");
2173                 goto cleanup_dev;
2174         }
2175
2176         ret = ip_vs_register_nl_ioctl();
2177         if (ret < 0) {
2178                 pr_err("can't register netlink/ioctl.\n");
2179                 goto cleanup_hooks;
2180         }
2181
2182         pr_info("ipvs loaded.\n");
2183
2184         return ret;
2185
2186 cleanup_hooks:
2187         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2188 cleanup_dev:
2189         unregister_pernet_device(&ipvs_core_dev_ops);
2190 cleanup_sub:
2191         unregister_pernet_subsys(&ipvs_core_ops);
2192 cleanup_conn:
2193         ip_vs_conn_cleanup();
2194 cleanup_protocol:
2195         ip_vs_protocol_cleanup();
2196         ip_vs_control_cleanup();
2197 exit:
2198         return ret;
2199 }
2200
2201 static void __exit ip_vs_cleanup(void)
2202 {
2203         ip_vs_unregister_nl_ioctl();
2204         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2205         unregister_pernet_device(&ipvs_core_dev_ops);
2206         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
2207         ip_vs_conn_cleanup();
2208         ip_vs_protocol_cleanup();
2209         ip_vs_control_cleanup();
2210         pr_info("ipvs unloaded.\n");
2211 }
2212
2213 module_init(ip_vs_init);
2214 module_exit(ip_vs_cleanup);
2215 MODULE_LICENSE("GPL");