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