net/netfilter/ipvs/ip_vs_core.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * IPVS         An implementation of the IP virtual server support for the
   4  *              LINUX operating system.  IPVS is now implemented as a module
   5  *              over the Netfilter framework. IPVS can be used to build a
   6  *              high-performance and highly available server based on a
   7  *              cluster of servers.
   8  *
   9  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
  10  *              Peter Kese <peter.kese@ijs.si>
  11  *              Julian Anastasov <ja@ssi.bg>
  12  *
  13  * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
  14  * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
  15  * and others.
  16  *
  17  * Changes:
  18  *      Paul `Rusty' Russell            properly handle non-linear skbs
  19  *      Harald Welte                    don't use nfcache
  20  */
  21
  22 #define KMSG_COMPONENT "IPVS"
  23 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  24
  25 #include <linux/module.h>
  26 #include <linux/kernel.h>
  27 #include <linux/ip.h>
  28 #include <linux/tcp.h>
  29 #include <linux/sctp.h>
  30 #include <linux/icmp.h>
  31 #include <linux/slab.h>
  32
  33 #include <net/ip.h>
  34 #include <net/tcp.h>
  35 #include <net/udp.h>
  36 #include <net/icmp.h>                   /* for icmp_send */
  37 #include <net/gue.h>
  38 #include <net/gre.h>
  39 #include <net/route.h>
  40 #include <net/ip6_checksum.h>
  41 #include <net/netns/generic.h>          /* net_generic() */
  42
  43 #include <linux/netfilter.h>
  44 #include <linux/netfilter_ipv4.h>
  45
  46 #ifdef CONFIG_IP_VS_IPV6
  47 #include <net/ipv6.h>
  48 #include <linux/netfilter_ipv6.h>
  49 #include <net/ip6_route.h>
  50 #endif
  51
  52 #include <net/ip_vs.h>
  53 #include <linux/indirect_call_wrapper.h>
  54
  55
  56 EXPORT_SYMBOL(register_ip_vs_scheduler);
  57 EXPORT_SYMBOL(unregister_ip_vs_scheduler);
  58 EXPORT_SYMBOL(ip_vs_proto_name);
  59 EXPORT_SYMBOL(ip_vs_conn_new);
  60 EXPORT_SYMBOL(ip_vs_conn_in_get);
  61 EXPORT_SYMBOL(ip_vs_conn_out_get);
  62 #ifdef CONFIG_IP_VS_PROTO_TCP
  63 EXPORT_SYMBOL(ip_vs_tcp_conn_listen);
  64 #endif
  65 EXPORT_SYMBOL(ip_vs_conn_put);
  66 #ifdef CONFIG_IP_VS_DEBUG
  67 EXPORT_SYMBOL(ip_vs_get_debug_level);
  68 #endif
  69 EXPORT_SYMBOL(ip_vs_new_conn_out);
  70
  71 #ifdef CONFIG_IP_VS_PROTO_TCP
  72 INDIRECT_CALLABLE_DECLARE(int
  73         tcp_snat_handler(struct sk_buff *skb, struct ip_vs_protocol *pp,
  74                          struct ip_vs_conn *cp, struct ip_vs_iphdr *iph));
  75 #endif
  76
  77 #ifdef CONFIG_IP_VS_PROTO_UDP
  78 INDIRECT_CALLABLE_DECLARE(int
  79         udp_snat_handler(struct sk_buff *skb, struct ip_vs_protocol *pp,
  80                          struct ip_vs_conn *cp, struct ip_vs_iphdr *iph));
  81 #endif
  82
  83 #if defined(CONFIG_IP_VS_PROTO_TCP) && defined(CONFIG_IP_VS_PROTO_UDP)
  84 #define SNAT_CALL(f, ...) \
  85         INDIRECT_CALL_2(f, tcp_snat_handler, udp_snat_handler, __VA_ARGS__)
  86 #elif defined(CONFIG_IP_VS_PROTO_TCP)
  87 #define SNAT_CALL(f, ...) INDIRECT_CALL_1(f, tcp_snat_handler, __VA_ARGS__)
  88 #elif defined(CONFIG_IP_VS_PROTO_UDP)
  89 #define SNAT_CALL(f, ...) INDIRECT_CALL_1(f, udp_snat_handler, __VA_ARGS__)
  90 #else
  91 #define SNAT_CALL(f, ...) f(__VA_ARGS__)
  92 #endif
  93
  94 static unsigned int ip_vs_net_id __read_mostly;
  95 /* netns cnt used for uniqueness */
  96 static atomic_t ipvs_netns_cnt = ATOMIC_INIT(0);
  97
  98 /* ID used in ICMP lookups */
  99 #define icmp_id(icmph)          (((icmph)->un).echo.id)
 100 #define icmpv6_id(icmph)        (icmph->icmp6_dataun.u_echo.identifier)
 101
 102 const char *ip_vs_proto_name(unsigned int proto)
 103 {
 104         static char buf[20];
 105
 106         switch (proto) {
 107         case IPPROTO_IP:
 108                 return "IP";
 109         case IPPROTO_UDP:
 110                 return "UDP";
 111         case IPPROTO_TCP:
 112                 return "TCP";
 113         case IPPROTO_SCTP:
 114                 return "SCTP";
 115         case IPPROTO_ICMP:
 116                 return "ICMP";
 117 #ifdef CONFIG_IP_VS_IPV6
 118         case IPPROTO_ICMPV6:
 119                 return "ICMPv6";
 120 #endif
 121         default:
 122                 sprintf(buf, "IP_%u", proto);
 123                 return buf;
 124         }
 125 }
 126
 127 void ip_vs_init_hash_table(struct list_head *table, int rows)
 128 {
 129         while (--rows >= 0)
 130                 INIT_LIST_HEAD(&table[rows]);
 131 }
 132
 133 static inline void
 134 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 135 {
 136         struct ip_vs_dest *dest = cp->dest;
 137         struct netns_ipvs *ipvs = cp->ipvs;
 138
 139         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 140                 struct ip_vs_cpu_stats *s;
 141                 struct ip_vs_service *svc;
 142
 143                 local_bh_disable();
 144
 145                 s = this_cpu_ptr(dest->stats.cpustats);
 146                 u64_stats_update_begin(&s->syncp);
 147                 s->cnt.inpkts++;
 148                 s->cnt.inbytes += skb->len;
 149                 u64_stats_update_end(&s->syncp);
 150
 151                 svc = rcu_dereference(dest->svc);
 152                 s = this_cpu_ptr(svc->stats.cpustats);
 153                 u64_stats_update_begin(&s->syncp);
 154                 s->cnt.inpkts++;
 155                 s->cnt.inbytes += skb->len;
 156                 u64_stats_update_end(&s->syncp);
 157
 158                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 159                 u64_stats_update_begin(&s->syncp);
 160                 s->cnt.inpkts++;
 161                 s->cnt.inbytes += skb->len;
 162                 u64_stats_update_end(&s->syncp);
 163
 164                 local_bh_enable();
 165         }
 166 }
 167
 168
 169 static inline void
 170 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 171 {
 172         struct ip_vs_dest *dest = cp->dest;
 173         struct netns_ipvs *ipvs = cp->ipvs;
 174
 175         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 176                 struct ip_vs_cpu_stats *s;
 177                 struct ip_vs_service *svc;
 178
 179                 local_bh_disable();
 180
 181                 s = this_cpu_ptr(dest->stats.cpustats);
 182                 u64_stats_update_begin(&s->syncp);
 183                 s->cnt.outpkts++;
 184                 s->cnt.outbytes += skb->len;
 185                 u64_stats_update_end(&s->syncp);
 186
 187                 svc = rcu_dereference(dest->svc);
 188                 s = this_cpu_ptr(svc->stats.cpustats);
 189                 u64_stats_update_begin(&s->syncp);
 190                 s->cnt.outpkts++;
 191                 s->cnt.outbytes += skb->len;
 192                 u64_stats_update_end(&s->syncp);
 193
 194                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 195                 u64_stats_update_begin(&s->syncp);
 196                 s->cnt.outpkts++;
 197                 s->cnt.outbytes += skb->len;
 198                 u64_stats_update_end(&s->syncp);
 199
 200                 local_bh_enable();
 201         }
 202 }
 203
 204
 205 static inline void
 206 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
 207 {
 208         struct netns_ipvs *ipvs = svc->ipvs;
 209         struct ip_vs_cpu_stats *s;
 210
 211         local_bh_disable();
 212
 213         s = this_cpu_ptr(cp->dest->stats.cpustats);
 214         u64_stats_update_begin(&s->syncp);
 215         s->cnt.conns++;
 216         u64_stats_update_end(&s->syncp);
 217
 218         s = this_cpu_ptr(svc->stats.cpustats);
 219         u64_stats_update_begin(&s->syncp);
 220         s->cnt.conns++;
 221         u64_stats_update_end(&s->syncp);
 222
 223         s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 224         u64_stats_update_begin(&s->syncp);
 225         s->cnt.conns++;
 226         u64_stats_update_end(&s->syncp);
 227
 228         local_bh_enable();
 229 }
 230
 231
 232 static inline void
 233 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
 234                 const struct sk_buff *skb,
 235                 struct ip_vs_proto_data *pd)
 236 {
 237         if (likely(pd->pp->state_transition))
 238                 pd->pp->state_transition(cp, direction, skb, pd);
 239 }
 240
 241 static inline int
 242 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
 243                               struct sk_buff *skb, int protocol,
 244                               const union nf_inet_addr *caddr, __be16 cport,
 245                               const union nf_inet_addr *vaddr, __be16 vport,
 246                               struct ip_vs_conn_param *p)
 247 {
 248         ip_vs_conn_fill_param(svc->ipvs, svc->af, protocol, caddr, cport, vaddr,
 249                               vport, p);
 250         p->pe = rcu_dereference(svc->pe);
 251         if (p->pe && p->pe->fill_param)
 252                 return p->pe->fill_param(p, skb);
 253
 254         return 0;
 255 }
 256
 257 /*
 258  *  IPVS persistent scheduling function
 259  *  It creates a connection entry according to its template if exists,
 260  *  or selects a server and creates a connection entry plus a template.
 261  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
 262  *  Protocols supported: TCP, UDP
 263  */
 264 static struct ip_vs_conn *
 265 ip_vs_sched_persist(struct ip_vs_service *svc,
 266                     struct sk_buff *skb, __be16 src_port, __be16 dst_port,
 267                     int *ignored, struct ip_vs_iphdr *iph)
 268 {
 269         struct ip_vs_conn *cp = NULL;
 270         struct ip_vs_dest *dest;
 271         struct ip_vs_conn *ct;
 272         __be16 dport = 0;               /* destination port to forward */
 273         unsigned int flags;
 274         struct ip_vs_conn_param param;
 275         const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
 276         union nf_inet_addr snet;        /* source network of the client,
 277                                            after masking */
 278         const union nf_inet_addr *src_addr, *dst_addr;
 279
 280         if (likely(!ip_vs_iph_inverse(iph))) {
 281                 src_addr = &iph->saddr;
 282                 dst_addr = &iph->daddr;
 283         } else {
 284                 src_addr = &iph->daddr;
 285                 dst_addr = &iph->saddr;
 286         }
 287
 288
 289         /* Mask saddr with the netmask to adjust template granularity */
 290 #ifdef CONFIG_IP_VS_IPV6
 291         if (svc->af == AF_INET6)
 292                 ipv6_addr_prefix(&snet.in6, &src_addr->in6,
 293                                  (__force __u32) svc->netmask);
 294         else
 295 #endif
 296                 snet.ip = src_addr->ip & svc->netmask;
 297
 298         IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
 299                       "mnet %s\n",
 300                       IP_VS_DBG_ADDR(svc->af, src_addr), ntohs(src_port),
 301                       IP_VS_DBG_ADDR(svc->af, dst_addr), ntohs(dst_port),
 302                       IP_VS_DBG_ADDR(svc->af, &snet));
 303
 304         /*
 305          * As far as we know, FTP is a very complicated network protocol, and
 306          * it uses control connection and data connections. For active FTP,
 307          * FTP server initialize data connection to the client, its source port
 308          * is often 20. For passive FTP, FTP server tells the clients the port
 309          * that it passively listens to,  and the client issues the data
 310          * connection. In the tunneling or direct routing mode, the load
 311          * balancer is on the client-to-server half of connection, the port
 312          * number is unknown to the load balancer. So, a conn template like
 313          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
 314          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
 315          * is created for other persistent services.
 316          */
 317         {
 318                 int protocol = iph->protocol;
 319                 const union nf_inet_addr *vaddr = dst_addr;
 320                 __be16 vport = 0;
 321
 322                 if (dst_port == svc->port) {
 323                         /* non-FTP template:
 324                          * <protocol, caddr, 0, vaddr, vport, daddr, dport>
 325                          * FTP template:
 326                          * <protocol, caddr, 0, vaddr, 0, daddr, 0>
 327                          */
 328                         if (svc->port != FTPPORT)
 329                                 vport = dst_port;
 330                 } else {
 331                         /* Note: persistent fwmark-based services and
 332                          * persistent port zero service are handled here.
 333                          * fwmark template:
 334                          * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
 335                          * port zero template:
 336                          * <protocol,caddr,0,vaddr,0,daddr,0>
 337                          */
 338                         if (svc->fwmark) {
 339                                 protocol = IPPROTO_IP;
 340                                 vaddr = &fwmark;
 341                         }
 342                 }
 343                 /* return *ignored = -1 so NF_DROP can be used */
 344                 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
 345                                                   vaddr, vport, &param) < 0) {
 346                         *ignored = -1;
 347                         return NULL;
 348                 }
 349         }
 350
 351         /* Check if a template already exists */
 352         ct = ip_vs_ct_in_get(&param);
 353         if (!ct || !ip_vs_check_template(ct, NULL)) {
 354                 struct ip_vs_scheduler *sched;
 355
 356                 /*
 357                  * No template found or the dest of the connection
 358                  * template is not available.
 359                  * return *ignored=0 i.e. ICMP and NF_DROP
 360                  */
 361                 sched = rcu_dereference(svc->scheduler);
 362                 if (sched) {
 363                         /* read svc->sched_data after svc->scheduler */
 364                         smp_rmb();
 365                         dest = sched->schedule(svc, skb, iph);
 366                 } else {
 367                         dest = NULL;
 368                 }
 369                 if (!dest) {
 370                         IP_VS_DBG(1, "p-schedule: no dest found.\n");
 371                         kfree(param.pe_data);
 372                         *ignored = 0;
 373                         return NULL;
 374                 }
 375
 376                 if (dst_port == svc->port && svc->port != FTPPORT)
 377                         dport = dest->port;
 378
 379                 /* Create a template
 380                  * This adds param.pe_data to the template,
 381                  * and thus param.pe_data will be destroyed
 382                  * when the template expires */
 383                 ct = ip_vs_conn_new(&param, dest->af, &dest->addr, dport,
 384                                     IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
 385                 if (ct == NULL) {
 386                         kfree(param.pe_data);
 387                         *ignored = -1;
 388                         return NULL;
 389                 }
 390
 391                 ct->timeout = svc->timeout;
 392         } else {
 393                 /* set destination with the found template */
 394                 dest = ct->dest;
 395                 kfree(param.pe_data);
 396         }
 397
 398         dport = dst_port;
 399         if (dport == svc->port && dest->port)
 400                 dport = dest->port;
 401
 402         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 403                  && iph->protocol == IPPROTO_UDP) ?
 404                 IP_VS_CONN_F_ONE_PACKET : 0;
 405
 406         /*
 407          *    Create a new connection according to the template
 408          */
 409         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol, src_addr,
 410                               src_port, dst_addr, dst_port, &param);
 411
 412         cp = ip_vs_conn_new(&param, dest->af, &dest->addr, dport, flags, dest,
 413                             skb->mark);
 414         if (cp == NULL) {
 415                 ip_vs_conn_put(ct);
 416                 *ignored = -1;
 417                 return NULL;
 418         }
 419
 420         /*
 421          *    Add its control
 422          */
 423         ip_vs_control_add(cp, ct);
 424         ip_vs_conn_put(ct);
 425
 426         ip_vs_conn_stats(cp, svc);
 427         return cp;
 428 }
 429
 430
 431 /*
 432  *  IPVS main scheduling function
 433  *  It selects a server according to the virtual service, and
 434  *  creates a connection entry.
 435  *  Protocols supported: TCP, UDP
 436  *
 437  *  Usage of *ignored
 438  *
 439  * 1 :   protocol tried to schedule (eg. on SYN), found svc but the
 440  *       svc/scheduler decides that this packet should be accepted with
 441  *       NF_ACCEPT because it must not be scheduled.
 442  *
 443  * 0 :   scheduler can not find destination, so try bypass or
 444  *       return ICMP and then NF_DROP (ip_vs_leave).
 445  *
 446  * -1 :  scheduler tried to schedule but fatal error occurred, eg.
 447  *       ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
 448  *       failure such as missing Call-ID, ENOMEM on skb_linearize
 449  *       or pe_data. In this case we should return NF_DROP without
 450  *       any attempts to send ICMP with ip_vs_leave.
 451  */
 452 struct ip_vs_conn *
 453 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
 454                struct ip_vs_proto_data *pd, int *ignored,
 455                struct ip_vs_iphdr *iph)
 456 {
 457         struct ip_vs_protocol *pp = pd->pp;
 458         struct ip_vs_conn *cp = NULL;
 459         struct ip_vs_scheduler *sched;
 460         struct ip_vs_dest *dest;
 461         __be16 _ports[2], *pptr, cport, vport;
 462         const void *caddr, *vaddr;
 463         unsigned int flags;
 464
 465         *ignored = 1;
 466         /*
 467          * IPv6 frags, only the first hit here.
 468          */
 469         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports);
 470         if (pptr == NULL)
 471                 return NULL;
 472
 473         if (likely(!ip_vs_iph_inverse(iph))) {
 474                 cport = pptr[0];
 475                 caddr = &iph->saddr;
 476                 vport = pptr[1];
 477                 vaddr = &iph->daddr;
 478         } else {
 479                 cport = pptr[1];
 480                 caddr = &iph->daddr;
 481                 vport = pptr[0];
 482                 vaddr = &iph->saddr;
 483         }
 484
 485         /*
 486          * FTPDATA needs this check when using local real server.
 487          * Never schedule Active FTPDATA connections from real server.
 488          * For LVS-NAT they must be already created. For other methods
 489          * with persistence the connection is created on SYN+ACK.
 490          */
 491         if (cport == FTPDATA) {
 492                 IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
 493                               "Not scheduling FTPDATA");
 494                 return NULL;
 495         }
 496
 497         /*
 498          *    Do not schedule replies from local real server.
 499          */
 500         if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK)) {
 501                 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
 502                 cp = INDIRECT_CALL_1(pp->conn_in_get,
 503                                      ip_vs_conn_in_get_proto, svc->ipvs,
 504                                      svc->af, skb, iph);
 505                 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
 506
 507                 if (cp) {
 508                         IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
 509                                       "Not scheduling reply for existing"
 510                                       " connection");
 511                         __ip_vs_conn_put(cp);
 512                         return NULL;
 513                 }
 514         }
 515
 516         /*
 517          *    Persistent service
 518          */
 519         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 520                 return ip_vs_sched_persist(svc, skb, cport, vport, ignored,
 521                                            iph);
 522
 523         *ignored = 0;
 524
 525         /*
 526          *    Non-persistent service
 527          */
 528         if (!svc->fwmark && vport != svc->port) {
 529                 if (!svc->port)
 530                         pr_err("Schedule: port zero only supported "
 531                                "in persistent services, "
 532                                "check your ipvs configuration\n");
 533                 return NULL;
 534         }
 535
 536         sched = rcu_dereference(svc->scheduler);
 537         if (sched) {
 538                 /* read svc->sched_data after svc->scheduler */
 539                 smp_rmb();
 540                 dest = sched->schedule(svc, skb, iph);
 541         } else {
 542                 dest = NULL;
 543         }
 544         if (dest == NULL) {
 545                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 546                 return NULL;
 547         }
 548
 549         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 550                  && iph->protocol == IPPROTO_UDP) ?
 551                 IP_VS_CONN_F_ONE_PACKET : 0;
 552
 553         /*
 554          *    Create a connection entry.
 555          */
 556         {
 557                 struct ip_vs_conn_param p;
 558
 559                 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
 560                                       caddr, cport, vaddr, vport, &p);
 561                 cp = ip_vs_conn_new(&p, dest->af, &dest->addr,
 562                                     dest->port ? dest->port : vport,
 563                                     flags, dest, skb->mark);
 564                 if (!cp) {
 565                         *ignored = -1;
 566                         return NULL;
 567                 }
 568         }
 569
 570         IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
 571                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
 572                       ip_vs_fwd_tag(cp),
 573                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 574                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 575                       IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
 576                       cp->flags, refcount_read(&cp->refcnt));
 577
 578         ip_vs_conn_stats(cp, svc);
 579         return cp;
 580 }
 581
 582 static inline int ip_vs_addr_is_unicast(struct net *net, int af,
 583                                         union nf_inet_addr *addr)
 584 {
 585 #ifdef CONFIG_IP_VS_IPV6
 586         if (af == AF_INET6)
 587                 return ipv6_addr_type(&addr->in6) & IPV6_ADDR_UNICAST;
 588 #endif
 589         return (inet_addr_type(net, addr->ip) == RTN_UNICAST);
 590 }
 591
 592 /*
 593  *  Pass or drop the packet.
 594  *  Called by ip_vs_in, when the virtual service is available but
 595  *  no destination is available for a new connection.
 596  */
 597 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 598                 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
 599 {
 600         __be16 _ports[2], *pptr, dport;
 601         struct netns_ipvs *ipvs = svc->ipvs;
 602         struct net *net = ipvs->net;
 603
 604         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports);
 605         if (!pptr)
 606                 return NF_DROP;
 607         dport = likely(!ip_vs_iph_inverse(iph)) ? pptr[1] : pptr[0];
 608
 609         /* if it is fwmark-based service, the cache_bypass sysctl is up
 610            and the destination is a non-local unicast, then create
 611            a cache_bypass connection entry */
 612         if (sysctl_cache_bypass(ipvs) && svc->fwmark &&
 613             !(iph->hdr_flags & (IP_VS_HDR_INVERSE | IP_VS_HDR_ICMP)) &&
 614             ip_vs_addr_is_unicast(net, svc->af, &iph->daddr)) {
 615                 int ret;
 616                 struct ip_vs_conn *cp;
 617                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
 618                                       iph->protocol == IPPROTO_UDP) ?
 619                                       IP_VS_CONN_F_ONE_PACKET : 0;
 620                 union nf_inet_addr daddr = { .all = { 0, 0, 0, 0 } };
 621
 622                 /* create a new connection entry */
 623                 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
 624                 {
 625                         struct ip_vs_conn_param p;
 626                         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
 627                                               &iph->saddr, pptr[0],
 628                                               &iph->daddr, pptr[1], &p);
 629                         cp = ip_vs_conn_new(&p, svc->af, &daddr, 0,
 630                                             IP_VS_CONN_F_BYPASS | flags,
 631                                             NULL, skb->mark);
 632                         if (!cp)
 633                                 return NF_DROP;
 634                 }
 635
 636                 /* statistics */
 637                 ip_vs_in_stats(cp, skb);
 638
 639                 /* set state */
 640                 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
 641
 642                 /* transmit the first SYN packet */
 643                 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
 644                 /* do not touch skb anymore */
 645
 646                 if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) && cp->control)
 647                         atomic_inc(&cp->control->in_pkts);
 648                 else
 649                         atomic_inc(&cp->in_pkts);
 650                 ip_vs_conn_put(cp);
 651                 return ret;
 652         }
 653
 654         /*
 655          * When the virtual ftp service is presented, packets destined
 656          * for other services on the VIP may get here (except services
 657          * listed in the ipvs table), pass the packets, because it is
 658          * not ipvs job to decide to drop the packets.
 659          */
 660         if (svc->port == FTPPORT && dport != FTPPORT)
 661                 return NF_ACCEPT;
 662
 663         if (unlikely(ip_vs_iph_icmp(iph)))
 664                 return NF_DROP;
 665
 666         /*
 667          * Notify the client that the destination is unreachable, and
 668          * release the socket buffer.
 669          * Since it is in IP layer, the TCP socket is not actually
 670          * created, the TCP RST packet cannot be sent, instead that
 671          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 672          */
 673 #ifdef CONFIG_IP_VS_IPV6
 674         if (svc->af == AF_INET6) {
 675                 if (!skb->dev)
 676                         skb->dev = net->loopback_dev;
 677                 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 678         } else
 679 #endif
 680                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 681
 682         return NF_DROP;
 683 }
 684
 685 #ifdef CONFIG_SYSCTL
 686
 687 static int sysctl_snat_reroute(struct netns_ipvs *ipvs)
 688 {
 689         return ipvs->sysctl_snat_reroute;
 690 }
 691
 692 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs)
 693 {
 694         return ipvs->sysctl_nat_icmp_send;
 695 }
 696
 697 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
 698 {
 699         return ipvs->sysctl_expire_nodest_conn;
 700 }
 701
 702 #else
 703
 704 static int sysctl_snat_reroute(struct netns_ipvs *ipvs) { return 0; }
 705 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs) { return 0; }
 706 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
 707
 708 #endif
 709
 710 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 711 {
 712         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 713 }
 714
 715 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
 716 {
 717         if (NF_INET_LOCAL_IN == hooknum)
 718                 return IP_DEFRAG_VS_IN;
 719         if (NF_INET_FORWARD == hooknum)
 720                 return IP_DEFRAG_VS_FWD;
 721         return IP_DEFRAG_VS_OUT;
 722 }
 723
 724 static inline int ip_vs_gather_frags(struct netns_ipvs *ipvs,
 725                                      struct sk_buff *skb, u_int32_t user)
 726 {
 727         int err;
 728
 729         local_bh_disable();
 730         err = ip_defrag(ipvs->net, skb, user);
 731         local_bh_enable();
 732         if (!err)
 733                 ip_send_check(ip_hdr(skb));
 734
 735         return err;
 736 }
 737
 738 static int ip_vs_route_me_harder(struct netns_ipvs *ipvs, int af,
 739                                  struct sk_buff *skb, unsigned int hooknum)
 740 {
 741         if (!sysctl_snat_reroute(ipvs))
 742                 return 0;
 743         /* Reroute replies only to remote clients (FORWARD and LOCAL_OUT) */
 744         if (NF_INET_LOCAL_IN == hooknum)
 745                 return 0;
 746 #ifdef CONFIG_IP_VS_IPV6
 747         if (af == AF_INET6) {
 748                 struct dst_entry *dst = skb_dst(skb);
 749
 750                 if (dst->dev && !(dst->dev->flags & IFF_LOOPBACK) &&
 751                     ip6_route_me_harder(ipvs->net, skb) != 0)
 752                         return 1;
 753         } else
 754 #endif
 755                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
 756                     ip_route_me_harder(ipvs->net, skb, RTN_LOCAL) != 0)
 757                         return 1;
 758
 759         return 0;
 760 }
 761
 762 /*
 763  * Packet has been made sufficiently writable in caller
 764  * - inout: 1=in->out, 0=out->in
 765  */
 766 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 767                     struct ip_vs_conn *cp, int inout)
 768 {
 769         struct iphdr *iph        = ip_hdr(skb);
 770         unsigned int icmp_offset = iph->ihl*4;
 771         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
 772                                                       icmp_offset);
 773         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 774
 775         if (inout) {
 776                 iph->saddr = cp->vaddr.ip;
 777                 ip_send_check(iph);
 778                 ciph->daddr = cp->vaddr.ip;
 779                 ip_send_check(ciph);
 780         } else {
 781                 iph->daddr = cp->daddr.ip;
 782                 ip_send_check(iph);
 783                 ciph->saddr = cp->daddr.ip;
 784                 ip_send_check(ciph);
 785         }
 786
 787         /* the TCP/UDP/SCTP port */
 788         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
 789             IPPROTO_SCTP == ciph->protocol) {
 790                 __be16 *ports = (void *)ciph + ciph->ihl*4;
 791
 792                 if (inout)
 793                         ports[1] = cp->vport;
 794                 else
 795                         ports[0] = cp->dport;
 796         }
 797
 798         /* And finally the ICMP checksum */
 799         icmph->checksum = 0;
 800         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 801         skb->ip_summed = CHECKSUM_UNNECESSARY;
 802
 803         if (inout)
 804                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 805                         "Forwarding altered outgoing ICMP");
 806         else
 807                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 808                         "Forwarding altered incoming ICMP");
 809 }
 810
 811 #ifdef CONFIG_IP_VS_IPV6
 812 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
 813                     struct ip_vs_conn *cp, int inout)
 814 {
 815         struct ipv6hdr *iph      = ipv6_hdr(skb);
 816         unsigned int icmp_offset = 0;
 817         unsigned int offs        = 0; /* header offset*/
 818         int protocol;
 819         struct icmp6hdr *icmph;
 820         struct ipv6hdr *ciph;
 821         unsigned short fragoffs;
 822
 823         ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
 824         icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
 825         offs = icmp_offset + sizeof(struct icmp6hdr);
 826         ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
 827
 828         protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
 829
 830         if (inout) {
 831                 iph->saddr = cp->vaddr.in6;
 832                 ciph->daddr = cp->vaddr.in6;
 833         } else {
 834                 iph->daddr = cp->daddr.in6;
 835                 ciph->saddr = cp->daddr.in6;
 836         }
 837
 838         /* the TCP/UDP/SCTP port */
 839         if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 840                           IPPROTO_SCTP == protocol)) {
 841                 __be16 *ports = (void *)(skb_network_header(skb) + offs);
 842
 843                 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
 844                               ntohs(inout ? ports[1] : ports[0]),
 845                               ntohs(inout ? cp->vport : cp->dport));
 846                 if (inout)
 847                         ports[1] = cp->vport;
 848                 else
 849                         ports[0] = cp->dport;
 850         }
 851
 852         /* And finally the ICMP checksum */
 853         icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
 854                                               skb->len - icmp_offset,
 855                                               IPPROTO_ICMPV6, 0);
 856         skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
 857         skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
 858         skb->ip_summed = CHECKSUM_PARTIAL;
 859
 860         if (inout)
 861                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 862                               (void *)ciph - (void *)iph,
 863                               "Forwarding altered outgoing ICMPv6");
 864         else
 865                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 866                               (void *)ciph - (void *)iph,
 867                               "Forwarding altered incoming ICMPv6");
 868 }
 869 #endif
 870
 871 /* Handle relevant response ICMP messages - forward to the right
 872  * destination host.
 873  */
 874 static int handle_response_icmp(int af, struct sk_buff *skb,
 875                                 union nf_inet_addr *snet,
 876                                 __u8 protocol, struct ip_vs_conn *cp,
 877                                 struct ip_vs_protocol *pp,
 878                                 unsigned int offset, unsigned int ihl,
 879                                 unsigned int hooknum)
 880 {
 881         unsigned int verdict = NF_DROP;
 882
 883         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
 884                 goto ignore_cp;
 885
 886         /* Ensure the checksum is correct */
 887         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 888                 /* Failed checksum! */
 889                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
 890                               IP_VS_DBG_ADDR(af, snet));
 891                 goto out;
 892         }
 893
 894         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 895             IPPROTO_SCTP == protocol)
 896                 offset += 2 * sizeof(__u16);
 897         if (skb_ensure_writable(skb, offset))
 898                 goto out;
 899
 900 #ifdef CONFIG_IP_VS_IPV6
 901         if (af == AF_INET6)
 902                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
 903         else
 904 #endif
 905                 ip_vs_nat_icmp(skb, pp, cp, 1);
 906
 907         if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
 908                 goto out;
 909
 910         /* do the statistics and put it back */
 911         ip_vs_out_stats(cp, skb);
 912
 913         skb->ipvs_property = 1;
 914         if (!(cp->flags & IP_VS_CONN_F_NFCT))
 915                 ip_vs_notrack(skb);
 916         else
 917                 ip_vs_update_conntrack(skb, cp, 0);
 918
 919 ignore_cp:
 920         verdict = NF_ACCEPT;
 921
 922 out:
 923         __ip_vs_conn_put(cp);
 924
 925         return verdict;
 926 }
 927
 928 /*
 929  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 930  *      Find any that might be relevant, check against existing connections.
 931  *      Currently handles error types - unreachable, quench, ttl exceeded.
 932  */
 933 static int ip_vs_out_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb,
 934                           int *related, unsigned int hooknum)
 935 {
 936         struct iphdr *iph;
 937         struct icmphdr  _icmph, *ic;
 938         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 939         struct ip_vs_iphdr ciph;
 940         struct ip_vs_conn *cp;
 941         struct ip_vs_protocol *pp;
 942         unsigned int offset, ihl;
 943         union nf_inet_addr snet;
 944
 945         *related = 1;
 946
 947         /* reassemble IP fragments */
 948         if (ip_is_fragment(ip_hdr(skb))) {
 949                 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
 950                         return NF_STOLEN;
 951         }
 952
 953         iph = ip_hdr(skb);
 954         offset = ihl = iph->ihl * 4;
 955         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 956         if (ic == NULL)
 957                 return NF_DROP;
 958
 959         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
 960                   ic->type, ntohs(icmp_id(ic)),
 961                   &iph->saddr, &iph->daddr);
 962
 963         /*
 964          * Work through seeing if this is for us.
 965          * These checks are supposed to be in an order that means easy
 966          * things are checked first to speed up processing.... however
 967          * this means that some packets will manage to get a long way
 968          * down this stack and then be rejected, but that's life.
 969          */
 970         if ((ic->type != ICMP_DEST_UNREACH) &&
 971             (ic->type != ICMP_SOURCE_QUENCH) &&
 972             (ic->type != ICMP_TIME_EXCEEDED)) {
 973                 *related = 0;
 974                 return NF_ACCEPT;
 975         }
 976
 977         /* Now find the contained IP header */
 978         offset += sizeof(_icmph);
 979         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 980         if (cih == NULL)
 981                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 982
 983         pp = ip_vs_proto_get(cih->protocol);
 984         if (!pp)
 985                 return NF_ACCEPT;
 986
 987         /* Is the embedded protocol header present? */
 988         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 989                      pp->dont_defrag))
 990                 return NF_ACCEPT;
 991
 992         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
 993                       "Checking outgoing ICMP for");
 994
 995         ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, true, &ciph);
 996
 997         /* The embedded headers contain source and dest in reverse order */
 998         cp = INDIRECT_CALL_1(pp->conn_out_get, ip_vs_conn_out_get_proto,
 999                              ipvs, AF_INET, skb, &ciph);
1000         if (!cp)
1001                 return NF_ACCEPT;
1002
1003         snet.ip = iph->saddr;
1004         return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
1005                                     pp, ciph.len, ihl, hooknum);
1006 }
1007
1008 #ifdef CONFIG_IP_VS_IPV6
1009 static int ip_vs_out_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
1010                              int *related,  unsigned int hooknum,
1011                              struct ip_vs_iphdr *ipvsh)
1012 {
1013         struct icmp6hdr _icmph, *ic;
1014         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1015         struct ip_vs_conn *cp;
1016         struct ip_vs_protocol *pp;
1017         union nf_inet_addr snet;
1018         unsigned int offset;
1019
1020         *related = 1;
1021         ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph);
1022         if (ic == NULL)
1023                 return NF_DROP;
1024
1025         /*
1026          * Work through seeing if this is for us.
1027          * These checks are supposed to be in an order that means easy
1028          * things are checked first to speed up processing.... however
1029          * this means that some packets will manage to get a long way
1030          * down this stack and then be rejected, but that's life.
1031          */
1032         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1033                 *related = 0;
1034                 return NF_ACCEPT;
1035         }
1036         /* Fragment header that is before ICMP header tells us that:
1037          * it's not an error message since they can't be fragmented.
1038          */
1039         if (ipvsh->flags & IP6_FH_F_FRAG)
1040                 return NF_DROP;
1041
1042         IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1043                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1044                   &ipvsh->saddr, &ipvsh->daddr);
1045
1046         if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, ipvsh->len + sizeof(_icmph),
1047                                      true, &ciph))
1048                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1049
1050         pp = ip_vs_proto_get(ciph.protocol);
1051         if (!pp)
1052                 return NF_ACCEPT;
1053
1054         /* The embedded headers contain source and dest in reverse order */
1055         cp = INDIRECT_CALL_1(pp->conn_out_get, ip_vs_conn_out_get_proto,
1056                              ipvs, AF_INET6, skb, &ciph);
1057         if (!cp)
1058                 return NF_ACCEPT;
1059
1060         snet.in6 = ciph.saddr.in6;
1061         offset = ciph.len;
1062         return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
1063                                     pp, offset, sizeof(struct ipv6hdr),
1064                                     hooknum);
1065 }
1066 #endif
1067
1068 /*
1069  * Check if sctp chunc is ABORT chunk
1070  */
1071 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
1072 {
1073         struct sctp_chunkhdr *sch, schunk;
1074         sch = skb_header_pointer(skb, nh_len + sizeof(struct sctphdr),
1075                                  sizeof(schunk), &schunk);
1076         if (sch == NULL)
1077                 return 0;
1078         if (sch->type == SCTP_CID_ABORT)
1079                 return 1;
1080         return 0;
1081 }
1082
1083 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
1084 {
1085         struct tcphdr _tcph, *th;
1086
1087         th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
1088         if (th == NULL)
1089                 return 0;
1090         return th->rst;
1091 }
1092
1093 static inline bool is_new_conn(const struct sk_buff *skb,
1094                                struct ip_vs_iphdr *iph)
1095 {
1096         switch (iph->protocol) {
1097         case IPPROTO_TCP: {
1098                 struct tcphdr _tcph, *th;
1099
1100                 th = skb_header_pointer(skb, iph->len, sizeof(_tcph), &_tcph);
1101                 if (th == NULL)
1102                         return false;
1103                 return th->syn;
1104         }
1105         case IPPROTO_SCTP: {
1106                 struct sctp_chunkhdr *sch, schunk;
1107
1108                 sch = skb_header_pointer(skb, iph->len + sizeof(struct sctphdr),
1109                                          sizeof(schunk), &schunk);
1110                 if (sch == NULL)
1111                         return false;
1112                 return sch->type == SCTP_CID_INIT;
1113         }
1114         default:
1115                 return false;
1116         }
1117 }
1118
1119 static inline bool is_new_conn_expected(const struct ip_vs_conn *cp,
1120                                         int conn_reuse_mode)
1121 {
1122         /* Controlled (FTP DATA or persistence)? */
1123         if (cp->control)
1124                 return false;
1125
1126         switch (cp->protocol) {
1127         case IPPROTO_TCP:
1128                 return (cp->state == IP_VS_TCP_S_TIME_WAIT) ||
1129                        (cp->state == IP_VS_TCP_S_CLOSE) ||
1130                         ((conn_reuse_mode & 2) &&
1131                          (cp->state == IP_VS_TCP_S_FIN_WAIT) &&
1132                          (cp->flags & IP_VS_CONN_F_NOOUTPUT));
1133         case IPPROTO_SCTP:
1134                 return cp->state == IP_VS_SCTP_S_CLOSED;
1135         default:
1136                 return false;
1137         }
1138 }
1139
1140 /* Generic function to create new connections for outgoing RS packets
1141  *
1142  * Pre-requisites for successful connection creation:
1143  * 1) Virtual Service is NOT fwmark based:
1144  *    In fwmark-VS actual vaddr and vport are unknown to IPVS
1145  * 2) Real Server and Virtual Service were NOT configured without port:
1146  *    This is to allow match of different VS to the same RS ip-addr
1147  */
1148 struct ip_vs_conn *ip_vs_new_conn_out(struct ip_vs_service *svc,
1149                                       struct ip_vs_dest *dest,
1150                                       struct sk_buff *skb,
1151                                       const struct ip_vs_iphdr *iph,
1152                                       __be16 dport,
1153                                       __be16 cport)
1154 {
1155         struct ip_vs_conn_param param;
1156         struct ip_vs_conn *ct = NULL, *cp = NULL;
1157         const union nf_inet_addr *vaddr, *daddr, *caddr;
1158         union nf_inet_addr snet;
1159         __be16 vport;
1160         unsigned int flags;
1161
1162         EnterFunction(12);
1163         vaddr = &svc->addr;
1164         vport = svc->port;
1165         daddr = &iph->saddr;
1166         caddr = &iph->daddr;
1167
1168         /* check pre-requisites are satisfied */
1169         if (svc->fwmark)
1170                 return NULL;
1171         if (!vport || !dport)
1172                 return NULL;
1173
1174         /* for persistent service first create connection template */
1175         if (svc->flags & IP_VS_SVC_F_PERSISTENT) {
1176                 /* apply netmask the same way ingress-side does */
1177 #ifdef CONFIG_IP_VS_IPV6
1178                 if (svc->af == AF_INET6)
1179                         ipv6_addr_prefix(&snet.in6, &caddr->in6,
1180                                          (__force __u32)svc->netmask);
1181                 else
1182 #endif
1183                         snet.ip = caddr->ip & svc->netmask;
1184                 /* fill params and create template if not existent */
1185                 if (ip_vs_conn_fill_param_persist(svc, skb, iph->protocol,
1186                                                   &snet, 0, vaddr,
1187                                                   vport, &param) < 0)
1188                         return NULL;
1189                 ct = ip_vs_ct_in_get(&param);
1190                 /* check if template exists and points to the same dest */
1191                 if (!ct || !ip_vs_check_template(ct, dest)) {
1192                         ct = ip_vs_conn_new(&param, dest->af, daddr, dport,
1193                                             IP_VS_CONN_F_TEMPLATE, dest, 0);
1194                         if (!ct) {
1195                                 kfree(param.pe_data);
1196                                 return NULL;
1197                         }
1198                         ct->timeout = svc->timeout;
1199                 } else {
1200                         kfree(param.pe_data);
1201                 }
1202         }
1203
1204         /* connection flags */
1205         flags = ((svc->flags & IP_VS_SVC_F_ONEPACKET) &&
1206                  iph->protocol == IPPROTO_UDP) ? IP_VS_CONN_F_ONE_PACKET : 0;
1207         /* create connection */
1208         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
1209                               caddr, cport, vaddr, vport, &param);
1210         cp = ip_vs_conn_new(&param, dest->af, daddr, dport, flags, dest, 0);
1211         if (!cp) {
1212                 if (ct)
1213                         ip_vs_conn_put(ct);
1214                 return NULL;
1215         }
1216         if (ct) {
1217                 ip_vs_control_add(cp, ct);
1218                 ip_vs_conn_put(ct);
1219         }
1220         ip_vs_conn_stats(cp, svc);
1221
1222         /* return connection (will be used to handle outgoing packet) */
1223         IP_VS_DBG_BUF(6, "New connection RS-initiated:%c c:%s:%u v:%s:%u "
1224                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
1225                       ip_vs_fwd_tag(cp),
1226                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
1227                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
1228                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
1229                       cp->flags, refcount_read(&cp->refcnt));
1230         LeaveFunction(12);
1231         return cp;
1232 }
1233
1234 /* Handle outgoing packets which are considered requests initiated by
1235  * real servers, so that subsequent responses from external client can be
1236  * routed to the right real server.
1237  * Used also for outgoing responses in OPS mode.
1238  *
1239  * Connection management is handled by persistent-engine specific callback.
1240  */
1241 static struct ip_vs_conn *__ip_vs_rs_conn_out(unsigned int hooknum,
1242                                               struct netns_ipvs *ipvs,
1243                                               int af, struct sk_buff *skb,
1244                                               const struct ip_vs_iphdr *iph)
1245 {
1246         struct ip_vs_dest *dest;
1247         struct ip_vs_conn *cp = NULL;
1248         __be16 _ports[2], *pptr;
1249
1250         if (hooknum == NF_INET_LOCAL_IN)
1251                 return NULL;
1252
1253         pptr = frag_safe_skb_hp(skb, iph->len,
1254                                 sizeof(_ports), _ports);
1255         if (!pptr)
1256                 return NULL;
1257
1258         dest = ip_vs_find_real_service(ipvs, af, iph->protocol,
1259                                        &iph->saddr, pptr[0]);
1260         if (dest) {
1261                 struct ip_vs_service *svc;
1262                 struct ip_vs_pe *pe;
1263
1264                 svc = rcu_dereference(dest->svc);
1265                 if (svc) {
1266                         pe = rcu_dereference(svc->pe);
1267                         if (pe && pe->conn_out)
1268                                 cp = pe->conn_out(svc, dest, skb, iph,
1269                                                   pptr[0], pptr[1]);
1270                 }
1271         }
1272
1273         return cp;
1274 }
1275
1276 /* Handle response packets: rewrite addresses and send away...
1277  */
1278 static unsigned int
1279 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1280                 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph,
1281                 unsigned int hooknum)
1282 {
1283         struct ip_vs_protocol *pp = pd->pp;
1284
1285         IP_VS_DBG_PKT(11, af, pp, skb, iph->off, "Outgoing packet");
1286
1287         if (skb_ensure_writable(skb, iph->len))
1288                 goto drop;
1289
1290         /* mangle the packet */
1291         if (pp->snat_handler &&
1292             !SNAT_CALL(pp->snat_handler, skb, pp, cp, iph))
1293                 goto drop;
1294
1295 #ifdef CONFIG_IP_VS_IPV6
1296         if (af == AF_INET6)
1297                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1298         else
1299 #endif
1300         {
1301                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1302                 ip_send_check(ip_hdr(skb));
1303         }
1304
1305         /*
1306          * nf_iterate does not expect change in the skb->dst->dev.
1307          * It looks like it is not fatal to enable this code for hooks
1308          * where our handlers are at the end of the chain list and
1309          * when all next handlers use skb->dst->dev and not outdev.
1310          * It will definitely route properly the inout NAT traffic
1311          * when multiple paths are used.
1312          */
1313
1314         /* For policy routing, packets originating from this
1315          * machine itself may be routed differently to packets
1316          * passing through.  We want this packet to be routed as
1317          * if it came from this machine itself.  So re-compute
1318          * the routing information.
1319          */
1320         if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
1321                 goto drop;
1322
1323         IP_VS_DBG_PKT(10, af, pp, skb, iph->off, "After SNAT");
1324
1325         ip_vs_out_stats(cp, skb);
1326         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1327         skb->ipvs_property = 1;
1328         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1329                 ip_vs_notrack(skb);
1330         else
1331                 ip_vs_update_conntrack(skb, cp, 0);
1332         ip_vs_conn_put(cp);
1333
1334         LeaveFunction(11);
1335         return NF_ACCEPT;
1336
1337 drop:
1338         ip_vs_conn_put(cp);
1339         kfree_skb(skb);
1340         LeaveFunction(11);
1341         return NF_STOLEN;
1342 }
1343
1344 /*
1345  *      Check if outgoing packet belongs to the established ip_vs_conn.
1346  */
1347 static unsigned int
1348 ip_vs_out(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1349 {
1350         struct ip_vs_iphdr iph;
1351         struct ip_vs_protocol *pp;
1352         struct ip_vs_proto_data *pd;
1353         struct ip_vs_conn *cp;
1354         struct sock *sk;
1355
1356         EnterFunction(11);
1357
1358         /* Already marked as IPVS request or reply? */
1359         if (skb->ipvs_property)
1360                 return NF_ACCEPT;
1361
1362         sk = skb_to_full_sk(skb);
1363         /* Bad... Do not break raw sockets */
1364         if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
1365                      af == AF_INET)) {
1366
1367                 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
1368                         return NF_ACCEPT;
1369         }
1370
1371         if (unlikely(!skb_dst(skb)))
1372                 return NF_ACCEPT;
1373
1374         if (!ipvs->enable)
1375                 return NF_ACCEPT;
1376
1377         ip_vs_fill_iph_skb(af, skb, false, &iph);
1378 #ifdef CONFIG_IP_VS_IPV6
1379         if (af == AF_INET6) {
1380                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1381                         int related;
1382                         int verdict = ip_vs_out_icmp_v6(ipvs, skb, &related,
1383                                                         hooknum, &iph);
1384
1385                         if (related)
1386                                 return verdict;
1387                 }
1388         } else
1389 #endif
1390                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1391                         int related;
1392                         int verdict = ip_vs_out_icmp(ipvs, skb, &related, hooknum);
1393
1394                         if (related)
1395                                 return verdict;
1396                 }
1397
1398         pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1399         if (unlikely(!pd))
1400                 return NF_ACCEPT;
1401         pp = pd->pp;
1402
1403         /* reassemble IP fragments */
1404 #ifdef CONFIG_IP_VS_IPV6
1405         if (af == AF_INET)
1406 #endif
1407                 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1408                         if (ip_vs_gather_frags(ipvs, skb,
1409                                                ip_vs_defrag_user(hooknum)))
1410                                 return NF_STOLEN;
1411
1412                         ip_vs_fill_iph_skb(AF_INET, skb, false, &iph);
1413                 }
1414
1415         /*
1416          * Check if the packet belongs to an existing entry
1417          */
1418         cp = INDIRECT_CALL_1(pp->conn_out_get, ip_vs_conn_out_get_proto,
1419                              ipvs, af, skb, &iph);
1420
1421         if (likely(cp)) {
1422                 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
1423                         goto ignore_cp;
1424                 return handle_response(af, skb, pd, cp, &iph, hooknum);
1425         }
1426
1427         /* Check for real-server-started requests */
1428         if (atomic_read(&ipvs->conn_out_counter)) {
1429                 /* Currently only for UDP:
1430                  * connection oriented protocols typically use
1431                  * ephemeral ports for outgoing connections, so
1432                  * related incoming responses would not match any VS
1433                  */
1434                 if (pp->protocol == IPPROTO_UDP) {
1435                         cp = __ip_vs_rs_conn_out(hooknum, ipvs, af, skb, &iph);
1436                         if (likely(cp))
1437                                 return handle_response(af, skb, pd, cp, &iph,
1438                                                        hooknum);
1439                 }
1440         }
1441
1442         if (sysctl_nat_icmp_send(ipvs) &&
1443             (pp->protocol == IPPROTO_TCP ||
1444              pp->protocol == IPPROTO_UDP ||
1445              pp->protocol == IPPROTO_SCTP)) {
1446                 __be16 _ports[2], *pptr;
1447
1448                 pptr = frag_safe_skb_hp(skb, iph.len,
1449                                          sizeof(_ports), _ports);
1450                 if (pptr == NULL)
1451                         return NF_ACCEPT;       /* Not for me */
1452                 if (ip_vs_has_real_service(ipvs, af, iph.protocol, &iph.saddr,
1453                                            pptr[0])) {
1454                         /*
1455                          * Notify the real server: there is no
1456                          * existing entry if it is not RST
1457                          * packet or not TCP packet.
1458                          */
1459                         if ((iph.protocol != IPPROTO_TCP &&
1460                              iph.protocol != IPPROTO_SCTP)
1461                              || ((iph.protocol == IPPROTO_TCP
1462                                   && !is_tcp_reset(skb, iph.len))
1463                                  || (iph.protocol == IPPROTO_SCTP
1464                                         && !is_sctp_abort(skb,
1465                                                 iph.len)))) {
1466 #ifdef CONFIG_IP_VS_IPV6
1467                                 if (af == AF_INET6) {
1468                                         if (!skb->dev)
1469                                                 skb->dev = ipvs->net->loopback_dev;
1470                                         icmpv6_send(skb,
1471                                                     ICMPV6_DEST_UNREACH,
1472                                                     ICMPV6_PORT_UNREACH,
1473                                                     0);
1474                                 } else
1475 #endif
1476                                         icmp_send(skb,
1477                                                   ICMP_DEST_UNREACH,
1478                                                   ICMP_PORT_UNREACH, 0);
1479                                 return NF_DROP;
1480                         }
1481                 }
1482         }
1483
1484 out:
1485         IP_VS_DBG_PKT(12, af, pp, skb, iph.off,
1486                       "ip_vs_out: packet continues traversal as normal");
1487         return NF_ACCEPT;
1488
1489 ignore_cp:
1490         __ip_vs_conn_put(cp);
1491         goto out;
1492 }
1493
1494 /*
1495  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1496  *      used only for VS/NAT.
1497  *      Check if packet is reply for established ip_vs_conn.
1498  */
1499 static unsigned int
1500 ip_vs_reply4(void *priv, struct sk_buff *skb,
1501              const struct nf_hook_state *state)
1502 {
1503         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1504 }
1505
1506 /*
1507  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1508  *      Check if packet is reply for established ip_vs_conn.
1509  */
1510 static unsigned int
1511 ip_vs_local_reply4(void *priv, struct sk_buff *skb,
1512                    const struct nf_hook_state *state)
1513 {
1514         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1515 }
1516
1517 #ifdef CONFIG_IP_VS_IPV6
1518
1519 /*
1520  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1521  *      used only for VS/NAT.
1522  *      Check if packet is reply for established ip_vs_conn.
1523  */
1524 static unsigned int
1525 ip_vs_reply6(void *priv, struct sk_buff *skb,
1526              const struct nf_hook_state *state)
1527 {
1528         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1529 }
1530
1531 /*
1532  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1533  *      Check if packet is reply for established ip_vs_conn.
1534  */
1535 static unsigned int
1536 ip_vs_local_reply6(void *priv, struct sk_buff *skb,
1537                    const struct nf_hook_state *state)
1538 {
1539         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1540 }
1541
1542 #endif
1543
1544 static unsigned int
1545 ip_vs_try_to_schedule(struct netns_ipvs *ipvs, int af, struct sk_buff *skb,
1546                       struct ip_vs_proto_data *pd,
1547                       int *verdict, struct ip_vs_conn **cpp,
1548                       struct ip_vs_iphdr *iph)
1549 {
1550         struct ip_vs_protocol *pp = pd->pp;
1551
1552         if (!iph->fragoffs) {
1553                 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1554                  * replayed fragment zero will already have created the cp
1555                  */
1556
1557                 /* Schedule and create new connection entry into cpp */
1558                 if (!pp->conn_schedule(ipvs, af, skb, pd, verdict, cpp, iph))
1559                         return 0;
1560         }
1561
1562         if (unlikely(!*cpp)) {
1563                 /* sorry, all this trouble for a no-hit :) */
1564                 IP_VS_DBG_PKT(12, af, pp, skb, iph->off,
1565                               "ip_vs_in: packet continues traversal as normal");
1566
1567                 /* Fragment couldn't be mapped to a conn entry */
1568                 if (iph->fragoffs)
1569                         IP_VS_DBG_PKT(7, af, pp, skb, iph->off,
1570                                       "unhandled fragment");
1571
1572                 *verdict = NF_ACCEPT;
1573                 return 0;
1574         }
1575
1576         return 1;
1577 }
1578
1579 /* Check the UDP tunnel and return its header length */
1580 static int ipvs_udp_decap(struct netns_ipvs *ipvs, struct sk_buff *skb,
1581                           unsigned int offset, __u16 af,
1582                           const union nf_inet_addr *daddr, __u8 *proto)
1583 {
1584         struct udphdr _udph, *udph;
1585         struct ip_vs_dest *dest;
1586
1587         udph = skb_header_pointer(skb, offset, sizeof(_udph), &_udph);
1588         if (!udph)
1589                 goto unk;
1590         offset += sizeof(struct udphdr);
1591         dest = ip_vs_find_tunnel(ipvs, af, daddr, udph->dest);
1592         if (!dest)
1593                 goto unk;
1594         if (dest->tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GUE) {
1595                 struct guehdr _gueh, *gueh;
1596
1597                 gueh = skb_header_pointer(skb, offset, sizeof(_gueh), &_gueh);
1598                 if (!gueh)
1599                         goto unk;
1600                 if (gueh->control != 0 || gueh->version != 0)
1601                         goto unk;
1602                 /* Later we can support also IPPROTO_IPV6 */
1603                 if (gueh->proto_ctype != IPPROTO_IPIP)
1604                         goto unk;
1605                 *proto = gueh->proto_ctype;
1606                 return sizeof(struct udphdr) + sizeof(struct guehdr) +
1607                        (gueh->hlen << 2);
1608         }
1609
1610 unk:
1611         return 0;
1612 }
1613
1614 /* Check the GRE tunnel and return its header length */
1615 static int ipvs_gre_decap(struct netns_ipvs *ipvs, struct sk_buff *skb,
1616                           unsigned int offset, __u16 af,
1617                           const union nf_inet_addr *daddr, __u8 *proto)
1618 {
1619         struct gre_base_hdr _greh, *greh;
1620         struct ip_vs_dest *dest;
1621
1622         greh = skb_header_pointer(skb, offset, sizeof(_greh), &_greh);
1623         if (!greh)
1624                 goto unk;
1625         dest = ip_vs_find_tunnel(ipvs, af, daddr, 0);
1626         if (!dest)
1627                 goto unk;
1628         if (dest->tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GRE) {
1629                 __be16 type;
1630
1631                 /* Only support version 0 and C (csum) */
1632                 if ((greh->flags & ~GRE_CSUM) != 0)
1633                         goto unk;
1634                 type = greh->protocol;
1635                 /* Later we can support also IPPROTO_IPV6 */
1636                 if (type != htons(ETH_P_IP))
1637                         goto unk;
1638                 *proto = IPPROTO_IPIP;
1639                 return gre_calc_hlen(gre_flags_to_tnl_flags(greh->flags));
1640         }
1641
1642 unk:
1643         return 0;
1644 }
1645
1646 /*
1647  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1648  *      Find any that might be relevant, check against existing connections,
1649  *      forward to the right destination host if relevant.
1650  *      Currently handles error types - unreachable, quench, ttl exceeded.
1651  */
1652 static int
1653 ip_vs_in_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb, int *related,
1654               unsigned int hooknum)
1655 {
1656         struct iphdr *iph;
1657         struct icmphdr  _icmph, *ic;
1658         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1659         struct ip_vs_iphdr ciph;
1660         struct ip_vs_conn *cp;
1661         struct ip_vs_protocol *pp;
1662         struct ip_vs_proto_data *pd;
1663         unsigned int offset, offset2, ihl, verdict;
1664         bool ipip, new_cp = false;
1665         union nf_inet_addr *raddr;
1666
1667         *related = 1;
1668
1669         /* reassemble IP fragments */
1670         if (ip_is_fragment(ip_hdr(skb))) {
1671                 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
1672                         return NF_STOLEN;
1673         }
1674
1675         iph = ip_hdr(skb);
1676         offset = ihl = iph->ihl * 4;
1677         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1678         if (ic == NULL)
1679                 return NF_DROP;
1680
1681         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1682                   ic->type, ntohs(icmp_id(ic)),
1683                   &iph->saddr, &iph->daddr);
1684
1685         /*
1686          * Work through seeing if this is for us.
1687          * These checks are supposed to be in an order that means easy
1688          * things are checked first to speed up processing.... however
1689          * this means that some packets will manage to get a long way
1690          * down this stack and then be rejected, but that's life.
1691          */
1692         if ((ic->type != ICMP_DEST_UNREACH) &&
1693             (ic->type != ICMP_SOURCE_QUENCH) &&
1694             (ic->type != ICMP_TIME_EXCEEDED)) {
1695                 *related = 0;
1696                 return NF_ACCEPT;
1697         }
1698
1699         /* Now find the contained IP header */
1700         offset += sizeof(_icmph);
1701         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1702         if (cih == NULL)
1703                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1704         raddr = (union nf_inet_addr *)&cih->daddr;
1705
1706         /* Special case for errors for IPIP packets */
1707         ipip = false;
1708         if (cih->protocol == IPPROTO_IPIP) {
1709                 struct ip_vs_dest *dest;
1710
1711                 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1712                         return NF_ACCEPT;
1713                 /* Error for our IPIP must arrive at LOCAL_IN */
1714                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1715                         return NF_ACCEPT;
1716                 dest = ip_vs_find_tunnel(ipvs, AF_INET, raddr, 0);
1717                 /* Only for known tunnel */
1718                 if (!dest || dest->tun_type != IP_VS_CONN_F_TUNNEL_TYPE_IPIP)
1719                         return NF_ACCEPT;
1720                 offset += cih->ihl * 4;
1721                 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1722                 if (cih == NULL)
1723                         return NF_ACCEPT; /* The packet looks wrong, ignore */
1724                 ipip = true;
1725         } else if ((cih->protocol == IPPROTO_UDP ||     /* Can be UDP encap */
1726                     cih->protocol == IPPROTO_GRE) &&    /* Can be GRE encap */
1727                    /* Error for our tunnel must arrive at LOCAL_IN */
1728                    (skb_rtable(skb)->rt_flags & RTCF_LOCAL)) {
1729                 __u8 iproto;
1730                 int ulen;
1731
1732                 /* Non-first fragment has no UDP header */
1733                 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1734                         return NF_ACCEPT;
1735                 offset2 = offset + cih->ihl * 4;
1736                 if (cih->protocol == IPPROTO_UDP)
1737                         ulen = ipvs_udp_decap(ipvs, skb, offset2, AF_INET,
1738                                               raddr, &iproto);
1739                 else
1740                         ulen = ipvs_gre_decap(ipvs, skb, offset2, AF_INET,
1741                                               raddr, &iproto);
1742                 if (ulen > 0) {
1743                         /* Skip IP and UDP/GRE tunnel headers */
1744                         offset = offset2 + ulen;
1745                         /* Now we should be at the original IP header */
1746                         cih = skb_header_pointer(skb, offset, sizeof(_ciph),
1747                                                  &_ciph);
1748                         if (cih && cih->version == 4 && cih->ihl >= 5 &&
1749                             iproto == IPPROTO_IPIP)
1750                                 ipip = true;
1751                         else
1752                                 return NF_ACCEPT;
1753                 }
1754         }
1755
1756         pd = ip_vs_proto_data_get(ipvs, cih->protocol);
1757         if (!pd)
1758                 return NF_ACCEPT;
1759         pp = pd->pp;
1760
1761         /* Is the embedded protocol header present? */
1762         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1763                      pp->dont_defrag))
1764                 return NF_ACCEPT;
1765
1766         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1767                       "Checking incoming ICMP for");
1768
1769         offset2 = offset;
1770         ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, !ipip, &ciph);
1771         offset = ciph.len;
1772
1773         /* The embedded headers contain source and dest in reverse order.
1774          * For IPIP this is error for request, not for reply.
1775          */
1776         cp = INDIRECT_CALL_1(pp->conn_in_get, ip_vs_conn_in_get_proto,
1777                              ipvs, AF_INET, skb, &ciph);
1778
1779         if (!cp) {
1780                 int v;
1781
1782                 if (ipip || !sysctl_schedule_icmp(ipvs))
1783                         return NF_ACCEPT;
1784
1785                 if (!ip_vs_try_to_schedule(ipvs, AF_INET, skb, pd, &v, &cp, &ciph))
1786                         return v;
1787                 new_cp = true;
1788         }
1789
1790         verdict = NF_DROP;
1791
1792         /* Ensure the checksum is correct */
1793         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1794                 /* Failed checksum! */
1795                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1796                           &iph->saddr);
1797                 goto out;
1798         }
1799
1800         if (ipip) {
1801                 __be32 info = ic->un.gateway;
1802                 __u8 type = ic->type;
1803                 __u8 code = ic->code;
1804
1805                 /* Update the MTU */
1806                 if (ic->type == ICMP_DEST_UNREACH &&
1807                     ic->code == ICMP_FRAG_NEEDED) {
1808                         struct ip_vs_dest *dest = cp->dest;
1809                         u32 mtu = ntohs(ic->un.frag.mtu);
1810                         __be16 frag_off = cih->frag_off;
1811
1812                         /* Strip outer IP and ICMP, go to IPIP header */
1813                         if (pskb_pull(skb, ihl + sizeof(_icmph)) == NULL)
1814                                 goto ignore_ipip;
1815                         offset2 -= ihl + sizeof(_icmph);
1816                         skb_reset_network_header(skb);
1817                         IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1818                                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1819                         ipv4_update_pmtu(skb, ipvs->net, mtu, 0, 0);
1820                         /* Client uses PMTUD? */
1821                         if (!(frag_off & htons(IP_DF)))
1822                                 goto ignore_ipip;
1823                         /* Prefer the resulting PMTU */
1824                         if (dest) {
1825                                 struct ip_vs_dest_dst *dest_dst;
1826
1827                                 dest_dst = rcu_dereference(dest->dest_dst);
1828                                 if (dest_dst)
1829                                         mtu = dst_mtu(dest_dst->dst_cache);
1830                         }
1831                         if (mtu > 68 + sizeof(struct iphdr))
1832                                 mtu -= sizeof(struct iphdr);
1833                         info = htonl(mtu);
1834                 }
1835                 /* Strip outer IP, ICMP and IPIP, go to IP header of
1836                  * original request.
1837                  */
1838                 if (pskb_pull(skb, offset2) == NULL)
1839                         goto ignore_ipip;
1840                 skb_reset_network_header(skb);
1841                 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1842                         &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1843                         type, code, ntohl(info));
1844                 icmp_send(skb, type, code, info);
1845                 /* ICMP can be shorter but anyways, account it */
1846                 ip_vs_out_stats(cp, skb);
1847
1848 ignore_ipip:
1849                 consume_skb(skb);
1850                 verdict = NF_STOLEN;
1851                 goto out;
1852         }
1853
1854         /* do the statistics and put it back */
1855         ip_vs_in_stats(cp, skb);
1856         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol ||
1857             IPPROTO_SCTP == cih->protocol)
1858                 offset += 2 * sizeof(__u16);
1859         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1860
1861 out:
1862         if (likely(!new_cp))
1863                 __ip_vs_conn_put(cp);
1864         else
1865                 ip_vs_conn_put(cp);
1866
1867         return verdict;
1868 }
1869
1870 #ifdef CONFIG_IP_VS_IPV6
1871 static int ip_vs_in_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
1872                             int *related, unsigned int hooknum,
1873                             struct ip_vs_iphdr *iph)
1874 {
1875         struct icmp6hdr _icmph, *ic;
1876         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1877         struct ip_vs_conn *cp;
1878         struct ip_vs_protocol *pp;
1879         struct ip_vs_proto_data *pd;
1880         unsigned int offset, verdict;
1881         bool new_cp = false;
1882
1883         *related = 1;
1884
1885         ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph);
1886         if (ic == NULL)
1887                 return NF_DROP;
1888
1889         /*
1890          * Work through seeing if this is for us.
1891          * These checks are supposed to be in an order that means easy
1892          * things are checked first to speed up processing.... however
1893          * this means that some packets will manage to get a long way
1894          * down this stack and then be rejected, but that's life.
1895          */
1896         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1897                 *related = 0;
1898                 return NF_ACCEPT;
1899         }
1900         /* Fragment header that is before ICMP header tells us that:
1901          * it's not an error message since they can't be fragmented.
1902          */
1903         if (iph->flags & IP6_FH_F_FRAG)
1904                 return NF_DROP;
1905
1906         IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1907                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1908                   &iph->saddr, &iph->daddr);
1909
1910         offset = iph->len + sizeof(_icmph);
1911         if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, offset, true, &ciph))
1912                 return NF_ACCEPT;
1913
1914         pd = ip_vs_proto_data_get(ipvs, ciph.protocol);
1915         if (!pd)
1916                 return NF_ACCEPT;
1917         pp = pd->pp;
1918
1919         /* Cannot handle fragmented embedded protocol */
1920         if (ciph.fragoffs)
1921                 return NF_ACCEPT;
1922
1923         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
1924                       "Checking incoming ICMPv6 for");
1925
1926         /* The embedded headers contain source and dest in reverse order
1927          * if not from localhost
1928          */
1929         cp = INDIRECT_CALL_1(pp->conn_in_get, ip_vs_conn_in_get_proto,
1930                              ipvs, AF_INET6, skb, &ciph);
1931
1932         if (!cp) {
1933                 int v;
1934
1935                 if (!sysctl_schedule_icmp(ipvs))
1936                         return NF_ACCEPT;
1937
1938                 if (!ip_vs_try_to_schedule(ipvs, AF_INET6, skb, pd, &v, &cp, &ciph))
1939                         return v;
1940
1941                 new_cp = true;
1942         }
1943
1944         /* VS/TUN, VS/DR and LOCALNODE just let it go */
1945         if ((hooknum == NF_INET_LOCAL_OUT) &&
1946             (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1947                 verdict = NF_ACCEPT;
1948                 goto out;
1949         }
1950
1951         /* do the statistics and put it back */
1952         ip_vs_in_stats(cp, skb);
1953
1954         /* Need to mangle contained IPv6 header in ICMPv6 packet */
1955         offset = ciph.len;
1956         if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1957             IPPROTO_SCTP == ciph.protocol)
1958                 offset += 2 * sizeof(__u16); /* Also mangle ports */
1959
1960         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset, hooknum, &ciph);
1961
1962 out:
1963         if (likely(!new_cp))
1964                 __ip_vs_conn_put(cp);
1965         else
1966                 ip_vs_conn_put(cp);
1967
1968         return verdict;
1969 }
1970 #endif
1971
1972
1973 /*
1974  *      Check if it's for virtual services, look it up,
1975  *      and send it on its way...
1976  */
1977 static unsigned int
1978 ip_vs_in(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1979 {
1980         struct ip_vs_iphdr iph;
1981         struct ip_vs_protocol *pp;
1982         struct ip_vs_proto_data *pd;
1983         struct ip_vs_conn *cp;
1984         int ret, pkts;
1985         int conn_reuse_mode;
1986         struct sock *sk;
1987
1988         /* Already marked as IPVS request or reply? */
1989         if (skb->ipvs_property)
1990                 return NF_ACCEPT;
1991
1992         /*
1993          *      Big tappo:
1994          *      - remote client: only PACKET_HOST
1995          *      - route: used for struct net when skb->dev is unset
1996          */
1997         if (unlikely((skb->pkt_type != PACKET_HOST &&
1998                       hooknum != NF_INET_LOCAL_OUT) ||
1999                      !skb_dst(skb))) {
2000                 ip_vs_fill_iph_skb(af, skb, false, &iph);
2001                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
2002                               " ignored in hook %u\n",
2003                               skb->pkt_type, iph.protocol,
2004                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
2005                 return NF_ACCEPT;
2006         }
2007         /* ipvs enabled in this netns ? */
2008         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
2009                 return NF_ACCEPT;
2010
2011         ip_vs_fill_iph_skb(af, skb, false, &iph);
2012
2013         /* Bad... Do not break raw sockets */
2014         sk = skb_to_full_sk(skb);
2015         if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
2016                      af == AF_INET)) {
2017
2018                 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
2019                         return NF_ACCEPT;
2020         }
2021
2022 #ifdef CONFIG_IP_VS_IPV6
2023         if (af == AF_INET6) {
2024                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
2025                         int related;
2026                         int verdict = ip_vs_in_icmp_v6(ipvs, skb, &related,
2027                                                        hooknum, &iph);
2028
2029                         if (related)
2030                                 return verdict;
2031                 }
2032         } else
2033 #endif
2034                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
2035                         int related;
2036                         int verdict = ip_vs_in_icmp(ipvs, skb, &related,
2037                                                     hooknum);
2038
2039                         if (related)
2040                                 return verdict;
2041                 }
2042
2043         /* Protocol supported? */
2044         pd = ip_vs_proto_data_get(ipvs, iph.protocol);
2045         if (unlikely(!pd)) {
2046                 /* The only way we'll see this packet again is if it's
2047                  * encapsulated, so mark it with ipvs_property=1 so we
2048                  * skip it if we're ignoring tunneled packets
2049                  */
2050                 if (sysctl_ignore_tunneled(ipvs))
2051                         skb->ipvs_property = 1;
2052
2053                 return NF_ACCEPT;
2054         }
2055         pp = pd->pp;
2056         /*
2057          * Check if the packet belongs to an existing connection entry
2058          */
2059         cp = INDIRECT_CALL_1(pp->conn_in_get, ip_vs_conn_in_get_proto,
2060                              ipvs, af, skb, &iph);
2061
2062         conn_reuse_mode = sysctl_conn_reuse_mode(ipvs);
2063         if (conn_reuse_mode && !iph.fragoffs && is_new_conn(skb, &iph) && cp) {
2064                 bool uses_ct = false, resched = false;
2065
2066                 if (unlikely(sysctl_expire_nodest_conn(ipvs)) && cp->dest &&
2067                     unlikely(!atomic_read(&cp->dest->weight))) {
2068                         resched = true;
2069                         uses_ct = ip_vs_conn_uses_conntrack(cp, skb);
2070                 } else if (is_new_conn_expected(cp, conn_reuse_mode)) {
2071                         uses_ct = ip_vs_conn_uses_conntrack(cp, skb);
2072                         if (!atomic_read(&cp->n_control)) {
2073                                 resched = true;
2074                         } else {
2075                                 /* Do not reschedule controlling connection
2076                                  * that uses conntrack while it is still
2077                                  * referenced by controlled connection(s).
2078                                  */
2079                                 resched = !uses_ct;
2080                         }
2081                 }
2082
2083                 if (resched) {
2084                         if (!atomic_read(&cp->n_control))
2085                                 ip_vs_conn_expire_now(cp);
2086                         __ip_vs_conn_put(cp);
2087                         if (uses_ct)
2088                                 return NF_DROP;
2089                         cp = NULL;
2090                 }
2091         }
2092
2093         if (unlikely(!cp)) {
2094                 int v;
2095
2096                 if (!ip_vs_try_to_schedule(ipvs, af, skb, pd, &v, &cp, &iph))
2097                         return v;
2098         }
2099
2100         IP_VS_DBG_PKT(11, af, pp, skb, iph.off, "Incoming packet");
2101
2102         /* Check the server status */
2103         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
2104                 /* the destination server is not available */
2105
2106                 __u32 flags = cp->flags;
2107
2108                 /* when timer already started, silently drop the packet.*/
2109                 if (timer_pending(&cp->timer))
2110                         __ip_vs_conn_put(cp);
2111                 else
2112                         ip_vs_conn_put(cp);
2113
2114                 if (sysctl_expire_nodest_conn(ipvs) &&
2115                     !(flags & IP_VS_CONN_F_ONE_PACKET)) {
2116                         /* try to expire the connection immediately */
2117                         ip_vs_conn_expire_now(cp);
2118                 }
2119
2120                 return NF_DROP;
2121         }
2122
2123         ip_vs_in_stats(cp, skb);
2124         ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
2125         if (cp->packet_xmit)
2126                 ret = cp->packet_xmit(skb, cp, pp, &iph);
2127                 /* do not touch skb anymore */
2128         else {
2129                 IP_VS_DBG_RL("warning: packet_xmit is null");
2130                 ret = NF_ACCEPT;
2131         }
2132
2133         /* Increase its packet counter and check if it is needed
2134          * to be synchronized
2135          *
2136          * Sync connection if it is about to close to
2137          * encorage the standby servers to update the connections timeout
2138          *
2139          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
2140          */
2141
2142         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
2143                 pkts = sysctl_sync_threshold(ipvs);
2144         else
2145                 pkts = atomic_add_return(1, &cp->in_pkts);
2146
2147         if (ipvs->sync_state & IP_VS_STATE_MASTER)
2148                 ip_vs_sync_conn(ipvs, cp, pkts);
2149         else if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) && cp->control)
2150                 /* increment is done inside ip_vs_sync_conn too */
2151                 atomic_inc(&cp->control->in_pkts);
2152
2153         ip_vs_conn_put(cp);
2154         return ret;
2155 }
2156
2157 /*
2158  *      AF_INET handler in NF_INET_LOCAL_IN chain
2159  *      Schedule and forward packets from remote clients
2160  */
2161 static unsigned int
2162 ip_vs_remote_request4(void *priv, struct sk_buff *skb,
2163                       const struct nf_hook_state *state)
2164 {
2165         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
2166 }
2167
2168 /*
2169  *      AF_INET handler in NF_INET_LOCAL_OUT chain
2170  *      Schedule and forward packets from local clients
2171  */
2172 static unsigned int
2173 ip_vs_local_request4(void *priv, struct sk_buff *skb,
2174                      const struct nf_hook_state *state)
2175 {
2176         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
2177 }
2178
2179 #ifdef CONFIG_IP_VS_IPV6
2180
2181 /*
2182  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
2183  *      Schedule and forward packets from remote clients
2184  */
2185 static unsigned int
2186 ip_vs_remote_request6(void *priv, struct sk_buff *skb,
2187                       const struct nf_hook_state *state)
2188 {
2189         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
2190 }
2191
2192 /*
2193  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
2194  *      Schedule and forward packets from local clients
2195  */
2196 static unsigned int
2197 ip_vs_local_request6(void *priv, struct sk_buff *skb,
2198                      const struct nf_hook_state *state)
2199 {
2200         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
2201 }
2202
2203 #endif
2204
2205
2206 /*
2207  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
2208  *      related packets destined for 0.0.0.0/0.
2209  *      When fwmark-based virtual service is used, such as transparent
2210  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
2211  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
2212  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
2213  *      and send them to ip_vs_in_icmp.
2214  */
2215 static unsigned int
2216 ip_vs_forward_icmp(void *priv, struct sk_buff *skb,
2217                    const struct nf_hook_state *state)
2218 {
2219         int r;
2220         struct netns_ipvs *ipvs = net_ipvs(state->net);
2221
2222         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
2223                 return NF_ACCEPT;
2224
2225         /* ipvs enabled in this netns ? */
2226         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
2227                 return NF_ACCEPT;
2228
2229         return ip_vs_in_icmp(ipvs, skb, &r, state->hook);
2230 }
2231
2232 #ifdef CONFIG_IP_VS_IPV6
2233 static unsigned int
2234 ip_vs_forward_icmp_v6(void *priv, struct sk_buff *skb,
2235                       const struct nf_hook_state *state)
2236 {
2237         int r;
2238         struct netns_ipvs *ipvs = net_ipvs(state->net);
2239         struct ip_vs_iphdr iphdr;
2240
2241         ip_vs_fill_iph_skb(AF_INET6, skb, false, &iphdr);
2242         if (iphdr.protocol != IPPROTO_ICMPV6)
2243                 return NF_ACCEPT;
2244
2245         /* ipvs enabled in this netns ? */
2246         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
2247                 return NF_ACCEPT;
2248
2249         return ip_vs_in_icmp_v6(ipvs, skb, &r, state->hook, &iphdr);
2250 }
2251 #endif
2252
2253
2254 static const struct nf_hook_ops ip_vs_ops[] = {
2255         /* After packet filtering, change source only for VS/NAT */
2256         {
2257                 .hook           = ip_vs_reply4,
2258                 .pf             = NFPROTO_IPV4,
2259                 .hooknum        = NF_INET_LOCAL_IN,
2260                 .priority       = NF_IP_PRI_NAT_SRC - 2,
2261         },
2262         /* After packet filtering, forward packet through VS/DR, VS/TUN,
2263          * or VS/NAT(change destination), so that filtering rules can be
2264          * applied to IPVS. */
2265         {
2266                 .hook           = ip_vs_remote_request4,
2267                 .pf             = NFPROTO_IPV4,
2268                 .hooknum        = NF_INET_LOCAL_IN,
2269                 .priority       = NF_IP_PRI_NAT_SRC - 1,
2270         },
2271         /* Before ip_vs_in, change source only for VS/NAT */
2272         {
2273                 .hook           = ip_vs_local_reply4,
2274                 .pf             = NFPROTO_IPV4,
2275                 .hooknum        = NF_INET_LOCAL_OUT,
2276                 .priority       = NF_IP_PRI_NAT_DST + 1,
2277         },
2278         /* After mangle, schedule and forward local requests */
2279         {
2280                 .hook           = ip_vs_local_request4,
2281                 .pf             = NFPROTO_IPV4,
2282                 .hooknum        = NF_INET_LOCAL_OUT,
2283                 .priority       = NF_IP_PRI_NAT_DST + 2,
2284         },
2285         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2286          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2287         {
2288                 .hook           = ip_vs_forward_icmp,
2289                 .pf             = NFPROTO_IPV4,
2290                 .hooknum        = NF_INET_FORWARD,
2291                 .priority       = 99,
2292         },
2293         /* After packet filtering, change source only for VS/NAT */
2294         {
2295                 .hook           = ip_vs_reply4,
2296                 .pf             = NFPROTO_IPV4,
2297                 .hooknum        = NF_INET_FORWARD,
2298                 .priority       = 100,
2299         },
2300 #ifdef CONFIG_IP_VS_IPV6
2301         /* After packet filtering, change source only for VS/NAT */
2302         {
2303                 .hook           = ip_vs_reply6,
2304                 .pf             = NFPROTO_IPV6,
2305                 .hooknum        = NF_INET_LOCAL_IN,
2306                 .priority       = NF_IP6_PRI_NAT_SRC - 2,
2307         },
2308         /* After packet filtering, forward packet through VS/DR, VS/TUN,
2309          * or VS/NAT(change destination), so that filtering rules can be
2310          * applied to IPVS. */
2311         {
2312                 .hook           = ip_vs_remote_request6,
2313                 .pf             = NFPROTO_IPV6,
2314                 .hooknum        = NF_INET_LOCAL_IN,
2315                 .priority       = NF_IP6_PRI_NAT_SRC - 1,
2316         },
2317         /* Before ip_vs_in, change source only for VS/NAT */
2318         {
2319                 .hook           = ip_vs_local_reply6,
2320                 .pf             = NFPROTO_IPV6,
2321                 .hooknum        = NF_INET_LOCAL_OUT,
2322                 .priority       = NF_IP6_PRI_NAT_DST + 1,
2323         },
2324         /* After mangle, schedule and forward local requests */
2325         {
2326                 .hook           = ip_vs_local_request6,
2327                 .pf             = NFPROTO_IPV6,
2328                 .hooknum        = NF_INET_LOCAL_OUT,
2329                 .priority       = NF_IP6_PRI_NAT_DST + 2,
2330         },
2331         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2332          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2333         {
2334                 .hook           = ip_vs_forward_icmp_v6,
2335                 .pf             = NFPROTO_IPV6,
2336                 .hooknum        = NF_INET_FORWARD,
2337                 .priority       = 99,
2338         },
2339         /* After packet filtering, change source only for VS/NAT */
2340         {
2341                 .hook           = ip_vs_reply6,
2342                 .pf             = NFPROTO_IPV6,
2343                 .hooknum        = NF_INET_FORWARD,
2344                 .priority       = 100,
2345         },
2346 #endif
2347 };
2348 /*
2349  *      Initialize IP Virtual Server netns mem.
2350  */
2351 static int __net_init __ip_vs_init(struct net *net)
2352 {
2353         struct netns_ipvs *ipvs;
2354
2355         ipvs = net_generic(net, ip_vs_net_id);
2356         if (ipvs == NULL)
2357                 return -ENOMEM;
2358
2359         /* Hold the beast until a service is registerd */
2360         ipvs->enable = 0;
2361         ipvs->net = net;
2362         /* Counters used for creating unique names */
2363         ipvs->gen = atomic_read(&ipvs_netns_cnt);
2364         atomic_inc(&ipvs_netns_cnt);
2365         net->ipvs = ipvs;
2366
2367         if (ip_vs_estimator_net_init(ipvs) < 0)
2368                 goto estimator_fail;
2369
2370         if (ip_vs_control_net_init(ipvs) < 0)
2371                 goto control_fail;
2372
2373         if (ip_vs_protocol_net_init(ipvs) < 0)
2374                 goto protocol_fail;
2375
2376         if (ip_vs_app_net_init(ipvs) < 0)
2377                 goto app_fail;
2378
2379         if (ip_vs_conn_net_init(ipvs) < 0)
2380                 goto conn_fail;
2381
2382         if (ip_vs_sync_net_init(ipvs) < 0)
2383                 goto sync_fail;
2384
2385         return 0;
2386 /*
2387  * Error handling
2388  */
2389
2390 sync_fail:
2391         ip_vs_conn_net_cleanup(ipvs);
2392 conn_fail:
2393         ip_vs_app_net_cleanup(ipvs);
2394 app_fail:
2395         ip_vs_protocol_net_cleanup(ipvs);
2396 protocol_fail:
2397         ip_vs_control_net_cleanup(ipvs);
2398 control_fail:
2399         ip_vs_estimator_net_cleanup(ipvs);
2400 estimator_fail:
2401         net->ipvs = NULL;
2402         return -ENOMEM;
2403 }
2404
2405 static void __net_exit __ip_vs_cleanup_batch(struct list_head *net_list)
2406 {
2407         struct netns_ipvs *ipvs;
2408         struct net *net;
2409
2410         ip_vs_service_nets_cleanup(net_list);   /* ip_vs_flush() with locks */
2411         list_for_each_entry(net, net_list, exit_list) {
2412                 ipvs = net_ipvs(net);
2413                 ip_vs_conn_net_cleanup(ipvs);
2414                 ip_vs_app_net_cleanup(ipvs);
2415                 ip_vs_protocol_net_cleanup(ipvs);
2416                 ip_vs_control_net_cleanup(ipvs);
2417                 ip_vs_estimator_net_cleanup(ipvs);
2418                 IP_VS_DBG(2, "ipvs netns %d released\n", ipvs->gen);
2419                 net->ipvs = NULL;
2420         }
2421 }
2422
2423 static int __net_init __ip_vs_dev_init(struct net *net)
2424 {
2425         int ret;
2426
2427         ret = nf_register_net_hooks(net, ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2428         if (ret < 0)
2429                 goto hook_fail;
2430         return 0;
2431
2432 hook_fail:
2433         return ret;
2434 }
2435
2436 static void __net_exit __ip_vs_dev_cleanup_batch(struct list_head *net_list)
2437 {
2438         struct netns_ipvs *ipvs;
2439         struct net *net;
2440
2441         EnterFunction(2);
2442         list_for_each_entry(net, net_list, exit_list) {
2443                 ipvs = net_ipvs(net);
2444                 nf_unregister_net_hooks(net, ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2445                 ipvs->enable = 0;       /* Disable packet reception */
2446                 smp_wmb();
2447                 ip_vs_sync_net_cleanup(ipvs);
2448         }
2449         LeaveFunction(2);
2450 }
2451
2452 static struct pernet_operations ipvs_core_ops = {
2453         .init = __ip_vs_init,
2454         .exit_batch = __ip_vs_cleanup_batch,
2455         .id   = &ip_vs_net_id,
2456         .size = sizeof(struct netns_ipvs),
2457 };
2458
2459 static struct pernet_operations ipvs_core_dev_ops = {
2460         .init = __ip_vs_dev_init,
2461         .exit_batch = __ip_vs_dev_cleanup_batch,
2462 };
2463
2464 /*
2465  *      Initialize IP Virtual Server
2466  */
2467 static int __init ip_vs_init(void)
2468 {
2469         int ret;
2470
2471         ret = ip_vs_control_init();
2472         if (ret < 0) {
2473                 pr_err("can't setup control.\n");
2474                 goto exit;
2475         }
2476
2477         ip_vs_protocol_init();
2478
2479         ret = ip_vs_conn_init();
2480         if (ret < 0) {
2481                 pr_err("can't setup connection table.\n");
2482                 goto cleanup_protocol;
2483         }
2484
2485         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
2486         if (ret < 0)
2487                 goto cleanup_conn;
2488
2489         ret = register_pernet_device(&ipvs_core_dev_ops);
2490         if (ret < 0)
2491                 goto cleanup_sub;
2492
2493         ret = ip_vs_register_nl_ioctl();
2494         if (ret < 0) {
2495                 pr_err("can't register netlink/ioctl.\n");
2496                 goto cleanup_dev;
2497         }
2498
2499         pr_info("ipvs loaded.\n");
2500
2501         return ret;
2502
2503 cleanup_dev:
2504         unregister_pernet_device(&ipvs_core_dev_ops);
2505 cleanup_sub:
2506         unregister_pernet_subsys(&ipvs_core_ops);
2507 cleanup_conn:
2508         ip_vs_conn_cleanup();
2509 cleanup_protocol:
2510         ip_vs_protocol_cleanup();
2511         ip_vs_control_cleanup();
2512 exit:
2513         return ret;
2514 }
2515
2516 static void __exit ip_vs_cleanup(void)
2517 {
2518         ip_vs_unregister_nl_ioctl();
2519         unregister_pernet_device(&ipvs_core_dev_ops);
2520         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
2521         ip_vs_conn_cleanup();
2522         ip_vs_protocol_cleanup();
2523         ip_vs_control_cleanup();
2524         pr_info("ipvs unloaded.\n");
2525 }
2526
2527 module_init(ip_vs_init);
2528 module_exit(ip_vs_cleanup);
2529 MODULE_LICENSE("GPL");