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 #if defined(CONFIG_IP_VS_PROTO_TCP) && defined(CONFIG_IP_VS_PROTO_UDP)
  72 #define SNAT_CALL(f, ...) \
  73         INDIRECT_CALL_2(f, tcp_snat_handler, udp_snat_handler, __VA_ARGS__)
  74 #elif defined(CONFIG_IP_VS_PROTO_TCP)
  75 #define SNAT_CALL(f, ...) INDIRECT_CALL_1(f, tcp_snat_handler, __VA_ARGS__)
  76 #elif defined(CONFIG_IP_VS_PROTO_UDP)
  77 #define SNAT_CALL(f, ...) INDIRECT_CALL_1(f, udp_snat_handler, __VA_ARGS__)
  78 #else
  79 #define SNAT_CALL(f, ...) f(__VA_ARGS__)
  80 #endif
  81
  82 static unsigned int ip_vs_net_id __read_mostly;
  83 /* netns cnt used for uniqueness */
  84 static atomic_t ipvs_netns_cnt = ATOMIC_INIT(0);
  85
  86 /* ID used in ICMP lookups */
  87 #define icmp_id(icmph)          (((icmph)->un).echo.id)
  88 #define icmpv6_id(icmph)        (icmph->icmp6_dataun.u_echo.identifier)
  89
  90 const char *ip_vs_proto_name(unsigned int proto)
  91 {
  92         static char buf[20];
  93
  94         switch (proto) {
  95         case IPPROTO_IP:
  96                 return "IP";
  97         case IPPROTO_UDP:
  98                 return "UDP";
  99         case IPPROTO_TCP:
 100                 return "TCP";
 101         case IPPROTO_SCTP:
 102                 return "SCTP";
 103         case IPPROTO_ICMP:
 104                 return "ICMP";
 105 #ifdef CONFIG_IP_VS_IPV6
 106         case IPPROTO_ICMPV6:
 107                 return "ICMPv6";
 108 #endif
 109         default:
 110                 sprintf(buf, "IP_%u", proto);
 111                 return buf;
 112         }
 113 }
 114
 115 void ip_vs_init_hash_table(struct list_head *table, int rows)
 116 {
 117         while (--rows >= 0)
 118                 INIT_LIST_HEAD(&table[rows]);
 119 }
 120
 121 static inline void
 122 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 123 {
 124         struct ip_vs_dest *dest = cp->dest;
 125         struct netns_ipvs *ipvs = cp->ipvs;
 126
 127         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 128                 struct ip_vs_cpu_stats *s;
 129                 struct ip_vs_service *svc;
 130
 131                 local_bh_disable();
 132
 133                 s = this_cpu_ptr(dest->stats.cpustats);
 134                 u64_stats_update_begin(&s->syncp);
 135                 s->cnt.inpkts++;
 136                 s->cnt.inbytes += skb->len;
 137                 u64_stats_update_end(&s->syncp);
 138
 139                 svc = rcu_dereference(dest->svc);
 140                 s = this_cpu_ptr(svc->stats.cpustats);
 141                 u64_stats_update_begin(&s->syncp);
 142                 s->cnt.inpkts++;
 143                 s->cnt.inbytes += skb->len;
 144                 u64_stats_update_end(&s->syncp);
 145
 146                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 147                 u64_stats_update_begin(&s->syncp);
 148                 s->cnt.inpkts++;
 149                 s->cnt.inbytes += skb->len;
 150                 u64_stats_update_end(&s->syncp);
 151
 152                 local_bh_enable();
 153         }
 154 }
 155
 156
 157 static inline void
 158 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 159 {
 160         struct ip_vs_dest *dest = cp->dest;
 161         struct netns_ipvs *ipvs = cp->ipvs;
 162
 163         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 164                 struct ip_vs_cpu_stats *s;
 165                 struct ip_vs_service *svc;
 166
 167                 local_bh_disable();
 168
 169                 s = this_cpu_ptr(dest->stats.cpustats);
 170                 u64_stats_update_begin(&s->syncp);
 171                 s->cnt.outpkts++;
 172                 s->cnt.outbytes += skb->len;
 173                 u64_stats_update_end(&s->syncp);
 174
 175                 svc = rcu_dereference(dest->svc);
 176                 s = this_cpu_ptr(svc->stats.cpustats);
 177                 u64_stats_update_begin(&s->syncp);
 178                 s->cnt.outpkts++;
 179                 s->cnt.outbytes += skb->len;
 180                 u64_stats_update_end(&s->syncp);
 181
 182                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 183                 u64_stats_update_begin(&s->syncp);
 184                 s->cnt.outpkts++;
 185                 s->cnt.outbytes += skb->len;
 186                 u64_stats_update_end(&s->syncp);
 187
 188                 local_bh_enable();
 189         }
 190 }
 191
 192
 193 static inline void
 194 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
 195 {
 196         struct netns_ipvs *ipvs = svc->ipvs;
 197         struct ip_vs_cpu_stats *s;
 198
 199         local_bh_disable();
 200
 201         s = this_cpu_ptr(cp->dest->stats.cpustats);
 202         u64_stats_update_begin(&s->syncp);
 203         s->cnt.conns++;
 204         u64_stats_update_end(&s->syncp);
 205
 206         s = this_cpu_ptr(svc->stats.cpustats);
 207         u64_stats_update_begin(&s->syncp);
 208         s->cnt.conns++;
 209         u64_stats_update_end(&s->syncp);
 210
 211         s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 212         u64_stats_update_begin(&s->syncp);
 213         s->cnt.conns++;
 214         u64_stats_update_end(&s->syncp);
 215
 216         local_bh_enable();
 217 }
 218
 219
 220 static inline void
 221 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
 222                 const struct sk_buff *skb,
 223                 struct ip_vs_proto_data *pd)
 224 {
 225         if (likely(pd->pp->state_transition))
 226                 pd->pp->state_transition(cp, direction, skb, pd);
 227 }
 228
 229 static inline int
 230 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
 231                               struct sk_buff *skb, int protocol,
 232                               const union nf_inet_addr *caddr, __be16 cport,
 233                               const union nf_inet_addr *vaddr, __be16 vport,
 234                               struct ip_vs_conn_param *p)
 235 {
 236         ip_vs_conn_fill_param(svc->ipvs, svc->af, protocol, caddr, cport, vaddr,
 237                               vport, p);
 238         p->pe = rcu_dereference(svc->pe);
 239         if (p->pe && p->pe->fill_param)
 240                 return p->pe->fill_param(p, skb);
 241
 242         return 0;
 243 }
 244
 245 /*
 246  *  IPVS persistent scheduling function
 247  *  It creates a connection entry according to its template if exists,
 248  *  or selects a server and creates a connection entry plus a template.
 249  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
 250  *  Protocols supported: TCP, UDP
 251  */
 252 static struct ip_vs_conn *
 253 ip_vs_sched_persist(struct ip_vs_service *svc,
 254                     struct sk_buff *skb, __be16 src_port, __be16 dst_port,
 255                     int *ignored, struct ip_vs_iphdr *iph)
 256 {
 257         struct ip_vs_conn *cp = NULL;
 258         struct ip_vs_dest *dest;
 259         struct ip_vs_conn *ct;
 260         __be16 dport = 0;               /* destination port to forward */
 261         unsigned int flags;
 262         struct ip_vs_conn_param param;
 263         const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
 264         union nf_inet_addr snet;        /* source network of the client,
 265                                            after masking */
 266         const union nf_inet_addr *src_addr, *dst_addr;
 267
 268         if (likely(!ip_vs_iph_inverse(iph))) {
 269                 src_addr = &iph->saddr;
 270                 dst_addr = &iph->daddr;
 271         } else {
 272                 src_addr = &iph->daddr;
 273                 dst_addr = &iph->saddr;
 274         }
 275
 276
 277         /* Mask saddr with the netmask to adjust template granularity */
 278 #ifdef CONFIG_IP_VS_IPV6
 279         if (svc->af == AF_INET6)
 280                 ipv6_addr_prefix(&snet.in6, &src_addr->in6,
 281                                  (__force __u32) svc->netmask);
 282         else
 283 #endif
 284                 snet.ip = src_addr->ip & svc->netmask;
 285
 286         IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
 287                       "mnet %s\n",
 288                       IP_VS_DBG_ADDR(svc->af, src_addr), ntohs(src_port),
 289                       IP_VS_DBG_ADDR(svc->af, dst_addr), ntohs(dst_port),
 290                       IP_VS_DBG_ADDR(svc->af, &snet));
 291
 292         /*
 293          * As far as we know, FTP is a very complicated network protocol, and
 294          * it uses control connection and data connections. For active FTP,
 295          * FTP server initialize data connection to the client, its source port
 296          * is often 20. For passive FTP, FTP server tells the clients the port
 297          * that it passively listens to,  and the client issues the data
 298          * connection. In the tunneling or direct routing mode, the load
 299          * balancer is on the client-to-server half of connection, the port
 300          * number is unknown to the load balancer. So, a conn template like
 301          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
 302          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
 303          * is created for other persistent services.
 304          */
 305         {
 306                 int protocol = iph->protocol;
 307                 const union nf_inet_addr *vaddr = dst_addr;
 308                 __be16 vport = 0;
 309
 310                 if (dst_port == svc->port) {
 311                         /* non-FTP template:
 312                          * <protocol, caddr, 0, vaddr, vport, daddr, dport>
 313                          * FTP template:
 314                          * <protocol, caddr, 0, vaddr, 0, daddr, 0>
 315                          */
 316                         if (svc->port != FTPPORT)
 317                                 vport = dst_port;
 318                 } else {
 319                         /* Note: persistent fwmark-based services and
 320                          * persistent port zero service are handled here.
 321                          * fwmark template:
 322                          * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
 323                          * port zero template:
 324                          * <protocol,caddr,0,vaddr,0,daddr,0>
 325                          */
 326                         if (svc->fwmark) {
 327                                 protocol = IPPROTO_IP;
 328                                 vaddr = &fwmark;
 329                         }
 330                 }
 331                 /* return *ignored = -1 so NF_DROP can be used */
 332                 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
 333                                                   vaddr, vport, &param) < 0) {
 334                         *ignored = -1;
 335                         return NULL;
 336                 }
 337         }
 338
 339         /* Check if a template already exists */
 340         ct = ip_vs_ct_in_get(&param);
 341         if (!ct || !ip_vs_check_template(ct, NULL)) {
 342                 struct ip_vs_scheduler *sched;
 343
 344                 /*
 345                  * No template found or the dest of the connection
 346                  * template is not available.
 347                  * return *ignored=0 i.e. ICMP and NF_DROP
 348                  */
 349                 sched = rcu_dereference(svc->scheduler);
 350                 if (sched) {
 351                         /* read svc->sched_data after svc->scheduler */
 352                         smp_rmb();
 353                         dest = sched->schedule(svc, skb, iph);
 354                 } else {
 355                         dest = NULL;
 356                 }
 357                 if (!dest) {
 358                         IP_VS_DBG(1, "p-schedule: no dest found.\n");
 359                         kfree(param.pe_data);
 360                         *ignored = 0;
 361                         return NULL;
 362                 }
 363
 364                 if (dst_port == svc->port && svc->port != FTPPORT)
 365                         dport = dest->port;
 366
 367                 /* Create a template
 368                  * This adds param.pe_data to the template,
 369                  * and thus param.pe_data will be destroyed
 370                  * when the template expires */
 371                 ct = ip_vs_conn_new(&param, dest->af, &dest->addr, dport,
 372                                     IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
 373                 if (ct == NULL) {
 374                         kfree(param.pe_data);
 375                         *ignored = -1;
 376                         return NULL;
 377                 }
 378
 379                 ct->timeout = svc->timeout;
 380         } else {
 381                 /* set destination with the found template */
 382                 dest = ct->dest;
 383                 kfree(param.pe_data);
 384         }
 385
 386         dport = dst_port;
 387         if (dport == svc->port && dest->port)
 388                 dport = dest->port;
 389
 390         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 391                  && iph->protocol == IPPROTO_UDP) ?
 392                 IP_VS_CONN_F_ONE_PACKET : 0;
 393
 394         /*
 395          *    Create a new connection according to the template
 396          */
 397         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol, src_addr,
 398                               src_port, dst_addr, dst_port, &param);
 399
 400         cp = ip_vs_conn_new(&param, dest->af, &dest->addr, dport, flags, dest,
 401                             skb->mark);
 402         if (cp == NULL) {
 403                 ip_vs_conn_put(ct);
 404                 *ignored = -1;
 405                 return NULL;
 406         }
 407
 408         /*
 409          *    Add its control
 410          */
 411         ip_vs_control_add(cp, ct);
 412         ip_vs_conn_put(ct);
 413
 414         ip_vs_conn_stats(cp, svc);
 415         return cp;
 416 }
 417
 418
 419 /*
 420  *  IPVS main scheduling function
 421  *  It selects a server according to the virtual service, and
 422  *  creates a connection entry.
 423  *  Protocols supported: TCP, UDP
 424  *
 425  *  Usage of *ignored
 426  *
 427  * 1 :   protocol tried to schedule (eg. on SYN), found svc but the
 428  *       svc/scheduler decides that this packet should be accepted with
 429  *       NF_ACCEPT because it must not be scheduled.
 430  *
 431  * 0 :   scheduler can not find destination, so try bypass or
 432  *       return ICMP and then NF_DROP (ip_vs_leave).
 433  *
 434  * -1 :  scheduler tried to schedule but fatal error occurred, eg.
 435  *       ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
 436  *       failure such as missing Call-ID, ENOMEM on skb_linearize
 437  *       or pe_data. In this case we should return NF_DROP without
 438  *       any attempts to send ICMP with ip_vs_leave.
 439  */
 440 struct ip_vs_conn *
 441 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
 442                struct ip_vs_proto_data *pd, int *ignored,
 443                struct ip_vs_iphdr *iph)
 444 {
 445         struct ip_vs_protocol *pp = pd->pp;
 446         struct ip_vs_conn *cp = NULL;
 447         struct ip_vs_scheduler *sched;
 448         struct ip_vs_dest *dest;
 449         __be16 _ports[2], *pptr, cport, vport;
 450         const void *caddr, *vaddr;
 451         unsigned int flags;
 452
 453         *ignored = 1;
 454         /*
 455          * IPv6 frags, only the first hit here.
 456          */
 457         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports);
 458         if (pptr == NULL)
 459                 return NULL;
 460
 461         if (likely(!ip_vs_iph_inverse(iph))) {
 462                 cport = pptr[0];
 463                 caddr = &iph->saddr;
 464                 vport = pptr[1];
 465                 vaddr = &iph->daddr;
 466         } else {
 467                 cport = pptr[1];
 468                 caddr = &iph->daddr;
 469                 vport = pptr[0];
 470                 vaddr = &iph->saddr;
 471         }
 472
 473         /*
 474          * FTPDATA needs this check when using local real server.
 475          * Never schedule Active FTPDATA connections from real server.
 476          * For LVS-NAT they must be already created. For other methods
 477          * with persistence the connection is created on SYN+ACK.
 478          */
 479         if (cport == FTPDATA) {
 480                 IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
 481                               "Not scheduling FTPDATA");
 482                 return NULL;
 483         }
 484
 485         /*
 486          *    Do not schedule replies from local real server.
 487          */
 488         if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK)) {
 489                 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
 490                 cp = INDIRECT_CALL_1(pp->conn_in_get,
 491                                      ip_vs_conn_in_get_proto, svc->ipvs,
 492                                      svc->af, skb, iph);
 493                 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
 494
 495                 if (cp) {
 496                         IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
 497                                       "Not scheduling reply for existing"
 498                                       " connection");
 499                         __ip_vs_conn_put(cp);
 500                         return NULL;
 501                 }
 502         }
 503
 504         /*
 505          *    Persistent service
 506          */
 507         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 508                 return ip_vs_sched_persist(svc, skb, cport, vport, ignored,
 509                                            iph);
 510
 511         *ignored = 0;
 512
 513         /*
 514          *    Non-persistent service
 515          */
 516         if (!svc->fwmark && vport != svc->port) {
 517                 if (!svc->port)
 518                         pr_err("Schedule: port zero only supported "
 519                                "in persistent services, "
 520                                "check your ipvs configuration\n");
 521                 return NULL;
 522         }
 523
 524         sched = rcu_dereference(svc->scheduler);
 525         if (sched) {
 526                 /* read svc->sched_data after svc->scheduler */
 527                 smp_rmb();
 528                 dest = sched->schedule(svc, skb, iph);
 529         } else {
 530                 dest = NULL;
 531         }
 532         if (dest == NULL) {
 533                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 534                 return NULL;
 535         }
 536
 537         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 538                  && iph->protocol == IPPROTO_UDP) ?
 539                 IP_VS_CONN_F_ONE_PACKET : 0;
 540
 541         /*
 542          *    Create a connection entry.
 543          */
 544         {
 545                 struct ip_vs_conn_param p;
 546
 547                 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
 548                                       caddr, cport, vaddr, vport, &p);
 549                 cp = ip_vs_conn_new(&p, dest->af, &dest->addr,
 550                                     dest->port ? dest->port : vport,
 551                                     flags, dest, skb->mark);
 552                 if (!cp) {
 553                         *ignored = -1;
 554                         return NULL;
 555                 }
 556         }
 557
 558         IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
 559                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
 560                       ip_vs_fwd_tag(cp),
 561                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 562                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 563                       IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
 564                       cp->flags, refcount_read(&cp->refcnt));
 565
 566         ip_vs_conn_stats(cp, svc);
 567         return cp;
 568 }
 569
 570 static inline int ip_vs_addr_is_unicast(struct net *net, int af,
 571                                         union nf_inet_addr *addr)
 572 {
 573 #ifdef CONFIG_IP_VS_IPV6
 574         if (af == AF_INET6)
 575                 return ipv6_addr_type(&addr->in6) & IPV6_ADDR_UNICAST;
 576 #endif
 577         return (inet_addr_type(net, addr->ip) == RTN_UNICAST);
 578 }
 579
 580 /*
 581  *  Pass or drop the packet.
 582  *  Called by ip_vs_in, when the virtual service is available but
 583  *  no destination is available for a new connection.
 584  */
 585 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 586                 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
 587 {
 588         __be16 _ports[2], *pptr, dport;
 589         struct netns_ipvs *ipvs = svc->ipvs;
 590         struct net *net = ipvs->net;
 591
 592         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports);
 593         if (!pptr)
 594                 return NF_DROP;
 595         dport = likely(!ip_vs_iph_inverse(iph)) ? pptr[1] : pptr[0];
 596
 597         /* if it is fwmark-based service, the cache_bypass sysctl is up
 598            and the destination is a non-local unicast, then create
 599            a cache_bypass connection entry */
 600         if (sysctl_cache_bypass(ipvs) && svc->fwmark &&
 601             !(iph->hdr_flags & (IP_VS_HDR_INVERSE | IP_VS_HDR_ICMP)) &&
 602             ip_vs_addr_is_unicast(net, svc->af, &iph->daddr)) {
 603                 int ret;
 604                 struct ip_vs_conn *cp;
 605                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
 606                                       iph->protocol == IPPROTO_UDP) ?
 607                                       IP_VS_CONN_F_ONE_PACKET : 0;
 608                 union nf_inet_addr daddr = { .all = { 0, 0, 0, 0 } };
 609
 610                 /* create a new connection entry */
 611                 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
 612                 {
 613                         struct ip_vs_conn_param p;
 614                         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
 615                                               &iph->saddr, pptr[0],
 616                                               &iph->daddr, pptr[1], &p);
 617                         cp = ip_vs_conn_new(&p, svc->af, &daddr, 0,
 618                                             IP_VS_CONN_F_BYPASS | flags,
 619                                             NULL, skb->mark);
 620                         if (!cp)
 621                                 return NF_DROP;
 622                 }
 623
 624                 /* statistics */
 625                 ip_vs_in_stats(cp, skb);
 626
 627                 /* set state */
 628                 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
 629
 630                 /* transmit the first SYN packet */
 631                 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
 632                 /* do not touch skb anymore */
 633
 634                 if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) && cp->control)
 635                         atomic_inc(&cp->control->in_pkts);
 636                 else
 637                         atomic_inc(&cp->in_pkts);
 638                 ip_vs_conn_put(cp);
 639                 return ret;
 640         }
 641
 642         /*
 643          * When the virtual ftp service is presented, packets destined
 644          * for other services on the VIP may get here (except services
 645          * listed in the ipvs table), pass the packets, because it is
 646          * not ipvs job to decide to drop the packets.
 647          */
 648         if (svc->port == FTPPORT && dport != FTPPORT)
 649                 return NF_ACCEPT;
 650
 651         if (unlikely(ip_vs_iph_icmp(iph)))
 652                 return NF_DROP;
 653
 654         /*
 655          * Notify the client that the destination is unreachable, and
 656          * release the socket buffer.
 657          * Since it is in IP layer, the TCP socket is not actually
 658          * created, the TCP RST packet cannot be sent, instead that
 659          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 660          */
 661 #ifdef CONFIG_IP_VS_IPV6
 662         if (svc->af == AF_INET6) {
 663                 if (!skb->dev)
 664                         skb->dev = net->loopback_dev;
 665                 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 666         } else
 667 #endif
 668                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 669
 670         return NF_DROP;
 671 }
 672
 673 #ifdef CONFIG_SYSCTL
 674
 675 static int sysctl_snat_reroute(struct netns_ipvs *ipvs)
 676 {
 677         return ipvs->sysctl_snat_reroute;
 678 }
 679
 680 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs)
 681 {
 682         return ipvs->sysctl_nat_icmp_send;
 683 }
 684
 685 #else
 686
 687 static int sysctl_snat_reroute(struct netns_ipvs *ipvs) { return 0; }
 688 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs) { return 0; }
 689
 690 #endif
 691
 692 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 693 {
 694         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 695 }
 696
 697 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
 698 {
 699         if (NF_INET_LOCAL_IN == hooknum)
 700                 return IP_DEFRAG_VS_IN;
 701         if (NF_INET_FORWARD == hooknum)
 702                 return IP_DEFRAG_VS_FWD;
 703         return IP_DEFRAG_VS_OUT;
 704 }
 705
 706 static inline int ip_vs_gather_frags(struct netns_ipvs *ipvs,
 707                                      struct sk_buff *skb, u_int32_t user)
 708 {
 709         int err;
 710
 711         local_bh_disable();
 712         err = ip_defrag(ipvs->net, skb, user);
 713         local_bh_enable();
 714         if (!err)
 715                 ip_send_check(ip_hdr(skb));
 716
 717         return err;
 718 }
 719
 720 static int ip_vs_route_me_harder(struct netns_ipvs *ipvs, int af,
 721                                  struct sk_buff *skb, unsigned int hooknum)
 722 {
 723         if (!sysctl_snat_reroute(ipvs))
 724                 return 0;
 725         /* Reroute replies only to remote clients (FORWARD and LOCAL_OUT) */
 726         if (NF_INET_LOCAL_IN == hooknum)
 727                 return 0;
 728 #ifdef CONFIG_IP_VS_IPV6
 729         if (af == AF_INET6) {
 730                 struct dst_entry *dst = skb_dst(skb);
 731
 732                 if (dst->dev && !(dst->dev->flags & IFF_LOOPBACK) &&
 733                     ip6_route_me_harder(ipvs->net, skb->sk, skb) != 0)
 734                         return 1;
 735         } else
 736 #endif
 737                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
 738                     ip_route_me_harder(ipvs->net, skb->sk, skb, RTN_LOCAL) != 0)
 739                         return 1;
 740
 741         return 0;
 742 }
 743
 744 /*
 745  * Packet has been made sufficiently writable in caller
 746  * - inout: 1=in->out, 0=out->in
 747  */
 748 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 749                     struct ip_vs_conn *cp, int inout)
 750 {
 751         struct iphdr *iph        = ip_hdr(skb);
 752         unsigned int icmp_offset = iph->ihl*4;
 753         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
 754                                                       icmp_offset);
 755         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 756
 757         if (inout) {
 758                 iph->saddr = cp->vaddr.ip;
 759                 ip_send_check(iph);
 760                 ciph->daddr = cp->vaddr.ip;
 761                 ip_send_check(ciph);
 762         } else {
 763                 iph->daddr = cp->daddr.ip;
 764                 ip_send_check(iph);
 765                 ciph->saddr = cp->daddr.ip;
 766                 ip_send_check(ciph);
 767         }
 768
 769         /* the TCP/UDP/SCTP port */
 770         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
 771             IPPROTO_SCTP == ciph->protocol) {
 772                 __be16 *ports = (void *)ciph + ciph->ihl*4;
 773
 774                 if (inout)
 775                         ports[1] = cp->vport;
 776                 else
 777                         ports[0] = cp->dport;
 778         }
 779
 780         /* And finally the ICMP checksum */
 781         icmph->checksum = 0;
 782         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 783         skb->ip_summed = CHECKSUM_UNNECESSARY;
 784
 785         if (inout)
 786                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 787                         "Forwarding altered outgoing ICMP");
 788         else
 789                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 790                         "Forwarding altered incoming ICMP");
 791 }
 792
 793 #ifdef CONFIG_IP_VS_IPV6
 794 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
 795                     struct ip_vs_conn *cp, int inout)
 796 {
 797         struct ipv6hdr *iph      = ipv6_hdr(skb);
 798         unsigned int icmp_offset = 0;
 799         unsigned int offs        = 0; /* header offset*/
 800         int protocol;
 801         struct icmp6hdr *icmph;
 802         struct ipv6hdr *ciph;
 803         unsigned short fragoffs;
 804
 805         ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
 806         icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
 807         offs = icmp_offset + sizeof(struct icmp6hdr);
 808         ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
 809
 810         protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
 811
 812         if (inout) {
 813                 iph->saddr = cp->vaddr.in6;
 814                 ciph->daddr = cp->vaddr.in6;
 815         } else {
 816                 iph->daddr = cp->daddr.in6;
 817                 ciph->saddr = cp->daddr.in6;
 818         }
 819
 820         /* the TCP/UDP/SCTP port */
 821         if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 822                           IPPROTO_SCTP == protocol)) {
 823                 __be16 *ports = (void *)(skb_network_header(skb) + offs);
 824
 825                 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
 826                               ntohs(inout ? ports[1] : ports[0]),
 827                               ntohs(inout ? cp->vport : cp->dport));
 828                 if (inout)
 829                         ports[1] = cp->vport;
 830                 else
 831                         ports[0] = cp->dport;
 832         }
 833
 834         /* And finally the ICMP checksum */
 835         icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
 836                                               skb->len - icmp_offset,
 837                                               IPPROTO_ICMPV6, 0);
 838         skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
 839         skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
 840         skb->ip_summed = CHECKSUM_PARTIAL;
 841
 842         if (inout)
 843                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 844                               (void *)ciph - (void *)iph,
 845                               "Forwarding altered outgoing ICMPv6");
 846         else
 847                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 848                               (void *)ciph - (void *)iph,
 849                               "Forwarding altered incoming ICMPv6");
 850 }
 851 #endif
 852
 853 /* Handle relevant response ICMP messages - forward to the right
 854  * destination host.
 855  */
 856 static int handle_response_icmp(int af, struct sk_buff *skb,
 857                                 union nf_inet_addr *snet,
 858                                 __u8 protocol, struct ip_vs_conn *cp,
 859                                 struct ip_vs_protocol *pp,
 860                                 unsigned int offset, unsigned int ihl,
 861                                 unsigned int hooknum)
 862 {
 863         unsigned int verdict = NF_DROP;
 864
 865         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
 866                 goto after_nat;
 867
 868         /* Ensure the checksum is correct */
 869         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 870                 /* Failed checksum! */
 871                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
 872                               IP_VS_DBG_ADDR(af, snet));
 873                 goto out;
 874         }
 875
 876         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 877             IPPROTO_SCTP == protocol)
 878                 offset += 2 * sizeof(__u16);
 879         if (skb_ensure_writable(skb, offset))
 880                 goto out;
 881
 882 #ifdef CONFIG_IP_VS_IPV6
 883         if (af == AF_INET6)
 884                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
 885         else
 886 #endif
 887                 ip_vs_nat_icmp(skb, pp, cp, 1);
 888
 889         if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
 890                 goto out;
 891
 892 after_nat:
 893         /* do the statistics and put it back */
 894         ip_vs_out_stats(cp, skb);
 895
 896         skb->ipvs_property = 1;
 897         if (!(cp->flags & IP_VS_CONN_F_NFCT))
 898                 ip_vs_notrack(skb);
 899         else
 900                 ip_vs_update_conntrack(skb, cp, 0);
 901         verdict = NF_ACCEPT;
 902
 903 out:
 904         __ip_vs_conn_put(cp);
 905
 906         return verdict;
 907 }
 908
 909 /*
 910  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 911  *      Find any that might be relevant, check against existing connections.
 912  *      Currently handles error types - unreachable, quench, ttl exceeded.
 913  */
 914 static int ip_vs_out_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb,
 915                           int *related, unsigned int hooknum)
 916 {
 917         struct iphdr *iph;
 918         struct icmphdr  _icmph, *ic;
 919         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 920         struct ip_vs_iphdr ciph;
 921         struct ip_vs_conn *cp;
 922         struct ip_vs_protocol *pp;
 923         unsigned int offset, ihl;
 924         union nf_inet_addr snet;
 925
 926         *related = 1;
 927
 928         /* reassemble IP fragments */
 929         if (ip_is_fragment(ip_hdr(skb))) {
 930                 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
 931                         return NF_STOLEN;
 932         }
 933
 934         iph = ip_hdr(skb);
 935         offset = ihl = iph->ihl * 4;
 936         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 937         if (ic == NULL)
 938                 return NF_DROP;
 939
 940         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
 941                   ic->type, ntohs(icmp_id(ic)),
 942                   &iph->saddr, &iph->daddr);
 943
 944         /*
 945          * Work through seeing if this is for us.
 946          * These checks are supposed to be in an order that means easy
 947          * things are checked first to speed up processing.... however
 948          * this means that some packets will manage to get a long way
 949          * down this stack and then be rejected, but that's life.
 950          */
 951         if ((ic->type != ICMP_DEST_UNREACH) &&
 952             (ic->type != ICMP_SOURCE_QUENCH) &&
 953             (ic->type != ICMP_TIME_EXCEEDED)) {
 954                 *related = 0;
 955                 return NF_ACCEPT;
 956         }
 957
 958         /* Now find the contained IP header */
 959         offset += sizeof(_icmph);
 960         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 961         if (cih == NULL)
 962                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 963
 964         pp = ip_vs_proto_get(cih->protocol);
 965         if (!pp)
 966                 return NF_ACCEPT;
 967
 968         /* Is the embedded protocol header present? */
 969         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 970                      pp->dont_defrag))
 971                 return NF_ACCEPT;
 972
 973         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
 974                       "Checking outgoing ICMP for");
 975
 976         ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, true, &ciph);
 977
 978         /* The embedded headers contain source and dest in reverse order */
 979         cp = INDIRECT_CALL_1(pp->conn_out_get, ip_vs_conn_out_get_proto,
 980                              ipvs, AF_INET, skb, &ciph);
 981         if (!cp)
 982                 return NF_ACCEPT;
 983
 984         snet.ip = iph->saddr;
 985         return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
 986                                     pp, ciph.len, ihl, hooknum);
 987 }
 988
 989 #ifdef CONFIG_IP_VS_IPV6
 990 static int ip_vs_out_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
 991                              int *related,  unsigned int hooknum,
 992                              struct ip_vs_iphdr *ipvsh)
 993 {
 994         struct icmp6hdr _icmph, *ic;
 995         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
 996         struct ip_vs_conn *cp;
 997         struct ip_vs_protocol *pp;
 998         union nf_inet_addr snet;
 999         unsigned int offset;
1000
1001         *related = 1;
1002         ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph);
1003         if (ic == NULL)
1004                 return NF_DROP;
1005
1006         /*
1007          * Work through seeing if this is for us.
1008          * These checks are supposed to be in an order that means easy
1009          * things are checked first to speed up processing.... however
1010          * this means that some packets will manage to get a long way
1011          * down this stack and then be rejected, but that's life.
1012          */
1013         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1014                 *related = 0;
1015                 return NF_ACCEPT;
1016         }
1017         /* Fragment header that is before ICMP header tells us that:
1018          * it's not an error message since they can't be fragmented.
1019          */
1020         if (ipvsh->flags & IP6_FH_F_FRAG)
1021                 return NF_DROP;
1022
1023         IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1024                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1025                   &ipvsh->saddr, &ipvsh->daddr);
1026
1027         if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, ipvsh->len + sizeof(_icmph),
1028                                      true, &ciph))
1029                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1030
1031         pp = ip_vs_proto_get(ciph.protocol);
1032         if (!pp)
1033                 return NF_ACCEPT;
1034
1035         /* The embedded headers contain source and dest in reverse order */
1036         cp = INDIRECT_CALL_1(pp->conn_out_get, ip_vs_conn_out_get_proto,
1037                              ipvs, AF_INET6, skb, &ciph);
1038         if (!cp)
1039                 return NF_ACCEPT;
1040
1041         snet.in6 = ciph.saddr.in6;
1042         offset = ciph.len;
1043         return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
1044                                     pp, offset, sizeof(struct ipv6hdr),
1045                                     hooknum);
1046 }
1047 #endif
1048
1049 /*
1050  * Check if sctp chunc is ABORT chunk
1051  */
1052 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
1053 {
1054         struct sctp_chunkhdr *sch, schunk;
1055         sch = skb_header_pointer(skb, nh_len + sizeof(struct sctphdr),
1056                                  sizeof(schunk), &schunk);
1057         if (sch == NULL)
1058                 return 0;
1059         if (sch->type == SCTP_CID_ABORT)
1060                 return 1;
1061         return 0;
1062 }
1063
1064 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
1065 {
1066         struct tcphdr _tcph, *th;
1067
1068         th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
1069         if (th == NULL)
1070                 return 0;
1071         return th->rst;
1072 }
1073
1074 static inline bool is_new_conn(const struct sk_buff *skb,
1075                                struct ip_vs_iphdr *iph)
1076 {
1077         switch (iph->protocol) {
1078         case IPPROTO_TCP: {
1079                 struct tcphdr _tcph, *th;
1080
1081                 th = skb_header_pointer(skb, iph->len, sizeof(_tcph), &_tcph);
1082                 if (th == NULL)
1083                         return false;
1084                 return th->syn;
1085         }
1086         case IPPROTO_SCTP: {
1087                 struct sctp_chunkhdr *sch, schunk;
1088
1089                 sch = skb_header_pointer(skb, iph->len + sizeof(struct sctphdr),
1090                                          sizeof(schunk), &schunk);
1091                 if (sch == NULL)
1092                         return false;
1093                 return sch->type == SCTP_CID_INIT;
1094         }
1095         default:
1096                 return false;
1097         }
1098 }
1099
1100 static inline bool is_new_conn_expected(const struct ip_vs_conn *cp,
1101                                         int conn_reuse_mode)
1102 {
1103         /* Controlled (FTP DATA or persistence)? */
1104         if (cp->control)
1105                 return false;
1106
1107         switch (cp->protocol) {
1108         case IPPROTO_TCP:
1109                 return (cp->state == IP_VS_TCP_S_TIME_WAIT) ||
1110                        (cp->state == IP_VS_TCP_S_CLOSE) ||
1111                         ((conn_reuse_mode & 2) &&
1112                          (cp->state == IP_VS_TCP_S_FIN_WAIT) &&
1113                          (cp->flags & IP_VS_CONN_F_NOOUTPUT));
1114         case IPPROTO_SCTP:
1115                 return cp->state == IP_VS_SCTP_S_CLOSED;
1116         default:
1117                 return false;
1118         }
1119 }
1120
1121 /* Generic function to create new connections for outgoing RS packets
1122  *
1123  * Pre-requisites for successful connection creation:
1124  * 1) Virtual Service is NOT fwmark based:
1125  *    In fwmark-VS actual vaddr and vport are unknown to IPVS
1126  * 2) Real Server and Virtual Service were NOT configured without port:
1127  *    This is to allow match of different VS to the same RS ip-addr
1128  */
1129 struct ip_vs_conn *ip_vs_new_conn_out(struct ip_vs_service *svc,
1130                                       struct ip_vs_dest *dest,
1131                                       struct sk_buff *skb,
1132                                       const struct ip_vs_iphdr *iph,
1133                                       __be16 dport,
1134                                       __be16 cport)
1135 {
1136         struct ip_vs_conn_param param;
1137         struct ip_vs_conn *ct = NULL, *cp = NULL;
1138         const union nf_inet_addr *vaddr, *daddr, *caddr;
1139         union nf_inet_addr snet;
1140         __be16 vport;
1141         unsigned int flags;
1142
1143         EnterFunction(12);
1144         vaddr = &svc->addr;
1145         vport = svc->port;
1146         daddr = &iph->saddr;
1147         caddr = &iph->daddr;
1148
1149         /* check pre-requisites are satisfied */
1150         if (svc->fwmark)
1151                 return NULL;
1152         if (!vport || !dport)
1153                 return NULL;
1154
1155         /* for persistent service first create connection template */
1156         if (svc->flags & IP_VS_SVC_F_PERSISTENT) {
1157                 /* apply netmask the same way ingress-side does */
1158 #ifdef CONFIG_IP_VS_IPV6
1159                 if (svc->af == AF_INET6)
1160                         ipv6_addr_prefix(&snet.in6, &caddr->in6,
1161                                          (__force __u32)svc->netmask);
1162                 else
1163 #endif
1164                         snet.ip = caddr->ip & svc->netmask;
1165                 /* fill params and create template if not existent */
1166                 if (ip_vs_conn_fill_param_persist(svc, skb, iph->protocol,
1167                                                   &snet, 0, vaddr,
1168                                                   vport, &param) < 0)
1169                         return NULL;
1170                 ct = ip_vs_ct_in_get(&param);
1171                 /* check if template exists and points to the same dest */
1172                 if (!ct || !ip_vs_check_template(ct, dest)) {
1173                         ct = ip_vs_conn_new(&param, dest->af, daddr, dport,
1174                                             IP_VS_CONN_F_TEMPLATE, dest, 0);
1175                         if (!ct) {
1176                                 kfree(param.pe_data);
1177                                 return NULL;
1178                         }
1179                         ct->timeout = svc->timeout;
1180                 } else {
1181                         kfree(param.pe_data);
1182                 }
1183         }
1184
1185         /* connection flags */
1186         flags = ((svc->flags & IP_VS_SVC_F_ONEPACKET) &&
1187                  iph->protocol == IPPROTO_UDP) ? IP_VS_CONN_F_ONE_PACKET : 0;
1188         /* create connection */
1189         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
1190                               caddr, cport, vaddr, vport, &param);
1191         cp = ip_vs_conn_new(&param, dest->af, daddr, dport, flags, dest, 0);
1192         if (!cp) {
1193                 if (ct)
1194                         ip_vs_conn_put(ct);
1195                 return NULL;
1196         }
1197         if (ct) {
1198                 ip_vs_control_add(cp, ct);
1199                 ip_vs_conn_put(ct);
1200         }
1201         ip_vs_conn_stats(cp, svc);
1202
1203         /* return connection (will be used to handle outgoing packet) */
1204         IP_VS_DBG_BUF(6, "New connection RS-initiated:%c c:%s:%u v:%s:%u "
1205                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
1206                       ip_vs_fwd_tag(cp),
1207                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
1208                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
1209                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
1210                       cp->flags, refcount_read(&cp->refcnt));
1211         LeaveFunction(12);
1212         return cp;
1213 }
1214
1215 /* Handle outgoing packets which are considered requests initiated by
1216  * real servers, so that subsequent responses from external client can be
1217  * routed to the right real server.
1218  * Used also for outgoing responses in OPS mode.
1219  *
1220  * Connection management is handled by persistent-engine specific callback.
1221  */
1222 static struct ip_vs_conn *__ip_vs_rs_conn_out(unsigned int hooknum,
1223                                               struct netns_ipvs *ipvs,
1224                                               int af, struct sk_buff *skb,
1225                                               const struct ip_vs_iphdr *iph)
1226 {
1227         struct ip_vs_dest *dest;
1228         struct ip_vs_conn *cp = NULL;
1229         __be16 _ports[2], *pptr;
1230
1231         if (hooknum == NF_INET_LOCAL_IN)
1232                 return NULL;
1233
1234         pptr = frag_safe_skb_hp(skb, iph->len,
1235                                 sizeof(_ports), _ports);
1236         if (!pptr)
1237                 return NULL;
1238
1239         dest = ip_vs_find_real_service(ipvs, af, iph->protocol,
1240                                        &iph->saddr, pptr[0]);
1241         if (dest) {
1242                 struct ip_vs_service *svc;
1243                 struct ip_vs_pe *pe;
1244
1245                 svc = rcu_dereference(dest->svc);
1246                 if (svc) {
1247                         pe = rcu_dereference(svc->pe);
1248                         if (pe && pe->conn_out)
1249                                 cp = pe->conn_out(svc, dest, skb, iph,
1250                                                   pptr[0], pptr[1]);
1251                 }
1252         }
1253
1254         return cp;
1255 }
1256
1257 /* Handle response packets: rewrite addresses and send away...
1258  */
1259 static unsigned int
1260 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1261                 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph,
1262                 unsigned int hooknum)
1263 {
1264         struct ip_vs_protocol *pp = pd->pp;
1265
1266         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
1267                 goto after_nat;
1268
1269         IP_VS_DBG_PKT(11, af, pp, skb, iph->off, "Outgoing packet");
1270
1271         if (skb_ensure_writable(skb, iph->len))
1272                 goto drop;
1273
1274         /* mangle the packet */
1275         if (pp->snat_handler &&
1276             !SNAT_CALL(pp->snat_handler, skb, pp, cp, iph))
1277                 goto drop;
1278
1279 #ifdef CONFIG_IP_VS_IPV6
1280         if (af == AF_INET6)
1281                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1282         else
1283 #endif
1284         {
1285                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1286                 ip_send_check(ip_hdr(skb));
1287         }
1288
1289         /*
1290          * nf_iterate does not expect change in the skb->dst->dev.
1291          * It looks like it is not fatal to enable this code for hooks
1292          * where our handlers are at the end of the chain list and
1293          * when all next handlers use skb->dst->dev and not outdev.
1294          * It will definitely route properly the inout NAT traffic
1295          * when multiple paths are used.
1296          */
1297
1298         /* For policy routing, packets originating from this
1299          * machine itself may be routed differently to packets
1300          * passing through.  We want this packet to be routed as
1301          * if it came from this machine itself.  So re-compute
1302          * the routing information.
1303          */
1304         if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
1305                 goto drop;
1306
1307         IP_VS_DBG_PKT(10, af, pp, skb, iph->off, "After SNAT");
1308
1309 after_nat:
1310         ip_vs_out_stats(cp, skb);
1311         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1312         skb->ipvs_property = 1;
1313         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1314                 ip_vs_notrack(skb);
1315         else
1316                 ip_vs_update_conntrack(skb, cp, 0);
1317         ip_vs_conn_put(cp);
1318
1319         LeaveFunction(11);
1320         return NF_ACCEPT;
1321
1322 drop:
1323         ip_vs_conn_put(cp);
1324         kfree_skb(skb);
1325         LeaveFunction(11);
1326         return NF_STOLEN;
1327 }
1328
1329 /*
1330  *      Check if outgoing packet belongs to the established ip_vs_conn.
1331  */
1332 static unsigned int
1333 ip_vs_out(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1334 {
1335         struct ip_vs_iphdr iph;
1336         struct ip_vs_protocol *pp;
1337         struct ip_vs_proto_data *pd;
1338         struct ip_vs_conn *cp;
1339         struct sock *sk;
1340
1341         EnterFunction(11);
1342
1343         /* Already marked as IPVS request or reply? */
1344         if (skb->ipvs_property)
1345                 return NF_ACCEPT;
1346
1347         sk = skb_to_full_sk(skb);
1348         /* Bad... Do not break raw sockets */
1349         if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
1350                      af == AF_INET)) {
1351
1352                 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
1353                         return NF_ACCEPT;
1354         }
1355
1356         if (unlikely(!skb_dst(skb)))
1357                 return NF_ACCEPT;
1358
1359         if (!ipvs->enable)
1360                 return NF_ACCEPT;
1361
1362         ip_vs_fill_iph_skb(af, skb, false, &iph);
1363 #ifdef CONFIG_IP_VS_IPV6
1364         if (af == AF_INET6) {
1365                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1366                         int related;
1367                         int verdict = ip_vs_out_icmp_v6(ipvs, skb, &related,
1368                                                         hooknum, &iph);
1369
1370                         if (related)
1371                                 return verdict;
1372                 }
1373         } else
1374 #endif
1375                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1376                         int related;
1377                         int verdict = ip_vs_out_icmp(ipvs, skb, &related, hooknum);
1378
1379                         if (related)
1380                                 return verdict;
1381                 }
1382
1383         pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1384         if (unlikely(!pd))
1385                 return NF_ACCEPT;
1386         pp = pd->pp;
1387
1388         /* reassemble IP fragments */
1389 #ifdef CONFIG_IP_VS_IPV6
1390         if (af == AF_INET)
1391 #endif
1392                 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1393                         if (ip_vs_gather_frags(ipvs, skb,
1394                                                ip_vs_defrag_user(hooknum)))
1395                                 return NF_STOLEN;
1396
1397                         ip_vs_fill_iph_skb(AF_INET, skb, false, &iph);
1398                 }
1399
1400         /*
1401          * Check if the packet belongs to an existing entry
1402          */
1403         cp = INDIRECT_CALL_1(pp->conn_out_get, ip_vs_conn_out_get_proto,
1404                              ipvs, af, skb, &iph);
1405
1406         if (likely(cp))
1407                 return handle_response(af, skb, pd, cp, &iph, hooknum);
1408
1409         /* Check for real-server-started requests */
1410         if (atomic_read(&ipvs->conn_out_counter)) {
1411                 /* Currently only for UDP:
1412                  * connection oriented protocols typically use
1413                  * ephemeral ports for outgoing connections, so
1414                  * related incoming responses would not match any VS
1415                  */
1416                 if (pp->protocol == IPPROTO_UDP) {
1417                         cp = __ip_vs_rs_conn_out(hooknum, ipvs, af, skb, &iph);
1418                         if (likely(cp))
1419                                 return handle_response(af, skb, pd, cp, &iph,
1420                                                        hooknum);
1421                 }
1422         }
1423
1424         if (sysctl_nat_icmp_send(ipvs) &&
1425             (pp->protocol == IPPROTO_TCP ||
1426              pp->protocol == IPPROTO_UDP ||
1427              pp->protocol == IPPROTO_SCTP)) {
1428                 __be16 _ports[2], *pptr;
1429
1430                 pptr = frag_safe_skb_hp(skb, iph.len,
1431                                          sizeof(_ports), _ports);
1432                 if (pptr == NULL)
1433                         return NF_ACCEPT;       /* Not for me */
1434                 if (ip_vs_has_real_service(ipvs, af, iph.protocol, &iph.saddr,
1435                                            pptr[0])) {
1436                         /*
1437                          * Notify the real server: there is no
1438                          * existing entry if it is not RST
1439                          * packet or not TCP packet.
1440                          */
1441                         if ((iph.protocol != IPPROTO_TCP &&
1442                              iph.protocol != IPPROTO_SCTP)
1443                              || ((iph.protocol == IPPROTO_TCP
1444                                   && !is_tcp_reset(skb, iph.len))
1445                                  || (iph.protocol == IPPROTO_SCTP
1446                                         && !is_sctp_abort(skb,
1447                                                 iph.len)))) {
1448 #ifdef CONFIG_IP_VS_IPV6
1449                                 if (af == AF_INET6) {
1450                                         if (!skb->dev)
1451                                                 skb->dev = ipvs->net->loopback_dev;
1452                                         icmpv6_send(skb,
1453                                                     ICMPV6_DEST_UNREACH,
1454                                                     ICMPV6_PORT_UNREACH,
1455                                                     0);
1456                                 } else
1457 #endif
1458                                         icmp_send(skb,
1459                                                   ICMP_DEST_UNREACH,
1460                                                   ICMP_PORT_UNREACH, 0);
1461                                 return NF_DROP;
1462                         }
1463                 }
1464         }
1465
1466         IP_VS_DBG_PKT(12, af, pp, skb, iph.off,
1467                       "ip_vs_out: packet continues traversal as normal");
1468         return NF_ACCEPT;
1469 }
1470
1471 /*
1472  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1473  *      used only for VS/NAT.
1474  *      Check if packet is reply for established ip_vs_conn.
1475  */
1476 static unsigned int
1477 ip_vs_reply4(void *priv, struct sk_buff *skb,
1478              const struct nf_hook_state *state)
1479 {
1480         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1481 }
1482
1483 /*
1484  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1485  *      Check if packet is reply for established ip_vs_conn.
1486  */
1487 static unsigned int
1488 ip_vs_local_reply4(void *priv, struct sk_buff *skb,
1489                    const struct nf_hook_state *state)
1490 {
1491         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1492 }
1493
1494 #ifdef CONFIG_IP_VS_IPV6
1495
1496 /*
1497  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1498  *      used only for VS/NAT.
1499  *      Check if packet is reply for established ip_vs_conn.
1500  */
1501 static unsigned int
1502 ip_vs_reply6(void *priv, struct sk_buff *skb,
1503              const struct nf_hook_state *state)
1504 {
1505         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1506 }
1507
1508 /*
1509  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1510  *      Check if packet is reply for established ip_vs_conn.
1511  */
1512 static unsigned int
1513 ip_vs_local_reply6(void *priv, struct sk_buff *skb,
1514                    const struct nf_hook_state *state)
1515 {
1516         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1517 }
1518
1519 #endif
1520
1521 static unsigned int
1522 ip_vs_try_to_schedule(struct netns_ipvs *ipvs, int af, struct sk_buff *skb,
1523                       struct ip_vs_proto_data *pd,
1524                       int *verdict, struct ip_vs_conn **cpp,
1525                       struct ip_vs_iphdr *iph)
1526 {
1527         struct ip_vs_protocol *pp = pd->pp;
1528
1529         if (!iph->fragoffs) {
1530                 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1531                  * replayed fragment zero will already have created the cp
1532                  */
1533
1534                 /* Schedule and create new connection entry into cpp */
1535                 if (!pp->conn_schedule(ipvs, af, skb, pd, verdict, cpp, iph))
1536                         return 0;
1537         }
1538
1539         if (unlikely(!*cpp)) {
1540                 /* sorry, all this trouble for a no-hit :) */
1541                 IP_VS_DBG_PKT(12, af, pp, skb, iph->off,
1542                               "ip_vs_in: packet continues traversal as normal");
1543
1544                 /* Fragment couldn't be mapped to a conn entry */
1545                 if (iph->fragoffs)
1546                         IP_VS_DBG_PKT(7, af, pp, skb, iph->off,
1547                                       "unhandled fragment");
1548
1549                 *verdict = NF_ACCEPT;
1550                 return 0;
1551         }
1552
1553         return 1;
1554 }
1555
1556 /* Check the UDP tunnel and return its header length */
1557 static int ipvs_udp_decap(struct netns_ipvs *ipvs, struct sk_buff *skb,
1558                           unsigned int offset, __u16 af,
1559                           const union nf_inet_addr *daddr, __u8 *proto)
1560 {
1561         struct udphdr _udph, *udph;
1562         struct ip_vs_dest *dest;
1563
1564         udph = skb_header_pointer(skb, offset, sizeof(_udph), &_udph);
1565         if (!udph)
1566                 goto unk;
1567         offset += sizeof(struct udphdr);
1568         dest = ip_vs_find_tunnel(ipvs, af, daddr, udph->dest);
1569         if (!dest)
1570                 goto unk;
1571         if (dest->tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GUE) {
1572                 struct guehdr _gueh, *gueh;
1573
1574                 gueh = skb_header_pointer(skb, offset, sizeof(_gueh), &_gueh);
1575                 if (!gueh)
1576                         goto unk;
1577                 if (gueh->control != 0 || gueh->version != 0)
1578                         goto unk;
1579                 /* Later we can support also IPPROTO_IPV6 */
1580                 if (gueh->proto_ctype != IPPROTO_IPIP)
1581                         goto unk;
1582                 *proto = gueh->proto_ctype;
1583                 return sizeof(struct udphdr) + sizeof(struct guehdr) +
1584                        (gueh->hlen << 2);
1585         }
1586
1587 unk:
1588         return 0;
1589 }
1590
1591 /* Check the GRE tunnel and return its header length */
1592 static int ipvs_gre_decap(struct netns_ipvs *ipvs, struct sk_buff *skb,
1593                           unsigned int offset, __u16 af,
1594                           const union nf_inet_addr *daddr, __u8 *proto)
1595 {
1596         struct gre_base_hdr _greh, *greh;
1597         struct ip_vs_dest *dest;
1598
1599         greh = skb_header_pointer(skb, offset, sizeof(_greh), &_greh);
1600         if (!greh)
1601                 goto unk;
1602         dest = ip_vs_find_tunnel(ipvs, af, daddr, 0);
1603         if (!dest)
1604                 goto unk;
1605         if (dest->tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GRE) {
1606                 __be16 type;
1607
1608                 /* Only support version 0 and C (csum) */
1609                 if ((greh->flags & ~GRE_CSUM) != 0)
1610                         goto unk;
1611                 type = greh->protocol;
1612                 /* Later we can support also IPPROTO_IPV6 */
1613                 if (type != htons(ETH_P_IP))
1614                         goto unk;
1615                 *proto = IPPROTO_IPIP;
1616                 return gre_calc_hlen(gre_flags_to_tnl_flags(greh->flags));
1617         }
1618
1619 unk:
1620         return 0;
1621 }
1622
1623 /*
1624  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1625  *      Find any that might be relevant, check against existing connections,
1626  *      forward to the right destination host if relevant.
1627  *      Currently handles error types - unreachable, quench, ttl exceeded.
1628  */
1629 static int
1630 ip_vs_in_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb, int *related,
1631               unsigned int hooknum)
1632 {
1633         struct iphdr *iph;
1634         struct icmphdr  _icmph, *ic;
1635         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1636         struct ip_vs_iphdr ciph;
1637         struct ip_vs_conn *cp;
1638         struct ip_vs_protocol *pp;
1639         struct ip_vs_proto_data *pd;
1640         unsigned int offset, offset2, ihl, verdict;
1641         bool tunnel, new_cp = false;
1642         union nf_inet_addr *raddr;
1643         char *outer_proto = "IPIP";
1644
1645         *related = 1;
1646
1647         /* reassemble IP fragments */
1648         if (ip_is_fragment(ip_hdr(skb))) {
1649                 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
1650                         return NF_STOLEN;
1651         }
1652
1653         iph = ip_hdr(skb);
1654         offset = ihl = iph->ihl * 4;
1655         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1656         if (ic == NULL)
1657                 return NF_DROP;
1658
1659         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1660                   ic->type, ntohs(icmp_id(ic)),
1661                   &iph->saddr, &iph->daddr);
1662
1663         /*
1664          * Work through seeing if this is for us.
1665          * These checks are supposed to be in an order that means easy
1666          * things are checked first to speed up processing.... however
1667          * this means that some packets will manage to get a long way
1668          * down this stack and then be rejected, but that's life.
1669          */
1670         if ((ic->type != ICMP_DEST_UNREACH) &&
1671             (ic->type != ICMP_SOURCE_QUENCH) &&
1672             (ic->type != ICMP_TIME_EXCEEDED)) {
1673                 *related = 0;
1674                 return NF_ACCEPT;
1675         }
1676
1677         /* Now find the contained IP header */
1678         offset += sizeof(_icmph);
1679         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1680         if (cih == NULL)
1681                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1682         raddr = (union nf_inet_addr *)&cih->daddr;
1683
1684         /* Special case for errors for IPIP/UDP/GRE tunnel packets */
1685         tunnel = false;
1686         if (cih->protocol == IPPROTO_IPIP) {
1687                 struct ip_vs_dest *dest;
1688
1689                 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1690                         return NF_ACCEPT;
1691                 /* Error for our IPIP must arrive at LOCAL_IN */
1692                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1693                         return NF_ACCEPT;
1694                 dest = ip_vs_find_tunnel(ipvs, AF_INET, raddr, 0);
1695                 /* Only for known tunnel */
1696                 if (!dest || dest->tun_type != IP_VS_CONN_F_TUNNEL_TYPE_IPIP)
1697                         return NF_ACCEPT;
1698                 offset += cih->ihl * 4;
1699                 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1700                 if (cih == NULL)
1701                         return NF_ACCEPT; /* The packet looks wrong, ignore */
1702                 tunnel = true;
1703         } else if ((cih->protocol == IPPROTO_UDP ||     /* Can be UDP encap */
1704                     cih->protocol == IPPROTO_GRE) &&    /* Can be GRE encap */
1705                    /* Error for our tunnel must arrive at LOCAL_IN */
1706                    (skb_rtable(skb)->rt_flags & RTCF_LOCAL)) {
1707                 __u8 iproto;
1708                 int ulen;
1709
1710                 /* Non-first fragment has no UDP/GRE header */
1711                 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1712                         return NF_ACCEPT;
1713                 offset2 = offset + cih->ihl * 4;
1714                 if (cih->protocol == IPPROTO_UDP) {
1715                         ulen = ipvs_udp_decap(ipvs, skb, offset2, AF_INET,
1716                                               raddr, &iproto);
1717                         outer_proto = "UDP";
1718                 } else {
1719                         ulen = ipvs_gre_decap(ipvs, skb, offset2, AF_INET,
1720                                               raddr, &iproto);
1721                         outer_proto = "GRE";
1722                 }
1723                 if (ulen > 0) {
1724                         /* Skip IP and UDP/GRE tunnel headers */
1725                         offset = offset2 + ulen;
1726                         /* Now we should be at the original IP header */
1727                         cih = skb_header_pointer(skb, offset, sizeof(_ciph),
1728                                                  &_ciph);
1729                         if (cih && cih->version == 4 && cih->ihl >= 5 &&
1730                             iproto == IPPROTO_IPIP)
1731                                 tunnel = true;
1732                         else
1733                                 return NF_ACCEPT;
1734                 }
1735         }
1736
1737         pd = ip_vs_proto_data_get(ipvs, cih->protocol);
1738         if (!pd)
1739                 return NF_ACCEPT;
1740         pp = pd->pp;
1741
1742         /* Is the embedded protocol header present? */
1743         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1744                      pp->dont_defrag))
1745                 return NF_ACCEPT;
1746
1747         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1748                       "Checking incoming ICMP for");
1749
1750         offset2 = offset;
1751         ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, !tunnel, &ciph);
1752         offset = ciph.len;
1753
1754         /* The embedded headers contain source and dest in reverse order.
1755          * For IPIP/UDP/GRE tunnel this is error for request, not for reply.
1756          */
1757         cp = INDIRECT_CALL_1(pp->conn_in_get, ip_vs_conn_in_get_proto,
1758                              ipvs, AF_INET, skb, &ciph);
1759
1760         if (!cp) {
1761                 int v;
1762
1763                 if (tunnel || !sysctl_schedule_icmp(ipvs))
1764                         return NF_ACCEPT;
1765
1766                 if (!ip_vs_try_to_schedule(ipvs, AF_INET, skb, pd, &v, &cp, &ciph))
1767                         return v;
1768                 new_cp = true;
1769         }
1770
1771         verdict = NF_DROP;
1772
1773         /* Ensure the checksum is correct */
1774         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1775                 /* Failed checksum! */
1776                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1777                           &iph->saddr);
1778                 goto out;
1779         }
1780
1781         if (tunnel) {
1782                 __be32 info = ic->un.gateway;
1783                 __u8 type = ic->type;
1784                 __u8 code = ic->code;
1785
1786                 /* Update the MTU */
1787                 if (ic->type == ICMP_DEST_UNREACH &&
1788                     ic->code == ICMP_FRAG_NEEDED) {
1789                         struct ip_vs_dest *dest = cp->dest;
1790                         u32 mtu = ntohs(ic->un.frag.mtu);
1791                         __be16 frag_off = cih->frag_off;
1792
1793                         /* Strip outer IP and ICMP, go to IPIP/UDP/GRE header */
1794                         if (pskb_pull(skb, ihl + sizeof(_icmph)) == NULL)
1795                                 goto ignore_tunnel;
1796                         offset2 -= ihl + sizeof(_icmph);
1797                         skb_reset_network_header(skb);
1798                         IP_VS_DBG(12, "ICMP for %s %pI4->%pI4: mtu=%u\n",
1799                                   outer_proto, &ip_hdr(skb)->saddr,
1800                                   &ip_hdr(skb)->daddr, mtu);
1801                         ipv4_update_pmtu(skb, ipvs->net, mtu, 0, 0);
1802                         /* Client uses PMTUD? */
1803                         if (!(frag_off & htons(IP_DF)))
1804                                 goto ignore_tunnel;
1805                         /* Prefer the resulting PMTU */
1806                         if (dest) {
1807                                 struct ip_vs_dest_dst *dest_dst;
1808
1809                                 dest_dst = rcu_dereference(dest->dest_dst);
1810                                 if (dest_dst)
1811                                         mtu = dst_mtu(dest_dst->dst_cache);
1812                         }
1813                         if (mtu > 68 + sizeof(struct iphdr))
1814                                 mtu -= sizeof(struct iphdr);
1815                         info = htonl(mtu);
1816                 }
1817                 /* Strip outer IP, ICMP and IPIP/UDP/GRE, go to IP header of
1818                  * original request.
1819                  */
1820                 if (pskb_pull(skb, offset2) == NULL)
1821                         goto ignore_tunnel;
1822                 skb_reset_network_header(skb);
1823                 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1824                         &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1825                         type, code, ntohl(info));
1826                 icmp_send(skb, type, code, info);
1827                 /* ICMP can be shorter but anyways, account it */
1828                 ip_vs_out_stats(cp, skb);
1829
1830 ignore_tunnel:
1831                 consume_skb(skb);
1832                 verdict = NF_STOLEN;
1833                 goto out;
1834         }
1835
1836         /* do the statistics and put it back */
1837         ip_vs_in_stats(cp, skb);
1838         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol ||
1839             IPPROTO_SCTP == cih->protocol)
1840                 offset += 2 * sizeof(__u16);
1841         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1842
1843 out:
1844         if (likely(!new_cp))
1845                 __ip_vs_conn_put(cp);
1846         else
1847                 ip_vs_conn_put(cp);
1848
1849         return verdict;
1850 }
1851
1852 #ifdef CONFIG_IP_VS_IPV6
1853 static int ip_vs_in_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
1854                             int *related, unsigned int hooknum,
1855                             struct ip_vs_iphdr *iph)
1856 {
1857         struct icmp6hdr _icmph, *ic;
1858         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1859         struct ip_vs_conn *cp;
1860         struct ip_vs_protocol *pp;
1861         struct ip_vs_proto_data *pd;
1862         unsigned int offset, verdict;
1863         bool new_cp = false;
1864
1865         *related = 1;
1866
1867         ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph);
1868         if (ic == NULL)
1869                 return NF_DROP;
1870
1871         /*
1872          * Work through seeing if this is for us.
1873          * These checks are supposed to be in an order that means easy
1874          * things are checked first to speed up processing.... however
1875          * this means that some packets will manage to get a long way
1876          * down this stack and then be rejected, but that's life.
1877          */
1878         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1879                 *related = 0;
1880                 return NF_ACCEPT;
1881         }
1882         /* Fragment header that is before ICMP header tells us that:
1883          * it's not an error message since they can't be fragmented.
1884          */
1885         if (iph->flags & IP6_FH_F_FRAG)
1886                 return NF_DROP;
1887
1888         IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1889                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1890                   &iph->saddr, &iph->daddr);
1891
1892         offset = iph->len + sizeof(_icmph);
1893         if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, offset, true, &ciph))
1894                 return NF_ACCEPT;
1895
1896         pd = ip_vs_proto_data_get(ipvs, ciph.protocol);
1897         if (!pd)
1898                 return NF_ACCEPT;
1899         pp = pd->pp;
1900
1901         /* Cannot handle fragmented embedded protocol */
1902         if (ciph.fragoffs)
1903                 return NF_ACCEPT;
1904
1905         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
1906                       "Checking incoming ICMPv6 for");
1907
1908         /* The embedded headers contain source and dest in reverse order
1909          * if not from localhost
1910          */
1911         cp = INDIRECT_CALL_1(pp->conn_in_get, ip_vs_conn_in_get_proto,
1912                              ipvs, AF_INET6, skb, &ciph);
1913
1914         if (!cp) {
1915                 int v;
1916
1917                 if (!sysctl_schedule_icmp(ipvs))
1918                         return NF_ACCEPT;
1919
1920                 if (!ip_vs_try_to_schedule(ipvs, AF_INET6, skb, pd, &v, &cp, &ciph))
1921                         return v;
1922
1923                 new_cp = true;
1924         }
1925
1926         /* VS/TUN, VS/DR and LOCALNODE just let it go */
1927         if ((hooknum == NF_INET_LOCAL_OUT) &&
1928             (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1929                 verdict = NF_ACCEPT;
1930                 goto out;
1931         }
1932
1933         /* do the statistics and put it back */
1934         ip_vs_in_stats(cp, skb);
1935
1936         /* Need to mangle contained IPv6 header in ICMPv6 packet */
1937         offset = ciph.len;
1938         if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1939             IPPROTO_SCTP == ciph.protocol)
1940                 offset += 2 * sizeof(__u16); /* Also mangle ports */
1941
1942         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset, hooknum, &ciph);
1943
1944 out:
1945         if (likely(!new_cp))
1946                 __ip_vs_conn_put(cp);
1947         else
1948                 ip_vs_conn_put(cp);
1949
1950         return verdict;
1951 }
1952 #endif
1953
1954
1955 /*
1956  *      Check if it's for virtual services, look it up,
1957  *      and send it on its way...
1958  */
1959 static unsigned int
1960 ip_vs_in(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1961 {
1962         struct ip_vs_iphdr iph;
1963         struct ip_vs_protocol *pp;
1964         struct ip_vs_proto_data *pd;
1965         struct ip_vs_conn *cp;
1966         int ret, pkts;
1967         int conn_reuse_mode;
1968         struct sock *sk;
1969
1970         /* Already marked as IPVS request or reply? */
1971         if (skb->ipvs_property)
1972                 return NF_ACCEPT;
1973
1974         /*
1975          *      Big tappo:
1976          *      - remote client: only PACKET_HOST
1977          *      - route: used for struct net when skb->dev is unset
1978          */
1979         if (unlikely((skb->pkt_type != PACKET_HOST &&
1980                       hooknum != NF_INET_LOCAL_OUT) ||
1981                      !skb_dst(skb))) {
1982                 ip_vs_fill_iph_skb(af, skb, false, &iph);
1983                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1984                               " ignored in hook %u\n",
1985                               skb->pkt_type, iph.protocol,
1986                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1987                 return NF_ACCEPT;
1988         }
1989         /* ipvs enabled in this netns ? */
1990         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1991                 return NF_ACCEPT;
1992
1993         ip_vs_fill_iph_skb(af, skb, false, &iph);
1994
1995         /* Bad... Do not break raw sockets */
1996         sk = skb_to_full_sk(skb);
1997         if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
1998                      af == AF_INET)) {
1999
2000                 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
2001                         return NF_ACCEPT;
2002         }
2003
2004 #ifdef CONFIG_IP_VS_IPV6
2005         if (af == AF_INET6) {
2006                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
2007                         int related;
2008                         int verdict = ip_vs_in_icmp_v6(ipvs, skb, &related,
2009                                                        hooknum, &iph);
2010
2011                         if (related)
2012                                 return verdict;
2013                 }
2014         } else
2015 #endif
2016                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
2017                         int related;
2018                         int verdict = ip_vs_in_icmp(ipvs, skb, &related,
2019                                                     hooknum);
2020
2021                         if (related)
2022                                 return verdict;
2023                 }
2024
2025         /* Protocol supported? */
2026         pd = ip_vs_proto_data_get(ipvs, iph.protocol);
2027         if (unlikely(!pd)) {
2028                 /* The only way we'll see this packet again is if it's
2029                  * encapsulated, so mark it with ipvs_property=1 so we
2030                  * skip it if we're ignoring tunneled packets
2031                  */
2032                 if (sysctl_ignore_tunneled(ipvs))
2033                         skb->ipvs_property = 1;
2034
2035                 return NF_ACCEPT;
2036         }
2037         pp = pd->pp;
2038         /*
2039          * Check if the packet belongs to an existing connection entry
2040          */
2041         cp = INDIRECT_CALL_1(pp->conn_in_get, ip_vs_conn_in_get_proto,
2042                              ipvs, af, skb, &iph);
2043
2044         conn_reuse_mode = sysctl_conn_reuse_mode(ipvs);
2045         if (conn_reuse_mode && !iph.fragoffs && is_new_conn(skb, &iph) && cp) {
2046                 bool old_ct = false, resched = false;
2047
2048                 if (unlikely(sysctl_expire_nodest_conn(ipvs)) && cp->dest &&
2049                     unlikely(!atomic_read(&cp->dest->weight))) {
2050                         resched = true;
2051                         old_ct = ip_vs_conn_uses_old_conntrack(cp, skb);
2052                 } else if (is_new_conn_expected(cp, conn_reuse_mode)) {
2053                         old_ct = ip_vs_conn_uses_old_conntrack(cp, skb);
2054                         if (!atomic_read(&cp->n_control)) {
2055                                 resched = true;
2056                         } else {
2057                                 /* Do not reschedule controlling connection
2058                                  * that uses conntrack while it is still
2059                                  * referenced by controlled connection(s).
2060                                  */
2061                                 resched = !old_ct;
2062                         }
2063                 }
2064
2065                 if (resched) {
2066                         if (!old_ct)
2067                                 cp->flags &= ~IP_VS_CONN_F_NFCT;
2068                         if (!atomic_read(&cp->n_control))
2069                                 ip_vs_conn_expire_now(cp);
2070                         __ip_vs_conn_put(cp);
2071                         if (old_ct)
2072                                 return NF_DROP;
2073                         cp = NULL;
2074                 }
2075         }
2076
2077         /* Check the server status */
2078         if (cp && cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
2079                 /* the destination server is not available */
2080                 if (sysctl_expire_nodest_conn(ipvs)) {
2081                         bool old_ct = ip_vs_conn_uses_old_conntrack(cp, skb);
2082
2083                         if (!old_ct)
2084                                 cp->flags &= ~IP_VS_CONN_F_NFCT;
2085
2086                         ip_vs_conn_expire_now(cp);
2087                         __ip_vs_conn_put(cp);
2088                         if (old_ct)
2089                                 return NF_DROP;
2090                         cp = NULL;
2091                 } else {
2092                         __ip_vs_conn_put(cp);
2093                         return NF_DROP;
2094                 }
2095         }
2096
2097         if (unlikely(!cp)) {
2098                 int v;
2099
2100                 if (!ip_vs_try_to_schedule(ipvs, af, skb, pd, &v, &cp, &iph))
2101                         return v;
2102         }
2103
2104         IP_VS_DBG_PKT(11, af, pp, skb, iph.off, "Incoming packet");
2105
2106         ip_vs_in_stats(cp, skb);
2107         ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
2108         if (cp->packet_xmit)
2109                 ret = cp->packet_xmit(skb, cp, pp, &iph);
2110                 /* do not touch skb anymore */
2111         else {
2112                 IP_VS_DBG_RL("warning: packet_xmit is null");
2113                 ret = NF_ACCEPT;
2114         }
2115
2116         /* Increase its packet counter and check if it is needed
2117          * to be synchronized
2118          *
2119          * Sync connection if it is about to close to
2120          * encorage the standby servers to update the connections timeout
2121          *
2122          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
2123          */
2124
2125         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
2126                 pkts = sysctl_sync_threshold(ipvs);
2127         else
2128                 pkts = atomic_inc_return(&cp->in_pkts);
2129
2130         if (ipvs->sync_state & IP_VS_STATE_MASTER)
2131                 ip_vs_sync_conn(ipvs, cp, pkts);
2132         else if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) && cp->control)
2133                 /* increment is done inside ip_vs_sync_conn too */
2134                 atomic_inc(&cp->control->in_pkts);
2135
2136         ip_vs_conn_put(cp);
2137         return ret;
2138 }
2139
2140 /*
2141  *      AF_INET handler in NF_INET_LOCAL_IN chain
2142  *      Schedule and forward packets from remote clients
2143  */
2144 static unsigned int
2145 ip_vs_remote_request4(void *priv, struct sk_buff *skb,
2146                       const struct nf_hook_state *state)
2147 {
2148         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
2149 }
2150
2151 /*
2152  *      AF_INET handler in NF_INET_LOCAL_OUT chain
2153  *      Schedule and forward packets from local clients
2154  */
2155 static unsigned int
2156 ip_vs_local_request4(void *priv, struct sk_buff *skb,
2157                      const struct nf_hook_state *state)
2158 {
2159         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
2160 }
2161
2162 #ifdef CONFIG_IP_VS_IPV6
2163
2164 /*
2165  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
2166  *      Schedule and forward packets from remote clients
2167  */
2168 static unsigned int
2169 ip_vs_remote_request6(void *priv, struct sk_buff *skb,
2170                       const struct nf_hook_state *state)
2171 {
2172         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
2173 }
2174
2175 /*
2176  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
2177  *      Schedule and forward packets from local clients
2178  */
2179 static unsigned int
2180 ip_vs_local_request6(void *priv, struct sk_buff *skb,
2181                      const struct nf_hook_state *state)
2182 {
2183         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
2184 }
2185
2186 #endif
2187
2188
2189 /*
2190  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
2191  *      related packets destined for 0.0.0.0/0.
2192  *      When fwmark-based virtual service is used, such as transparent
2193  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
2194  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
2195  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
2196  *      and send them to ip_vs_in_icmp.
2197  */
2198 static unsigned int
2199 ip_vs_forward_icmp(void *priv, struct sk_buff *skb,
2200                    const struct nf_hook_state *state)
2201 {
2202         int r;
2203         struct netns_ipvs *ipvs = net_ipvs(state->net);
2204
2205         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
2206                 return NF_ACCEPT;
2207
2208         /* ipvs enabled in this netns ? */
2209         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
2210                 return NF_ACCEPT;
2211
2212         return ip_vs_in_icmp(ipvs, skb, &r, state->hook);
2213 }
2214
2215 #ifdef CONFIG_IP_VS_IPV6
2216 static unsigned int
2217 ip_vs_forward_icmp_v6(void *priv, struct sk_buff *skb,
2218                       const struct nf_hook_state *state)
2219 {
2220         int r;
2221         struct netns_ipvs *ipvs = net_ipvs(state->net);
2222         struct ip_vs_iphdr iphdr;
2223
2224         ip_vs_fill_iph_skb(AF_INET6, skb, false, &iphdr);
2225         if (iphdr.protocol != IPPROTO_ICMPV6)
2226                 return NF_ACCEPT;
2227
2228         /* ipvs enabled in this netns ? */
2229         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
2230                 return NF_ACCEPT;
2231
2232         return ip_vs_in_icmp_v6(ipvs, skb, &r, state->hook, &iphdr);
2233 }
2234 #endif
2235
2236
2237 static const struct nf_hook_ops ip_vs_ops4[] = {
2238         /* After packet filtering, change source only for VS/NAT */
2239         {
2240                 .hook           = ip_vs_reply4,
2241                 .pf             = NFPROTO_IPV4,
2242                 .hooknum        = NF_INET_LOCAL_IN,
2243                 .priority       = NF_IP_PRI_NAT_SRC - 2,
2244         },
2245         /* After packet filtering, forward packet through VS/DR, VS/TUN,
2246          * or VS/NAT(change destination), so that filtering rules can be
2247          * applied to IPVS. */
2248         {
2249                 .hook           = ip_vs_remote_request4,
2250                 .pf             = NFPROTO_IPV4,
2251                 .hooknum        = NF_INET_LOCAL_IN,
2252                 .priority       = NF_IP_PRI_NAT_SRC - 1,
2253         },
2254         /* Before ip_vs_in, change source only for VS/NAT */
2255         {
2256                 .hook           = ip_vs_local_reply4,
2257                 .pf             = NFPROTO_IPV4,
2258                 .hooknum        = NF_INET_LOCAL_OUT,
2259                 .priority       = NF_IP_PRI_NAT_DST + 1,
2260         },
2261         /* After mangle, schedule and forward local requests */
2262         {
2263                 .hook           = ip_vs_local_request4,
2264                 .pf             = NFPROTO_IPV4,
2265                 .hooknum        = NF_INET_LOCAL_OUT,
2266                 .priority       = NF_IP_PRI_NAT_DST + 2,
2267         },
2268         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2269          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2270         {
2271                 .hook           = ip_vs_forward_icmp,
2272                 .pf             = NFPROTO_IPV4,
2273                 .hooknum        = NF_INET_FORWARD,
2274                 .priority       = 99,
2275         },
2276         /* After packet filtering, change source only for VS/NAT */
2277         {
2278                 .hook           = ip_vs_reply4,
2279                 .pf             = NFPROTO_IPV4,
2280                 .hooknum        = NF_INET_FORWARD,
2281                 .priority       = 100,
2282         },
2283 };
2284
2285 #ifdef CONFIG_IP_VS_IPV6
2286 static const struct nf_hook_ops ip_vs_ops6[] = {
2287         /* After packet filtering, change source only for VS/NAT */
2288         {
2289                 .hook           = ip_vs_reply6,
2290                 .pf             = NFPROTO_IPV6,
2291                 .hooknum        = NF_INET_LOCAL_IN,
2292                 .priority       = NF_IP6_PRI_NAT_SRC - 2,
2293         },
2294         /* After packet filtering, forward packet through VS/DR, VS/TUN,
2295          * or VS/NAT(change destination), so that filtering rules can be
2296          * applied to IPVS. */
2297         {
2298                 .hook           = ip_vs_remote_request6,
2299                 .pf             = NFPROTO_IPV6,
2300                 .hooknum        = NF_INET_LOCAL_IN,
2301                 .priority       = NF_IP6_PRI_NAT_SRC - 1,
2302         },
2303         /* Before ip_vs_in, change source only for VS/NAT */
2304         {
2305                 .hook           = ip_vs_local_reply6,
2306                 .pf             = NFPROTO_IPV6,
2307                 .hooknum        = NF_INET_LOCAL_OUT,
2308                 .priority       = NF_IP6_PRI_NAT_DST + 1,
2309         },
2310         /* After mangle, schedule and forward local requests */
2311         {
2312                 .hook           = ip_vs_local_request6,
2313                 .pf             = NFPROTO_IPV6,
2314                 .hooknum        = NF_INET_LOCAL_OUT,
2315                 .priority       = NF_IP6_PRI_NAT_DST + 2,
2316         },
2317         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2318          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2319         {
2320                 .hook           = ip_vs_forward_icmp_v6,
2321                 .pf             = NFPROTO_IPV6,
2322                 .hooknum        = NF_INET_FORWARD,
2323                 .priority       = 99,
2324         },
2325         /* After packet filtering, change source only for VS/NAT */
2326         {
2327                 .hook           = ip_vs_reply6,
2328                 .pf             = NFPROTO_IPV6,
2329                 .hooknum        = NF_INET_FORWARD,
2330                 .priority       = 100,
2331         },
2332 };
2333 #endif
2334
2335 int ip_vs_register_hooks(struct netns_ipvs *ipvs, unsigned int af)
2336 {
2337         const struct nf_hook_ops *ops;
2338         unsigned int count;
2339         unsigned int afmask;
2340         int ret = 0;
2341
2342         if (af == AF_INET6) {
2343 #ifdef CONFIG_IP_VS_IPV6
2344                 ops = ip_vs_ops6;
2345                 count = ARRAY_SIZE(ip_vs_ops6);
2346                 afmask = 2;
2347 #else
2348                 return -EINVAL;
2349 #endif
2350         } else {
2351                 ops = ip_vs_ops4;
2352                 count = ARRAY_SIZE(ip_vs_ops4);
2353                 afmask = 1;
2354         }
2355
2356         if (!(ipvs->hooks_afmask & afmask)) {
2357                 ret = nf_register_net_hooks(ipvs->net, ops, count);
2358                 if (ret >= 0)
2359                         ipvs->hooks_afmask |= afmask;
2360         }
2361         return ret;
2362 }
2363
2364 void ip_vs_unregister_hooks(struct netns_ipvs *ipvs, unsigned int af)
2365 {
2366         const struct nf_hook_ops *ops;
2367         unsigned int count;
2368         unsigned int afmask;
2369
2370         if (af == AF_INET6) {
2371 #ifdef CONFIG_IP_VS_IPV6
2372                 ops = ip_vs_ops6;
2373                 count = ARRAY_SIZE(ip_vs_ops6);
2374                 afmask = 2;
2375 #else
2376                 return;
2377 #endif
2378         } else {
2379                 ops = ip_vs_ops4;
2380                 count = ARRAY_SIZE(ip_vs_ops4);
2381                 afmask = 1;
2382         }
2383
2384         if (ipvs->hooks_afmask & afmask) {
2385                 nf_unregister_net_hooks(ipvs->net, ops, count);
2386                 ipvs->hooks_afmask &= ~afmask;
2387         }
2388 }
2389
2390 /*
2391  *      Initialize IP Virtual Server netns mem.
2392  */
2393 static int __net_init __ip_vs_init(struct net *net)
2394 {
2395         struct netns_ipvs *ipvs;
2396
2397         ipvs = net_generic(net, ip_vs_net_id);
2398         if (ipvs == NULL)
2399                 return -ENOMEM;
2400
2401         /* Hold the beast until a service is registered */
2402         ipvs->enable = 0;
2403         ipvs->net = net;
2404         /* Counters used for creating unique names */
2405         ipvs->gen = atomic_read(&ipvs_netns_cnt);
2406         atomic_inc(&ipvs_netns_cnt);
2407         net->ipvs = ipvs;
2408
2409         if (ip_vs_estimator_net_init(ipvs) < 0)
2410                 goto estimator_fail;
2411
2412         if (ip_vs_control_net_init(ipvs) < 0)
2413                 goto control_fail;
2414
2415         if (ip_vs_protocol_net_init(ipvs) < 0)
2416                 goto protocol_fail;
2417
2418         if (ip_vs_app_net_init(ipvs) < 0)
2419                 goto app_fail;
2420
2421         if (ip_vs_conn_net_init(ipvs) < 0)
2422                 goto conn_fail;
2423
2424         if (ip_vs_sync_net_init(ipvs) < 0)
2425                 goto sync_fail;
2426
2427         return 0;
2428 /*
2429  * Error handling
2430  */
2431
2432 sync_fail:
2433         ip_vs_conn_net_cleanup(ipvs);
2434 conn_fail:
2435         ip_vs_app_net_cleanup(ipvs);
2436 app_fail:
2437         ip_vs_protocol_net_cleanup(ipvs);
2438 protocol_fail:
2439         ip_vs_control_net_cleanup(ipvs);
2440 control_fail:
2441         ip_vs_estimator_net_cleanup(ipvs);
2442 estimator_fail:
2443         net->ipvs = NULL;
2444         return -ENOMEM;
2445 }
2446
2447 static void __net_exit __ip_vs_cleanup_batch(struct list_head *net_list)
2448 {
2449         struct netns_ipvs *ipvs;
2450         struct net *net;
2451
2452         ip_vs_service_nets_cleanup(net_list);   /* ip_vs_flush() with locks */
2453         list_for_each_entry(net, net_list, exit_list) {
2454                 ipvs = net_ipvs(net);
2455                 ip_vs_conn_net_cleanup(ipvs);
2456                 ip_vs_app_net_cleanup(ipvs);
2457                 ip_vs_protocol_net_cleanup(ipvs);
2458                 ip_vs_control_net_cleanup(ipvs);
2459                 ip_vs_estimator_net_cleanup(ipvs);
2460                 IP_VS_DBG(2, "ipvs netns %d released\n", ipvs->gen);
2461                 net->ipvs = NULL;
2462         }
2463 }
2464
2465 static void __net_exit __ip_vs_dev_cleanup_batch(struct list_head *net_list)
2466 {
2467         struct netns_ipvs *ipvs;
2468         struct net *net;
2469
2470         EnterFunction(2);
2471         list_for_each_entry(net, net_list, exit_list) {
2472                 ipvs = net_ipvs(net);
2473                 ip_vs_unregister_hooks(ipvs, AF_INET);
2474                 ip_vs_unregister_hooks(ipvs, AF_INET6);
2475                 ipvs->enable = 0;       /* Disable packet reception */
2476                 smp_wmb();
2477                 ip_vs_sync_net_cleanup(ipvs);
2478         }
2479         LeaveFunction(2);
2480 }
2481
2482 static struct pernet_operations ipvs_core_ops = {
2483         .init = __ip_vs_init,
2484         .exit_batch = __ip_vs_cleanup_batch,
2485         .id   = &ip_vs_net_id,
2486         .size = sizeof(struct netns_ipvs),
2487 };
2488
2489 static struct pernet_operations ipvs_core_dev_ops = {
2490         .exit_batch = __ip_vs_dev_cleanup_batch,
2491 };
2492
2493 /*
2494  *      Initialize IP Virtual Server
2495  */
2496 static int __init ip_vs_init(void)
2497 {
2498         int ret;
2499
2500         ret = ip_vs_control_init();
2501         if (ret < 0) {
2502                 pr_err("can't setup control.\n");
2503                 goto exit;
2504         }
2505
2506         ip_vs_protocol_init();
2507
2508         ret = ip_vs_conn_init();
2509         if (ret < 0) {
2510                 pr_err("can't setup connection table.\n");
2511                 goto cleanup_protocol;
2512         }
2513
2514         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
2515         if (ret < 0)
2516                 goto cleanup_conn;
2517
2518         ret = register_pernet_device(&ipvs_core_dev_ops);
2519         if (ret < 0)
2520                 goto cleanup_sub;
2521
2522         ret = ip_vs_register_nl_ioctl();
2523         if (ret < 0) {
2524                 pr_err("can't register netlink/ioctl.\n");
2525                 goto cleanup_dev;
2526         }
2527
2528         pr_info("ipvs loaded.\n");
2529
2530         return ret;
2531
2532 cleanup_dev:
2533         unregister_pernet_device(&ipvs_core_dev_ops);
2534 cleanup_sub:
2535         unregister_pernet_subsys(&ipvs_core_ops);
2536 cleanup_conn:
2537         ip_vs_conn_cleanup();
2538 cleanup_protocol:
2539         ip_vs_protocol_cleanup();
2540         ip_vs_control_cleanup();
2541 exit:
2542         return ret;
2543 }
2544
2545 static void __exit ip_vs_cleanup(void)
2546 {
2547         ip_vs_unregister_nl_ioctl();
2548         unregister_pernet_device(&ipvs_core_dev_ops);
2549         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
2550         ip_vs_conn_cleanup();
2551         ip_vs_protocol_cleanup();
2552         ip_vs_control_cleanup();
2553         pr_info("ipvs unloaded.\n");
2554 }
2555
2556 module_init(ip_vs_init);
2557 module_exit(ip_vs_cleanup);
2558 MODULE_LICENSE("GPL");