net/netfilter/ipvs/ip_vs_core.c

   1 /*
   2  * IPVS         An implementation of the IP virtual server support for the
   3  *              LINUX operating system.  IPVS is now implemented as a module
   4  *              over the Netfilter framework. IPVS can be used to build a
   5  *              high-performance and highly available server based on a
   6  *              cluster of servers.
   7  *
   8  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
   9  *              Peter Kese <peter.kese@ijs.si>
  10  *              Julian Anastasov <ja@ssi.bg>
  11  *
  12  *              This program is free software; you can redistribute it and/or
  13  *              modify it under the terms of the GNU General Public License
  14  *              as published by the Free Software Foundation; either version
  15  *              2 of the License, or (at your option) any later version.
  16  *
  17  * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
  18  * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
  19  * and others.
  20  *
  21  * Changes:
  22  *      Paul `Rusty' Russell            properly handle non-linear skbs
  23  *      Harald Welte                    don't use nfcache
  24  *
  25  */
  26
  27 #define KMSG_COMPONENT "IPVS"
  28 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  29
  30 #include <linux/module.h>
  31 #include <linux/kernel.h>
  32 #include <linux/ip.h>
  33 #include <linux/tcp.h>
  34 #include <linux/sctp.h>
  35 #include <linux/icmp.h>
  36 #include <linux/slab.h>
  37
  38 #include <net/ip.h>
  39 #include <net/tcp.h>
  40 #include <net/udp.h>
  41 #include <net/icmp.h>                   /* for icmp_send */
  42 #include <net/route.h>
  43 #include <net/ip6_checksum.h>
  44 #include <net/netns/generic.h>          /* net_generic() */
  45
  46 #include <linux/netfilter.h>
  47 #include <linux/netfilter_ipv4.h>
  48
  49 #ifdef CONFIG_IP_VS_IPV6
  50 #include <net/ipv6.h>
  51 #include <linux/netfilter_ipv6.h>
  52 #include <net/ip6_route.h>
  53 #endif
  54
  55 #include <net/ip_vs.h>
  56
  57
  58 EXPORT_SYMBOL(register_ip_vs_scheduler);
  59 EXPORT_SYMBOL(unregister_ip_vs_scheduler);
  60 EXPORT_SYMBOL(ip_vs_proto_name);
  61 EXPORT_SYMBOL(ip_vs_conn_new);
  62 EXPORT_SYMBOL(ip_vs_conn_in_get);
  63 EXPORT_SYMBOL(ip_vs_conn_out_get);
  64 #ifdef CONFIG_IP_VS_PROTO_TCP
  65 EXPORT_SYMBOL(ip_vs_tcp_conn_listen);
  66 #endif
  67 EXPORT_SYMBOL(ip_vs_conn_put);
  68 #ifdef CONFIG_IP_VS_DEBUG
  69 EXPORT_SYMBOL(ip_vs_get_debug_level);
  70 #endif
  71
  72 static int ip_vs_net_id __read_mostly;
  73 /* netns cnt used for uniqueness */
  74 static atomic_t ipvs_netns_cnt = ATOMIC_INIT(0);
  75
  76 /* ID used in ICMP lookups */
  77 #define icmp_id(icmph)          (((icmph)->un).echo.id)
  78 #define icmpv6_id(icmph)        (icmph->icmp6_dataun.u_echo.identifier)
  79
  80 const char *ip_vs_proto_name(unsigned int proto)
  81 {
  82         static char buf[20];
  83
  84         switch (proto) {
  85         case IPPROTO_IP:
  86                 return "IP";
  87         case IPPROTO_UDP:
  88                 return "UDP";
  89         case IPPROTO_TCP:
  90                 return "TCP";
  91         case IPPROTO_SCTP:
  92                 return "SCTP";
  93         case IPPROTO_ICMP:
  94                 return "ICMP";
  95 #ifdef CONFIG_IP_VS_IPV6
  96         case IPPROTO_ICMPV6:
  97                 return "ICMPv6";
  98 #endif
  99         default:
 100                 sprintf(buf, "IP_%u", proto);
 101                 return buf;
 102         }
 103 }
 104
 105 void ip_vs_init_hash_table(struct list_head *table, int rows)
 106 {
 107         while (--rows >= 0)
 108                 INIT_LIST_HEAD(&table[rows]);
 109 }
 110
 111 static inline void
 112 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 113 {
 114         struct ip_vs_dest *dest = cp->dest;
 115         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 116
 117         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 118                 struct ip_vs_cpu_stats *s;
 119                 struct ip_vs_service *svc;
 120
 121                 s = this_cpu_ptr(dest->stats.cpustats);
 122                 s->ustats.inpkts++;
 123                 u64_stats_update_begin(&s->syncp);
 124                 s->ustats.inbytes += skb->len;
 125                 u64_stats_update_end(&s->syncp);
 126
 127                 rcu_read_lock();
 128                 svc = rcu_dereference(dest->svc);
 129                 s = this_cpu_ptr(svc->stats.cpustats);
 130                 s->ustats.inpkts++;
 131                 u64_stats_update_begin(&s->syncp);
 132                 s->ustats.inbytes += skb->len;
 133                 u64_stats_update_end(&s->syncp);
 134                 rcu_read_unlock();
 135
 136                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 137                 s->ustats.inpkts++;
 138                 u64_stats_update_begin(&s->syncp);
 139                 s->ustats.inbytes += skb->len;
 140                 u64_stats_update_end(&s->syncp);
 141         }
 142 }
 143
 144
 145 static inline void
 146 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 147 {
 148         struct ip_vs_dest *dest = cp->dest;
 149         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 150
 151         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 152                 struct ip_vs_cpu_stats *s;
 153                 struct ip_vs_service *svc;
 154
 155                 s = this_cpu_ptr(dest->stats.cpustats);
 156                 s->ustats.outpkts++;
 157                 u64_stats_update_begin(&s->syncp);
 158                 s->ustats.outbytes += skb->len;
 159                 u64_stats_update_end(&s->syncp);
 160
 161                 rcu_read_lock();
 162                 svc = rcu_dereference(dest->svc);
 163                 s = this_cpu_ptr(svc->stats.cpustats);
 164                 s->ustats.outpkts++;
 165                 u64_stats_update_begin(&s->syncp);
 166                 s->ustats.outbytes += skb->len;
 167                 u64_stats_update_end(&s->syncp);
 168                 rcu_read_unlock();
 169
 170                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 171                 s->ustats.outpkts++;
 172                 u64_stats_update_begin(&s->syncp);
 173                 s->ustats.outbytes += skb->len;
 174                 u64_stats_update_end(&s->syncp);
 175         }
 176 }
 177
 178
 179 static inline void
 180 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
 181 {
 182         struct netns_ipvs *ipvs = net_ipvs(svc->net);
 183         struct ip_vs_cpu_stats *s;
 184
 185         s = this_cpu_ptr(cp->dest->stats.cpustats);
 186         s->ustats.conns++;
 187
 188         s = this_cpu_ptr(svc->stats.cpustats);
 189         s->ustats.conns++;
 190
 191         s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 192         s->ustats.conns++;
 193 }
 194
 195
 196 static inline void
 197 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
 198                 const struct sk_buff *skb,
 199                 struct ip_vs_proto_data *pd)
 200 {
 201         if (likely(pd->pp->state_transition))
 202                 pd->pp->state_transition(cp, direction, skb, pd);
 203 }
 204
 205 static inline int
 206 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
 207                               struct sk_buff *skb, int protocol,
 208                               const union nf_inet_addr *caddr, __be16 cport,
 209                               const union nf_inet_addr *vaddr, __be16 vport,
 210                               struct ip_vs_conn_param *p)
 211 {
 212         ip_vs_conn_fill_param(svc->net, svc->af, protocol, caddr, cport, vaddr,
 213                               vport, p);
 214         p->pe = rcu_dereference(svc->pe);
 215         if (p->pe && p->pe->fill_param)
 216                 return p->pe->fill_param(p, skb);
 217
 218         return 0;
 219 }
 220
 221 /*
 222  *  IPVS persistent scheduling function
 223  *  It creates a connection entry according to its template if exists,
 224  *  or selects a server and creates a connection entry plus a template.
 225  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
 226  *  Protocols supported: TCP, UDP
 227  */
 228 static struct ip_vs_conn *
 229 ip_vs_sched_persist(struct ip_vs_service *svc,
 230                     struct sk_buff *skb, __be16 src_port, __be16 dst_port,
 231                     int *ignored, struct ip_vs_iphdr *iph)
 232 {
 233         struct ip_vs_conn *cp = NULL;
 234         struct ip_vs_dest *dest;
 235         struct ip_vs_conn *ct;
 236         __be16 dport = 0;               /* destination port to forward */
 237         unsigned int flags;
 238         struct ip_vs_conn_param param;
 239         const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
 240         union nf_inet_addr snet;        /* source network of the client,
 241                                            after masking */
 242
 243         /* Mask saddr with the netmask to adjust template granularity */
 244 #ifdef CONFIG_IP_VS_IPV6
 245         if (svc->af == AF_INET6)
 246                 ipv6_addr_prefix(&snet.in6, &iph->saddr.in6,
 247                                  (__force __u32) svc->netmask);
 248         else
 249 #endif
 250                 snet.ip = iph->saddr.ip & svc->netmask;
 251
 252         IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
 253                       "mnet %s\n",
 254                       IP_VS_DBG_ADDR(svc->af, &iph->saddr), ntohs(src_port),
 255                       IP_VS_DBG_ADDR(svc->af, &iph->daddr), ntohs(dst_port),
 256                       IP_VS_DBG_ADDR(svc->af, &snet));
 257
 258         /*
 259          * As far as we know, FTP is a very complicated network protocol, and
 260          * it uses control connection and data connections. For active FTP,
 261          * FTP server initialize data connection to the client, its source port
 262          * is often 20. For passive FTP, FTP server tells the clients the port
 263          * that it passively listens to,  and the client issues the data
 264          * connection. In the tunneling or direct routing mode, the load
 265          * balancer is on the client-to-server half of connection, the port
 266          * number is unknown to the load balancer. So, a conn template like
 267          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
 268          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
 269          * is created for other persistent services.
 270          */
 271         {
 272                 int protocol = iph->protocol;
 273                 const union nf_inet_addr *vaddr = &iph->daddr;
 274                 __be16 vport = 0;
 275
 276                 if (dst_port == svc->port) {
 277                         /* non-FTP template:
 278                          * <protocol, caddr, 0, vaddr, vport, daddr, dport>
 279                          * FTP template:
 280                          * <protocol, caddr, 0, vaddr, 0, daddr, 0>
 281                          */
 282                         if (svc->port != FTPPORT)
 283                                 vport = dst_port;
 284                 } else {
 285                         /* Note: persistent fwmark-based services and
 286                          * persistent port zero service are handled here.
 287                          * fwmark template:
 288                          * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
 289                          * port zero template:
 290                          * <protocol,caddr,0,vaddr,0,daddr,0>
 291                          */
 292                         if (svc->fwmark) {
 293                                 protocol = IPPROTO_IP;
 294                                 vaddr = &fwmark;
 295                         }
 296                 }
 297                 /* return *ignored = -1 so NF_DROP can be used */
 298                 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
 299                                                   vaddr, vport, &param) < 0) {
 300                         *ignored = -1;
 301                         return NULL;
 302                 }
 303         }
 304
 305         /* Check if a template already exists */
 306         ct = ip_vs_ct_in_get(&param);
 307         if (!ct || !ip_vs_check_template(ct)) {
 308                 struct ip_vs_scheduler *sched;
 309
 310                 /*
 311                  * No template found or the dest of the connection
 312                  * template is not available.
 313                  * return *ignored=0 i.e. ICMP and NF_DROP
 314                  */
 315                 sched = rcu_dereference(svc->scheduler);
 316                 dest = sched->schedule(svc, skb, iph);
 317                 if (!dest) {
 318                         IP_VS_DBG(1, "p-schedule: no dest found.\n");
 319                         kfree(param.pe_data);
 320                         *ignored = 0;
 321                         return NULL;
 322                 }
 323
 324                 if (dst_port == svc->port && svc->port != FTPPORT)
 325                         dport = dest->port;
 326
 327                 /* Create a template
 328                  * This adds param.pe_data to the template,
 329                  * and thus param.pe_data will be destroyed
 330                  * when the template expires */
 331                 ct = ip_vs_conn_new(&param, dest->af, &dest->addr, dport,
 332                                     IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
 333                 if (ct == NULL) {
 334                         kfree(param.pe_data);
 335                         *ignored = -1;
 336                         return NULL;
 337                 }
 338
 339                 ct->timeout = svc->timeout;
 340         } else {
 341                 /* set destination with the found template */
 342                 dest = ct->dest;
 343                 kfree(param.pe_data);
 344         }
 345
 346         dport = dst_port;
 347         if (dport == svc->port && dest->port)
 348                 dport = dest->port;
 349
 350         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 351                  && iph->protocol == IPPROTO_UDP) ?
 352                 IP_VS_CONN_F_ONE_PACKET : 0;
 353
 354         /*
 355          *    Create a new connection according to the template
 356          */
 357         ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol, &iph->saddr,
 358                               src_port, &iph->daddr, dst_port, &param);
 359
 360         cp = ip_vs_conn_new(&param, dest->af, &dest->addr, dport, flags, dest,
 361                             skb->mark);
 362         if (cp == NULL) {
 363                 ip_vs_conn_put(ct);
 364                 *ignored = -1;
 365                 return NULL;
 366         }
 367
 368         /*
 369          *    Add its control
 370          */
 371         ip_vs_control_add(cp, ct);
 372         ip_vs_conn_put(ct);
 373
 374         ip_vs_conn_stats(cp, svc);
 375         return cp;
 376 }
 377
 378
 379 /*
 380  *  IPVS main scheduling function
 381  *  It selects a server according to the virtual service, and
 382  *  creates a connection entry.
 383  *  Protocols supported: TCP, UDP
 384  *
 385  *  Usage of *ignored
 386  *
 387  * 1 :   protocol tried to schedule (eg. on SYN), found svc but the
 388  *       svc/scheduler decides that this packet should be accepted with
 389  *       NF_ACCEPT because it must not be scheduled.
 390  *
 391  * 0 :   scheduler can not find destination, so try bypass or
 392  *       return ICMP and then NF_DROP (ip_vs_leave).
 393  *
 394  * -1 :  scheduler tried to schedule but fatal error occurred, eg.
 395  *       ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
 396  *       failure such as missing Call-ID, ENOMEM on skb_linearize
 397  *       or pe_data. In this case we should return NF_DROP without
 398  *       any attempts to send ICMP with ip_vs_leave.
 399  */
 400 struct ip_vs_conn *
 401 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
 402                struct ip_vs_proto_data *pd, int *ignored,
 403                struct ip_vs_iphdr *iph)
 404 {
 405         struct ip_vs_protocol *pp = pd->pp;
 406         struct ip_vs_conn *cp = NULL;
 407         struct ip_vs_scheduler *sched;
 408         struct ip_vs_dest *dest;
 409         __be16 _ports[2], *pptr;
 410         unsigned int flags;
 411
 412         *ignored = 1;
 413         /*
 414          * IPv6 frags, only the first hit here.
 415          */
 416         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 417         if (pptr == NULL)
 418                 return NULL;
 419
 420         /*
 421          * FTPDATA needs this check when using local real server.
 422          * Never schedule Active FTPDATA connections from real server.
 423          * For LVS-NAT they must be already created. For other methods
 424          * with persistence the connection is created on SYN+ACK.
 425          */
 426         if (pptr[0] == FTPDATA) {
 427                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 428                               "Not scheduling FTPDATA");
 429                 return NULL;
 430         }
 431
 432         /*
 433          *    Do not schedule replies from local real server.
 434          */
 435         if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
 436             (cp = pp->conn_in_get(svc->af, skb, iph, 1))) {
 437                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 438                               "Not scheduling reply for existing connection");
 439                 __ip_vs_conn_put(cp);
 440                 return NULL;
 441         }
 442
 443         /*
 444          *    Persistent service
 445          */
 446         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 447                 return ip_vs_sched_persist(svc, skb, pptr[0], pptr[1], ignored,
 448                                            iph);
 449
 450         *ignored = 0;
 451
 452         /*
 453          *    Non-persistent service
 454          */
 455         if (!svc->fwmark && pptr[1] != svc->port) {
 456                 if (!svc->port)
 457                         pr_err("Schedule: port zero only supported "
 458                                "in persistent services, "
 459                                "check your ipvs configuration\n");
 460                 return NULL;
 461         }
 462
 463         sched = rcu_dereference(svc->scheduler);
 464         dest = sched->schedule(svc, skb, iph);
 465         if (dest == NULL) {
 466                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 467                 return NULL;
 468         }
 469
 470         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 471                  && iph->protocol == IPPROTO_UDP) ?
 472                 IP_VS_CONN_F_ONE_PACKET : 0;
 473
 474         /*
 475          *    Create a connection entry.
 476          */
 477         {
 478                 struct ip_vs_conn_param p;
 479
 480                 ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol,
 481                                       &iph->saddr, pptr[0], &iph->daddr,
 482                                       pptr[1], &p);
 483                 cp = ip_vs_conn_new(&p, dest->af, &dest->addr,
 484                                     dest->port ? dest->port : pptr[1],
 485                                     flags, dest, skb->mark);
 486                 if (!cp) {
 487                         *ignored = -1;
 488                         return NULL;
 489                 }
 490         }
 491
 492         IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
 493                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
 494                       ip_vs_fwd_tag(cp),
 495                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 496                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 497                       IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
 498                       cp->flags, atomic_read(&cp->refcnt));
 499
 500         ip_vs_conn_stats(cp, svc);
 501         return cp;
 502 }
 503
 504
 505 /*
 506  *  Pass or drop the packet.
 507  *  Called by ip_vs_in, when the virtual service is available but
 508  *  no destination is available for a new connection.
 509  */
 510 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 511                 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
 512 {
 513         __be16 _ports[2], *pptr;
 514 #ifdef CONFIG_SYSCTL
 515         struct net *net;
 516         struct netns_ipvs *ipvs;
 517         int unicast;
 518 #endif
 519
 520         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 521         if (pptr == NULL) {
 522                 return NF_DROP;
 523         }
 524
 525 #ifdef CONFIG_SYSCTL
 526         net = skb_net(skb);
 527
 528 #ifdef CONFIG_IP_VS_IPV6
 529         if (svc->af == AF_INET6)
 530                 unicast = ipv6_addr_type(&iph->daddr.in6) & IPV6_ADDR_UNICAST;
 531         else
 532 #endif
 533                 unicast = (inet_addr_type(net, iph->daddr.ip) == RTN_UNICAST);
 534
 535         /* if it is fwmark-based service, the cache_bypass sysctl is up
 536            and the destination is a non-local unicast, then create
 537            a cache_bypass connection entry */
 538         ipvs = net_ipvs(net);
 539         if (ipvs->sysctl_cache_bypass && svc->fwmark && unicast) {
 540                 int ret;
 541                 struct ip_vs_conn *cp;
 542                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
 543                                       iph->protocol == IPPROTO_UDP) ?
 544                                       IP_VS_CONN_F_ONE_PACKET : 0;
 545                 union nf_inet_addr daddr =  { .all = { 0, 0, 0, 0 } };
 546
 547                 /* create a new connection entry */
 548                 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
 549                 {
 550                         struct ip_vs_conn_param p;
 551                         ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol,
 552                                               &iph->saddr, pptr[0],
 553                                               &iph->daddr, pptr[1], &p);
 554                         cp = ip_vs_conn_new(&p, svc->af, &daddr, 0,
 555                                             IP_VS_CONN_F_BYPASS | flags,
 556                                             NULL, skb->mark);
 557                         if (!cp)
 558                                 return NF_DROP;
 559                 }
 560
 561                 /* statistics */
 562                 ip_vs_in_stats(cp, skb);
 563
 564                 /* set state */
 565                 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
 566
 567                 /* transmit the first SYN packet */
 568                 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
 569                 /* do not touch skb anymore */
 570
 571                 atomic_inc(&cp->in_pkts);
 572                 ip_vs_conn_put(cp);
 573                 return ret;
 574         }
 575 #endif
 576
 577         /*
 578          * When the virtual ftp service is presented, packets destined
 579          * for other services on the VIP may get here (except services
 580          * listed in the ipvs table), pass the packets, because it is
 581          * not ipvs job to decide to drop the packets.
 582          */
 583         if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT))
 584                 return NF_ACCEPT;
 585
 586         /*
 587          * Notify the client that the destination is unreachable, and
 588          * release the socket buffer.
 589          * Since it is in IP layer, the TCP socket is not actually
 590          * created, the TCP RST packet cannot be sent, instead that
 591          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 592          */
 593 #ifdef CONFIG_IP_VS_IPV6
 594         if (svc->af == AF_INET6) {
 595                 if (!skb->dev) {
 596                         struct net *net_ = dev_net(skb_dst(skb)->dev);
 597
 598                         skb->dev = net_->loopback_dev;
 599                 }
 600                 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 601         } else
 602 #endif
 603                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 604
 605         return NF_DROP;
 606 }
 607
 608 #ifdef CONFIG_SYSCTL
 609
 610 static int sysctl_snat_reroute(struct sk_buff *skb)
 611 {
 612         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 613         return ipvs->sysctl_snat_reroute;
 614 }
 615
 616 static int sysctl_nat_icmp_send(struct net *net)
 617 {
 618         struct netns_ipvs *ipvs = net_ipvs(net);
 619         return ipvs->sysctl_nat_icmp_send;
 620 }
 621
 622 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
 623 {
 624         return ipvs->sysctl_expire_nodest_conn;
 625 }
 626
 627 #else
 628
 629 static int sysctl_snat_reroute(struct sk_buff *skb) { return 0; }
 630 static int sysctl_nat_icmp_send(struct net *net) { return 0; }
 631 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
 632
 633 #endif
 634
 635 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 636 {
 637         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 638 }
 639
 640 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
 641 {
 642         if (NF_INET_LOCAL_IN == hooknum)
 643                 return IP_DEFRAG_VS_IN;
 644         if (NF_INET_FORWARD == hooknum)
 645                 return IP_DEFRAG_VS_FWD;
 646         return IP_DEFRAG_VS_OUT;
 647 }
 648
 649 static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user)
 650 {
 651         int err;
 652
 653         local_bh_disable();
 654         err = ip_defrag(skb, user);
 655         local_bh_enable();
 656         if (!err)
 657                 ip_send_check(ip_hdr(skb));
 658
 659         return err;
 660 }
 661
 662 static int ip_vs_route_me_harder(int af, struct sk_buff *skb,
 663                                  unsigned int hooknum)
 664 {
 665         if (!sysctl_snat_reroute(skb))
 666                 return 0;
 667         /* Reroute replies only to remote clients (FORWARD and LOCAL_OUT) */
 668         if (NF_INET_LOCAL_IN == hooknum)
 669                 return 0;
 670 #ifdef CONFIG_IP_VS_IPV6
 671         if (af == AF_INET6) {
 672                 struct dst_entry *dst = skb_dst(skb);
 673
 674                 if (dst->dev && !(dst->dev->flags & IFF_LOOPBACK) &&
 675                     ip6_route_me_harder(skb) != 0)
 676                         return 1;
 677         } else
 678 #endif
 679                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
 680                     ip_route_me_harder(skb, RTN_LOCAL) != 0)
 681                         return 1;
 682
 683         return 0;
 684 }
 685
 686 /*
 687  * Packet has been made sufficiently writable in caller
 688  * - inout: 1=in->out, 0=out->in
 689  */
 690 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 691                     struct ip_vs_conn *cp, int inout)
 692 {
 693         struct iphdr *iph        = ip_hdr(skb);
 694         unsigned int icmp_offset = iph->ihl*4;
 695         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
 696                                                       icmp_offset);
 697         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 698
 699         if (inout) {
 700                 iph->saddr = cp->vaddr.ip;
 701                 ip_send_check(iph);
 702                 ciph->daddr = cp->vaddr.ip;
 703                 ip_send_check(ciph);
 704         } else {
 705                 iph->daddr = cp->daddr.ip;
 706                 ip_send_check(iph);
 707                 ciph->saddr = cp->daddr.ip;
 708                 ip_send_check(ciph);
 709         }
 710
 711         /* the TCP/UDP/SCTP port */
 712         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
 713             IPPROTO_SCTP == ciph->protocol) {
 714                 __be16 *ports = (void *)ciph + ciph->ihl*4;
 715
 716                 if (inout)
 717                         ports[1] = cp->vport;
 718                 else
 719                         ports[0] = cp->dport;
 720         }
 721
 722         /* And finally the ICMP checksum */
 723         icmph->checksum = 0;
 724         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 725         skb->ip_summed = CHECKSUM_UNNECESSARY;
 726
 727         if (inout)
 728                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 729                         "Forwarding altered outgoing ICMP");
 730         else
 731                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 732                         "Forwarding altered incoming ICMP");
 733 }
 734
 735 #ifdef CONFIG_IP_VS_IPV6
 736 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
 737                     struct ip_vs_conn *cp, int inout)
 738 {
 739         struct ipv6hdr *iph      = ipv6_hdr(skb);
 740         unsigned int icmp_offset = 0;
 741         unsigned int offs        = 0; /* header offset*/
 742         int protocol;
 743         struct icmp6hdr *icmph;
 744         struct ipv6hdr *ciph;
 745         unsigned short fragoffs;
 746
 747         ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
 748         icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
 749         offs = icmp_offset + sizeof(struct icmp6hdr);
 750         ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
 751
 752         protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
 753
 754         if (inout) {
 755                 iph->saddr = cp->vaddr.in6;
 756                 ciph->daddr = cp->vaddr.in6;
 757         } else {
 758                 iph->daddr = cp->daddr.in6;
 759                 ciph->saddr = cp->daddr.in6;
 760         }
 761
 762         /* the TCP/UDP/SCTP port */
 763         if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 764                           IPPROTO_SCTP == protocol)) {
 765                 __be16 *ports = (void *)(skb_network_header(skb) + offs);
 766
 767                 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
 768                               ntohs(inout ? ports[1] : ports[0]),
 769                               ntohs(inout ? cp->vport : cp->dport));
 770                 if (inout)
 771                         ports[1] = cp->vport;
 772                 else
 773                         ports[0] = cp->dport;
 774         }
 775
 776         /* And finally the ICMP checksum */
 777         icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
 778                                               skb->len - icmp_offset,
 779                                               IPPROTO_ICMPV6, 0);
 780         skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
 781         skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
 782         skb->ip_summed = CHECKSUM_PARTIAL;
 783
 784         if (inout)
 785                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 786                               (void *)ciph - (void *)iph,
 787                               "Forwarding altered outgoing ICMPv6");
 788         else
 789                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 790                               (void *)ciph - (void *)iph,
 791                               "Forwarding altered incoming ICMPv6");
 792 }
 793 #endif
 794
 795 /* Handle relevant response ICMP messages - forward to the right
 796  * destination host.
 797  */
 798 static int handle_response_icmp(int af, struct sk_buff *skb,
 799                                 union nf_inet_addr *snet,
 800                                 __u8 protocol, struct ip_vs_conn *cp,
 801                                 struct ip_vs_protocol *pp,
 802                                 unsigned int offset, unsigned int ihl,
 803                                 unsigned int hooknum)
 804 {
 805         unsigned int verdict = NF_DROP;
 806
 807         if (IP_VS_FWD_METHOD(cp) != 0) {
 808                 pr_err("shouldn't reach here, because the box is on the "
 809                        "half connection in the tun/dr module.\n");
 810         }
 811
 812         /* Ensure the checksum is correct */
 813         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 814                 /* Failed checksum! */
 815                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
 816                               IP_VS_DBG_ADDR(af, snet));
 817                 goto out;
 818         }
 819
 820         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 821             IPPROTO_SCTP == protocol)
 822                 offset += 2 * sizeof(__u16);
 823         if (!skb_make_writable(skb, offset))
 824                 goto out;
 825
 826 #ifdef CONFIG_IP_VS_IPV6
 827         if (af == AF_INET6)
 828                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
 829         else
 830 #endif
 831                 ip_vs_nat_icmp(skb, pp, cp, 1);
 832
 833         if (ip_vs_route_me_harder(af, skb, hooknum))
 834                 goto out;
 835
 836         /* do the statistics and put it back */
 837         ip_vs_out_stats(cp, skb);
 838
 839         skb->ipvs_property = 1;
 840         if (!(cp->flags & IP_VS_CONN_F_NFCT))
 841                 ip_vs_notrack(skb);
 842         else
 843                 ip_vs_update_conntrack(skb, cp, 0);
 844         verdict = NF_ACCEPT;
 845
 846 out:
 847         __ip_vs_conn_put(cp);
 848
 849         return verdict;
 850 }
 851
 852 /*
 853  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 854  *      Find any that might be relevant, check against existing connections.
 855  *      Currently handles error types - unreachable, quench, ttl exceeded.
 856  */
 857 static int ip_vs_out_icmp(struct sk_buff *skb, int *related,
 858                           unsigned int hooknum)
 859 {
 860         struct iphdr *iph;
 861         struct icmphdr  _icmph, *ic;
 862         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 863         struct ip_vs_iphdr ciph;
 864         struct ip_vs_conn *cp;
 865         struct ip_vs_protocol *pp;
 866         unsigned int offset, ihl;
 867         union nf_inet_addr snet;
 868
 869         *related = 1;
 870
 871         /* reassemble IP fragments */
 872         if (ip_is_fragment(ip_hdr(skb))) {
 873                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
 874                         return NF_STOLEN;
 875         }
 876
 877         iph = ip_hdr(skb);
 878         offset = ihl = iph->ihl * 4;
 879         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 880         if (ic == NULL)
 881                 return NF_DROP;
 882
 883         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
 884                   ic->type, ntohs(icmp_id(ic)),
 885                   &iph->saddr, &iph->daddr);
 886
 887         /*
 888          * Work through seeing if this is for us.
 889          * These checks are supposed to be in an order that means easy
 890          * things are checked first to speed up processing.... however
 891          * this means that some packets will manage to get a long way
 892          * down this stack and then be rejected, but that's life.
 893          */
 894         if ((ic->type != ICMP_DEST_UNREACH) &&
 895             (ic->type != ICMP_SOURCE_QUENCH) &&
 896             (ic->type != ICMP_TIME_EXCEEDED)) {
 897                 *related = 0;
 898                 return NF_ACCEPT;
 899         }
 900
 901         /* Now find the contained IP header */
 902         offset += sizeof(_icmph);
 903         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 904         if (cih == NULL)
 905                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 906
 907         pp = ip_vs_proto_get(cih->protocol);
 908         if (!pp)
 909                 return NF_ACCEPT;
 910
 911         /* Is the embedded protocol header present? */
 912         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 913                      pp->dont_defrag))
 914                 return NF_ACCEPT;
 915
 916         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
 917                       "Checking outgoing ICMP for");
 918
 919         ip_vs_fill_ip4hdr(cih, &ciph);
 920         ciph.len += offset;
 921         /* The embedded headers contain source and dest in reverse order */
 922         cp = pp->conn_out_get(AF_INET, skb, &ciph, 1);
 923         if (!cp)
 924                 return NF_ACCEPT;
 925
 926         snet.ip = iph->saddr;
 927         return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
 928                                     pp, ciph.len, ihl, hooknum);
 929 }
 930
 931 #ifdef CONFIG_IP_VS_IPV6
 932 static int ip_vs_out_icmp_v6(struct sk_buff *skb, int *related,
 933                              unsigned int hooknum, struct ip_vs_iphdr *ipvsh)
 934 {
 935         struct icmp6hdr _icmph, *ic;
 936         struct ipv6hdr _ip6h, *ip6h; /* The ip header contained within ICMP */
 937         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
 938         struct ip_vs_conn *cp;
 939         struct ip_vs_protocol *pp;
 940         union nf_inet_addr snet;
 941         unsigned int writable;
 942
 943         *related = 1;
 944         ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph, ipvsh);
 945         if (ic == NULL)
 946                 return NF_DROP;
 947
 948         /*
 949          * Work through seeing if this is for us.
 950          * These checks are supposed to be in an order that means easy
 951          * things are checked first to speed up processing.... however
 952          * this means that some packets will manage to get a long way
 953          * down this stack and then be rejected, but that's life.
 954          */
 955         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
 956                 *related = 0;
 957                 return NF_ACCEPT;
 958         }
 959         /* Fragment header that is before ICMP header tells us that:
 960          * it's not an error message since they can't be fragmented.
 961          */
 962         if (ipvsh->flags & IP6_FH_F_FRAG)
 963                 return NF_DROP;
 964
 965         IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
 966                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
 967                   &ipvsh->saddr, &ipvsh->daddr);
 968
 969         /* Now find the contained IP header */
 970         ciph.len = ipvsh->len + sizeof(_icmph);
 971         ip6h = skb_header_pointer(skb, ciph.len, sizeof(_ip6h), &_ip6h);
 972         if (ip6h == NULL)
 973                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 974         ciph.saddr.in6 = ip6h->saddr; /* conn_out_get() handles reverse order */
 975         ciph.daddr.in6 = ip6h->daddr;
 976         /* skip possible IPv6 exthdrs of contained IPv6 packet */
 977         ciph.protocol = ipv6_find_hdr(skb, &ciph.len, -1, &ciph.fragoffs, NULL);
 978         if (ciph.protocol < 0)
 979                 return NF_ACCEPT; /* Contained IPv6 hdr looks wrong, ignore */
 980
 981         pp = ip_vs_proto_get(ciph.protocol);
 982         if (!pp)
 983                 return NF_ACCEPT;
 984
 985         /* The embedded headers contain source and dest in reverse order */
 986         cp = pp->conn_out_get(AF_INET6, skb, &ciph, 1);
 987         if (!cp)
 988                 return NF_ACCEPT;
 989
 990         snet.in6 = ciph.saddr.in6;
 991         writable = ciph.len;
 992         return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
 993                                     pp, writable, sizeof(struct ipv6hdr),
 994                                     hooknum);
 995 }
 996 #endif
 997
 998 /*
 999  * Check if sctp chunc is ABORT chunk
1000  */
1001 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
1002 {
1003         sctp_chunkhdr_t *sch, schunk;
1004         sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
1005                         sizeof(schunk), &schunk);
1006         if (sch == NULL)
1007                 return 0;
1008         if (sch->type == SCTP_CID_ABORT)
1009                 return 1;
1010         return 0;
1011 }
1012
1013 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
1014 {
1015         struct tcphdr _tcph, *th;
1016
1017         th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
1018         if (th == NULL)
1019                 return 0;
1020         return th->rst;
1021 }
1022
1023 static inline bool is_new_conn(const struct sk_buff *skb,
1024                                struct ip_vs_iphdr *iph)
1025 {
1026         switch (iph->protocol) {
1027         case IPPROTO_TCP: {
1028                 struct tcphdr _tcph, *th;
1029
1030                 th = skb_header_pointer(skb, iph->len, sizeof(_tcph), &_tcph);
1031                 if (th == NULL)
1032                         return false;
1033                 return th->syn;
1034         }
1035         case IPPROTO_SCTP: {
1036                 sctp_chunkhdr_t *sch, schunk;
1037
1038                 sch = skb_header_pointer(skb, iph->len + sizeof(sctp_sctphdr_t),
1039                                          sizeof(schunk), &schunk);
1040                 if (sch == NULL)
1041                         return false;
1042                 return sch->type == SCTP_CID_INIT;
1043         }
1044         default:
1045                 return false;
1046         }
1047 }
1048
1049 /* Handle response packets: rewrite addresses and send away...
1050  */
1051 static unsigned int
1052 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1053                 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph,
1054                 unsigned int hooknum)
1055 {
1056         struct ip_vs_protocol *pp = pd->pp;
1057
1058         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Outgoing packet");
1059
1060         if (!skb_make_writable(skb, iph->len))
1061                 goto drop;
1062
1063         /* mangle the packet */
1064         if (pp->snat_handler && !pp->snat_handler(skb, pp, cp, iph))
1065                 goto drop;
1066
1067 #ifdef CONFIG_IP_VS_IPV6
1068         if (af == AF_INET6)
1069                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1070         else
1071 #endif
1072         {
1073                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1074                 ip_send_check(ip_hdr(skb));
1075         }
1076
1077         /*
1078          * nf_iterate does not expect change in the skb->dst->dev.
1079          * It looks like it is not fatal to enable this code for hooks
1080          * where our handlers are at the end of the chain list and
1081          * when all next handlers use skb->dst->dev and not outdev.
1082          * It will definitely route properly the inout NAT traffic
1083          * when multiple paths are used.
1084          */
1085
1086         /* For policy routing, packets originating from this
1087          * machine itself may be routed differently to packets
1088          * passing through.  We want this packet to be routed as
1089          * if it came from this machine itself.  So re-compute
1090          * the routing information.
1091          */
1092         if (ip_vs_route_me_harder(af, skb, hooknum))
1093                 goto drop;
1094
1095         IP_VS_DBG_PKT(10, af, pp, skb, 0, "After SNAT");
1096
1097         ip_vs_out_stats(cp, skb);
1098         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1099         skb->ipvs_property = 1;
1100         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1101                 ip_vs_notrack(skb);
1102         else
1103                 ip_vs_update_conntrack(skb, cp, 0);
1104         ip_vs_conn_put(cp);
1105
1106         LeaveFunction(11);
1107         return NF_ACCEPT;
1108
1109 drop:
1110         ip_vs_conn_put(cp);
1111         kfree_skb(skb);
1112         LeaveFunction(11);
1113         return NF_STOLEN;
1114 }
1115
1116 /*
1117  *      Check if outgoing packet belongs to the established ip_vs_conn.
1118  */
1119 static unsigned int
1120 ip_vs_out(unsigned int hooknum, struct sk_buff *skb, int af)
1121 {
1122         struct net *net = NULL;
1123         struct ip_vs_iphdr iph;
1124         struct ip_vs_protocol *pp;
1125         struct ip_vs_proto_data *pd;
1126         struct ip_vs_conn *cp;
1127
1128         EnterFunction(11);
1129
1130         /* Already marked as IPVS request or reply? */
1131         if (skb->ipvs_property)
1132                 return NF_ACCEPT;
1133
1134         /* Bad... Do not break raw sockets */
1135         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1136                      af == AF_INET)) {
1137                 struct sock *sk = skb->sk;
1138                 struct inet_sock *inet = inet_sk(skb->sk);
1139
1140                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1141                         return NF_ACCEPT;
1142         }
1143
1144         if (unlikely(!skb_dst(skb)))
1145                 return NF_ACCEPT;
1146
1147         net = skb_net(skb);
1148         if (!net_ipvs(net)->enable)
1149                 return NF_ACCEPT;
1150
1151         ip_vs_fill_iph_skb(af, skb, &iph);
1152 #ifdef CONFIG_IP_VS_IPV6
1153         if (af == AF_INET6) {
1154                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1155                         int related;
1156                         int verdict = ip_vs_out_icmp_v6(skb, &related,
1157                                                         hooknum, &iph);
1158
1159                         if (related)
1160                                 return verdict;
1161                 }
1162         } else
1163 #endif
1164                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1165                         int related;
1166                         int verdict = ip_vs_out_icmp(skb, &related, hooknum);
1167
1168                         if (related)
1169                                 return verdict;
1170                 }
1171
1172         pd = ip_vs_proto_data_get(net, iph.protocol);
1173         if (unlikely(!pd))
1174                 return NF_ACCEPT;
1175         pp = pd->pp;
1176
1177         /* reassemble IP fragments */
1178 #ifdef CONFIG_IP_VS_IPV6
1179         if (af == AF_INET)
1180 #endif
1181                 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1182                         if (ip_vs_gather_frags(skb,
1183                                                ip_vs_defrag_user(hooknum)))
1184                                 return NF_STOLEN;
1185
1186                         ip_vs_fill_ip4hdr(skb_network_header(skb), &iph);
1187                 }
1188
1189         /*
1190          * Check if the packet belongs to an existing entry
1191          */
1192         cp = pp->conn_out_get(af, skb, &iph, 0);
1193
1194         if (likely(cp))
1195                 return handle_response(af, skb, pd, cp, &iph, hooknum);
1196         if (sysctl_nat_icmp_send(net) &&
1197             (pp->protocol == IPPROTO_TCP ||
1198              pp->protocol == IPPROTO_UDP ||
1199              pp->protocol == IPPROTO_SCTP)) {
1200                 __be16 _ports[2], *pptr;
1201
1202                 pptr = frag_safe_skb_hp(skb, iph.len,
1203                                          sizeof(_ports), _ports, &iph);
1204                 if (pptr == NULL)
1205                         return NF_ACCEPT;       /* Not for me */
1206                 if (ip_vs_has_real_service(net, af, iph.protocol, &iph.saddr,
1207                                            pptr[0])) {
1208                         /*
1209                          * Notify the real server: there is no
1210                          * existing entry if it is not RST
1211                          * packet or not TCP packet.
1212                          */
1213                         if ((iph.protocol != IPPROTO_TCP &&
1214                              iph.protocol != IPPROTO_SCTP)
1215                              || ((iph.protocol == IPPROTO_TCP
1216                                   && !is_tcp_reset(skb, iph.len))
1217                                  || (iph.protocol == IPPROTO_SCTP
1218                                         && !is_sctp_abort(skb,
1219                                                 iph.len)))) {
1220 #ifdef CONFIG_IP_VS_IPV6
1221                                 if (af == AF_INET6) {
1222                                         if (!skb->dev)
1223                                                 skb->dev = net->loopback_dev;
1224                                         icmpv6_send(skb,
1225                                                     ICMPV6_DEST_UNREACH,
1226                                                     ICMPV6_PORT_UNREACH,
1227                                                     0);
1228                                 } else
1229 #endif
1230                                         icmp_send(skb,
1231                                                   ICMP_DEST_UNREACH,
1232                                                   ICMP_PORT_UNREACH, 0);
1233                                 return NF_DROP;
1234                         }
1235                 }
1236         }
1237         IP_VS_DBG_PKT(12, af, pp, skb, 0,
1238                       "ip_vs_out: packet continues traversal as normal");
1239         return NF_ACCEPT;
1240 }
1241
1242 /*
1243  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1244  *      used only for VS/NAT.
1245  *      Check if packet is reply for established ip_vs_conn.
1246  */
1247 static unsigned int
1248 ip_vs_reply4(const struct nf_hook_ops *ops, struct sk_buff *skb,
1249              const struct net_device *in, const struct net_device *out,
1250              int (*okfn)(struct sk_buff *))
1251 {
1252         return ip_vs_out(ops->hooknum, skb, AF_INET);
1253 }
1254
1255 /*
1256  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1257  *      Check if packet is reply for established ip_vs_conn.
1258  */
1259 static unsigned int
1260 ip_vs_local_reply4(const struct nf_hook_ops *ops, struct sk_buff *skb,
1261                    const struct net_device *in, const struct net_device *out,
1262                    int (*okfn)(struct sk_buff *))
1263 {
1264         return ip_vs_out(ops->hooknum, skb, AF_INET);
1265 }
1266
1267 #ifdef CONFIG_IP_VS_IPV6
1268
1269 /*
1270  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1271  *      used only for VS/NAT.
1272  *      Check if packet is reply for established ip_vs_conn.
1273  */
1274 static unsigned int
1275 ip_vs_reply6(const struct nf_hook_ops *ops, struct sk_buff *skb,
1276              const struct net_device *in, const struct net_device *out,
1277              int (*okfn)(struct sk_buff *))
1278 {
1279         return ip_vs_out(ops->hooknum, skb, AF_INET6);
1280 }
1281
1282 /*
1283  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1284  *      Check if packet is reply for established ip_vs_conn.
1285  */
1286 static unsigned int
1287 ip_vs_local_reply6(const struct nf_hook_ops *ops, struct sk_buff *skb,
1288                    const struct net_device *in, const struct net_device *out,
1289                    int (*okfn)(struct sk_buff *))
1290 {
1291         return ip_vs_out(ops->hooknum, skb, AF_INET6);
1292 }
1293
1294 #endif
1295
1296 /*
1297  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1298  *      Find any that might be relevant, check against existing connections,
1299  *      forward to the right destination host if relevant.
1300  *      Currently handles error types - unreachable, quench, ttl exceeded.
1301  */
1302 static int
1303 ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum)
1304 {
1305         struct net *net = NULL;
1306         struct iphdr *iph;
1307         struct icmphdr  _icmph, *ic;
1308         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1309         struct ip_vs_iphdr ciph;
1310         struct ip_vs_conn *cp;
1311         struct ip_vs_protocol *pp;
1312         struct ip_vs_proto_data *pd;
1313         unsigned int offset, offset2, ihl, verdict;
1314         bool ipip;
1315
1316         *related = 1;
1317
1318         /* reassemble IP fragments */
1319         if (ip_is_fragment(ip_hdr(skb))) {
1320                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
1321                         return NF_STOLEN;
1322         }
1323
1324         iph = ip_hdr(skb);
1325         offset = ihl = iph->ihl * 4;
1326         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1327         if (ic == NULL)
1328                 return NF_DROP;
1329
1330         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1331                   ic->type, ntohs(icmp_id(ic)),
1332                   &iph->saddr, &iph->daddr);
1333
1334         /*
1335          * Work through seeing if this is for us.
1336          * These checks are supposed to be in an order that means easy
1337          * things are checked first to speed up processing.... however
1338          * this means that some packets will manage to get a long way
1339          * down this stack and then be rejected, but that's life.
1340          */
1341         if ((ic->type != ICMP_DEST_UNREACH) &&
1342             (ic->type != ICMP_SOURCE_QUENCH) &&
1343             (ic->type != ICMP_TIME_EXCEEDED)) {
1344                 *related = 0;
1345                 return NF_ACCEPT;
1346         }
1347
1348         /* Now find the contained IP header */
1349         offset += sizeof(_icmph);
1350         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1351         if (cih == NULL)
1352                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1353
1354         net = skb_net(skb);
1355
1356         /* Special case for errors for IPIP packets */
1357         ipip = false;
1358         if (cih->protocol == IPPROTO_IPIP) {
1359                 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1360                         return NF_ACCEPT;
1361                 /* Error for our IPIP must arrive at LOCAL_IN */
1362                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1363                         return NF_ACCEPT;
1364                 offset += cih->ihl * 4;
1365                 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1366                 if (cih == NULL)
1367                         return NF_ACCEPT; /* The packet looks wrong, ignore */
1368                 ipip = true;
1369         }
1370
1371         pd = ip_vs_proto_data_get(net, cih->protocol);
1372         if (!pd)
1373                 return NF_ACCEPT;
1374         pp = pd->pp;
1375
1376         /* Is the embedded protocol header present? */
1377         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1378                      pp->dont_defrag))
1379                 return NF_ACCEPT;
1380
1381         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1382                       "Checking incoming ICMP for");
1383
1384         offset2 = offset;
1385         ip_vs_fill_ip4hdr(cih, &ciph);
1386         ciph.len += offset;
1387         offset = ciph.len;
1388         /* The embedded headers contain source and dest in reverse order.
1389          * For IPIP this is error for request, not for reply.
1390          */
1391         cp = pp->conn_in_get(AF_INET, skb, &ciph, ipip ? 0 : 1);
1392         if (!cp)
1393                 return NF_ACCEPT;
1394
1395         verdict = NF_DROP;
1396
1397         /* Ensure the checksum is correct */
1398         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1399                 /* Failed checksum! */
1400                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1401                           &iph->saddr);
1402                 goto out;
1403         }
1404
1405         if (ipip) {
1406                 __be32 info = ic->un.gateway;
1407                 __u8 type = ic->type;
1408                 __u8 code = ic->code;
1409
1410                 /* Update the MTU */
1411                 if (ic->type == ICMP_DEST_UNREACH &&
1412                     ic->code == ICMP_FRAG_NEEDED) {
1413                         struct ip_vs_dest *dest = cp->dest;
1414                         u32 mtu = ntohs(ic->un.frag.mtu);
1415                         __be16 frag_off = cih->frag_off;
1416
1417                         /* Strip outer IP and ICMP, go to IPIP header */
1418                         if (pskb_pull(skb, ihl + sizeof(_icmph)) == NULL)
1419                                 goto ignore_ipip;
1420                         offset2 -= ihl + sizeof(_icmph);
1421                         skb_reset_network_header(skb);
1422                         IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1423                                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1424                         ipv4_update_pmtu(skb, dev_net(skb->dev),
1425                                          mtu, 0, 0, 0, 0);
1426                         /* Client uses PMTUD? */
1427                         if (!(frag_off & htons(IP_DF)))
1428                                 goto ignore_ipip;
1429                         /* Prefer the resulting PMTU */
1430                         if (dest) {
1431                                 struct ip_vs_dest_dst *dest_dst;
1432
1433                                 rcu_read_lock();
1434                                 dest_dst = rcu_dereference(dest->dest_dst);
1435                                 if (dest_dst)
1436                                         mtu = dst_mtu(dest_dst->dst_cache);
1437                                 rcu_read_unlock();
1438                         }
1439                         if (mtu > 68 + sizeof(struct iphdr))
1440                                 mtu -= sizeof(struct iphdr);
1441                         info = htonl(mtu);
1442                 }
1443                 /* Strip outer IP, ICMP and IPIP, go to IP header of
1444                  * original request.
1445                  */
1446                 if (pskb_pull(skb, offset2) == NULL)
1447                         goto ignore_ipip;
1448                 skb_reset_network_header(skb);
1449                 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1450                         &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1451                         type, code, ntohl(info));
1452                 icmp_send(skb, type, code, info);
1453                 /* ICMP can be shorter but anyways, account it */
1454                 ip_vs_out_stats(cp, skb);
1455
1456 ignore_ipip:
1457                 consume_skb(skb);
1458                 verdict = NF_STOLEN;
1459                 goto out;
1460         }
1461
1462         /* do the statistics and put it back */
1463         ip_vs_in_stats(cp, skb);
1464         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol ||
1465             IPPROTO_SCTP == cih->protocol)
1466                 offset += 2 * sizeof(__u16);
1467         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1468
1469 out:
1470         __ip_vs_conn_put(cp);
1471
1472         return verdict;
1473 }
1474
1475 #ifdef CONFIG_IP_VS_IPV6
1476 static int ip_vs_in_icmp_v6(struct sk_buff *skb, int *related,
1477                             unsigned int hooknum, struct ip_vs_iphdr *iph)
1478 {
1479         struct net *net = NULL;
1480         struct ipv6hdr _ip6h, *ip6h;
1481         struct icmp6hdr _icmph, *ic;
1482         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1483         struct ip_vs_conn *cp;
1484         struct ip_vs_protocol *pp;
1485         struct ip_vs_proto_data *pd;
1486         unsigned int offs_ciph, writable, verdict;
1487
1488         *related = 1;
1489
1490         ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph, iph);
1491         if (ic == NULL)
1492                 return NF_DROP;
1493
1494         /*
1495          * Work through seeing if this is for us.
1496          * These checks are supposed to be in an order that means easy
1497          * things are checked first to speed up processing.... however
1498          * this means that some packets will manage to get a long way
1499          * down this stack and then be rejected, but that's life.
1500          */
1501         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1502                 *related = 0;
1503                 return NF_ACCEPT;
1504         }
1505         /* Fragment header that is before ICMP header tells us that:
1506          * it's not an error message since they can't be fragmented.
1507          */
1508         if (iph->flags & IP6_FH_F_FRAG)
1509                 return NF_DROP;
1510
1511         IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1512                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1513                   &iph->saddr, &iph->daddr);
1514
1515         /* Now find the contained IP header */
1516         ciph.len = iph->len + sizeof(_icmph);
1517         offs_ciph = ciph.len; /* Save ip header offset */
1518         ip6h = skb_header_pointer(skb, ciph.len, sizeof(_ip6h), &_ip6h);
1519         if (ip6h == NULL)
1520                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1521         ciph.saddr.in6 = ip6h->saddr; /* conn_in_get() handles reverse order */
1522         ciph.daddr.in6 = ip6h->daddr;
1523         /* skip possible IPv6 exthdrs of contained IPv6 packet */
1524         ciph.protocol = ipv6_find_hdr(skb, &ciph.len, -1, &ciph.fragoffs, NULL);
1525         if (ciph.protocol < 0)
1526                 return NF_ACCEPT; /* Contained IPv6 hdr looks wrong, ignore */
1527
1528         net = skb_net(skb);
1529         pd = ip_vs_proto_data_get(net, ciph.protocol);
1530         if (!pd)
1531                 return NF_ACCEPT;
1532         pp = pd->pp;
1533
1534         /* Cannot handle fragmented embedded protocol */
1535         if (ciph.fragoffs)
1536                 return NF_ACCEPT;
1537
1538         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offs_ciph,
1539                       "Checking incoming ICMPv6 for");
1540
1541         /* The embedded headers contain source and dest in reverse order
1542          * if not from localhost
1543          */
1544         cp = pp->conn_in_get(AF_INET6, skb, &ciph,
1545                              (hooknum == NF_INET_LOCAL_OUT) ? 0 : 1);
1546
1547         if (!cp)
1548                 return NF_ACCEPT;
1549         /* VS/TUN, VS/DR and LOCALNODE just let it go */
1550         if ((hooknum == NF_INET_LOCAL_OUT) &&
1551             (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1552                 __ip_vs_conn_put(cp);
1553                 return NF_ACCEPT;
1554         }
1555
1556         /* do the statistics and put it back */
1557         ip_vs_in_stats(cp, skb);
1558
1559         /* Need to mangle contained IPv6 header in ICMPv6 packet */
1560         writable = ciph.len;
1561         if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1562             IPPROTO_SCTP == ciph.protocol)
1563                 writable += 2 * sizeof(__u16); /* Also mangle ports */
1564
1565         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, writable, hooknum, &ciph);
1566
1567         __ip_vs_conn_put(cp);
1568
1569         return verdict;
1570 }
1571 #endif
1572
1573
1574 /*
1575  *      Check if it's for virtual services, look it up,
1576  *      and send it on its way...
1577  */
1578 static unsigned int
1579 ip_vs_in(unsigned int hooknum, struct sk_buff *skb, int af)
1580 {
1581         struct net *net;
1582         struct ip_vs_iphdr iph;
1583         struct ip_vs_protocol *pp;
1584         struct ip_vs_proto_data *pd;
1585         struct ip_vs_conn *cp;
1586         int ret, pkts;
1587         struct netns_ipvs *ipvs;
1588
1589         /* Already marked as IPVS request or reply? */
1590         if (skb->ipvs_property)
1591                 return NF_ACCEPT;
1592
1593         /*
1594          *      Big tappo:
1595          *      - remote client: only PACKET_HOST
1596          *      - route: used for struct net when skb->dev is unset
1597          */
1598         if (unlikely((skb->pkt_type != PACKET_HOST &&
1599                       hooknum != NF_INET_LOCAL_OUT) ||
1600                      !skb_dst(skb))) {
1601                 ip_vs_fill_iph_skb(af, skb, &iph);
1602                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1603                               " ignored in hook %u\n",
1604                               skb->pkt_type, iph.protocol,
1605                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1606                 return NF_ACCEPT;
1607         }
1608         /* ipvs enabled in this netns ? */
1609         net = skb_net(skb);
1610         ipvs = net_ipvs(net);
1611         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1612                 return NF_ACCEPT;
1613
1614         ip_vs_fill_iph_skb(af, skb, &iph);
1615
1616         /* Bad... Do not break raw sockets */
1617         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1618                      af == AF_INET)) {
1619                 struct sock *sk = skb->sk;
1620                 struct inet_sock *inet = inet_sk(skb->sk);
1621
1622                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1623                         return NF_ACCEPT;
1624         }
1625
1626 #ifdef CONFIG_IP_VS_IPV6
1627         if (af == AF_INET6) {
1628                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1629                         int related;
1630                         int verdict = ip_vs_in_icmp_v6(skb, &related, hooknum,
1631                                                        &iph);
1632
1633                         if (related)
1634                                 return verdict;
1635                 }
1636         } else
1637 #endif
1638                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1639                         int related;
1640                         int verdict = ip_vs_in_icmp(skb, &related, hooknum);
1641
1642                         if (related)
1643                                 return verdict;
1644                 }
1645
1646         /* Protocol supported? */
1647         pd = ip_vs_proto_data_get(net, iph.protocol);
1648         if (unlikely(!pd))
1649                 return NF_ACCEPT;
1650         pp = pd->pp;
1651         /*
1652          * Check if the packet belongs to an existing connection entry
1653          */
1654         cp = pp->conn_in_get(af, skb, &iph, 0);
1655
1656         if (unlikely(sysctl_expire_nodest_conn(ipvs)) && cp && cp->dest &&
1657             unlikely(!atomic_read(&cp->dest->weight)) && !iph.fragoffs &&
1658             is_new_conn(skb, &iph)) {
1659                 ip_vs_conn_expire_now(cp);
1660                 __ip_vs_conn_put(cp);
1661                 cp = NULL;
1662         }
1663
1664         if (unlikely(!cp) && !iph.fragoffs) {
1665                 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1666                  * replayed fragment zero will already have created the cp
1667                  */
1668                 int v;
1669
1670                 /* Schedule and create new connection entry into &cp */
1671                 if (!pp->conn_schedule(af, skb, pd, &v, &cp, &iph))
1672                         return v;
1673         }
1674
1675         if (unlikely(!cp)) {
1676                 /* sorry, all this trouble for a no-hit :) */
1677                 IP_VS_DBG_PKT(12, af, pp, skb, 0,
1678                               "ip_vs_in: packet continues traversal as normal");
1679                 if (iph.fragoffs) {
1680                         /* Fragment that couldn't be mapped to a conn entry
1681                          * is missing module nf_defrag_ipv6
1682                          */
1683                         IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1684                         IP_VS_DBG_PKT(7, af, pp, skb, 0, "unhandled fragment");
1685                 }
1686                 return NF_ACCEPT;
1687         }
1688
1689         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Incoming packet");
1690         /* Check the server status */
1691         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1692                 /* the destination server is not available */
1693
1694                 if (sysctl_expire_nodest_conn(ipvs)) {
1695                         /* try to expire the connection immediately */
1696                         ip_vs_conn_expire_now(cp);
1697                 }
1698                 /* don't restart its timer, and silently
1699                    drop the packet. */
1700                 __ip_vs_conn_put(cp);
1701                 return NF_DROP;
1702         }
1703
1704         ip_vs_in_stats(cp, skb);
1705         ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1706         if (cp->packet_xmit)
1707                 ret = cp->packet_xmit(skb, cp, pp, &iph);
1708                 /* do not touch skb anymore */
1709         else {
1710                 IP_VS_DBG_RL("warning: packet_xmit is null");
1711                 ret = NF_ACCEPT;
1712         }
1713
1714         /* Increase its packet counter and check if it is needed
1715          * to be synchronized
1716          *
1717          * Sync connection if it is about to close to
1718          * encorage the standby servers to update the connections timeout
1719          *
1720          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1721          */
1722
1723         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
1724                 pkts = sysctl_sync_threshold(ipvs);
1725         else
1726                 pkts = atomic_add_return(1, &cp->in_pkts);
1727
1728         if (ipvs->sync_state & IP_VS_STATE_MASTER)
1729                 ip_vs_sync_conn(net, cp, pkts);
1730
1731         ip_vs_conn_put(cp);
1732         return ret;
1733 }
1734
1735 /*
1736  *      AF_INET handler in NF_INET_LOCAL_IN chain
1737  *      Schedule and forward packets from remote clients
1738  */
1739 static unsigned int
1740 ip_vs_remote_request4(const struct nf_hook_ops *ops, struct sk_buff *skb,
1741                       const struct net_device *in,
1742                       const struct net_device *out,
1743                       int (*okfn)(struct sk_buff *))
1744 {
1745         return ip_vs_in(ops->hooknum, skb, AF_INET);
1746 }
1747
1748 /*
1749  *      AF_INET handler in NF_INET_LOCAL_OUT chain
1750  *      Schedule and forward packets from local clients
1751  */
1752 static unsigned int
1753 ip_vs_local_request4(const struct nf_hook_ops *ops, struct sk_buff *skb,
1754                      const struct net_device *in, const struct net_device *out,
1755                      int (*okfn)(struct sk_buff *))
1756 {
1757         return ip_vs_in(ops->hooknum, skb, AF_INET);
1758 }
1759
1760 #ifdef CONFIG_IP_VS_IPV6
1761
1762 /*
1763  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
1764  *      Schedule and forward packets from remote clients
1765  */
1766 static unsigned int
1767 ip_vs_remote_request6(const struct nf_hook_ops *ops, struct sk_buff *skb,
1768                       const struct net_device *in,
1769                       const struct net_device *out,
1770                       int (*okfn)(struct sk_buff *))
1771 {
1772         return ip_vs_in(ops->hooknum, skb, AF_INET6);
1773 }
1774
1775 /*
1776  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
1777  *      Schedule and forward packets from local clients
1778  */
1779 static unsigned int
1780 ip_vs_local_request6(const struct nf_hook_ops *ops, struct sk_buff *skb,
1781                      const struct net_device *in, const struct net_device *out,
1782                      int (*okfn)(struct sk_buff *))
1783 {
1784         return ip_vs_in(ops->hooknum, skb, AF_INET6);
1785 }
1786
1787 #endif
1788
1789
1790 /*
1791  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1792  *      related packets destined for 0.0.0.0/0.
1793  *      When fwmark-based virtual service is used, such as transparent
1794  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
1795  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1796  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1797  *      and send them to ip_vs_in_icmp.
1798  */
1799 static unsigned int
1800 ip_vs_forward_icmp(const struct nf_hook_ops *ops, struct sk_buff *skb,
1801                    const struct net_device *in, const struct net_device *out,
1802                    int (*okfn)(struct sk_buff *))
1803 {
1804         int r;
1805         struct net *net;
1806         struct netns_ipvs *ipvs;
1807
1808         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1809                 return NF_ACCEPT;
1810
1811         /* ipvs enabled in this netns ? */
1812         net = skb_net(skb);
1813         ipvs = net_ipvs(net);
1814         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1815                 return NF_ACCEPT;
1816
1817         return ip_vs_in_icmp(skb, &r, ops->hooknum);
1818 }
1819
1820 #ifdef CONFIG_IP_VS_IPV6
1821 static unsigned int
1822 ip_vs_forward_icmp_v6(const struct nf_hook_ops *ops, struct sk_buff *skb,
1823                       const struct net_device *in, const struct net_device *out,
1824                       int (*okfn)(struct sk_buff *))
1825 {
1826         int r;
1827         struct net *net;
1828         struct netns_ipvs *ipvs;
1829         struct ip_vs_iphdr iphdr;
1830
1831         ip_vs_fill_iph_skb(AF_INET6, skb, &iphdr);
1832         if (iphdr.protocol != IPPROTO_ICMPV6)
1833                 return NF_ACCEPT;
1834
1835         /* ipvs enabled in this netns ? */
1836         net = skb_net(skb);
1837         ipvs = net_ipvs(net);
1838         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1839                 return NF_ACCEPT;
1840
1841         return ip_vs_in_icmp_v6(skb, &r, ops->hooknum, &iphdr);
1842 }
1843 #endif
1844
1845
1846 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
1847         /* After packet filtering, change source only for VS/NAT */
1848         {
1849                 .hook           = ip_vs_reply4,
1850                 .owner          = THIS_MODULE,
1851                 .pf             = NFPROTO_IPV4,
1852                 .hooknum        = NF_INET_LOCAL_IN,
1853                 .priority       = NF_IP_PRI_NAT_SRC - 2,
1854         },
1855         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1856          * or VS/NAT(change destination), so that filtering rules can be
1857          * applied to IPVS. */
1858         {
1859                 .hook           = ip_vs_remote_request4,
1860                 .owner          = THIS_MODULE,
1861                 .pf             = NFPROTO_IPV4,
1862                 .hooknum        = NF_INET_LOCAL_IN,
1863                 .priority       = NF_IP_PRI_NAT_SRC - 1,
1864         },
1865         /* Before ip_vs_in, change source only for VS/NAT */
1866         {
1867                 .hook           = ip_vs_local_reply4,
1868                 .owner          = THIS_MODULE,
1869                 .pf             = NFPROTO_IPV4,
1870                 .hooknum        = NF_INET_LOCAL_OUT,
1871                 .priority       = NF_IP_PRI_NAT_DST + 1,
1872         },
1873         /* After mangle, schedule and forward local requests */
1874         {
1875                 .hook           = ip_vs_local_request4,
1876                 .owner          = THIS_MODULE,
1877                 .pf             = NFPROTO_IPV4,
1878                 .hooknum        = NF_INET_LOCAL_OUT,
1879                 .priority       = NF_IP_PRI_NAT_DST + 2,
1880         },
1881         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1882          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1883         {
1884                 .hook           = ip_vs_forward_icmp,
1885                 .owner          = THIS_MODULE,
1886                 .pf             = NFPROTO_IPV4,
1887                 .hooknum        = NF_INET_FORWARD,
1888                 .priority       = 99,
1889         },
1890         /* After packet filtering, change source only for VS/NAT */
1891         {
1892                 .hook           = ip_vs_reply4,
1893                 .owner          = THIS_MODULE,
1894                 .pf             = NFPROTO_IPV4,
1895                 .hooknum        = NF_INET_FORWARD,
1896                 .priority       = 100,
1897         },
1898 #ifdef CONFIG_IP_VS_IPV6
1899         /* After packet filtering, change source only for VS/NAT */
1900         {
1901                 .hook           = ip_vs_reply6,
1902                 .owner          = THIS_MODULE,
1903                 .pf             = NFPROTO_IPV6,
1904                 .hooknum        = NF_INET_LOCAL_IN,
1905                 .priority       = NF_IP6_PRI_NAT_SRC - 2,
1906         },
1907         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1908          * or VS/NAT(change destination), so that filtering rules can be
1909          * applied to IPVS. */
1910         {
1911                 .hook           = ip_vs_remote_request6,
1912                 .owner          = THIS_MODULE,
1913                 .pf             = NFPROTO_IPV6,
1914                 .hooknum        = NF_INET_LOCAL_IN,
1915                 .priority       = NF_IP6_PRI_NAT_SRC - 1,
1916         },
1917         /* Before ip_vs_in, change source only for VS/NAT */
1918         {
1919                 .hook           = ip_vs_local_reply6,
1920                 .owner          = THIS_MODULE,
1921                 .pf             = NFPROTO_IPV6,
1922                 .hooknum        = NF_INET_LOCAL_OUT,
1923                 .priority       = NF_IP6_PRI_NAT_DST + 1,
1924         },
1925         /* After mangle, schedule and forward local requests */
1926         {
1927                 .hook           = ip_vs_local_request6,
1928                 .owner          = THIS_MODULE,
1929                 .pf             = NFPROTO_IPV6,
1930                 .hooknum        = NF_INET_LOCAL_OUT,
1931                 .priority       = NF_IP6_PRI_NAT_DST + 2,
1932         },
1933         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1934          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1935         {
1936                 .hook           = ip_vs_forward_icmp_v6,
1937                 .owner          = THIS_MODULE,
1938                 .pf             = NFPROTO_IPV6,
1939                 .hooknum        = NF_INET_FORWARD,
1940                 .priority       = 99,
1941         },
1942         /* After packet filtering, change source only for VS/NAT */
1943         {
1944                 .hook           = ip_vs_reply6,
1945                 .owner          = THIS_MODULE,
1946                 .pf             = NFPROTO_IPV6,
1947                 .hooknum        = NF_INET_FORWARD,
1948                 .priority       = 100,
1949         },
1950 #endif
1951 };
1952 /*
1953  *      Initialize IP Virtual Server netns mem.
1954  */
1955 static int __net_init __ip_vs_init(struct net *net)
1956 {
1957         struct netns_ipvs *ipvs;
1958
1959         ipvs = net_generic(net, ip_vs_net_id);
1960         if (ipvs == NULL)
1961                 return -ENOMEM;
1962
1963         /* Hold the beast until a service is registerd */
1964         ipvs->enable = 0;
1965         ipvs->net = net;
1966         /* Counters used for creating unique names */
1967         ipvs->gen = atomic_read(&ipvs_netns_cnt);
1968         atomic_inc(&ipvs_netns_cnt);
1969         net->ipvs = ipvs;
1970
1971         if (ip_vs_estimator_net_init(net) < 0)
1972                 goto estimator_fail;
1973
1974         if (ip_vs_control_net_init(net) < 0)
1975                 goto control_fail;
1976
1977         if (ip_vs_protocol_net_init(net) < 0)
1978                 goto protocol_fail;
1979
1980         if (ip_vs_app_net_init(net) < 0)
1981                 goto app_fail;
1982
1983         if (ip_vs_conn_net_init(net) < 0)
1984                 goto conn_fail;
1985
1986         if (ip_vs_sync_net_init(net) < 0)
1987                 goto sync_fail;
1988
1989         printk(KERN_INFO "IPVS: Creating netns size=%zu id=%d\n",
1990                          sizeof(struct netns_ipvs), ipvs->gen);
1991         return 0;
1992 /*
1993  * Error handling
1994  */
1995
1996 sync_fail:
1997         ip_vs_conn_net_cleanup(net);
1998 conn_fail:
1999         ip_vs_app_net_cleanup(net);
2000 app_fail:
2001         ip_vs_protocol_net_cleanup(net);
2002 protocol_fail:
2003         ip_vs_control_net_cleanup(net);
2004 control_fail:
2005         ip_vs_estimator_net_cleanup(net);
2006 estimator_fail:
2007         net->ipvs = NULL;
2008         return -ENOMEM;
2009 }
2010
2011 static void __net_exit __ip_vs_cleanup(struct net *net)
2012 {
2013         ip_vs_service_net_cleanup(net); /* ip_vs_flush() with locks */
2014         ip_vs_conn_net_cleanup(net);
2015         ip_vs_app_net_cleanup(net);
2016         ip_vs_protocol_net_cleanup(net);
2017         ip_vs_control_net_cleanup(net);
2018         ip_vs_estimator_net_cleanup(net);
2019         IP_VS_DBG(2, "ipvs netns %d released\n", net_ipvs(net)->gen);
2020         net->ipvs = NULL;
2021 }
2022
2023 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
2024 {
2025         EnterFunction(2);
2026         net_ipvs(net)->enable = 0;      /* Disable packet reception */
2027         smp_wmb();
2028         ip_vs_sync_net_cleanup(net);
2029         LeaveFunction(2);
2030 }
2031
2032 static struct pernet_operations ipvs_core_ops = {
2033         .init = __ip_vs_init,
2034         .exit = __ip_vs_cleanup,
2035         .id   = &ip_vs_net_id,
2036         .size = sizeof(struct netns_ipvs),
2037 };
2038
2039 static struct pernet_operations ipvs_core_dev_ops = {
2040         .exit = __ip_vs_dev_cleanup,
2041 };
2042
2043 /*
2044  *      Initialize IP Virtual Server
2045  */
2046 static int __init ip_vs_init(void)
2047 {
2048         int ret;
2049
2050         ret = ip_vs_control_init();
2051         if (ret < 0) {
2052                 pr_err("can't setup control.\n");
2053                 goto exit;
2054         }
2055
2056         ip_vs_protocol_init();
2057
2058         ret = ip_vs_conn_init();
2059         if (ret < 0) {
2060                 pr_err("can't setup connection table.\n");
2061                 goto cleanup_protocol;
2062         }
2063
2064         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
2065         if (ret < 0)
2066                 goto cleanup_conn;
2067
2068         ret = register_pernet_device(&ipvs_core_dev_ops);
2069         if (ret < 0)
2070                 goto cleanup_sub;
2071
2072         ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2073         if (ret < 0) {
2074                 pr_err("can't register hooks.\n");
2075                 goto cleanup_dev;
2076         }
2077
2078         ret = ip_vs_register_nl_ioctl();
2079         if (ret < 0) {
2080                 pr_err("can't register netlink/ioctl.\n");
2081                 goto cleanup_hooks;
2082         }
2083
2084         pr_info("ipvs loaded.\n");
2085
2086         return ret;
2087
2088 cleanup_hooks:
2089         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2090 cleanup_dev:
2091         unregister_pernet_device(&ipvs_core_dev_ops);
2092 cleanup_sub:
2093         unregister_pernet_subsys(&ipvs_core_ops);
2094 cleanup_conn:
2095         ip_vs_conn_cleanup();
2096 cleanup_protocol:
2097         ip_vs_protocol_cleanup();
2098         ip_vs_control_cleanup();
2099 exit:
2100         return ret;
2101 }
2102
2103 static void __exit ip_vs_cleanup(void)
2104 {
2105         ip_vs_unregister_nl_ioctl();
2106         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2107         unregister_pernet_device(&ipvs_core_dev_ops);
2108         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
2109         ip_vs_conn_cleanup();
2110         ip_vs_protocol_cleanup();
2111         ip_vs_control_cleanup();
2112         pr_info("ipvs unloaded.\n");
2113 }
2114
2115 module_init(ip_vs_init);
2116 module_exit(ip_vs_cleanup);
2117 MODULE_LICENSE("GPL");