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_%d", 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
 120                 s = this_cpu_ptr(dest->stats.cpustats);
 121                 s->ustats.inpkts++;
 122                 u64_stats_update_begin(&s->syncp);
 123                 s->ustats.inbytes += skb->len;
 124                 u64_stats_update_end(&s->syncp);
 125
 126                 s = this_cpu_ptr(dest->svc->stats.cpustats);
 127                 s->ustats.inpkts++;
 128                 u64_stats_update_begin(&s->syncp);
 129                 s->ustats.inbytes += skb->len;
 130                 u64_stats_update_end(&s->syncp);
 131
 132                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 133                 s->ustats.inpkts++;
 134                 u64_stats_update_begin(&s->syncp);
 135                 s->ustats.inbytes += skb->len;
 136                 u64_stats_update_end(&s->syncp);
 137         }
 138 }
 139
 140
 141 static inline void
 142 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 143 {
 144         struct ip_vs_dest *dest = cp->dest;
 145         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 146
 147         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 148                 struct ip_vs_cpu_stats *s;
 149
 150                 s = this_cpu_ptr(dest->stats.cpustats);
 151                 s->ustats.outpkts++;
 152                 u64_stats_update_begin(&s->syncp);
 153                 s->ustats.outbytes += skb->len;
 154                 u64_stats_update_end(&s->syncp);
 155
 156                 s = this_cpu_ptr(dest->svc->stats.cpustats);
 157                 s->ustats.outpkts++;
 158                 u64_stats_update_begin(&s->syncp);
 159                 s->ustats.outbytes += skb->len;
 160                 u64_stats_update_end(&s->syncp);
 161
 162                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 163                 s->ustats.outpkts++;
 164                 u64_stats_update_begin(&s->syncp);
 165                 s->ustats.outbytes += skb->len;
 166                 u64_stats_update_end(&s->syncp);
 167         }
 168 }
 169
 170
 171 static inline void
 172 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
 173 {
 174         struct netns_ipvs *ipvs = net_ipvs(svc->net);
 175         struct ip_vs_cpu_stats *s;
 176
 177         s = this_cpu_ptr(cp->dest->stats.cpustats);
 178         s->ustats.conns++;
 179
 180         s = this_cpu_ptr(svc->stats.cpustats);
 181         s->ustats.conns++;
 182
 183         s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 184         s->ustats.conns++;
 185 }
 186
 187
 188 static inline void
 189 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
 190                 const struct sk_buff *skb,
 191                 struct ip_vs_proto_data *pd)
 192 {
 193         if (likely(pd->pp->state_transition))
 194                 pd->pp->state_transition(cp, direction, skb, pd);
 195 }
 196
 197 static inline int
 198 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
 199                               struct sk_buff *skb, int protocol,
 200                               const union nf_inet_addr *caddr, __be16 cport,
 201                               const union nf_inet_addr *vaddr, __be16 vport,
 202                               struct ip_vs_conn_param *p)
 203 {
 204         ip_vs_conn_fill_param(svc->net, svc->af, protocol, caddr, cport, vaddr,
 205                               vport, p);
 206         p->pe = rcu_dereference(svc->pe);
 207         if (p->pe && p->pe->fill_param)
 208                 return p->pe->fill_param(p, skb);
 209
 210         return 0;
 211 }
 212
 213 /*
 214  *  IPVS persistent scheduling function
 215  *  It creates a connection entry according to its template if exists,
 216  *  or selects a server and creates a connection entry plus a template.
 217  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
 218  *  Protocols supported: TCP, UDP
 219  */
 220 static struct ip_vs_conn *
 221 ip_vs_sched_persist(struct ip_vs_service *svc,
 222                     struct sk_buff *skb, __be16 src_port, __be16 dst_port,
 223                     int *ignored, struct ip_vs_iphdr *iph)
 224 {
 225         struct ip_vs_conn *cp = NULL;
 226         struct ip_vs_dest *dest;
 227         struct ip_vs_conn *ct;
 228         __be16 dport = 0;               /* destination port to forward */
 229         unsigned int flags;
 230         struct ip_vs_conn_param param;
 231         const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
 232         union nf_inet_addr snet;        /* source network of the client,
 233                                            after masking */
 234
 235         /* Mask saddr with the netmask to adjust template granularity */
 236 #ifdef CONFIG_IP_VS_IPV6
 237         if (svc->af == AF_INET6)
 238                 ipv6_addr_prefix(&snet.in6, &iph->saddr.in6, svc->netmask);
 239         else
 240 #endif
 241                 snet.ip = iph->saddr.ip & svc->netmask;
 242
 243         IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
 244                       "mnet %s\n",
 245                       IP_VS_DBG_ADDR(svc->af, &iph->saddr), ntohs(src_port),
 246                       IP_VS_DBG_ADDR(svc->af, &iph->daddr), ntohs(dst_port),
 247                       IP_VS_DBG_ADDR(svc->af, &snet));
 248
 249         /*
 250          * As far as we know, FTP is a very complicated network protocol, and
 251          * it uses control connection and data connections. For active FTP,
 252          * FTP server initialize data connection to the client, its source port
 253          * is often 20. For passive FTP, FTP server tells the clients the port
 254          * that it passively listens to,  and the client issues the data
 255          * connection. In the tunneling or direct routing mode, the load
 256          * balancer is on the client-to-server half of connection, the port
 257          * number is unknown to the load balancer. So, a conn template like
 258          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
 259          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
 260          * is created for other persistent services.
 261          */
 262         {
 263                 int protocol = iph->protocol;
 264                 const union nf_inet_addr *vaddr = &iph->daddr;
 265                 __be16 vport = 0;
 266
 267                 if (dst_port == svc->port) {
 268                         /* non-FTP template:
 269                          * <protocol, caddr, 0, vaddr, vport, daddr, dport>
 270                          * FTP template:
 271                          * <protocol, caddr, 0, vaddr, 0, daddr, 0>
 272                          */
 273                         if (svc->port != FTPPORT)
 274                                 vport = dst_port;
 275                 } else {
 276                         /* Note: persistent fwmark-based services and
 277                          * persistent port zero service are handled here.
 278                          * fwmark template:
 279                          * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
 280                          * port zero template:
 281                          * <protocol,caddr,0,vaddr,0,daddr,0>
 282                          */
 283                         if (svc->fwmark) {
 284                                 protocol = IPPROTO_IP;
 285                                 vaddr = &fwmark;
 286                         }
 287                 }
 288                 /* return *ignored = -1 so NF_DROP can be used */
 289                 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
 290                                                   vaddr, vport, &param) < 0) {
 291                         *ignored = -1;
 292                         return NULL;
 293                 }
 294         }
 295
 296         /* Check if a template already exists */
 297         ct = ip_vs_ct_in_get(&param);
 298         if (!ct || !ip_vs_check_template(ct)) {
 299                 struct ip_vs_scheduler *sched;
 300
 301                 /*
 302                  * No template found or the dest of the connection
 303                  * template is not available.
 304                  * return *ignored=0 i.e. ICMP and NF_DROP
 305                  */
 306                 sched = rcu_dereference(svc->scheduler);
 307                 dest = sched->schedule(svc, skb);
 308                 if (!dest) {
 309                         IP_VS_DBG(1, "p-schedule: no dest found.\n");
 310                         kfree(param.pe_data);
 311                         *ignored = 0;
 312                         return NULL;
 313                 }
 314
 315                 if (dst_port == svc->port && svc->port != FTPPORT)
 316                         dport = dest->port;
 317
 318                 /* Create a template
 319                  * This adds param.pe_data to the template,
 320                  * and thus param.pe_data will be destroyed
 321                  * when the template expires */
 322                 ct = ip_vs_conn_new(&param, &dest->addr, dport,
 323                                     IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
 324                 if (ct == NULL) {
 325                         kfree(param.pe_data);
 326                         *ignored = -1;
 327                         return NULL;
 328                 }
 329
 330                 ct->timeout = svc->timeout;
 331         } else {
 332                 /* set destination with the found template */
 333                 dest = ct->dest;
 334                 kfree(param.pe_data);
 335         }
 336
 337         dport = dst_port;
 338         if (dport == svc->port && dest->port)
 339                 dport = dest->port;
 340
 341         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 342                  && iph->protocol == IPPROTO_UDP) ?
 343                 IP_VS_CONN_F_ONE_PACKET : 0;
 344
 345         /*
 346          *    Create a new connection according to the template
 347          */
 348         ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol, &iph->saddr,
 349                               src_port, &iph->daddr, dst_port, &param);
 350
 351         cp = ip_vs_conn_new(&param, &dest->addr, dport, flags, dest, skb->mark);
 352         if (cp == NULL) {
 353                 ip_vs_conn_put(ct);
 354                 *ignored = -1;
 355                 return NULL;
 356         }
 357
 358         /*
 359          *    Add its control
 360          */
 361         ip_vs_control_add(cp, ct);
 362         ip_vs_conn_put(ct);
 363
 364         ip_vs_conn_stats(cp, svc);
 365         return cp;
 366 }
 367
 368
 369 /*
 370  *  IPVS main scheduling function
 371  *  It selects a server according to the virtual service, and
 372  *  creates a connection entry.
 373  *  Protocols supported: TCP, UDP
 374  *
 375  *  Usage of *ignored
 376  *
 377  * 1 :   protocol tried to schedule (eg. on SYN), found svc but the
 378  *       svc/scheduler decides that this packet should be accepted with
 379  *       NF_ACCEPT because it must not be scheduled.
 380  *
 381  * 0 :   scheduler can not find destination, so try bypass or
 382  *       return ICMP and then NF_DROP (ip_vs_leave).
 383  *
 384  * -1 :  scheduler tried to schedule but fatal error occurred, eg.
 385  *       ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
 386  *       failure such as missing Call-ID, ENOMEM on skb_linearize
 387  *       or pe_data. In this case we should return NF_DROP without
 388  *       any attempts to send ICMP with ip_vs_leave.
 389  */
 390 struct ip_vs_conn *
 391 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
 392                struct ip_vs_proto_data *pd, int *ignored,
 393                struct ip_vs_iphdr *iph)
 394 {
 395         struct ip_vs_protocol *pp = pd->pp;
 396         struct ip_vs_conn *cp = NULL;
 397         struct ip_vs_scheduler *sched;
 398         struct ip_vs_dest *dest;
 399         __be16 _ports[2], *pptr;
 400         unsigned int flags;
 401
 402         *ignored = 1;
 403         /*
 404          * IPv6 frags, only the first hit here.
 405          */
 406         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 407         if (pptr == NULL)
 408                 return NULL;
 409
 410         /*
 411          * FTPDATA needs this check when using local real server.
 412          * Never schedule Active FTPDATA connections from real server.
 413          * For LVS-NAT they must be already created. For other methods
 414          * with persistence the connection is created on SYN+ACK.
 415          */
 416         if (pptr[0] == FTPDATA) {
 417                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 418                               "Not scheduling FTPDATA");
 419                 return NULL;
 420         }
 421
 422         /*
 423          *    Do not schedule replies from local real server.
 424          */
 425         if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
 426             (cp = pp->conn_in_get(svc->af, skb, iph, 1))) {
 427                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 428                               "Not scheduling reply for existing connection");
 429                 __ip_vs_conn_put(cp);
 430                 return NULL;
 431         }
 432
 433         /*
 434          *    Persistent service
 435          */
 436         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 437                 return ip_vs_sched_persist(svc, skb, pptr[0], pptr[1], ignored,
 438                                            iph);
 439
 440         *ignored = 0;
 441
 442         /*
 443          *    Non-persistent service
 444          */
 445         if (!svc->fwmark && pptr[1] != svc->port) {
 446                 if (!svc->port)
 447                         pr_err("Schedule: port zero only supported "
 448                                "in persistent services, "
 449                                "check your ipvs configuration\n");
 450                 return NULL;
 451         }
 452
 453         sched = rcu_dereference(svc->scheduler);
 454         dest = sched->schedule(svc, skb);
 455         if (dest == NULL) {
 456                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 457                 return NULL;
 458         }
 459
 460         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 461                  && iph->protocol == IPPROTO_UDP) ?
 462                 IP_VS_CONN_F_ONE_PACKET : 0;
 463
 464         /*
 465          *    Create a connection entry.
 466          */
 467         {
 468                 struct ip_vs_conn_param p;
 469
 470                 ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol,
 471                                       &iph->saddr, pptr[0], &iph->daddr,
 472                                       pptr[1], &p);
 473                 cp = ip_vs_conn_new(&p, &dest->addr,
 474                                     dest->port ? dest->port : pptr[1],
 475                                     flags, dest, skb->mark);
 476                 if (!cp) {
 477                         *ignored = -1;
 478                         return NULL;
 479                 }
 480         }
 481
 482         IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
 483                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
 484                       ip_vs_fwd_tag(cp),
 485                       IP_VS_DBG_ADDR(svc->af, &cp->caddr), ntohs(cp->cport),
 486                       IP_VS_DBG_ADDR(svc->af, &cp->vaddr), ntohs(cp->vport),
 487                       IP_VS_DBG_ADDR(svc->af, &cp->daddr), ntohs(cp->dport),
 488                       cp->flags, atomic_read(&cp->refcnt));
 489
 490         ip_vs_conn_stats(cp, svc);
 491         return cp;
 492 }
 493
 494
 495 /*
 496  *  Pass or drop the packet.
 497  *  Called by ip_vs_in, when the virtual service is available but
 498  *  no destination is available for a new connection.
 499  */
 500 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 501                 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
 502 {
 503         __be16 _ports[2], *pptr;
 504 #ifdef CONFIG_SYSCTL
 505         struct net *net;
 506         struct netns_ipvs *ipvs;
 507         int unicast;
 508 #endif
 509
 510         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 511         if (pptr == NULL) {
 512                 return NF_DROP;
 513         }
 514
 515 #ifdef CONFIG_SYSCTL
 516         net = skb_net(skb);
 517
 518 #ifdef CONFIG_IP_VS_IPV6
 519         if (svc->af == AF_INET6)
 520                 unicast = ipv6_addr_type(&iph->daddr.in6) & IPV6_ADDR_UNICAST;
 521         else
 522 #endif
 523                 unicast = (inet_addr_type(net, iph->daddr.ip) == RTN_UNICAST);
 524
 525         /* if it is fwmark-based service, the cache_bypass sysctl is up
 526            and the destination is a non-local unicast, then create
 527            a cache_bypass connection entry */
 528         ipvs = net_ipvs(net);
 529         if (ipvs->sysctl_cache_bypass && svc->fwmark && unicast) {
 530                 int ret;
 531                 struct ip_vs_conn *cp;
 532                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
 533                                       iph->protocol == IPPROTO_UDP) ?
 534                                       IP_VS_CONN_F_ONE_PACKET : 0;
 535                 union nf_inet_addr daddr =  { .all = { 0, 0, 0, 0 } };
 536
 537                 /* create a new connection entry */
 538                 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
 539                 {
 540                         struct ip_vs_conn_param p;
 541                         ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol,
 542                                               &iph->saddr, pptr[0],
 543                                               &iph->daddr, pptr[1], &p);
 544                         cp = ip_vs_conn_new(&p, &daddr, 0,
 545                                             IP_VS_CONN_F_BYPASS | flags,
 546                                             NULL, skb->mark);
 547                         if (!cp)
 548                                 return NF_DROP;
 549                 }
 550
 551                 /* statistics */
 552                 ip_vs_in_stats(cp, skb);
 553
 554                 /* set state */
 555                 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
 556
 557                 /* transmit the first SYN packet */
 558                 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
 559                 /* do not touch skb anymore */
 560
 561                 atomic_inc(&cp->in_pkts);
 562                 ip_vs_conn_put(cp);
 563                 return ret;
 564         }
 565 #endif
 566
 567         /*
 568          * When the virtual ftp service is presented, packets destined
 569          * for other services on the VIP may get here (except services
 570          * listed in the ipvs table), pass the packets, because it is
 571          * not ipvs job to decide to drop the packets.
 572          */
 573         if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT))
 574                 return NF_ACCEPT;
 575
 576         /*
 577          * Notify the client that the destination is unreachable, and
 578          * release the socket buffer.
 579          * Since it is in IP layer, the TCP socket is not actually
 580          * created, the TCP RST packet cannot be sent, instead that
 581          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 582          */
 583 #ifdef CONFIG_IP_VS_IPV6
 584         if (svc->af == AF_INET6) {
 585                 if (!skb->dev) {
 586                         struct net *net = dev_net(skb_dst(skb)->dev);
 587
 588                         skb->dev = net->loopback_dev;
 589                 }
 590                 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 591         } else
 592 #endif
 593                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 594
 595         return NF_DROP;
 596 }
 597
 598 #ifdef CONFIG_SYSCTL
 599
 600 static int sysctl_snat_reroute(struct sk_buff *skb)
 601 {
 602         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 603         return ipvs->sysctl_snat_reroute;
 604 }
 605
 606 static int sysctl_nat_icmp_send(struct net *net)
 607 {
 608         struct netns_ipvs *ipvs = net_ipvs(net);
 609         return ipvs->sysctl_nat_icmp_send;
 610 }
 611
 612 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
 613 {
 614         return ipvs->sysctl_expire_nodest_conn;
 615 }
 616
 617 #else
 618
 619 static int sysctl_snat_reroute(struct sk_buff *skb) { return 0; }
 620 static int sysctl_nat_icmp_send(struct net *net) { return 0; }
 621 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
 622
 623 #endif
 624
 625 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 626 {
 627         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 628 }
 629
 630 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
 631 {
 632         if (NF_INET_LOCAL_IN == hooknum)
 633                 return IP_DEFRAG_VS_IN;
 634         if (NF_INET_FORWARD == hooknum)
 635                 return IP_DEFRAG_VS_FWD;
 636         return IP_DEFRAG_VS_OUT;
 637 }
 638
 639 static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user)
 640 {
 641         int err;
 642
 643         local_bh_disable();
 644         err = ip_defrag(skb, user);
 645         local_bh_enable();
 646         if (!err)
 647                 ip_send_check(ip_hdr(skb));
 648
 649         return err;
 650 }
 651
 652 static int ip_vs_route_me_harder(int af, struct sk_buff *skb)
 653 {
 654 #ifdef CONFIG_IP_VS_IPV6
 655         if (af == AF_INET6) {
 656                 if (sysctl_snat_reroute(skb) && ip6_route_me_harder(skb) != 0)
 657                         return 1;
 658         } else
 659 #endif
 660                 if ((sysctl_snat_reroute(skb) ||
 661                      skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
 662                     ip_route_me_harder(skb, RTN_LOCAL) != 0)
 663                         return 1;
 664
 665         return 0;
 666 }
 667
 668 /*
 669  * Packet has been made sufficiently writable in caller
 670  * - inout: 1=in->out, 0=out->in
 671  */
 672 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 673                     struct ip_vs_conn *cp, int inout)
 674 {
 675         struct iphdr *iph        = ip_hdr(skb);
 676         unsigned int icmp_offset = iph->ihl*4;
 677         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
 678                                                       icmp_offset);
 679         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 680
 681         if (inout) {
 682                 iph->saddr = cp->vaddr.ip;
 683                 ip_send_check(iph);
 684                 ciph->daddr = cp->vaddr.ip;
 685                 ip_send_check(ciph);
 686         } else {
 687                 iph->daddr = cp->daddr.ip;
 688                 ip_send_check(iph);
 689                 ciph->saddr = cp->daddr.ip;
 690                 ip_send_check(ciph);
 691         }
 692
 693         /* the TCP/UDP/SCTP port */
 694         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
 695             IPPROTO_SCTP == ciph->protocol) {
 696                 __be16 *ports = (void *)ciph + ciph->ihl*4;
 697
 698                 if (inout)
 699                         ports[1] = cp->vport;
 700                 else
 701                         ports[0] = cp->dport;
 702         }
 703
 704         /* And finally the ICMP checksum */
 705         icmph->checksum = 0;
 706         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 707         skb->ip_summed = CHECKSUM_UNNECESSARY;
 708
 709         if (inout)
 710                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 711                         "Forwarding altered outgoing ICMP");
 712         else
 713                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 714                         "Forwarding altered incoming ICMP");
 715 }
 716
 717 #ifdef CONFIG_IP_VS_IPV6
 718 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
 719                     struct ip_vs_conn *cp, int inout)
 720 {
 721         struct ipv6hdr *iph      = ipv6_hdr(skb);
 722         unsigned int icmp_offset = 0;
 723         unsigned int offs        = 0; /* header offset*/
 724         int protocol;
 725         struct icmp6hdr *icmph;
 726         struct ipv6hdr *ciph;
 727         unsigned short fragoffs;
 728
 729         ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
 730         icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
 731         offs = icmp_offset + sizeof(struct icmp6hdr);
 732         ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
 733
 734         protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
 735
 736         if (inout) {
 737                 iph->saddr = cp->vaddr.in6;
 738                 ciph->daddr = cp->vaddr.in6;
 739         } else {
 740                 iph->daddr = cp->daddr.in6;
 741                 ciph->saddr = cp->daddr.in6;
 742         }
 743
 744         /* the TCP/UDP/SCTP port */
 745         if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 746                           IPPROTO_SCTP == protocol)) {
 747                 __be16 *ports = (void *)(skb_network_header(skb) + offs);
 748
 749                 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
 750                               ntohs(inout ? ports[1] : ports[0]),
 751                               ntohs(inout ? cp->vport : cp->dport));
 752                 if (inout)
 753                         ports[1] = cp->vport;
 754                 else
 755                         ports[0] = cp->dport;
 756         }
 757
 758         /* And finally the ICMP checksum */
 759         icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
 760                                               skb->len - icmp_offset,
 761                                               IPPROTO_ICMPV6, 0);
 762         skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
 763         skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
 764         skb->ip_summed = CHECKSUM_PARTIAL;
 765
 766         if (inout)
 767                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 768                               (void *)ciph - (void *)iph,
 769                               "Forwarding altered outgoing ICMPv6");
 770         else
 771                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 772                               (void *)ciph - (void *)iph,
 773                               "Forwarding altered incoming ICMPv6");
 774 }
 775 #endif
 776
 777 /* Handle relevant response ICMP messages - forward to the right
 778  * destination host.
 779  */
 780 static int handle_response_icmp(int af, struct sk_buff *skb,
 781                                 union nf_inet_addr *snet,
 782                                 __u8 protocol, struct ip_vs_conn *cp,
 783                                 struct ip_vs_protocol *pp,
 784                                 unsigned int offset, unsigned int ihl)
 785 {
 786         unsigned int verdict = NF_DROP;
 787
 788         if (IP_VS_FWD_METHOD(cp) != 0) {
 789                 pr_err("shouldn't reach here, because the box is on the "
 790                        "half connection in the tun/dr module.\n");
 791         }
 792
 793         /* Ensure the checksum is correct */
 794         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 795                 /* Failed checksum! */
 796                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
 797                               IP_VS_DBG_ADDR(af, snet));
 798                 goto out;
 799         }
 800
 801         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 802             IPPROTO_SCTP == protocol)
 803                 offset += 2 * sizeof(__u16);
 804         if (!skb_make_writable(skb, offset))
 805                 goto out;
 806
 807 #ifdef CONFIG_IP_VS_IPV6
 808         if (af == AF_INET6)
 809                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
 810         else
 811 #endif
 812                 ip_vs_nat_icmp(skb, pp, cp, 1);
 813
 814         if (ip_vs_route_me_harder(af, skb))
 815                 goto out;
 816
 817         /* do the statistics and put it back */
 818         ip_vs_out_stats(cp, skb);
 819
 820         skb->ipvs_property = 1;
 821         if (!(cp->flags & IP_VS_CONN_F_NFCT))
 822                 ip_vs_notrack(skb);
 823         else
 824                 ip_vs_update_conntrack(skb, cp, 0);
 825         verdict = NF_ACCEPT;
 826
 827 out:
 828         __ip_vs_conn_put(cp);
 829
 830         return verdict;
 831 }
 832
 833 /*
 834  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 835  *      Find any that might be relevant, check against existing connections.
 836  *      Currently handles error types - unreachable, quench, ttl exceeded.
 837  */
 838 static int ip_vs_out_icmp(struct sk_buff *skb, int *related,
 839                           unsigned int hooknum)
 840 {
 841         struct iphdr *iph;
 842         struct icmphdr  _icmph, *ic;
 843         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 844         struct ip_vs_iphdr ciph;
 845         struct ip_vs_conn *cp;
 846         struct ip_vs_protocol *pp;
 847         unsigned int offset, ihl;
 848         union nf_inet_addr snet;
 849
 850         *related = 1;
 851
 852         /* reassemble IP fragments */
 853         if (ip_is_fragment(ip_hdr(skb))) {
 854                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
 855                         return NF_STOLEN;
 856         }
 857
 858         iph = ip_hdr(skb);
 859         offset = ihl = iph->ihl * 4;
 860         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 861         if (ic == NULL)
 862                 return NF_DROP;
 863
 864         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
 865                   ic->type, ntohs(icmp_id(ic)),
 866                   &iph->saddr, &iph->daddr);
 867
 868         /*
 869          * Work through seeing if this is for us.
 870          * These checks are supposed to be in an order that means easy
 871          * things are checked first to speed up processing.... however
 872          * this means that some packets will manage to get a long way
 873          * down this stack and then be rejected, but that's life.
 874          */
 875         if ((ic->type != ICMP_DEST_UNREACH) &&
 876             (ic->type != ICMP_SOURCE_QUENCH) &&
 877             (ic->type != ICMP_TIME_EXCEEDED)) {
 878                 *related = 0;
 879                 return NF_ACCEPT;
 880         }
 881
 882         /* Now find the contained IP header */
 883         offset += sizeof(_icmph);
 884         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 885         if (cih == NULL)
 886                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 887
 888         pp = ip_vs_proto_get(cih->protocol);
 889         if (!pp)
 890                 return NF_ACCEPT;
 891
 892         /* Is the embedded protocol header present? */
 893         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 894                      pp->dont_defrag))
 895                 return NF_ACCEPT;
 896
 897         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
 898                       "Checking outgoing ICMP for");
 899
 900         ip_vs_fill_ip4hdr(cih, &ciph);
 901         ciph.len += offset;
 902         /* The embedded headers contain source and dest in reverse order */
 903         cp = pp->conn_out_get(AF_INET, skb, &ciph, 1);
 904         if (!cp)
 905                 return NF_ACCEPT;
 906
 907         snet.ip = iph->saddr;
 908         return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
 909                                     pp, ciph.len, ihl);
 910 }
 911
 912 #ifdef CONFIG_IP_VS_IPV6
 913 static int ip_vs_out_icmp_v6(struct sk_buff *skb, int *related,
 914                              unsigned int hooknum, struct ip_vs_iphdr *ipvsh)
 915 {
 916         struct icmp6hdr _icmph, *ic;
 917         struct ipv6hdr _ip6h, *ip6h; /* The ip header contained within ICMP */
 918         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
 919         struct ip_vs_conn *cp;
 920         struct ip_vs_protocol *pp;
 921         union nf_inet_addr snet;
 922         unsigned int writable;
 923
 924         *related = 1;
 925         ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph, ipvsh);
 926         if (ic == NULL)
 927                 return NF_DROP;
 928
 929         /*
 930          * Work through seeing if this is for us.
 931          * These checks are supposed to be in an order that means easy
 932          * things are checked first to speed up processing.... however
 933          * this means that some packets will manage to get a long way
 934          * down this stack and then be rejected, but that's life.
 935          */
 936         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
 937                 *related = 0;
 938                 return NF_ACCEPT;
 939         }
 940         /* Fragment header that is before ICMP header tells us that:
 941          * it's not an error message since they can't be fragmented.
 942          */
 943         if (ipvsh->flags & IP6_FH_F_FRAG)
 944                 return NF_DROP;
 945
 946         IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
 947                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
 948                   &ipvsh->saddr, &ipvsh->daddr);
 949
 950         /* Now find the contained IP header */
 951         ciph.len = ipvsh->len + sizeof(_icmph);
 952         ip6h = skb_header_pointer(skb, ciph.len, sizeof(_ip6h), &_ip6h);
 953         if (ip6h == NULL)
 954                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 955         ciph.saddr.in6 = ip6h->saddr; /* conn_out_get() handles reverse order */
 956         ciph.daddr.in6 = ip6h->daddr;
 957         /* skip possible IPv6 exthdrs of contained IPv6 packet */
 958         ciph.protocol = ipv6_find_hdr(skb, &ciph.len, -1, &ciph.fragoffs, NULL);
 959         if (ciph.protocol < 0)
 960                 return NF_ACCEPT; /* Contained IPv6 hdr looks wrong, ignore */
 961
 962         pp = ip_vs_proto_get(ciph.protocol);
 963         if (!pp)
 964                 return NF_ACCEPT;
 965
 966         /* The embedded headers contain source and dest in reverse order */
 967         cp = pp->conn_out_get(AF_INET6, skb, &ciph, 1);
 968         if (!cp)
 969                 return NF_ACCEPT;
 970
 971         snet.in6 = ciph.saddr.in6;
 972         writable = ciph.len;
 973         return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
 974                                     pp, writable, sizeof(struct ipv6hdr));
 975 }
 976 #endif
 977
 978 /*
 979  * Check if sctp chunc is ABORT chunk
 980  */
 981 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
 982 {
 983         sctp_chunkhdr_t *sch, schunk;
 984         sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
 985                         sizeof(schunk), &schunk);
 986         if (sch == NULL)
 987                 return 0;
 988         if (sch->type == SCTP_CID_ABORT)
 989                 return 1;
 990         return 0;
 991 }
 992
 993 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
 994 {
 995         struct tcphdr _tcph, *th;
 996
 997         th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
 998         if (th == NULL)
 999                 return 0;
1000         return th->rst;
1001 }
1002
1003 /* Handle response packets: rewrite addresses and send away...
1004  */
1005 static unsigned int
1006 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1007                 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph)
1008 {
1009         struct ip_vs_protocol *pp = pd->pp;
1010
1011         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Outgoing packet");
1012
1013         if (!skb_make_writable(skb, iph->len))
1014                 goto drop;
1015
1016         /* mangle the packet */
1017         if (pp->snat_handler && !pp->snat_handler(skb, pp, cp, iph))
1018                 goto drop;
1019
1020 #ifdef CONFIG_IP_VS_IPV6
1021         if (af == AF_INET6)
1022                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1023         else
1024 #endif
1025         {
1026                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1027                 ip_send_check(ip_hdr(skb));
1028         }
1029
1030         /*
1031          * nf_iterate does not expect change in the skb->dst->dev.
1032          * It looks like it is not fatal to enable this code for hooks
1033          * where our handlers are at the end of the chain list and
1034          * when all next handlers use skb->dst->dev and not outdev.
1035          * It will definitely route properly the inout NAT traffic
1036          * when multiple paths are used.
1037          */
1038
1039         /* For policy routing, packets originating from this
1040          * machine itself may be routed differently to packets
1041          * passing through.  We want this packet to be routed as
1042          * if it came from this machine itself.  So re-compute
1043          * the routing information.
1044          */
1045         if (ip_vs_route_me_harder(af, skb))
1046                 goto drop;
1047
1048         IP_VS_DBG_PKT(10, af, pp, skb, 0, "After SNAT");
1049
1050         ip_vs_out_stats(cp, skb);
1051         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1052         skb->ipvs_property = 1;
1053         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1054                 ip_vs_notrack(skb);
1055         else
1056                 ip_vs_update_conntrack(skb, cp, 0);
1057         ip_vs_conn_put(cp);
1058
1059         LeaveFunction(11);
1060         return NF_ACCEPT;
1061
1062 drop:
1063         ip_vs_conn_put(cp);
1064         kfree_skb(skb);
1065         LeaveFunction(11);
1066         return NF_STOLEN;
1067 }
1068
1069 /*
1070  *      Check if outgoing packet belongs to the established ip_vs_conn.
1071  */
1072 static unsigned int
1073 ip_vs_out(unsigned int hooknum, struct sk_buff *skb, int af)
1074 {
1075         struct net *net = NULL;
1076         struct ip_vs_iphdr iph;
1077         struct ip_vs_protocol *pp;
1078         struct ip_vs_proto_data *pd;
1079         struct ip_vs_conn *cp;
1080
1081         EnterFunction(11);
1082
1083         /* Already marked as IPVS request or reply? */
1084         if (skb->ipvs_property)
1085                 return NF_ACCEPT;
1086
1087         /* Bad... Do not break raw sockets */
1088         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1089                      af == AF_INET)) {
1090                 struct sock *sk = skb->sk;
1091                 struct inet_sock *inet = inet_sk(skb->sk);
1092
1093                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1094                         return NF_ACCEPT;
1095         }
1096
1097         if (unlikely(!skb_dst(skb)))
1098                 return NF_ACCEPT;
1099
1100         net = skb_net(skb);
1101         if (!net_ipvs(net)->enable)
1102                 return NF_ACCEPT;
1103
1104         ip_vs_fill_iph_skb(af, skb, &iph);
1105 #ifdef CONFIG_IP_VS_IPV6
1106         if (af == AF_INET6) {
1107                 if (!iph.fragoffs && skb_nfct_reasm(skb)) {
1108                         struct sk_buff *reasm = skb_nfct_reasm(skb);
1109                         /* Save fw mark for coming frags */
1110                         reasm->ipvs_property = 1;
1111                         reasm->mark = skb->mark;
1112                 }
1113                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1114                         int related;
1115                         int verdict = ip_vs_out_icmp_v6(skb, &related,
1116                                                         hooknum, &iph);
1117
1118                         if (related)
1119                                 return verdict;
1120                 }
1121         } else
1122 #endif
1123                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1124                         int related;
1125                         int verdict = ip_vs_out_icmp(skb, &related, hooknum);
1126
1127                         if (related)
1128                                 return verdict;
1129                 }
1130
1131         pd = ip_vs_proto_data_get(net, iph.protocol);
1132         if (unlikely(!pd))
1133                 return NF_ACCEPT;
1134         pp = pd->pp;
1135
1136         /* reassemble IP fragments */
1137 #ifdef CONFIG_IP_VS_IPV6
1138         if (af == AF_INET)
1139 #endif
1140                 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1141                         if (ip_vs_gather_frags(skb,
1142                                                ip_vs_defrag_user(hooknum)))
1143                                 return NF_STOLEN;
1144
1145                         ip_vs_fill_ip4hdr(skb_network_header(skb), &iph);
1146                 }
1147
1148         /*
1149          * Check if the packet belongs to an existing entry
1150          */
1151         cp = pp->conn_out_get(af, skb, &iph, 0);
1152
1153         if (likely(cp))
1154                 return handle_response(af, skb, pd, cp, &iph);
1155         if (sysctl_nat_icmp_send(net) &&
1156             (pp->protocol == IPPROTO_TCP ||
1157              pp->protocol == IPPROTO_UDP ||
1158              pp->protocol == IPPROTO_SCTP)) {
1159                 __be16 _ports[2], *pptr;
1160
1161                 pptr = frag_safe_skb_hp(skb, iph.len,
1162                                          sizeof(_ports), _ports, &iph);
1163                 if (pptr == NULL)
1164                         return NF_ACCEPT;       /* Not for me */
1165                 if (ip_vs_has_real_service(net, af, iph.protocol, &iph.saddr,
1166                                            pptr[0])) {
1167                         /*
1168                          * Notify the real server: there is no
1169                          * existing entry if it is not RST
1170                          * packet or not TCP packet.
1171                          */
1172                         if ((iph.protocol != IPPROTO_TCP &&
1173                              iph.protocol != IPPROTO_SCTP)
1174                              || ((iph.protocol == IPPROTO_TCP
1175                                   && !is_tcp_reset(skb, iph.len))
1176                                  || (iph.protocol == IPPROTO_SCTP
1177                                         && !is_sctp_abort(skb,
1178                                                 iph.len)))) {
1179 #ifdef CONFIG_IP_VS_IPV6
1180                                 if (af == AF_INET6) {
1181                                         if (!skb->dev)
1182                                                 skb->dev = net->loopback_dev;
1183                                         icmpv6_send(skb,
1184                                                     ICMPV6_DEST_UNREACH,
1185                                                     ICMPV6_PORT_UNREACH,
1186                                                     0);
1187                                 } else
1188 #endif
1189                                         icmp_send(skb,
1190                                                   ICMP_DEST_UNREACH,
1191                                                   ICMP_PORT_UNREACH, 0);
1192                                 return NF_DROP;
1193                         }
1194                 }
1195         }
1196         IP_VS_DBG_PKT(12, af, pp, skb, 0,
1197                       "ip_vs_out: packet continues traversal as normal");
1198         return NF_ACCEPT;
1199 }
1200
1201 /*
1202  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1203  *      used only for VS/NAT.
1204  *      Check if packet is reply for established ip_vs_conn.
1205  */
1206 static unsigned int
1207 ip_vs_reply4(unsigned int hooknum, struct sk_buff *skb,
1208              const struct net_device *in, const struct net_device *out,
1209              int (*okfn)(struct sk_buff *))
1210 {
1211         return ip_vs_out(hooknum, skb, AF_INET);
1212 }
1213
1214 /*
1215  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1216  *      Check if packet is reply for established ip_vs_conn.
1217  */
1218 static unsigned int
1219 ip_vs_local_reply4(unsigned int hooknum, struct sk_buff *skb,
1220                    const struct net_device *in, const struct net_device *out,
1221                    int (*okfn)(struct sk_buff *))
1222 {
1223         return ip_vs_out(hooknum, skb, AF_INET);
1224 }
1225
1226 #ifdef CONFIG_IP_VS_IPV6
1227
1228 /*
1229  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1230  *      used only for VS/NAT.
1231  *      Check if packet is reply for established ip_vs_conn.
1232  */
1233 static unsigned int
1234 ip_vs_reply6(unsigned int hooknum, struct sk_buff *skb,
1235              const struct net_device *in, const struct net_device *out,
1236              int (*okfn)(struct sk_buff *))
1237 {
1238         return ip_vs_out(hooknum, skb, AF_INET6);
1239 }
1240
1241 /*
1242  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1243  *      Check if packet is reply for established ip_vs_conn.
1244  */
1245 static unsigned int
1246 ip_vs_local_reply6(unsigned int hooknum, struct sk_buff *skb,
1247                    const struct net_device *in, const struct net_device *out,
1248                    int (*okfn)(struct sk_buff *))
1249 {
1250         return ip_vs_out(hooknum, skb, AF_INET6);
1251 }
1252
1253 #endif
1254
1255 /*
1256  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1257  *      Find any that might be relevant, check against existing connections,
1258  *      forward to the right destination host if relevant.
1259  *      Currently handles error types - unreachable, quench, ttl exceeded.
1260  */
1261 static int
1262 ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum)
1263 {
1264         struct net *net = NULL;
1265         struct iphdr *iph;
1266         struct icmphdr  _icmph, *ic;
1267         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1268         struct ip_vs_iphdr ciph;
1269         struct ip_vs_conn *cp;
1270         struct ip_vs_protocol *pp;
1271         struct ip_vs_proto_data *pd;
1272         unsigned int offset, offset2, ihl, verdict;
1273         bool ipip;
1274
1275         *related = 1;
1276
1277         /* reassemble IP fragments */
1278         if (ip_is_fragment(ip_hdr(skb))) {
1279                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
1280                         return NF_STOLEN;
1281         }
1282
1283         iph = ip_hdr(skb);
1284         offset = ihl = iph->ihl * 4;
1285         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1286         if (ic == NULL)
1287                 return NF_DROP;
1288
1289         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1290                   ic->type, ntohs(icmp_id(ic)),
1291                   &iph->saddr, &iph->daddr);
1292
1293         /*
1294          * Work through seeing if this is for us.
1295          * These checks are supposed to be in an order that means easy
1296          * things are checked first to speed up processing.... however
1297          * this means that some packets will manage to get a long way
1298          * down this stack and then be rejected, but that's life.
1299          */
1300         if ((ic->type != ICMP_DEST_UNREACH) &&
1301             (ic->type != ICMP_SOURCE_QUENCH) &&
1302             (ic->type != ICMP_TIME_EXCEEDED)) {
1303                 *related = 0;
1304                 return NF_ACCEPT;
1305         }
1306
1307         /* Now find the contained IP header */
1308         offset += sizeof(_icmph);
1309         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1310         if (cih == NULL)
1311                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1312
1313         net = skb_net(skb);
1314
1315         /* Special case for errors for IPIP packets */
1316         ipip = false;
1317         if (cih->protocol == IPPROTO_IPIP) {
1318                 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1319                         return NF_ACCEPT;
1320                 /* Error for our IPIP must arrive at LOCAL_IN */
1321                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1322                         return NF_ACCEPT;
1323                 offset += cih->ihl * 4;
1324                 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1325                 if (cih == NULL)
1326                         return NF_ACCEPT; /* The packet looks wrong, ignore */
1327                 ipip = true;
1328         }
1329
1330         pd = ip_vs_proto_data_get(net, cih->protocol);
1331         if (!pd)
1332                 return NF_ACCEPT;
1333         pp = pd->pp;
1334
1335         /* Is the embedded protocol header present? */
1336         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1337                      pp->dont_defrag))
1338                 return NF_ACCEPT;
1339
1340         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1341                       "Checking incoming ICMP for");
1342
1343         offset2 = offset;
1344         ip_vs_fill_ip4hdr(cih, &ciph);
1345         ciph.len += offset;
1346         offset = ciph.len;
1347         /* The embedded headers contain source and dest in reverse order.
1348          * For IPIP this is error for request, not for reply.
1349          */
1350         cp = pp->conn_in_get(AF_INET, skb, &ciph, ipip ? 0 : 1);
1351         if (!cp)
1352                 return NF_ACCEPT;
1353
1354         verdict = NF_DROP;
1355
1356         /* Ensure the checksum is correct */
1357         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1358                 /* Failed checksum! */
1359                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1360                           &iph->saddr);
1361                 goto out;
1362         }
1363
1364         if (ipip) {
1365                 __be32 info = ic->un.gateway;
1366
1367                 /* Update the MTU */
1368                 if (ic->type == ICMP_DEST_UNREACH &&
1369                     ic->code == ICMP_FRAG_NEEDED) {
1370                         struct ip_vs_dest *dest = cp->dest;
1371                         u32 mtu = ntohs(ic->un.frag.mtu);
1372
1373                         /* Strip outer IP and ICMP, go to IPIP header */
1374                         __skb_pull(skb, ihl + sizeof(_icmph));
1375                         offset2 -= ihl + sizeof(_icmph);
1376                         skb_reset_network_header(skb);
1377                         IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1378                                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1379                         ipv4_update_pmtu(skb, dev_net(skb->dev),
1380                                          mtu, 0, 0, 0, 0);
1381                         /* Client uses PMTUD? */
1382                         if (!(cih->frag_off & htons(IP_DF)))
1383                                 goto ignore_ipip;
1384                         /* Prefer the resulting PMTU */
1385                         if (dest) {
1386                                 struct ip_vs_dest_dst *dest_dst;
1387
1388                                 rcu_read_lock();
1389                                 dest_dst = rcu_dereference(dest->dest_dst);
1390                                 if (dest_dst)
1391                                         mtu = dst_mtu(dest_dst->dst_cache);
1392                                 rcu_read_unlock();
1393                         }
1394                         if (mtu > 68 + sizeof(struct iphdr))
1395                                 mtu -= sizeof(struct iphdr);
1396                         info = htonl(mtu);
1397                 }
1398                 /* Strip outer IP, ICMP and IPIP, go to IP header of
1399                  * original request.
1400                  */
1401                 __skb_pull(skb, offset2);
1402                 skb_reset_network_header(skb);
1403                 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1404                         &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1405                         ic->type, ic->code, ntohl(info));
1406                 icmp_send(skb, ic->type, ic->code, info);
1407                 /* ICMP can be shorter but anyways, account it */
1408                 ip_vs_out_stats(cp, skb);
1409
1410 ignore_ipip:
1411                 consume_skb(skb);
1412                 verdict = NF_STOLEN;
1413                 goto out;
1414         }
1415
1416         /* do the statistics and put it back */
1417         ip_vs_in_stats(cp, skb);
1418         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol)
1419                 offset += 2 * sizeof(__u16);
1420         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1421
1422 out:
1423         __ip_vs_conn_put(cp);
1424
1425         return verdict;
1426 }
1427
1428 #ifdef CONFIG_IP_VS_IPV6
1429 static int ip_vs_in_icmp_v6(struct sk_buff *skb, int *related,
1430                             unsigned int hooknum, struct ip_vs_iphdr *iph)
1431 {
1432         struct net *net = NULL;
1433         struct ipv6hdr _ip6h, *ip6h;
1434         struct icmp6hdr _icmph, *ic;
1435         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1436         struct ip_vs_conn *cp;
1437         struct ip_vs_protocol *pp;
1438         struct ip_vs_proto_data *pd;
1439         unsigned int offs_ciph, writable, verdict;
1440
1441         *related = 1;
1442
1443         ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph, iph);
1444         if (ic == NULL)
1445                 return NF_DROP;
1446
1447         /*
1448          * Work through seeing if this is for us.
1449          * These checks are supposed to be in an order that means easy
1450          * things are checked first to speed up processing.... however
1451          * this means that some packets will manage to get a long way
1452          * down this stack and then be rejected, but that's life.
1453          */
1454         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1455                 *related = 0;
1456                 return NF_ACCEPT;
1457         }
1458         /* Fragment header that is before ICMP header tells us that:
1459          * it's not an error message since they can't be fragmented.
1460          */
1461         if (iph->flags & IP6_FH_F_FRAG)
1462                 return NF_DROP;
1463
1464         IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1465                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1466                   &iph->saddr, &iph->daddr);
1467
1468         /* Now find the contained IP header */
1469         ciph.len = iph->len + sizeof(_icmph);
1470         offs_ciph = ciph.len; /* Save ip header offset */
1471         ip6h = skb_header_pointer(skb, ciph.len, sizeof(_ip6h), &_ip6h);
1472         if (ip6h == NULL)
1473                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1474         ciph.saddr.in6 = ip6h->saddr; /* conn_in_get() handles reverse order */
1475         ciph.daddr.in6 = ip6h->daddr;
1476         /* skip possible IPv6 exthdrs of contained IPv6 packet */
1477         ciph.protocol = ipv6_find_hdr(skb, &ciph.len, -1, &ciph.fragoffs, NULL);
1478         if (ciph.protocol < 0)
1479                 return NF_ACCEPT; /* Contained IPv6 hdr looks wrong, ignore */
1480
1481         net = skb_net(skb);
1482         pd = ip_vs_proto_data_get(net, ciph.protocol);
1483         if (!pd)
1484                 return NF_ACCEPT;
1485         pp = pd->pp;
1486
1487         /* Cannot handle fragmented embedded protocol */
1488         if (ciph.fragoffs)
1489                 return NF_ACCEPT;
1490
1491         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offs_ciph,
1492                       "Checking incoming ICMPv6 for");
1493
1494         /* The embedded headers contain source and dest in reverse order
1495          * if not from localhost
1496          */
1497         cp = pp->conn_in_get(AF_INET6, skb, &ciph,
1498                              (hooknum == NF_INET_LOCAL_OUT) ? 0 : 1);
1499
1500         if (!cp)
1501                 return NF_ACCEPT;
1502         /* VS/TUN, VS/DR and LOCALNODE just let it go */
1503         if ((hooknum == NF_INET_LOCAL_OUT) &&
1504             (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1505                 __ip_vs_conn_put(cp);
1506                 return NF_ACCEPT;
1507         }
1508
1509         /* do the statistics and put it back */
1510         ip_vs_in_stats(cp, skb);
1511
1512         /* Need to mangle contained IPv6 header in ICMPv6 packet */
1513         writable = ciph.len;
1514         if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1515             IPPROTO_SCTP == ciph.protocol)
1516                 writable += 2 * sizeof(__u16); /* Also mangle ports */
1517
1518         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, writable, hooknum, &ciph);
1519
1520         __ip_vs_conn_put(cp);
1521
1522         return verdict;
1523 }
1524 #endif
1525
1526
1527 /*
1528  *      Check if it's for virtual services, look it up,
1529  *      and send it on its way...
1530  */
1531 static unsigned int
1532 ip_vs_in(unsigned int hooknum, struct sk_buff *skb, int af)
1533 {
1534         struct net *net;
1535         struct ip_vs_iphdr iph;
1536         struct ip_vs_protocol *pp;
1537         struct ip_vs_proto_data *pd;
1538         struct ip_vs_conn *cp;
1539         int ret, pkts;
1540         struct netns_ipvs *ipvs;
1541
1542         /* Already marked as IPVS request or reply? */
1543         if (skb->ipvs_property)
1544                 return NF_ACCEPT;
1545
1546         /*
1547          *      Big tappo:
1548          *      - remote client: only PACKET_HOST
1549          *      - route: used for struct net when skb->dev is unset
1550          */
1551         if (unlikely((skb->pkt_type != PACKET_HOST &&
1552                       hooknum != NF_INET_LOCAL_OUT) ||
1553                      !skb_dst(skb))) {
1554                 ip_vs_fill_iph_skb(af, skb, &iph);
1555                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1556                               " ignored in hook %u\n",
1557                               skb->pkt_type, iph.protocol,
1558                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1559                 return NF_ACCEPT;
1560         }
1561         /* ipvs enabled in this netns ? */
1562         net = skb_net(skb);
1563         ipvs = net_ipvs(net);
1564         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1565                 return NF_ACCEPT;
1566
1567         ip_vs_fill_iph_skb(af, skb, &iph);
1568
1569         /* Bad... Do not break raw sockets */
1570         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1571                      af == AF_INET)) {
1572                 struct sock *sk = skb->sk;
1573                 struct inet_sock *inet = inet_sk(skb->sk);
1574
1575                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1576                         return NF_ACCEPT;
1577         }
1578
1579 #ifdef CONFIG_IP_VS_IPV6
1580         if (af == AF_INET6) {
1581                 if (!iph.fragoffs && skb_nfct_reasm(skb)) {
1582                         struct sk_buff *reasm = skb_nfct_reasm(skb);
1583                         /* Save fw mark for coming frags. */
1584                         reasm->ipvs_property = 1;
1585                         reasm->mark = skb->mark;
1586                 }
1587                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1588                         int related;
1589                         int verdict = ip_vs_in_icmp_v6(skb, &related, hooknum,
1590                                                        &iph);
1591
1592                         if (related)
1593                                 return verdict;
1594                 }
1595         } else
1596 #endif
1597                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1598                         int related;
1599                         int verdict = ip_vs_in_icmp(skb, &related, hooknum);
1600
1601                         if (related)
1602                                 return verdict;
1603                 }
1604
1605         /* Protocol supported? */
1606         pd = ip_vs_proto_data_get(net, iph.protocol);
1607         if (unlikely(!pd))
1608                 return NF_ACCEPT;
1609         pp = pd->pp;
1610         /*
1611          * Check if the packet belongs to an existing connection entry
1612          */
1613         cp = pp->conn_in_get(af, skb, &iph, 0);
1614         if (unlikely(!cp) && !iph.fragoffs) {
1615                 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1616                  * replayed fragment zero will already have created the cp
1617                  */
1618                 int v;
1619
1620                 /* Schedule and create new connection entry into &cp */
1621                 if (!pp->conn_schedule(af, skb, pd, &v, &cp, &iph))
1622                         return v;
1623         }
1624
1625         if (unlikely(!cp)) {
1626                 /* sorry, all this trouble for a no-hit :) */
1627                 IP_VS_DBG_PKT(12, af, pp, skb, 0,
1628                               "ip_vs_in: packet continues traversal as normal");
1629                 if (iph.fragoffs && !skb_nfct_reasm(skb)) {
1630                         /* Fragment that couldn't be mapped to a conn entry
1631                          * and don't have any pointer to a reasm skb
1632                          * is missing module nf_defrag_ipv6
1633                          */
1634                         IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1635                         IP_VS_DBG_PKT(7, af, pp, skb, 0, "unhandled fragment");
1636                 }
1637                 return NF_ACCEPT;
1638         }
1639
1640         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Incoming packet");
1641         /* Check the server status */
1642         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1643                 /* the destination server is not available */
1644
1645                 if (sysctl_expire_nodest_conn(ipvs)) {
1646                         /* try to expire the connection immediately */
1647                         ip_vs_conn_expire_now(cp);
1648                 }
1649                 /* don't restart its timer, and silently
1650                    drop the packet. */
1651                 __ip_vs_conn_put(cp);
1652                 return NF_DROP;
1653         }
1654
1655         ip_vs_in_stats(cp, skb);
1656         ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1657         if (cp->packet_xmit)
1658                 ret = cp->packet_xmit(skb, cp, pp, &iph);
1659                 /* do not touch skb anymore */
1660         else {
1661                 IP_VS_DBG_RL("warning: packet_xmit is null");
1662                 ret = NF_ACCEPT;
1663         }
1664
1665         /* Increase its packet counter and check if it is needed
1666          * to be synchronized
1667          *
1668          * Sync connection if it is about to close to
1669          * encorage the standby servers to update the connections timeout
1670          *
1671          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1672          */
1673
1674         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
1675                 pkts = sysctl_sync_threshold(ipvs);
1676         else
1677                 pkts = atomic_add_return(1, &cp->in_pkts);
1678
1679         if (ipvs->sync_state & IP_VS_STATE_MASTER)
1680                 ip_vs_sync_conn(net, cp, pkts);
1681
1682         ip_vs_conn_put(cp);
1683         return ret;
1684 }
1685
1686 /*
1687  *      AF_INET handler in NF_INET_LOCAL_IN chain
1688  *      Schedule and forward packets from remote clients
1689  */
1690 static unsigned int
1691 ip_vs_remote_request4(unsigned int hooknum, struct sk_buff *skb,
1692                       const struct net_device *in,
1693                       const struct net_device *out,
1694                       int (*okfn)(struct sk_buff *))
1695 {
1696         return ip_vs_in(hooknum, skb, AF_INET);
1697 }
1698
1699 /*
1700  *      AF_INET handler in NF_INET_LOCAL_OUT chain
1701  *      Schedule and forward packets from local clients
1702  */
1703 static unsigned int
1704 ip_vs_local_request4(unsigned int hooknum, struct sk_buff *skb,
1705                      const struct net_device *in, const struct net_device *out,
1706                      int (*okfn)(struct sk_buff *))
1707 {
1708         return ip_vs_in(hooknum, skb, AF_INET);
1709 }
1710
1711 #ifdef CONFIG_IP_VS_IPV6
1712
1713 /*
1714  * AF_INET6 fragment handling
1715  * Copy info from first fragment, to the rest of them.
1716  */
1717 static unsigned int
1718 ip_vs_preroute_frag6(unsigned int hooknum, struct sk_buff *skb,
1719                      const struct net_device *in,
1720                      const struct net_device *out,
1721                      int (*okfn)(struct sk_buff *))
1722 {
1723         struct sk_buff *reasm = skb_nfct_reasm(skb);
1724         struct net *net;
1725
1726         /* Skip if not a "replay" from nf_ct_frag6_output or first fragment.
1727          * ipvs_property is set when checking first fragment
1728          * in ip_vs_in() and ip_vs_out().
1729          */
1730         if (reasm)
1731                 IP_VS_DBG(2, "Fragment recv prop:%d\n", reasm->ipvs_property);
1732         if (!reasm || !reasm->ipvs_property)
1733                 return NF_ACCEPT;
1734
1735         net = skb_net(skb);
1736         if (!net_ipvs(net)->enable)
1737                 return NF_ACCEPT;
1738
1739         /* Copy stored fw mark, saved in ip_vs_{in,out} */
1740         skb->mark = reasm->mark;
1741
1742         return NF_ACCEPT;
1743 }
1744
1745 /*
1746  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
1747  *      Schedule and forward packets from remote clients
1748  */
1749 static unsigned int
1750 ip_vs_remote_request6(unsigned int hooknum, struct sk_buff *skb,
1751                       const struct net_device *in,
1752                       const struct net_device *out,
1753                       int (*okfn)(struct sk_buff *))
1754 {
1755         return ip_vs_in(hooknum, skb, AF_INET6);
1756 }
1757
1758 /*
1759  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
1760  *      Schedule and forward packets from local clients
1761  */
1762 static unsigned int
1763 ip_vs_local_request6(unsigned int hooknum, struct sk_buff *skb,
1764                      const struct net_device *in, const struct net_device *out,
1765                      int (*okfn)(struct sk_buff *))
1766 {
1767         return ip_vs_in(hooknum, skb, AF_INET6);
1768 }
1769
1770 #endif
1771
1772
1773 /*
1774  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1775  *      related packets destined for 0.0.0.0/0.
1776  *      When fwmark-based virtual service is used, such as transparent
1777  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
1778  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1779  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1780  *      and send them to ip_vs_in_icmp.
1781  */
1782 static unsigned int
1783 ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff *skb,
1784                    const struct net_device *in, const struct net_device *out,
1785                    int (*okfn)(struct sk_buff *))
1786 {
1787         int r;
1788         struct net *net;
1789         struct netns_ipvs *ipvs;
1790
1791         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1792                 return NF_ACCEPT;
1793
1794         /* ipvs enabled in this netns ? */
1795         net = skb_net(skb);
1796         ipvs = net_ipvs(net);
1797         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1798                 return NF_ACCEPT;
1799
1800         return ip_vs_in_icmp(skb, &r, hooknum);
1801 }
1802
1803 #ifdef CONFIG_IP_VS_IPV6
1804 static unsigned int
1805 ip_vs_forward_icmp_v6(unsigned int hooknum, struct sk_buff *skb,
1806                       const struct net_device *in, const struct net_device *out,
1807                       int (*okfn)(struct sk_buff *))
1808 {
1809         int r;
1810         struct net *net;
1811         struct netns_ipvs *ipvs;
1812         struct ip_vs_iphdr iphdr;
1813
1814         ip_vs_fill_iph_skb(AF_INET6, skb, &iphdr);
1815         if (iphdr.protocol != IPPROTO_ICMPV6)
1816                 return NF_ACCEPT;
1817
1818         /* ipvs enabled in this netns ? */
1819         net = skb_net(skb);
1820         ipvs = net_ipvs(net);
1821         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1822                 return NF_ACCEPT;
1823
1824         return ip_vs_in_icmp_v6(skb, &r, hooknum, &iphdr);
1825 }
1826 #endif
1827
1828
1829 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
1830         /* After packet filtering, change source only for VS/NAT */
1831         {
1832                 .hook           = ip_vs_reply4,
1833                 .owner          = THIS_MODULE,
1834                 .pf             = NFPROTO_IPV4,
1835                 .hooknum        = NF_INET_LOCAL_IN,
1836                 .priority       = NF_IP_PRI_NAT_SRC - 2,
1837         },
1838         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1839          * or VS/NAT(change destination), so that filtering rules can be
1840          * applied to IPVS. */
1841         {
1842                 .hook           = ip_vs_remote_request4,
1843                 .owner          = THIS_MODULE,
1844                 .pf             = NFPROTO_IPV4,
1845                 .hooknum        = NF_INET_LOCAL_IN,
1846                 .priority       = NF_IP_PRI_NAT_SRC - 1,
1847         },
1848         /* Before ip_vs_in, change source only for VS/NAT */
1849         {
1850                 .hook           = ip_vs_local_reply4,
1851                 .owner          = THIS_MODULE,
1852                 .pf             = NFPROTO_IPV4,
1853                 .hooknum        = NF_INET_LOCAL_OUT,
1854                 .priority       = NF_IP_PRI_NAT_DST + 1,
1855         },
1856         /* After mangle, schedule and forward local requests */
1857         {
1858                 .hook           = ip_vs_local_request4,
1859                 .owner          = THIS_MODULE,
1860                 .pf             = NFPROTO_IPV4,
1861                 .hooknum        = NF_INET_LOCAL_OUT,
1862                 .priority       = NF_IP_PRI_NAT_DST + 2,
1863         },
1864         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1865          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1866         {
1867                 .hook           = ip_vs_forward_icmp,
1868                 .owner          = THIS_MODULE,
1869                 .pf             = NFPROTO_IPV4,
1870                 .hooknum        = NF_INET_FORWARD,
1871                 .priority       = 99,
1872         },
1873         /* After packet filtering, change source only for VS/NAT */
1874         {
1875                 .hook           = ip_vs_reply4,
1876                 .owner          = THIS_MODULE,
1877                 .pf             = NFPROTO_IPV4,
1878                 .hooknum        = NF_INET_FORWARD,
1879                 .priority       = 100,
1880         },
1881 #ifdef CONFIG_IP_VS_IPV6
1882         /* After mangle & nat fetch 2:nd fragment and following */
1883         {
1884                 .hook           = ip_vs_preroute_frag6,
1885                 .owner          = THIS_MODULE,
1886                 .pf             = NFPROTO_IPV6,
1887                 .hooknum        = NF_INET_PRE_ROUTING,
1888                 .priority       = NF_IP6_PRI_NAT_DST + 1,
1889         },
1890         /* After packet filtering, change source only for VS/NAT */
1891         {
1892                 .hook           = ip_vs_reply6,
1893                 .owner          = THIS_MODULE,
1894                 .pf             = NFPROTO_IPV6,
1895                 .hooknum        = NF_INET_LOCAL_IN,
1896                 .priority       = NF_IP6_PRI_NAT_SRC - 2,
1897         },
1898         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1899          * or VS/NAT(change destination), so that filtering rules can be
1900          * applied to IPVS. */
1901         {
1902                 .hook           = ip_vs_remote_request6,
1903                 .owner          = THIS_MODULE,
1904                 .pf             = NFPROTO_IPV6,
1905                 .hooknum        = NF_INET_LOCAL_IN,
1906                 .priority       = NF_IP6_PRI_NAT_SRC - 1,
1907         },
1908         /* Before ip_vs_in, change source only for VS/NAT */
1909         {
1910                 .hook           = ip_vs_local_reply6,
1911                 .owner          = THIS_MODULE,
1912                 .pf             = NFPROTO_IPV4,
1913                 .hooknum        = NF_INET_LOCAL_OUT,
1914                 .priority       = NF_IP6_PRI_NAT_DST + 1,
1915         },
1916         /* After mangle, schedule and forward local requests */
1917         {
1918                 .hook           = ip_vs_local_request6,
1919                 .owner          = THIS_MODULE,
1920                 .pf             = NFPROTO_IPV6,
1921                 .hooknum        = NF_INET_LOCAL_OUT,
1922                 .priority       = NF_IP6_PRI_NAT_DST + 2,
1923         },
1924         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1925          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1926         {
1927                 .hook           = ip_vs_forward_icmp_v6,
1928                 .owner          = THIS_MODULE,
1929                 .pf             = NFPROTO_IPV6,
1930                 .hooknum        = NF_INET_FORWARD,
1931                 .priority       = 99,
1932         },
1933         /* After packet filtering, change source only for VS/NAT */
1934         {
1935                 .hook           = ip_vs_reply6,
1936                 .owner          = THIS_MODULE,
1937                 .pf             = NFPROTO_IPV6,
1938                 .hooknum        = NF_INET_FORWARD,
1939                 .priority       = 100,
1940         },
1941 #endif
1942 };
1943 /*
1944  *      Initialize IP Virtual Server netns mem.
1945  */
1946 static int __net_init __ip_vs_init(struct net *net)
1947 {
1948         struct netns_ipvs *ipvs;
1949
1950         ipvs = net_generic(net, ip_vs_net_id);
1951         if (ipvs == NULL)
1952                 return -ENOMEM;
1953
1954         /* Hold the beast until a service is registerd */
1955         ipvs->enable = 0;
1956         ipvs->net = net;
1957         /* Counters used for creating unique names */
1958         ipvs->gen = atomic_read(&ipvs_netns_cnt);
1959         atomic_inc(&ipvs_netns_cnt);
1960         net->ipvs = ipvs;
1961
1962         if (ip_vs_estimator_net_init(net) < 0)
1963                 goto estimator_fail;
1964
1965         if (ip_vs_control_net_init(net) < 0)
1966                 goto control_fail;
1967
1968         if (ip_vs_protocol_net_init(net) < 0)
1969                 goto protocol_fail;
1970
1971         if (ip_vs_app_net_init(net) < 0)
1972                 goto app_fail;
1973
1974         if (ip_vs_conn_net_init(net) < 0)
1975                 goto conn_fail;
1976
1977         if (ip_vs_sync_net_init(net) < 0)
1978                 goto sync_fail;
1979
1980         printk(KERN_INFO "IPVS: Creating netns size=%zu id=%d\n",
1981                          sizeof(struct netns_ipvs), ipvs->gen);
1982         return 0;
1983 /*
1984  * Error handling
1985  */
1986
1987 sync_fail:
1988         ip_vs_conn_net_cleanup(net);
1989 conn_fail:
1990         ip_vs_app_net_cleanup(net);
1991 app_fail:
1992         ip_vs_protocol_net_cleanup(net);
1993 protocol_fail:
1994         ip_vs_control_net_cleanup(net);
1995 control_fail:
1996         ip_vs_estimator_net_cleanup(net);
1997 estimator_fail:
1998         net->ipvs = NULL;
1999         return -ENOMEM;
2000 }
2001
2002 static void __net_exit __ip_vs_cleanup(struct net *net)
2003 {
2004         ip_vs_service_net_cleanup(net); /* ip_vs_flush() with locks */
2005         ip_vs_conn_net_cleanup(net);
2006         ip_vs_app_net_cleanup(net);
2007         ip_vs_protocol_net_cleanup(net);
2008         ip_vs_control_net_cleanup(net);
2009         ip_vs_estimator_net_cleanup(net);
2010         IP_VS_DBG(2, "ipvs netns %d released\n", net_ipvs(net)->gen);
2011         net->ipvs = NULL;
2012 }
2013
2014 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
2015 {
2016         EnterFunction(2);
2017         net_ipvs(net)->enable = 0;      /* Disable packet reception */
2018         smp_wmb();
2019         ip_vs_sync_net_cleanup(net);
2020         LeaveFunction(2);
2021 }
2022
2023 static struct pernet_operations ipvs_core_ops = {
2024         .init = __ip_vs_init,
2025         .exit = __ip_vs_cleanup,
2026         .id   = &ip_vs_net_id,
2027         .size = sizeof(struct netns_ipvs),
2028 };
2029
2030 static struct pernet_operations ipvs_core_dev_ops = {
2031         .exit = __ip_vs_dev_cleanup,
2032 };
2033
2034 /*
2035  *      Initialize IP Virtual Server
2036  */
2037 static int __init ip_vs_init(void)
2038 {
2039         int ret;
2040
2041         ret = ip_vs_control_init();
2042         if (ret < 0) {
2043                 pr_err("can't setup control.\n");
2044                 goto exit;
2045         }
2046
2047         ip_vs_protocol_init();
2048
2049         ret = ip_vs_conn_init();
2050         if (ret < 0) {
2051                 pr_err("can't setup connection table.\n");
2052                 goto cleanup_protocol;
2053         }
2054
2055         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
2056         if (ret < 0)
2057                 goto cleanup_conn;
2058
2059         ret = register_pernet_device(&ipvs_core_dev_ops);
2060         if (ret < 0)
2061                 goto cleanup_sub;
2062
2063         ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2064         if (ret < 0) {
2065                 pr_err("can't register hooks.\n");
2066                 goto cleanup_dev;
2067         }
2068
2069         ret = ip_vs_register_nl_ioctl();
2070         if (ret < 0) {
2071                 pr_err("can't register netlink/ioctl.\n");
2072                 goto cleanup_hooks;
2073         }
2074
2075         pr_info("ipvs loaded.\n");
2076
2077         return ret;
2078
2079 cleanup_hooks:
2080         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2081 cleanup_dev:
2082         unregister_pernet_device(&ipvs_core_dev_ops);
2083 cleanup_sub:
2084         unregister_pernet_subsys(&ipvs_core_ops);
2085 cleanup_conn:
2086         ip_vs_conn_cleanup();
2087 cleanup_protocol:
2088         ip_vs_protocol_cleanup();
2089         ip_vs_control_cleanup();
2090 exit:
2091         return ret;
2092 }
2093
2094 static void __exit ip_vs_cleanup(void)
2095 {
2096         ip_vs_unregister_nl_ioctl();
2097         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2098         unregister_pernet_device(&ipvs_core_dev_ops);
2099         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
2100         ip_vs_conn_cleanup();
2101         ip_vs_protocol_cleanup();
2102         ip_vs_control_cleanup();
2103         pr_info("ipvs unloaded.\n");
2104 }
2105
2106 module_init(ip_vs_init);
2107 module_exit(ip_vs_cleanup);
2108 MODULE_LICENSE("GPL");