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,
 239                                  (__force __u32) svc->netmask);
 240         else
 241 #endif
 242                 snet.ip = iph->saddr.ip & svc->netmask;
 243
 244         IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
 245                       "mnet %s\n",
 246                       IP_VS_DBG_ADDR(svc->af, &iph->saddr), ntohs(src_port),
 247                       IP_VS_DBG_ADDR(svc->af, &iph->daddr), ntohs(dst_port),
 248                       IP_VS_DBG_ADDR(svc->af, &snet));
 249
 250         /*
 251          * As far as we know, FTP is a very complicated network protocol, and
 252          * it uses control connection and data connections. For active FTP,
 253          * FTP server initialize data connection to the client, its source port
 254          * is often 20. For passive FTP, FTP server tells the clients the port
 255          * that it passively listens to,  and the client issues the data
 256          * connection. In the tunneling or direct routing mode, the load
 257          * balancer is on the client-to-server half of connection, the port
 258          * number is unknown to the load balancer. So, a conn template like
 259          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
 260          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
 261          * is created for other persistent services.
 262          */
 263         {
 264                 int protocol = iph->protocol;
 265                 const union nf_inet_addr *vaddr = &iph->daddr;
 266                 __be16 vport = 0;
 267
 268                 if (dst_port == svc->port) {
 269                         /* non-FTP template:
 270                          * <protocol, caddr, 0, vaddr, vport, daddr, dport>
 271                          * FTP template:
 272                          * <protocol, caddr, 0, vaddr, 0, daddr, 0>
 273                          */
 274                         if (svc->port != FTPPORT)
 275                                 vport = dst_port;
 276                 } else {
 277                         /* Note: persistent fwmark-based services and
 278                          * persistent port zero service are handled here.
 279                          * fwmark template:
 280                          * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
 281                          * port zero template:
 282                          * <protocol,caddr,0,vaddr,0,daddr,0>
 283                          */
 284                         if (svc->fwmark) {
 285                                 protocol = IPPROTO_IP;
 286                                 vaddr = &fwmark;
 287                         }
 288                 }
 289                 /* return *ignored = -1 so NF_DROP can be used */
 290                 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
 291                                                   vaddr, vport, &param) < 0) {
 292                         *ignored = -1;
 293                         return NULL;
 294                 }
 295         }
 296
 297         /* Check if a template already exists */
 298         ct = ip_vs_ct_in_get(&param);
 299         if (!ct || !ip_vs_check_template(ct)) {
 300                 struct ip_vs_scheduler *sched;
 301
 302                 /*
 303                  * No template found or the dest of the connection
 304                  * template is not available.
 305                  * return *ignored=0 i.e. ICMP and NF_DROP
 306                  */
 307                 sched = rcu_dereference(svc->scheduler);
 308                 dest = sched->schedule(svc, skb, iph);
 309                 if (!dest) {
 310                         IP_VS_DBG(1, "p-schedule: no dest found.\n");
 311                         kfree(param.pe_data);
 312                         *ignored = 0;
 313                         return NULL;
 314                 }
 315
 316                 if (dst_port == svc->port && svc->port != FTPPORT)
 317                         dport = dest->port;
 318
 319                 /* Create a template
 320                  * This adds param.pe_data to the template,
 321                  * and thus param.pe_data will be destroyed
 322                  * when the template expires */
 323                 ct = ip_vs_conn_new(&param, &dest->addr, dport,
 324                                     IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
 325                 if (ct == NULL) {
 326                         kfree(param.pe_data);
 327                         *ignored = -1;
 328                         return NULL;
 329                 }
 330
 331                 ct->timeout = svc->timeout;
 332         } else {
 333                 /* set destination with the found template */
 334                 dest = ct->dest;
 335                 kfree(param.pe_data);
 336         }
 337
 338         dport = dst_port;
 339         if (dport == svc->port && dest->port)
 340                 dport = dest->port;
 341
 342         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 343                  && iph->protocol == IPPROTO_UDP) ?
 344                 IP_VS_CONN_F_ONE_PACKET : 0;
 345
 346         /*
 347          *    Create a new connection according to the template
 348          */
 349         ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol, &iph->saddr,
 350                               src_port, &iph->daddr, dst_port, &param);
 351
 352         cp = ip_vs_conn_new(&param, &dest->addr, dport, flags, dest, skb->mark);
 353         if (cp == NULL) {
 354                 ip_vs_conn_put(ct);
 355                 *ignored = -1;
 356                 return NULL;
 357         }
 358
 359         /*
 360          *    Add its control
 361          */
 362         ip_vs_control_add(cp, ct);
 363         ip_vs_conn_put(ct);
 364
 365         ip_vs_conn_stats(cp, svc);
 366         return cp;
 367 }
 368
 369
 370 /*
 371  *  IPVS main scheduling function
 372  *  It selects a server according to the virtual service, and
 373  *  creates a connection entry.
 374  *  Protocols supported: TCP, UDP
 375  *
 376  *  Usage of *ignored
 377  *
 378  * 1 :   protocol tried to schedule (eg. on SYN), found svc but the
 379  *       svc/scheduler decides that this packet should be accepted with
 380  *       NF_ACCEPT because it must not be scheduled.
 381  *
 382  * 0 :   scheduler can not find destination, so try bypass or
 383  *       return ICMP and then NF_DROP (ip_vs_leave).
 384  *
 385  * -1 :  scheduler tried to schedule but fatal error occurred, eg.
 386  *       ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
 387  *       failure such as missing Call-ID, ENOMEM on skb_linearize
 388  *       or pe_data. In this case we should return NF_DROP without
 389  *       any attempts to send ICMP with ip_vs_leave.
 390  */
 391 struct ip_vs_conn *
 392 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
 393                struct ip_vs_proto_data *pd, int *ignored,
 394                struct ip_vs_iphdr *iph)
 395 {
 396         struct ip_vs_protocol *pp = pd->pp;
 397         struct ip_vs_conn *cp = NULL;
 398         struct ip_vs_scheduler *sched;
 399         struct ip_vs_dest *dest;
 400         __be16 _ports[2], *pptr;
 401         unsigned int flags;
 402
 403         *ignored = 1;
 404         /*
 405          * IPv6 frags, only the first hit here.
 406          */
 407         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 408         if (pptr == NULL)
 409                 return NULL;
 410
 411         /*
 412          * FTPDATA needs this check when using local real server.
 413          * Never schedule Active FTPDATA connections from real server.
 414          * For LVS-NAT they must be already created. For other methods
 415          * with persistence the connection is created on SYN+ACK.
 416          */
 417         if (pptr[0] == FTPDATA) {
 418                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 419                               "Not scheduling FTPDATA");
 420                 return NULL;
 421         }
 422
 423         /*
 424          *    Do not schedule replies from local real server.
 425          */
 426         if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
 427             (cp = pp->conn_in_get(svc->af, skb, iph, 1))) {
 428                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 429                               "Not scheduling reply for existing connection");
 430                 __ip_vs_conn_put(cp);
 431                 return NULL;
 432         }
 433
 434         /*
 435          *    Persistent service
 436          */
 437         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 438                 return ip_vs_sched_persist(svc, skb, pptr[0], pptr[1], ignored,
 439                                            iph);
 440
 441         *ignored = 0;
 442
 443         /*
 444          *    Non-persistent service
 445          */
 446         if (!svc->fwmark && pptr[1] != svc->port) {
 447                 if (!svc->port)
 448                         pr_err("Schedule: port zero only supported "
 449                                "in persistent services, "
 450                                "check your ipvs configuration\n");
 451                 return NULL;
 452         }
 453
 454         sched = rcu_dereference(svc->scheduler);
 455         dest = sched->schedule(svc, skb, iph);
 456         if (dest == NULL) {
 457                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 458                 return NULL;
 459         }
 460
 461         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 462                  && iph->protocol == IPPROTO_UDP) ?
 463                 IP_VS_CONN_F_ONE_PACKET : 0;
 464
 465         /*
 466          *    Create a connection entry.
 467          */
 468         {
 469                 struct ip_vs_conn_param p;
 470
 471                 ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol,
 472                                       &iph->saddr, pptr[0], &iph->daddr,
 473                                       pptr[1], &p);
 474                 cp = ip_vs_conn_new(&p, &dest->addr,
 475                                     dest->port ? dest->port : pptr[1],
 476                                     flags, dest, skb->mark);
 477                 if (!cp) {
 478                         *ignored = -1;
 479                         return NULL;
 480                 }
 481         }
 482
 483         IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
 484                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
 485                       ip_vs_fwd_tag(cp),
 486                       IP_VS_DBG_ADDR(svc->af, &cp->caddr), ntohs(cp->cport),
 487                       IP_VS_DBG_ADDR(svc->af, &cp->vaddr), ntohs(cp->vport),
 488                       IP_VS_DBG_ADDR(svc->af, &cp->daddr), ntohs(cp->dport),
 489                       cp->flags, atomic_read(&cp->refcnt));
 490
 491         ip_vs_conn_stats(cp, svc);
 492         return cp;
 493 }
 494
 495
 496 /*
 497  *  Pass or drop the packet.
 498  *  Called by ip_vs_in, when the virtual service is available but
 499  *  no destination is available for a new connection.
 500  */
 501 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 502                 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
 503 {
 504         __be16 _ports[2], *pptr;
 505 #ifdef CONFIG_SYSCTL
 506         struct net *net;
 507         struct netns_ipvs *ipvs;
 508         int unicast;
 509 #endif
 510
 511         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 512         if (pptr == NULL) {
 513                 return NF_DROP;
 514         }
 515
 516 #ifdef CONFIG_SYSCTL
 517         net = skb_net(skb);
 518
 519 #ifdef CONFIG_IP_VS_IPV6
 520         if (svc->af == AF_INET6)
 521                 unicast = ipv6_addr_type(&iph->daddr.in6) & IPV6_ADDR_UNICAST;
 522         else
 523 #endif
 524                 unicast = (inet_addr_type(net, iph->daddr.ip) == RTN_UNICAST);
 525
 526         /* if it is fwmark-based service, the cache_bypass sysctl is up
 527            and the destination is a non-local unicast, then create
 528            a cache_bypass connection entry */
 529         ipvs = net_ipvs(net);
 530         if (ipvs->sysctl_cache_bypass && svc->fwmark && unicast) {
 531                 int ret;
 532                 struct ip_vs_conn *cp;
 533                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
 534                                       iph->protocol == IPPROTO_UDP) ?
 535                                       IP_VS_CONN_F_ONE_PACKET : 0;
 536                 union nf_inet_addr daddr =  { .all = { 0, 0, 0, 0 } };
 537
 538                 /* create a new connection entry */
 539                 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
 540                 {
 541                         struct ip_vs_conn_param p;
 542                         ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol,
 543                                               &iph->saddr, pptr[0],
 544                                               &iph->daddr, pptr[1], &p);
 545                         cp = ip_vs_conn_new(&p, &daddr, 0,
 546                                             IP_VS_CONN_F_BYPASS | flags,
 547                                             NULL, skb->mark);
 548                         if (!cp)
 549                                 return NF_DROP;
 550                 }
 551
 552                 /* statistics */
 553                 ip_vs_in_stats(cp, skb);
 554
 555                 /* set state */
 556                 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
 557
 558                 /* transmit the first SYN packet */
 559                 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
 560                 /* do not touch skb anymore */
 561
 562                 atomic_inc(&cp->in_pkts);
 563                 ip_vs_conn_put(cp);
 564                 return ret;
 565         }
 566 #endif
 567
 568         /*
 569          * When the virtual ftp service is presented, packets destined
 570          * for other services on the VIP may get here (except services
 571          * listed in the ipvs table), pass the packets, because it is
 572          * not ipvs job to decide to drop the packets.
 573          */
 574         if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT))
 575                 return NF_ACCEPT;
 576
 577         /*
 578          * Notify the client that the destination is unreachable, and
 579          * release the socket buffer.
 580          * Since it is in IP layer, the TCP socket is not actually
 581          * created, the TCP RST packet cannot be sent, instead that
 582          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 583          */
 584 #ifdef CONFIG_IP_VS_IPV6
 585         if (svc->af == AF_INET6) {
 586                 if (!skb->dev) {
 587                         struct net *net_ = dev_net(skb_dst(skb)->dev);
 588
 589                         skb->dev = net_->loopback_dev;
 590                 }
 591                 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 592         } else
 593 #endif
 594                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 595
 596         return NF_DROP;
 597 }
 598
 599 #ifdef CONFIG_SYSCTL
 600
 601 static int sysctl_snat_reroute(struct sk_buff *skb)
 602 {
 603         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 604         return ipvs->sysctl_snat_reroute;
 605 }
 606
 607 static int sysctl_nat_icmp_send(struct net *net)
 608 {
 609         struct netns_ipvs *ipvs = net_ipvs(net);
 610         return ipvs->sysctl_nat_icmp_send;
 611 }
 612
 613 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
 614 {
 615         return ipvs->sysctl_expire_nodest_conn;
 616 }
 617
 618 #else
 619
 620 static int sysctl_snat_reroute(struct sk_buff *skb) { return 0; }
 621 static int sysctl_nat_icmp_send(struct net *net) { return 0; }
 622 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
 623
 624 #endif
 625
 626 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 627 {
 628         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 629 }
 630
 631 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
 632 {
 633         if (NF_INET_LOCAL_IN == hooknum)
 634                 return IP_DEFRAG_VS_IN;
 635         if (NF_INET_FORWARD == hooknum)
 636                 return IP_DEFRAG_VS_FWD;
 637         return IP_DEFRAG_VS_OUT;
 638 }
 639
 640 static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user)
 641 {
 642         int err;
 643
 644         local_bh_disable();
 645         err = ip_defrag(skb, user);
 646         local_bh_enable();
 647         if (!err)
 648                 ip_send_check(ip_hdr(skb));
 649
 650         return err;
 651 }
 652
 653 static int ip_vs_route_me_harder(int af, struct sk_buff *skb)
 654 {
 655 #ifdef CONFIG_IP_VS_IPV6
 656         if (af == AF_INET6) {
 657                 if (sysctl_snat_reroute(skb) && ip6_route_me_harder(skb) != 0)
 658                         return 1;
 659         } else
 660 #endif
 661                 if ((sysctl_snat_reroute(skb) ||
 662                      skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
 663                     ip_route_me_harder(skb, RTN_LOCAL) != 0)
 664                         return 1;
 665
 666         return 0;
 667 }
 668
 669 /*
 670  * Packet has been made sufficiently writable in caller
 671  * - inout: 1=in->out, 0=out->in
 672  */
 673 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 674                     struct ip_vs_conn *cp, int inout)
 675 {
 676         struct iphdr *iph        = ip_hdr(skb);
 677         unsigned int icmp_offset = iph->ihl*4;
 678         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
 679                                                       icmp_offset);
 680         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 681
 682         if (inout) {
 683                 iph->saddr = cp->vaddr.ip;
 684                 ip_send_check(iph);
 685                 ciph->daddr = cp->vaddr.ip;
 686                 ip_send_check(ciph);
 687         } else {
 688                 iph->daddr = cp->daddr.ip;
 689                 ip_send_check(iph);
 690                 ciph->saddr = cp->daddr.ip;
 691                 ip_send_check(ciph);
 692         }
 693
 694         /* the TCP/UDP/SCTP port */
 695         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
 696             IPPROTO_SCTP == ciph->protocol) {
 697                 __be16 *ports = (void *)ciph + ciph->ihl*4;
 698
 699                 if (inout)
 700                         ports[1] = cp->vport;
 701                 else
 702                         ports[0] = cp->dport;
 703         }
 704
 705         /* And finally the ICMP checksum */
 706         icmph->checksum = 0;
 707         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 708         skb->ip_summed = CHECKSUM_UNNECESSARY;
 709
 710         if (inout)
 711                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 712                         "Forwarding altered outgoing ICMP");
 713         else
 714                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 715                         "Forwarding altered incoming ICMP");
 716 }
 717
 718 #ifdef CONFIG_IP_VS_IPV6
 719 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
 720                     struct ip_vs_conn *cp, int inout)
 721 {
 722         struct ipv6hdr *iph      = ipv6_hdr(skb);
 723         unsigned int icmp_offset = 0;
 724         unsigned int offs        = 0; /* header offset*/
 725         int protocol;
 726         struct icmp6hdr *icmph;
 727         struct ipv6hdr *ciph;
 728         unsigned short fragoffs;
 729
 730         ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
 731         icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
 732         offs = icmp_offset + sizeof(struct icmp6hdr);
 733         ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
 734
 735         protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
 736
 737         if (inout) {
 738                 iph->saddr = cp->vaddr.in6;
 739                 ciph->daddr = cp->vaddr.in6;
 740         } else {
 741                 iph->daddr = cp->daddr.in6;
 742                 ciph->saddr = cp->daddr.in6;
 743         }
 744
 745         /* the TCP/UDP/SCTP port */
 746         if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 747                           IPPROTO_SCTP == protocol)) {
 748                 __be16 *ports = (void *)(skb_network_header(skb) + offs);
 749
 750                 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
 751                               ntohs(inout ? ports[1] : ports[0]),
 752                               ntohs(inout ? cp->vport : cp->dport));
 753                 if (inout)
 754                         ports[1] = cp->vport;
 755                 else
 756                         ports[0] = cp->dport;
 757         }
 758
 759         /* And finally the ICMP checksum */
 760         icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
 761                                               skb->len - icmp_offset,
 762                                               IPPROTO_ICMPV6, 0);
 763         skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
 764         skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
 765         skb->ip_summed = CHECKSUM_PARTIAL;
 766
 767         if (inout)
 768                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 769                               (void *)ciph - (void *)iph,
 770                               "Forwarding altered outgoing ICMPv6");
 771         else
 772                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 773                               (void *)ciph - (void *)iph,
 774                               "Forwarding altered incoming ICMPv6");
 775 }
 776 #endif
 777
 778 /* Handle relevant response ICMP messages - forward to the right
 779  * destination host.
 780  */
 781 static int handle_response_icmp(int af, struct sk_buff *skb,
 782                                 union nf_inet_addr *snet,
 783                                 __u8 protocol, struct ip_vs_conn *cp,
 784                                 struct ip_vs_protocol *pp,
 785                                 unsigned int offset, unsigned int ihl)
 786 {
 787         unsigned int verdict = NF_DROP;
 788
 789         if (IP_VS_FWD_METHOD(cp) != 0) {
 790                 pr_err("shouldn't reach here, because the box is on the "
 791                        "half connection in the tun/dr module.\n");
 792         }
 793
 794         /* Ensure the checksum is correct */
 795         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 796                 /* Failed checksum! */
 797                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
 798                               IP_VS_DBG_ADDR(af, snet));
 799                 goto out;
 800         }
 801
 802         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 803             IPPROTO_SCTP == protocol)
 804                 offset += 2 * sizeof(__u16);
 805         if (!skb_make_writable(skb, offset))
 806                 goto out;
 807
 808 #ifdef CONFIG_IP_VS_IPV6
 809         if (af == AF_INET6)
 810                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
 811         else
 812 #endif
 813                 ip_vs_nat_icmp(skb, pp, cp, 1);
 814
 815         if (ip_vs_route_me_harder(af, skb))
 816                 goto out;
 817
 818         /* do the statistics and put it back */
 819         ip_vs_out_stats(cp, skb);
 820
 821         skb->ipvs_property = 1;
 822         if (!(cp->flags & IP_VS_CONN_F_NFCT))
 823                 ip_vs_notrack(skb);
 824         else
 825                 ip_vs_update_conntrack(skb, cp, 0);
 826         verdict = NF_ACCEPT;
 827
 828 out:
 829         __ip_vs_conn_put(cp);
 830
 831         return verdict;
 832 }
 833
 834 /*
 835  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 836  *      Find any that might be relevant, check against existing connections.
 837  *      Currently handles error types - unreachable, quench, ttl exceeded.
 838  */
 839 static int ip_vs_out_icmp(struct sk_buff *skb, int *related,
 840                           unsigned int hooknum)
 841 {
 842         struct iphdr *iph;
 843         struct icmphdr  _icmph, *ic;
 844         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 845         struct ip_vs_iphdr ciph;
 846         struct ip_vs_conn *cp;
 847         struct ip_vs_protocol *pp;
 848         unsigned int offset, ihl;
 849         union nf_inet_addr snet;
 850
 851         *related = 1;
 852
 853         /* reassemble IP fragments */
 854         if (ip_is_fragment(ip_hdr(skb))) {
 855                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
 856                         return NF_STOLEN;
 857         }
 858
 859         iph = ip_hdr(skb);
 860         offset = ihl = iph->ihl * 4;
 861         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 862         if (ic == NULL)
 863                 return NF_DROP;
 864
 865         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
 866                   ic->type, ntohs(icmp_id(ic)),
 867                   &iph->saddr, &iph->daddr);
 868
 869         /*
 870          * Work through seeing if this is for us.
 871          * These checks are supposed to be in an order that means easy
 872          * things are checked first to speed up processing.... however
 873          * this means that some packets will manage to get a long way
 874          * down this stack and then be rejected, but that's life.
 875          */
 876         if ((ic->type != ICMP_DEST_UNREACH) &&
 877             (ic->type != ICMP_SOURCE_QUENCH) &&
 878             (ic->type != ICMP_TIME_EXCEEDED)) {
 879                 *related = 0;
 880                 return NF_ACCEPT;
 881         }
 882
 883         /* Now find the contained IP header */
 884         offset += sizeof(_icmph);
 885         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 886         if (cih == NULL)
 887                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 888
 889         pp = ip_vs_proto_get(cih->protocol);
 890         if (!pp)
 891                 return NF_ACCEPT;
 892
 893         /* Is the embedded protocol header present? */
 894         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 895                      pp->dont_defrag))
 896                 return NF_ACCEPT;
 897
 898         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
 899                       "Checking outgoing ICMP for");
 900
 901         ip_vs_fill_ip4hdr(cih, &ciph);
 902         ciph.len += offset;
 903         /* The embedded headers contain source and dest in reverse order */
 904         cp = pp->conn_out_get(AF_INET, skb, &ciph, 1);
 905         if (!cp)
 906                 return NF_ACCEPT;
 907
 908         snet.ip = iph->saddr;
 909         return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
 910                                     pp, ciph.len, ihl);
 911 }
 912
 913 #ifdef CONFIG_IP_VS_IPV6
 914 static int ip_vs_out_icmp_v6(struct sk_buff *skb, int *related,
 915                              unsigned int hooknum, struct ip_vs_iphdr *ipvsh)
 916 {
 917         struct icmp6hdr _icmph, *ic;
 918         struct ipv6hdr _ip6h, *ip6h; /* The ip header contained within ICMP */
 919         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
 920         struct ip_vs_conn *cp;
 921         struct ip_vs_protocol *pp;
 922         union nf_inet_addr snet;
 923         unsigned int writable;
 924
 925         *related = 1;
 926         ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph, ipvsh);
 927         if (ic == NULL)
 928                 return NF_DROP;
 929
 930         /*
 931          * Work through seeing if this is for us.
 932          * These checks are supposed to be in an order that means easy
 933          * things are checked first to speed up processing.... however
 934          * this means that some packets will manage to get a long way
 935          * down this stack and then be rejected, but that's life.
 936          */
 937         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
 938                 *related = 0;
 939                 return NF_ACCEPT;
 940         }
 941         /* Fragment header that is before ICMP header tells us that:
 942          * it's not an error message since they can't be fragmented.
 943          */
 944         if (ipvsh->flags & IP6_FH_F_FRAG)
 945                 return NF_DROP;
 946
 947         IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
 948                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
 949                   &ipvsh->saddr, &ipvsh->daddr);
 950
 951         /* Now find the contained IP header */
 952         ciph.len = ipvsh->len + sizeof(_icmph);
 953         ip6h = skb_header_pointer(skb, ciph.len, sizeof(_ip6h), &_ip6h);
 954         if (ip6h == NULL)
 955                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 956         ciph.saddr.in6 = ip6h->saddr; /* conn_out_get() handles reverse order */
 957         ciph.daddr.in6 = ip6h->daddr;
 958         /* skip possible IPv6 exthdrs of contained IPv6 packet */
 959         ciph.protocol = ipv6_find_hdr(skb, &ciph.len, -1, &ciph.fragoffs, NULL);
 960         if (ciph.protocol < 0)
 961                 return NF_ACCEPT; /* Contained IPv6 hdr looks wrong, ignore */
 962
 963         pp = ip_vs_proto_get(ciph.protocol);
 964         if (!pp)
 965                 return NF_ACCEPT;
 966
 967         /* The embedded headers contain source and dest in reverse order */
 968         cp = pp->conn_out_get(AF_INET6, skb, &ciph, 1);
 969         if (!cp)
 970                 return NF_ACCEPT;
 971
 972         snet.in6 = ciph.saddr.in6;
 973         writable = ciph.len;
 974         return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
 975                                     pp, writable, sizeof(struct ipv6hdr));
 976 }
 977 #endif
 978
 979 /*
 980  * Check if sctp chunc is ABORT chunk
 981  */
 982 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
 983 {
 984         sctp_chunkhdr_t *sch, schunk;
 985         sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
 986                         sizeof(schunk), &schunk);
 987         if (sch == NULL)
 988                 return 0;
 989         if (sch->type == SCTP_CID_ABORT)
 990                 return 1;
 991         return 0;
 992 }
 993
 994 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
 995 {
 996         struct tcphdr _tcph, *th;
 997
 998         th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
 999         if (th == NULL)
1000                 return 0;
1001         return th->rst;
1002 }
1003
1004 static inline bool is_new_conn(const struct sk_buff *skb,
1005                                struct ip_vs_iphdr *iph)
1006 {
1007         switch (iph->protocol) {
1008         case IPPROTO_TCP: {
1009                 struct tcphdr _tcph, *th;
1010
1011                 th = skb_header_pointer(skb, iph->len, sizeof(_tcph), &_tcph);
1012                 if (th == NULL)
1013                         return false;
1014                 return th->syn;
1015         }
1016         case IPPROTO_SCTP: {
1017                 sctp_chunkhdr_t *sch, schunk;
1018
1019                 sch = skb_header_pointer(skb, iph->len + sizeof(sctp_sctphdr_t),
1020                                          sizeof(schunk), &schunk);
1021                 if (sch == NULL)
1022                         return false;
1023                 return sch->type == SCTP_CID_INIT;
1024         }
1025         default:
1026                 return false;
1027         }
1028 }
1029
1030 /* Handle response packets: rewrite addresses and send away...
1031  */
1032 static unsigned int
1033 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1034                 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph)
1035 {
1036         struct ip_vs_protocol *pp = pd->pp;
1037
1038         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Outgoing packet");
1039
1040         if (!skb_make_writable(skb, iph->len))
1041                 goto drop;
1042
1043         /* mangle the packet */
1044         if (pp->snat_handler && !pp->snat_handler(skb, pp, cp, iph))
1045                 goto drop;
1046
1047 #ifdef CONFIG_IP_VS_IPV6
1048         if (af == AF_INET6)
1049                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1050         else
1051 #endif
1052         {
1053                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1054                 ip_send_check(ip_hdr(skb));
1055         }
1056
1057         /*
1058          * nf_iterate does not expect change in the skb->dst->dev.
1059          * It looks like it is not fatal to enable this code for hooks
1060          * where our handlers are at the end of the chain list and
1061          * when all next handlers use skb->dst->dev and not outdev.
1062          * It will definitely route properly the inout NAT traffic
1063          * when multiple paths are used.
1064          */
1065
1066         /* For policy routing, packets originating from this
1067          * machine itself may be routed differently to packets
1068          * passing through.  We want this packet to be routed as
1069          * if it came from this machine itself.  So re-compute
1070          * the routing information.
1071          */
1072         if (ip_vs_route_me_harder(af, skb))
1073                 goto drop;
1074
1075         IP_VS_DBG_PKT(10, af, pp, skb, 0, "After SNAT");
1076
1077         ip_vs_out_stats(cp, skb);
1078         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1079         skb->ipvs_property = 1;
1080         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1081                 ip_vs_notrack(skb);
1082         else
1083                 ip_vs_update_conntrack(skb, cp, 0);
1084         ip_vs_conn_put(cp);
1085
1086         LeaveFunction(11);
1087         return NF_ACCEPT;
1088
1089 drop:
1090         ip_vs_conn_put(cp);
1091         kfree_skb(skb);
1092         LeaveFunction(11);
1093         return NF_STOLEN;
1094 }
1095
1096 /*
1097  *      Check if outgoing packet belongs to the established ip_vs_conn.
1098  */
1099 static unsigned int
1100 ip_vs_out(unsigned int hooknum, struct sk_buff *skb, int af)
1101 {
1102         struct net *net = NULL;
1103         struct ip_vs_iphdr iph;
1104         struct ip_vs_protocol *pp;
1105         struct ip_vs_proto_data *pd;
1106         struct ip_vs_conn *cp;
1107
1108         EnterFunction(11);
1109
1110         /* Already marked as IPVS request or reply? */
1111         if (skb->ipvs_property)
1112                 return NF_ACCEPT;
1113
1114         /* Bad... Do not break raw sockets */
1115         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1116                      af == AF_INET)) {
1117                 struct sock *sk = skb->sk;
1118                 struct inet_sock *inet = inet_sk(skb->sk);
1119
1120                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1121                         return NF_ACCEPT;
1122         }
1123
1124         if (unlikely(!skb_dst(skb)))
1125                 return NF_ACCEPT;
1126
1127         net = skb_net(skb);
1128         if (!net_ipvs(net)->enable)
1129                 return NF_ACCEPT;
1130
1131         ip_vs_fill_iph_skb(af, skb, &iph);
1132 #ifdef CONFIG_IP_VS_IPV6
1133         if (af == AF_INET6) {
1134                 if (!iph.fragoffs && skb_nfct_reasm(skb)) {
1135                         struct sk_buff *reasm = skb_nfct_reasm(skb);
1136                         /* Save fw mark for coming frags */
1137                         reasm->ipvs_property = 1;
1138                         reasm->mark = skb->mark;
1139                 }
1140                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1141                         int related;
1142                         int verdict = ip_vs_out_icmp_v6(skb, &related,
1143                                                         hooknum, &iph);
1144
1145                         if (related)
1146                                 return verdict;
1147                 }
1148         } else
1149 #endif
1150                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1151                         int related;
1152                         int verdict = ip_vs_out_icmp(skb, &related, hooknum);
1153
1154                         if (related)
1155                                 return verdict;
1156                 }
1157
1158         pd = ip_vs_proto_data_get(net, iph.protocol);
1159         if (unlikely(!pd))
1160                 return NF_ACCEPT;
1161         pp = pd->pp;
1162
1163         /* reassemble IP fragments */
1164 #ifdef CONFIG_IP_VS_IPV6
1165         if (af == AF_INET)
1166 #endif
1167                 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1168                         if (ip_vs_gather_frags(skb,
1169                                                ip_vs_defrag_user(hooknum)))
1170                                 return NF_STOLEN;
1171
1172                         ip_vs_fill_ip4hdr(skb_network_header(skb), &iph);
1173                 }
1174
1175         /*
1176          * Check if the packet belongs to an existing entry
1177          */
1178         cp = pp->conn_out_get(af, skb, &iph, 0);
1179
1180         if (likely(cp))
1181                 return handle_response(af, skb, pd, cp, &iph);
1182         if (sysctl_nat_icmp_send(net) &&
1183             (pp->protocol == IPPROTO_TCP ||
1184              pp->protocol == IPPROTO_UDP ||
1185              pp->protocol == IPPROTO_SCTP)) {
1186                 __be16 _ports[2], *pptr;
1187
1188                 pptr = frag_safe_skb_hp(skb, iph.len,
1189                                          sizeof(_ports), _ports, &iph);
1190                 if (pptr == NULL)
1191                         return NF_ACCEPT;       /* Not for me */
1192                 if (ip_vs_has_real_service(net, af, iph.protocol, &iph.saddr,
1193                                            pptr[0])) {
1194                         /*
1195                          * Notify the real server: there is no
1196                          * existing entry if it is not RST
1197                          * packet or not TCP packet.
1198                          */
1199                         if ((iph.protocol != IPPROTO_TCP &&
1200                              iph.protocol != IPPROTO_SCTP)
1201                              || ((iph.protocol == IPPROTO_TCP
1202                                   && !is_tcp_reset(skb, iph.len))
1203                                  || (iph.protocol == IPPROTO_SCTP
1204                                         && !is_sctp_abort(skb,
1205                                                 iph.len)))) {
1206 #ifdef CONFIG_IP_VS_IPV6
1207                                 if (af == AF_INET6) {
1208                                         if (!skb->dev)
1209                                                 skb->dev = net->loopback_dev;
1210                                         icmpv6_send(skb,
1211                                                     ICMPV6_DEST_UNREACH,
1212                                                     ICMPV6_PORT_UNREACH,
1213                                                     0);
1214                                 } else
1215 #endif
1216                                         icmp_send(skb,
1217                                                   ICMP_DEST_UNREACH,
1218                                                   ICMP_PORT_UNREACH, 0);
1219                                 return NF_DROP;
1220                         }
1221                 }
1222         }
1223         IP_VS_DBG_PKT(12, af, pp, skb, 0,
1224                       "ip_vs_out: packet continues traversal as normal");
1225         return NF_ACCEPT;
1226 }
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_reply4(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_INET);
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_reply4(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_INET);
1251 }
1252
1253 #ifdef CONFIG_IP_VS_IPV6
1254
1255 /*
1256  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1257  *      used only for VS/NAT.
1258  *      Check if packet is reply for established ip_vs_conn.
1259  */
1260 static unsigned int
1261 ip_vs_reply6(unsigned int hooknum, struct sk_buff *skb,
1262              const struct net_device *in, const struct net_device *out,
1263              int (*okfn)(struct sk_buff *))
1264 {
1265         return ip_vs_out(hooknum, skb, AF_INET6);
1266 }
1267
1268 /*
1269  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1270  *      Check if packet is reply for established ip_vs_conn.
1271  */
1272 static unsigned int
1273 ip_vs_local_reply6(unsigned int hooknum, struct sk_buff *skb,
1274                    const struct net_device *in, const struct net_device *out,
1275                    int (*okfn)(struct sk_buff *))
1276 {
1277         return ip_vs_out(hooknum, skb, AF_INET6);
1278 }
1279
1280 #endif
1281
1282 /*
1283  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1284  *      Find any that might be relevant, check against existing connections,
1285  *      forward to the right destination host if relevant.
1286  *      Currently handles error types - unreachable, quench, ttl exceeded.
1287  */
1288 static int
1289 ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum)
1290 {
1291         struct net *net = NULL;
1292         struct iphdr *iph;
1293         struct icmphdr  _icmph, *ic;
1294         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1295         struct ip_vs_iphdr ciph;
1296         struct ip_vs_conn *cp;
1297         struct ip_vs_protocol *pp;
1298         struct ip_vs_proto_data *pd;
1299         unsigned int offset, offset2, ihl, verdict;
1300         bool ipip;
1301
1302         *related = 1;
1303
1304         /* reassemble IP fragments */
1305         if (ip_is_fragment(ip_hdr(skb))) {
1306                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
1307                         return NF_STOLEN;
1308         }
1309
1310         iph = ip_hdr(skb);
1311         offset = ihl = iph->ihl * 4;
1312         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1313         if (ic == NULL)
1314                 return NF_DROP;
1315
1316         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1317                   ic->type, ntohs(icmp_id(ic)),
1318                   &iph->saddr, &iph->daddr);
1319
1320         /*
1321          * Work through seeing if this is for us.
1322          * These checks are supposed to be in an order that means easy
1323          * things are checked first to speed up processing.... however
1324          * this means that some packets will manage to get a long way
1325          * down this stack and then be rejected, but that's life.
1326          */
1327         if ((ic->type != ICMP_DEST_UNREACH) &&
1328             (ic->type != ICMP_SOURCE_QUENCH) &&
1329             (ic->type != ICMP_TIME_EXCEEDED)) {
1330                 *related = 0;
1331                 return NF_ACCEPT;
1332         }
1333
1334         /* Now find the contained IP header */
1335         offset += sizeof(_icmph);
1336         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1337         if (cih == NULL)
1338                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1339
1340         net = skb_net(skb);
1341
1342         /* Special case for errors for IPIP packets */
1343         ipip = false;
1344         if (cih->protocol == IPPROTO_IPIP) {
1345                 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1346                         return NF_ACCEPT;
1347                 /* Error for our IPIP must arrive at LOCAL_IN */
1348                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1349                         return NF_ACCEPT;
1350                 offset += cih->ihl * 4;
1351                 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1352                 if (cih == NULL)
1353                         return NF_ACCEPT; /* The packet looks wrong, ignore */
1354                 ipip = true;
1355         }
1356
1357         pd = ip_vs_proto_data_get(net, cih->protocol);
1358         if (!pd)
1359                 return NF_ACCEPT;
1360         pp = pd->pp;
1361
1362         /* Is the embedded protocol header present? */
1363         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1364                      pp->dont_defrag))
1365                 return NF_ACCEPT;
1366
1367         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1368                       "Checking incoming ICMP for");
1369
1370         offset2 = offset;
1371         ip_vs_fill_ip4hdr(cih, &ciph);
1372         ciph.len += offset;
1373         offset = ciph.len;
1374         /* The embedded headers contain source and dest in reverse order.
1375          * For IPIP this is error for request, not for reply.
1376          */
1377         cp = pp->conn_in_get(AF_INET, skb, &ciph, ipip ? 0 : 1);
1378         if (!cp)
1379                 return NF_ACCEPT;
1380
1381         verdict = NF_DROP;
1382
1383         /* Ensure the checksum is correct */
1384         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1385                 /* Failed checksum! */
1386                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1387                           &iph->saddr);
1388                 goto out;
1389         }
1390
1391         if (ipip) {
1392                 __be32 info = ic->un.gateway;
1393
1394                 /* Update the MTU */
1395                 if (ic->type == ICMP_DEST_UNREACH &&
1396                     ic->code == ICMP_FRAG_NEEDED) {
1397                         struct ip_vs_dest *dest = cp->dest;
1398                         u32 mtu = ntohs(ic->un.frag.mtu);
1399
1400                         /* Strip outer IP and ICMP, go to IPIP header */
1401                         __skb_pull(skb, ihl + sizeof(_icmph));
1402                         offset2 -= ihl + sizeof(_icmph);
1403                         skb_reset_network_header(skb);
1404                         IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1405                                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1406                         ipv4_update_pmtu(skb, dev_net(skb->dev),
1407                                          mtu, 0, 0, 0, 0);
1408                         /* Client uses PMTUD? */
1409                         if (!(cih->frag_off & htons(IP_DF)))
1410                                 goto ignore_ipip;
1411                         /* Prefer the resulting PMTU */
1412                         if (dest) {
1413                                 struct ip_vs_dest_dst *dest_dst;
1414
1415                                 rcu_read_lock();
1416                                 dest_dst = rcu_dereference(dest->dest_dst);
1417                                 if (dest_dst)
1418                                         mtu = dst_mtu(dest_dst->dst_cache);
1419                                 rcu_read_unlock();
1420                         }
1421                         if (mtu > 68 + sizeof(struct iphdr))
1422                                 mtu -= sizeof(struct iphdr);
1423                         info = htonl(mtu);
1424                 }
1425                 /* Strip outer IP, ICMP and IPIP, go to IP header of
1426                  * original request.
1427                  */
1428                 __skb_pull(skb, offset2);
1429                 skb_reset_network_header(skb);
1430                 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1431                         &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1432                         ic->type, ic->code, ntohl(info));
1433                 icmp_send(skb, ic->type, ic->code, info);
1434                 /* ICMP can be shorter but anyways, account it */
1435                 ip_vs_out_stats(cp, skb);
1436
1437 ignore_ipip:
1438                 consume_skb(skb);
1439                 verdict = NF_STOLEN;
1440                 goto out;
1441         }
1442
1443         /* do the statistics and put it back */
1444         ip_vs_in_stats(cp, skb);
1445         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol ||
1446             IPPROTO_SCTP == cih->protocol)
1447                 offset += 2 * sizeof(__u16);
1448         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1449
1450 out:
1451         __ip_vs_conn_put(cp);
1452
1453         return verdict;
1454 }
1455
1456 #ifdef CONFIG_IP_VS_IPV6
1457 static int ip_vs_in_icmp_v6(struct sk_buff *skb, int *related,
1458                             unsigned int hooknum, struct ip_vs_iphdr *iph)
1459 {
1460         struct net *net = NULL;
1461         struct ipv6hdr _ip6h, *ip6h;
1462         struct icmp6hdr _icmph, *ic;
1463         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1464         struct ip_vs_conn *cp;
1465         struct ip_vs_protocol *pp;
1466         struct ip_vs_proto_data *pd;
1467         unsigned int offs_ciph, writable, verdict;
1468
1469         *related = 1;
1470
1471         ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph, iph);
1472         if (ic == NULL)
1473                 return NF_DROP;
1474
1475         /*
1476          * Work through seeing if this is for us.
1477          * These checks are supposed to be in an order that means easy
1478          * things are checked first to speed up processing.... however
1479          * this means that some packets will manage to get a long way
1480          * down this stack and then be rejected, but that's life.
1481          */
1482         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1483                 *related = 0;
1484                 return NF_ACCEPT;
1485         }
1486         /* Fragment header that is before ICMP header tells us that:
1487          * it's not an error message since they can't be fragmented.
1488          */
1489         if (iph->flags & IP6_FH_F_FRAG)
1490                 return NF_DROP;
1491
1492         IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1493                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1494                   &iph->saddr, &iph->daddr);
1495
1496         /* Now find the contained IP header */
1497         ciph.len = iph->len + sizeof(_icmph);
1498         offs_ciph = ciph.len; /* Save ip header offset */
1499         ip6h = skb_header_pointer(skb, ciph.len, sizeof(_ip6h), &_ip6h);
1500         if (ip6h == NULL)
1501                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1502         ciph.saddr.in6 = ip6h->saddr; /* conn_in_get() handles reverse order */
1503         ciph.daddr.in6 = ip6h->daddr;
1504         /* skip possible IPv6 exthdrs of contained IPv6 packet */
1505         ciph.protocol = ipv6_find_hdr(skb, &ciph.len, -1, &ciph.fragoffs, NULL);
1506         if (ciph.protocol < 0)
1507                 return NF_ACCEPT; /* Contained IPv6 hdr looks wrong, ignore */
1508
1509         net = skb_net(skb);
1510         pd = ip_vs_proto_data_get(net, ciph.protocol);
1511         if (!pd)
1512                 return NF_ACCEPT;
1513         pp = pd->pp;
1514
1515         /* Cannot handle fragmented embedded protocol */
1516         if (ciph.fragoffs)
1517                 return NF_ACCEPT;
1518
1519         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offs_ciph,
1520                       "Checking incoming ICMPv6 for");
1521
1522         /* The embedded headers contain source and dest in reverse order
1523          * if not from localhost
1524          */
1525         cp = pp->conn_in_get(AF_INET6, skb, &ciph,
1526                              (hooknum == NF_INET_LOCAL_OUT) ? 0 : 1);
1527
1528         if (!cp)
1529                 return NF_ACCEPT;
1530         /* VS/TUN, VS/DR and LOCALNODE just let it go */
1531         if ((hooknum == NF_INET_LOCAL_OUT) &&
1532             (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1533                 __ip_vs_conn_put(cp);
1534                 return NF_ACCEPT;
1535         }
1536
1537         /* do the statistics and put it back */
1538         ip_vs_in_stats(cp, skb);
1539
1540         /* Need to mangle contained IPv6 header in ICMPv6 packet */
1541         writable = ciph.len;
1542         if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1543             IPPROTO_SCTP == ciph.protocol)
1544                 writable += 2 * sizeof(__u16); /* Also mangle ports */
1545
1546         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, writable, hooknum, &ciph);
1547
1548         __ip_vs_conn_put(cp);
1549
1550         return verdict;
1551 }
1552 #endif
1553
1554
1555 /*
1556  *      Check if it's for virtual services, look it up,
1557  *      and send it on its way...
1558  */
1559 static unsigned int
1560 ip_vs_in(unsigned int hooknum, struct sk_buff *skb, int af)
1561 {
1562         struct net *net;
1563         struct ip_vs_iphdr iph;
1564         struct ip_vs_protocol *pp;
1565         struct ip_vs_proto_data *pd;
1566         struct ip_vs_conn *cp;
1567         int ret, pkts;
1568         struct netns_ipvs *ipvs;
1569
1570         /* Already marked as IPVS request or reply? */
1571         if (skb->ipvs_property)
1572                 return NF_ACCEPT;
1573
1574         /*
1575          *      Big tappo:
1576          *      - remote client: only PACKET_HOST
1577          *      - route: used for struct net when skb->dev is unset
1578          */
1579         if (unlikely((skb->pkt_type != PACKET_HOST &&
1580                       hooknum != NF_INET_LOCAL_OUT) ||
1581                      !skb_dst(skb))) {
1582                 ip_vs_fill_iph_skb(af, skb, &iph);
1583                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1584                               " ignored in hook %u\n",
1585                               skb->pkt_type, iph.protocol,
1586                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1587                 return NF_ACCEPT;
1588         }
1589         /* ipvs enabled in this netns ? */
1590         net = skb_net(skb);
1591         ipvs = net_ipvs(net);
1592         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1593                 return NF_ACCEPT;
1594
1595         ip_vs_fill_iph_skb(af, skb, &iph);
1596
1597         /* Bad... Do not break raw sockets */
1598         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1599                      af == AF_INET)) {
1600                 struct sock *sk = skb->sk;
1601                 struct inet_sock *inet = inet_sk(skb->sk);
1602
1603                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1604                         return NF_ACCEPT;
1605         }
1606
1607 #ifdef CONFIG_IP_VS_IPV6
1608         if (af == AF_INET6) {
1609                 if (!iph.fragoffs && skb_nfct_reasm(skb)) {
1610                         struct sk_buff *reasm = skb_nfct_reasm(skb);
1611                         /* Save fw mark for coming frags. */
1612                         reasm->ipvs_property = 1;
1613                         reasm->mark = skb->mark;
1614                 }
1615                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1616                         int related;
1617                         int verdict = ip_vs_in_icmp_v6(skb, &related, hooknum,
1618                                                        &iph);
1619
1620                         if (related)
1621                                 return verdict;
1622                 }
1623         } else
1624 #endif
1625                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1626                         int related;
1627                         int verdict = ip_vs_in_icmp(skb, &related, hooknum);
1628
1629                         if (related)
1630                                 return verdict;
1631                 }
1632
1633         /* Protocol supported? */
1634         pd = ip_vs_proto_data_get(net, iph.protocol);
1635         if (unlikely(!pd))
1636                 return NF_ACCEPT;
1637         pp = pd->pp;
1638         /*
1639          * Check if the packet belongs to an existing connection entry
1640          */
1641         cp = pp->conn_in_get(af, skb, &iph, 0);
1642
1643         if (unlikely(sysctl_expire_nodest_conn(ipvs)) && cp && cp->dest &&
1644             unlikely(!atomic_read(&cp->dest->weight)) && !iph.fragoffs &&
1645             is_new_conn(skb, &iph)) {
1646                 ip_vs_conn_expire_now(cp);
1647                 __ip_vs_conn_put(cp);
1648                 cp = NULL;
1649         }
1650
1651         if (unlikely(!cp) && !iph.fragoffs) {
1652                 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1653                  * replayed fragment zero will already have created the cp
1654                  */
1655                 int v;
1656
1657                 /* Schedule and create new connection entry into &cp */
1658                 if (!pp->conn_schedule(af, skb, pd, &v, &cp, &iph))
1659                         return v;
1660         }
1661
1662         if (unlikely(!cp)) {
1663                 /* sorry, all this trouble for a no-hit :) */
1664                 IP_VS_DBG_PKT(12, af, pp, skb, 0,
1665                               "ip_vs_in: packet continues traversal as normal");
1666                 if (iph.fragoffs && !skb_nfct_reasm(skb)) {
1667                         /* Fragment that couldn't be mapped to a conn entry
1668                          * and don't have any pointer to a reasm skb
1669                          * is missing module nf_defrag_ipv6
1670                          */
1671                         IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1672                         IP_VS_DBG_PKT(7, af, pp, skb, 0, "unhandled fragment");
1673                 }
1674                 return NF_ACCEPT;
1675         }
1676
1677         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Incoming packet");
1678         /* Check the server status */
1679         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1680                 /* the destination server is not available */
1681
1682                 if (sysctl_expire_nodest_conn(ipvs)) {
1683                         /* try to expire the connection immediately */
1684                         ip_vs_conn_expire_now(cp);
1685                 }
1686                 /* don't restart its timer, and silently
1687                    drop the packet. */
1688                 __ip_vs_conn_put(cp);
1689                 return NF_DROP;
1690         }
1691
1692         ip_vs_in_stats(cp, skb);
1693         ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1694         if (cp->packet_xmit)
1695                 ret = cp->packet_xmit(skb, cp, pp, &iph);
1696                 /* do not touch skb anymore */
1697         else {
1698                 IP_VS_DBG_RL("warning: packet_xmit is null");
1699                 ret = NF_ACCEPT;
1700         }
1701
1702         /* Increase its packet counter and check if it is needed
1703          * to be synchronized
1704          *
1705          * Sync connection if it is about to close to
1706          * encorage the standby servers to update the connections timeout
1707          *
1708          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1709          */
1710
1711         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
1712                 pkts = sysctl_sync_threshold(ipvs);
1713         else
1714                 pkts = atomic_add_return(1, &cp->in_pkts);
1715
1716         if (ipvs->sync_state & IP_VS_STATE_MASTER)
1717                 ip_vs_sync_conn(net, cp, pkts);
1718
1719         ip_vs_conn_put(cp);
1720         return ret;
1721 }
1722
1723 /*
1724  *      AF_INET handler in NF_INET_LOCAL_IN chain
1725  *      Schedule and forward packets from remote clients
1726  */
1727 static unsigned int
1728 ip_vs_remote_request4(unsigned int hooknum, struct sk_buff *skb,
1729                       const struct net_device *in,
1730                       const struct net_device *out,
1731                       int (*okfn)(struct sk_buff *))
1732 {
1733         return ip_vs_in(hooknum, skb, AF_INET);
1734 }
1735
1736 /*
1737  *      AF_INET handler in NF_INET_LOCAL_OUT chain
1738  *      Schedule and forward packets from local clients
1739  */
1740 static unsigned int
1741 ip_vs_local_request4(unsigned int hooknum, struct sk_buff *skb,
1742                      const struct net_device *in, const struct net_device *out,
1743                      int (*okfn)(struct sk_buff *))
1744 {
1745         return ip_vs_in(hooknum, skb, AF_INET);
1746 }
1747
1748 #ifdef CONFIG_IP_VS_IPV6
1749
1750 /*
1751  * AF_INET6 fragment handling
1752  * Copy info from first fragment, to the rest of them.
1753  */
1754 static unsigned int
1755 ip_vs_preroute_frag6(unsigned int hooknum, struct sk_buff *skb,
1756                      const struct net_device *in,
1757                      const struct net_device *out,
1758                      int (*okfn)(struct sk_buff *))
1759 {
1760         struct sk_buff *reasm = skb_nfct_reasm(skb);
1761         struct net *net;
1762
1763         /* Skip if not a "replay" from nf_ct_frag6_output or first fragment.
1764          * ipvs_property is set when checking first fragment
1765          * in ip_vs_in() and ip_vs_out().
1766          */
1767         if (reasm)
1768                 IP_VS_DBG(2, "Fragment recv prop:%d\n", reasm->ipvs_property);
1769         if (!reasm || !reasm->ipvs_property)
1770                 return NF_ACCEPT;
1771
1772         net = skb_net(skb);
1773         if (!net_ipvs(net)->enable)
1774                 return NF_ACCEPT;
1775
1776         /* Copy stored fw mark, saved in ip_vs_{in,out} */
1777         skb->mark = reasm->mark;
1778
1779         return NF_ACCEPT;
1780 }
1781
1782 /*
1783  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
1784  *      Schedule and forward packets from remote clients
1785  */
1786 static unsigned int
1787 ip_vs_remote_request6(unsigned int hooknum, struct sk_buff *skb,
1788                       const struct net_device *in,
1789                       const struct net_device *out,
1790                       int (*okfn)(struct sk_buff *))
1791 {
1792         return ip_vs_in(hooknum, skb, AF_INET6);
1793 }
1794
1795 /*
1796  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
1797  *      Schedule and forward packets from local clients
1798  */
1799 static unsigned int
1800 ip_vs_local_request6(unsigned int hooknum, struct sk_buff *skb,
1801                      const struct net_device *in, const struct net_device *out,
1802                      int (*okfn)(struct sk_buff *))
1803 {
1804         return ip_vs_in(hooknum, skb, AF_INET6);
1805 }
1806
1807 #endif
1808
1809
1810 /*
1811  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1812  *      related packets destined for 0.0.0.0/0.
1813  *      When fwmark-based virtual service is used, such as transparent
1814  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
1815  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1816  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1817  *      and send them to ip_vs_in_icmp.
1818  */
1819 static unsigned int
1820 ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff *skb,
1821                    const struct net_device *in, const struct net_device *out,
1822                    int (*okfn)(struct sk_buff *))
1823 {
1824         int r;
1825         struct net *net;
1826         struct netns_ipvs *ipvs;
1827
1828         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1829                 return NF_ACCEPT;
1830
1831         /* ipvs enabled in this netns ? */
1832         net = skb_net(skb);
1833         ipvs = net_ipvs(net);
1834         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1835                 return NF_ACCEPT;
1836
1837         return ip_vs_in_icmp(skb, &r, hooknum);
1838 }
1839
1840 #ifdef CONFIG_IP_VS_IPV6
1841 static unsigned int
1842 ip_vs_forward_icmp_v6(unsigned int hooknum, struct sk_buff *skb,
1843                       const struct net_device *in, const struct net_device *out,
1844                       int (*okfn)(struct sk_buff *))
1845 {
1846         int r;
1847         struct net *net;
1848         struct netns_ipvs *ipvs;
1849         struct ip_vs_iphdr iphdr;
1850
1851         ip_vs_fill_iph_skb(AF_INET6, skb, &iphdr);
1852         if (iphdr.protocol != IPPROTO_ICMPV6)
1853                 return NF_ACCEPT;
1854
1855         /* ipvs enabled in this netns ? */
1856         net = skb_net(skb);
1857         ipvs = net_ipvs(net);
1858         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1859                 return NF_ACCEPT;
1860
1861         return ip_vs_in_icmp_v6(skb, &r, hooknum, &iphdr);
1862 }
1863 #endif
1864
1865
1866 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
1867         /* After packet filtering, change source only for VS/NAT */
1868         {
1869                 .hook           = ip_vs_reply4,
1870                 .owner          = THIS_MODULE,
1871                 .pf             = NFPROTO_IPV4,
1872                 .hooknum        = NF_INET_LOCAL_IN,
1873                 .priority       = NF_IP_PRI_NAT_SRC - 2,
1874         },
1875         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1876          * or VS/NAT(change destination), so that filtering rules can be
1877          * applied to IPVS. */
1878         {
1879                 .hook           = ip_vs_remote_request4,
1880                 .owner          = THIS_MODULE,
1881                 .pf             = NFPROTO_IPV4,
1882                 .hooknum        = NF_INET_LOCAL_IN,
1883                 .priority       = NF_IP_PRI_NAT_SRC - 1,
1884         },
1885         /* Before ip_vs_in, change source only for VS/NAT */
1886         {
1887                 .hook           = ip_vs_local_reply4,
1888                 .owner          = THIS_MODULE,
1889                 .pf             = NFPROTO_IPV4,
1890                 .hooknum        = NF_INET_LOCAL_OUT,
1891                 .priority       = NF_IP_PRI_NAT_DST + 1,
1892         },
1893         /* After mangle, schedule and forward local requests */
1894         {
1895                 .hook           = ip_vs_local_request4,
1896                 .owner          = THIS_MODULE,
1897                 .pf             = NFPROTO_IPV4,
1898                 .hooknum        = NF_INET_LOCAL_OUT,
1899                 .priority       = NF_IP_PRI_NAT_DST + 2,
1900         },
1901         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1902          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1903         {
1904                 .hook           = ip_vs_forward_icmp,
1905                 .owner          = THIS_MODULE,
1906                 .pf             = NFPROTO_IPV4,
1907                 .hooknum        = NF_INET_FORWARD,
1908                 .priority       = 99,
1909         },
1910         /* After packet filtering, change source only for VS/NAT */
1911         {
1912                 .hook           = ip_vs_reply4,
1913                 .owner          = THIS_MODULE,
1914                 .pf             = NFPROTO_IPV4,
1915                 .hooknum        = NF_INET_FORWARD,
1916                 .priority       = 100,
1917         },
1918 #ifdef CONFIG_IP_VS_IPV6
1919         /* After mangle & nat fetch 2:nd fragment and following */
1920         {
1921                 .hook           = ip_vs_preroute_frag6,
1922                 .owner          = THIS_MODULE,
1923                 .pf             = NFPROTO_IPV6,
1924                 .hooknum        = NF_INET_PRE_ROUTING,
1925                 .priority       = NF_IP6_PRI_NAT_DST + 1,
1926         },
1927         /* After packet filtering, change source only for VS/NAT */
1928         {
1929                 .hook           = ip_vs_reply6,
1930                 .owner          = THIS_MODULE,
1931                 .pf             = NFPROTO_IPV6,
1932                 .hooknum        = NF_INET_LOCAL_IN,
1933                 .priority       = NF_IP6_PRI_NAT_SRC - 2,
1934         },
1935         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1936          * or VS/NAT(change destination), so that filtering rules can be
1937          * applied to IPVS. */
1938         {
1939                 .hook           = ip_vs_remote_request6,
1940                 .owner          = THIS_MODULE,
1941                 .pf             = NFPROTO_IPV6,
1942                 .hooknum        = NF_INET_LOCAL_IN,
1943                 .priority       = NF_IP6_PRI_NAT_SRC - 1,
1944         },
1945         /* Before ip_vs_in, change source only for VS/NAT */
1946         {
1947                 .hook           = ip_vs_local_reply6,
1948                 .owner          = THIS_MODULE,
1949                 .pf             = NFPROTO_IPV4,
1950                 .hooknum        = NF_INET_LOCAL_OUT,
1951                 .priority       = NF_IP6_PRI_NAT_DST + 1,
1952         },
1953         /* After mangle, schedule and forward local requests */
1954         {
1955                 .hook           = ip_vs_local_request6,
1956                 .owner          = THIS_MODULE,
1957                 .pf             = NFPROTO_IPV6,
1958                 .hooknum        = NF_INET_LOCAL_OUT,
1959                 .priority       = NF_IP6_PRI_NAT_DST + 2,
1960         },
1961         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1962          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1963         {
1964                 .hook           = ip_vs_forward_icmp_v6,
1965                 .owner          = THIS_MODULE,
1966                 .pf             = NFPROTO_IPV6,
1967                 .hooknum        = NF_INET_FORWARD,
1968                 .priority       = 99,
1969         },
1970         /* After packet filtering, change source only for VS/NAT */
1971         {
1972                 .hook           = ip_vs_reply6,
1973                 .owner          = THIS_MODULE,
1974                 .pf             = NFPROTO_IPV6,
1975                 .hooknum        = NF_INET_FORWARD,
1976                 .priority       = 100,
1977         },
1978 #endif
1979 };
1980 /*
1981  *      Initialize IP Virtual Server netns mem.
1982  */
1983 static int __net_init __ip_vs_init(struct net *net)
1984 {
1985         struct netns_ipvs *ipvs;
1986
1987         ipvs = net_generic(net, ip_vs_net_id);
1988         if (ipvs == NULL)
1989                 return -ENOMEM;
1990
1991         /* Hold the beast until a service is registerd */
1992         ipvs->enable = 0;
1993         ipvs->net = net;
1994         /* Counters used for creating unique names */
1995         ipvs->gen = atomic_read(&ipvs_netns_cnt);
1996         atomic_inc(&ipvs_netns_cnt);
1997         net->ipvs = ipvs;
1998
1999         if (ip_vs_estimator_net_init(net) < 0)
2000                 goto estimator_fail;
2001
2002         if (ip_vs_control_net_init(net) < 0)
2003                 goto control_fail;
2004
2005         if (ip_vs_protocol_net_init(net) < 0)
2006                 goto protocol_fail;
2007
2008         if (ip_vs_app_net_init(net) < 0)
2009                 goto app_fail;
2010
2011         if (ip_vs_conn_net_init(net) < 0)
2012                 goto conn_fail;
2013
2014         if (ip_vs_sync_net_init(net) < 0)
2015                 goto sync_fail;
2016
2017         printk(KERN_INFO "IPVS: Creating netns size=%zu id=%d\n",
2018                          sizeof(struct netns_ipvs), ipvs->gen);
2019         return 0;
2020 /*
2021  * Error handling
2022  */
2023
2024 sync_fail:
2025         ip_vs_conn_net_cleanup(net);
2026 conn_fail:
2027         ip_vs_app_net_cleanup(net);
2028 app_fail:
2029         ip_vs_protocol_net_cleanup(net);
2030 protocol_fail:
2031         ip_vs_control_net_cleanup(net);
2032 control_fail:
2033         ip_vs_estimator_net_cleanup(net);
2034 estimator_fail:
2035         net->ipvs = NULL;
2036         return -ENOMEM;
2037 }
2038
2039 static void __net_exit __ip_vs_cleanup(struct net *net)
2040 {
2041         ip_vs_service_net_cleanup(net); /* ip_vs_flush() with locks */
2042         ip_vs_conn_net_cleanup(net);
2043         ip_vs_app_net_cleanup(net);
2044         ip_vs_protocol_net_cleanup(net);
2045         ip_vs_control_net_cleanup(net);
2046         ip_vs_estimator_net_cleanup(net);
2047         IP_VS_DBG(2, "ipvs netns %d released\n", net_ipvs(net)->gen);
2048         net->ipvs = NULL;
2049 }
2050
2051 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
2052 {
2053         EnterFunction(2);
2054         net_ipvs(net)->enable = 0;      /* Disable packet reception */
2055         smp_wmb();
2056         ip_vs_sync_net_cleanup(net);
2057         LeaveFunction(2);
2058 }
2059
2060 static struct pernet_operations ipvs_core_ops = {
2061         .init = __ip_vs_init,
2062         .exit = __ip_vs_cleanup,
2063         .id   = &ip_vs_net_id,
2064         .size = sizeof(struct netns_ipvs),
2065 };
2066
2067 static struct pernet_operations ipvs_core_dev_ops = {
2068         .exit = __ip_vs_dev_cleanup,
2069 };
2070
2071 /*
2072  *      Initialize IP Virtual Server
2073  */
2074 static int __init ip_vs_init(void)
2075 {
2076         int ret;
2077
2078         ret = ip_vs_control_init();
2079         if (ret < 0) {
2080                 pr_err("can't setup control.\n");
2081                 goto exit;
2082         }
2083
2084         ip_vs_protocol_init();
2085
2086         ret = ip_vs_conn_init();
2087         if (ret < 0) {
2088                 pr_err("can't setup connection table.\n");
2089                 goto cleanup_protocol;
2090         }
2091
2092         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
2093         if (ret < 0)
2094                 goto cleanup_conn;
2095
2096         ret = register_pernet_device(&ipvs_core_dev_ops);
2097         if (ret < 0)
2098                 goto cleanup_sub;
2099
2100         ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2101         if (ret < 0) {
2102                 pr_err("can't register hooks.\n");
2103                 goto cleanup_dev;
2104         }
2105
2106         ret = ip_vs_register_nl_ioctl();
2107         if (ret < 0) {
2108                 pr_err("can't register netlink/ioctl.\n");
2109                 goto cleanup_hooks;
2110         }
2111
2112         pr_info("ipvs loaded.\n");
2113
2114         return ret;
2115
2116 cleanup_hooks:
2117         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2118 cleanup_dev:
2119         unregister_pernet_device(&ipvs_core_dev_ops);
2120 cleanup_sub:
2121         unregister_pernet_subsys(&ipvs_core_ops);
2122 cleanup_conn:
2123         ip_vs_conn_cleanup();
2124 cleanup_protocol:
2125         ip_vs_protocol_cleanup();
2126         ip_vs_control_cleanup();
2127 exit:
2128         return ret;
2129 }
2130
2131 static void __exit ip_vs_cleanup(void)
2132 {
2133         ip_vs_unregister_nl_ioctl();
2134         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2135         unregister_pernet_device(&ipvs_core_dev_ops);
2136         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
2137         ip_vs_conn_cleanup();
2138         ip_vs_protocol_cleanup();
2139         ip_vs_control_cleanup();
2140         pr_info("ipvs unloaded.\n");
2141 }
2142
2143 module_init(ip_vs_init);
2144 module_exit(ip_vs_cleanup);
2145 MODULE_LICENSE("GPL");