Pileus Git - ~andy/linux/blob - 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 = 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                 /*
 300                  * No template found or the dest of the connection
 301                  * template is not available.
 302                  * return *ignored=0 i.e. ICMP and NF_DROP
 303                  */
 304                 dest = svc->scheduler->schedule(svc, skb);
 305                 if (!dest) {
 306                         IP_VS_DBG(1, "p-schedule: no dest found.\n");
 307                         kfree(param.pe_data);
 308                         *ignored = 0;
 309                         return NULL;
 310                 }
 311
 312                 if (dst_port == svc->port && svc->port != FTPPORT)
 313                         dport = dest->port;
 314
 315                 /* Create a template
 316                  * This adds param.pe_data to the template,
 317                  * and thus param.pe_data will be destroyed
 318                  * when the template expires */
 319                 ct = ip_vs_conn_new(&param, &dest->addr, dport,
 320                                     IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
 321                 if (ct == NULL) {
 322                         kfree(param.pe_data);
 323                         *ignored = -1;
 324                         return NULL;
 325                 }
 326
 327                 ct->timeout = svc->timeout;
 328         } else {
 329                 /* set destination with the found template */
 330                 dest = ct->dest;
 331                 kfree(param.pe_data);
 332         }
 333
 334         dport = dst_port;
 335         if (dport == svc->port && dest->port)
 336                 dport = dest->port;
 337
 338         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 339                  && iph->protocol == IPPROTO_UDP) ?
 340                 IP_VS_CONN_F_ONE_PACKET : 0;
 341
 342         /*
 343          *    Create a new connection according to the template
 344          */
 345         ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol, &iph->saddr,
 346                               src_port, &iph->daddr, dst_port, &param);
 347
 348         cp = ip_vs_conn_new(&param, &dest->addr, dport, flags, dest, skb->mark);
 349         if (cp == NULL) {
 350                 ip_vs_conn_put(ct);
 351                 *ignored = -1;
 352                 return NULL;
 353         }
 354
 355         /*
 356          *    Add its control
 357          */
 358         ip_vs_control_add(cp, ct);
 359         ip_vs_conn_put(ct);
 360
 361         ip_vs_conn_stats(cp, svc);
 362         return cp;
 363 }
 364
 365
 366 /*
 367  *  IPVS main scheduling function
 368  *  It selects a server according to the virtual service, and
 369  *  creates a connection entry.
 370  *  Protocols supported: TCP, UDP
 371  *
 372  *  Usage of *ignored
 373  *
 374  * 1 :   protocol tried to schedule (eg. on SYN), found svc but the
 375  *       svc/scheduler decides that this packet should be accepted with
 376  *       NF_ACCEPT because it must not be scheduled.
 377  *
 378  * 0 :   scheduler can not find destination, so try bypass or
 379  *       return ICMP and then NF_DROP (ip_vs_leave).
 380  *
 381  * -1 :  scheduler tried to schedule but fatal error occurred, eg.
 382  *       ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
 383  *       failure such as missing Call-ID, ENOMEM on skb_linearize
 384  *       or pe_data. In this case we should return NF_DROP without
 385  *       any attempts to send ICMP with ip_vs_leave.
 386  */
 387 struct ip_vs_conn *
 388 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
 389                struct ip_vs_proto_data *pd, int *ignored,
 390                struct ip_vs_iphdr *iph)
 391 {
 392         struct ip_vs_protocol *pp = pd->pp;
 393         struct ip_vs_conn *cp = NULL;
 394         struct ip_vs_dest *dest;
 395         __be16 _ports[2], *pptr;
 396         unsigned int flags;
 397
 398         *ignored = 1;
 399         /*
 400          * IPv6 frags, only the first hit here.
 401          */
 402         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 403         if (pptr == NULL)
 404                 return NULL;
 405
 406         /*
 407          * FTPDATA needs this check when using local real server.
 408          * Never schedule Active FTPDATA connections from real server.
 409          * For LVS-NAT they must be already created. For other methods
 410          * with persistence the connection is created on SYN+ACK.
 411          */
 412         if (pptr[0] == FTPDATA) {
 413                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 414                               "Not scheduling FTPDATA");
 415                 return NULL;
 416         }
 417
 418         /*
 419          *    Do not schedule replies from local real server.
 420          */
 421         if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
 422             (cp = pp->conn_in_get(svc->af, skb, iph, 1))) {
 423                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 424                               "Not scheduling reply for existing connection");
 425                 __ip_vs_conn_put(cp);
 426                 return NULL;
 427         }
 428
 429         /*
 430          *    Persistent service
 431          */
 432         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 433                 return ip_vs_sched_persist(svc, skb, pptr[0], pptr[1], ignored,
 434                                            iph);
 435
 436         *ignored = 0;
 437
 438         /*
 439          *    Non-persistent service
 440          */
 441         if (!svc->fwmark && pptr[1] != svc->port) {
 442                 if (!svc->port)
 443                         pr_err("Schedule: port zero only supported "
 444                                "in persistent services, "
 445                                "check your ipvs configuration\n");
 446                 return NULL;
 447         }
 448
 449         dest = svc->scheduler->schedule(svc, skb);
 450         if (dest == NULL) {
 451                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 452                 return NULL;
 453         }
 454
 455         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 456                  && iph->protocol == IPPROTO_UDP) ?
 457                 IP_VS_CONN_F_ONE_PACKET : 0;
 458
 459         /*
 460          *    Create a connection entry.
 461          */
 462         {
 463                 struct ip_vs_conn_param p;
 464
 465                 ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol,
 466                                       &iph->saddr, pptr[0], &iph->daddr,
 467                                       pptr[1], &p);
 468                 cp = ip_vs_conn_new(&p, &dest->addr,
 469                                     dest->port ? dest->port : pptr[1],
 470                                     flags, dest, skb->mark);
 471                 if (!cp) {
 472                         *ignored = -1;
 473                         return NULL;
 474                 }
 475         }
 476
 477         IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
 478                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
 479                       ip_vs_fwd_tag(cp),
 480                       IP_VS_DBG_ADDR(svc->af, &cp->caddr), ntohs(cp->cport),
 481                       IP_VS_DBG_ADDR(svc->af, &cp->vaddr), ntohs(cp->vport),
 482                       IP_VS_DBG_ADDR(svc->af, &cp->daddr), ntohs(cp->dport),
 483                       cp->flags, atomic_read(&cp->refcnt));
 484
 485         ip_vs_conn_stats(cp, svc);
 486         return cp;
 487 }
 488
 489
 490 /*
 491  *  Pass or drop the packet.
 492  *  Called by ip_vs_in, when the virtual service is available but
 493  *  no destination is available for a new connection.
 494  */
 495 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 496                 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
 497 {
 498         __be16 _ports[2], *pptr;
 499 #ifdef CONFIG_SYSCTL
 500         struct net *net;
 501         struct netns_ipvs *ipvs;
 502         int unicast;
 503 #endif
 504
 505         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 506         if (pptr == NULL) {
 507                 ip_vs_service_put(svc);
 508                 return NF_DROP;
 509         }
 510
 511 #ifdef CONFIG_SYSCTL
 512         net = skb_net(skb);
 513
 514 #ifdef CONFIG_IP_VS_IPV6
 515         if (svc->af == AF_INET6)
 516                 unicast = ipv6_addr_type(&iph->daddr.in6) & IPV6_ADDR_UNICAST;
 517         else
 518 #endif
 519                 unicast = (inet_addr_type(net, iph->daddr.ip) == RTN_UNICAST);
 520
 521         /* if it is fwmark-based service, the cache_bypass sysctl is up
 522            and the destination is a non-local unicast, then create
 523            a cache_bypass connection entry */
 524         ipvs = net_ipvs(net);
 525         if (ipvs->sysctl_cache_bypass && svc->fwmark && unicast) {
 526                 int ret;
 527                 struct ip_vs_conn *cp;
 528                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
 529                                       iph->protocol == IPPROTO_UDP) ?
 530                                       IP_VS_CONN_F_ONE_PACKET : 0;
 531                 union nf_inet_addr daddr =  { .all = { 0, 0, 0, 0 } };
 532
 533                 ip_vs_service_put(svc);
 534
 535                 /* create a new connection entry */
 536                 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
 537                 {
 538                         struct ip_vs_conn_param p;
 539                         ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol,
 540                                               &iph->saddr, pptr[0],
 541                                               &iph->daddr, pptr[1], &p);
 542                         cp = ip_vs_conn_new(&p, &daddr, 0,
 543                                             IP_VS_CONN_F_BYPASS | flags,
 544                                             NULL, skb->mark);
 545                         if (!cp)
 546                                 return NF_DROP;
 547                 }
 548
 549                 /* statistics */
 550                 ip_vs_in_stats(cp, skb);
 551
 552                 /* set state */
 553                 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
 554
 555                 /* transmit the first SYN packet */
 556                 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
 557                 /* do not touch skb anymore */
 558
 559                 atomic_inc(&cp->in_pkts);
 560                 ip_vs_conn_put(cp);
 561                 return ret;
 562         }
 563 #endif
 564
 565         /*
 566          * When the virtual ftp service is presented, packets destined
 567          * for other services on the VIP may get here (except services
 568          * listed in the ipvs table), pass the packets, because it is
 569          * not ipvs job to decide to drop the packets.
 570          */
 571         if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT)) {
 572                 ip_vs_service_put(svc);
 573                 return NF_ACCEPT;
 574         }
 575
 576         ip_vs_service_put(svc);
 577
 578         /*
 579          * Notify the client that the destination is unreachable, and
 580          * release the socket buffer.
 581          * Since it is in IP layer, the TCP socket is not actually
 582          * created, the TCP RST packet cannot be sent, instead that
 583          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 584          */
 585 #ifdef CONFIG_IP_VS_IPV6
 586         if (svc->af == AF_INET6) {
 587                 if (!skb->dev) {
 588                         struct net *net = dev_net(skb_dst(skb)->dev);
 589
 590                         skb->dev = net->loopback_dev;
 591                 }
 592                 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 593         } else
 594 #endif
 595                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 596
 597         return NF_DROP;
 598 }
 599
 600 #ifdef CONFIG_SYSCTL
 601
 602 static int sysctl_snat_reroute(struct sk_buff *skb)
 603 {
 604         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 605         return ipvs->sysctl_snat_reroute;
 606 }
 607
 608 static int sysctl_nat_icmp_send(struct net *net)
 609 {
 610         struct netns_ipvs *ipvs = net_ipvs(net);
 611         return ipvs->sysctl_nat_icmp_send;
 612 }
 613
 614 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
 615 {
 616         return ipvs->sysctl_expire_nodest_conn;
 617 }
 618
 619 #else
 620
 621 static int sysctl_snat_reroute(struct sk_buff *skb) { return 0; }
 622 static int sysctl_nat_icmp_send(struct net *net) { return 0; }
 623 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
 624
 625 #endif
 626
 627 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 628 {
 629         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 630 }
 631
 632 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
 633 {
 634         if (NF_INET_LOCAL_IN == hooknum)
 635                 return IP_DEFRAG_VS_IN;
 636         if (NF_INET_FORWARD == hooknum)
 637                 return IP_DEFRAG_VS_FWD;
 638         return IP_DEFRAG_VS_OUT;
 639 }
 640
 641 static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user)
 642 {
 643         int err = ip_defrag(skb, user);
 644
 645         if (!err)
 646                 ip_send_check(ip_hdr(skb));
 647
 648         return err;
 649 }
 650
 651 static int ip_vs_route_me_harder(int af, struct sk_buff *skb)
 652 {
 653 #ifdef CONFIG_IP_VS_IPV6
 654         if (af == AF_INET6) {
 655                 if (sysctl_snat_reroute(skb) && ip6_route_me_harder(skb) != 0)
 656                         return 1;
 657         } else
 658 #endif
 659                 if ((sysctl_snat_reroute(skb) ||
 660                      skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
 661                     ip_route_me_harder(skb, RTN_LOCAL) != 0)
 662                         return 1;
 663
 664         return 0;
 665 }
 666
 667 /*
 668  * Packet has been made sufficiently writable in caller
 669  * - inout: 1=in->out, 0=out->in
 670  */
 671 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 672                     struct ip_vs_conn *cp, int inout)
 673 {
 674         struct iphdr *iph        = ip_hdr(skb);
 675         unsigned int icmp_offset = iph->ihl*4;
 676         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
 677                                                       icmp_offset);
 678         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 679
 680         if (inout) {
 681                 iph->saddr = cp->vaddr.ip;
 682                 ip_send_check(iph);
 683                 ciph->daddr = cp->vaddr.ip;
 684                 ip_send_check(ciph);
 685         } else {
 686                 iph->daddr = cp->daddr.ip;
 687                 ip_send_check(iph);
 688                 ciph->saddr = cp->daddr.ip;
 689                 ip_send_check(ciph);
 690         }
 691
 692         /* the TCP/UDP/SCTP port */
 693         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
 694             IPPROTO_SCTP == ciph->protocol) {
 695                 __be16 *ports = (void *)ciph + ciph->ihl*4;
 696
 697                 if (inout)
 698                         ports[1] = cp->vport;
 699                 else
 700                         ports[0] = cp->dport;
 701         }
 702
 703         /* And finally the ICMP checksum */
 704         icmph->checksum = 0;
 705         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 706         skb->ip_summed = CHECKSUM_UNNECESSARY;
 707
 708         if (inout)
 709                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 710                         "Forwarding altered outgoing ICMP");
 711         else
 712                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 713                         "Forwarding altered incoming ICMP");
 714 }
 715
 716 #ifdef CONFIG_IP_VS_IPV6
 717 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
 718                     struct ip_vs_conn *cp, int inout)
 719 {
 720         struct ipv6hdr *iph      = ipv6_hdr(skb);
 721         unsigned int icmp_offset = 0;
 722         unsigned int offs        = 0; /* header offset*/
 723         int protocol;
 724         struct icmp6hdr *icmph;
 725         struct ipv6hdr *ciph;
 726         unsigned short fragoffs;
 727
 728         ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
 729         icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
 730         offs = icmp_offset + sizeof(struct icmp6hdr);
 731         ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
 732
 733         protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
 734
 735         if (inout) {
 736                 iph->saddr = cp->vaddr.in6;
 737                 ciph->daddr = cp->vaddr.in6;
 738         } else {
 739                 iph->daddr = cp->daddr.in6;
 740                 ciph->saddr = cp->daddr.in6;
 741         }
 742
 743         /* the TCP/UDP/SCTP port */
 744         if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 745                           IPPROTO_SCTP == protocol)) {
 746                 __be16 *ports = (void *)(skb_network_header(skb) + offs);
 747
 748                 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
 749                               ntohs(inout ? ports[1] : ports[0]),
 750                               ntohs(inout ? cp->vport : cp->dport));
 751                 if (inout)
 752                         ports[1] = cp->vport;
 753                 else
 754                         ports[0] = cp->dport;
 755         }
 756
 757         /* And finally the ICMP checksum */
 758         icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
 759                                               skb->len - icmp_offset,
 760                                               IPPROTO_ICMPV6, 0);
 761         skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
 762         skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
 763         skb->ip_summed = CHECKSUM_PARTIAL;
 764
 765         if (inout)
 766                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 767                               (void *)ciph - (void *)iph,
 768                               "Forwarding altered outgoing ICMPv6");
 769         else
 770                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 771                               (void *)ciph - (void *)iph,
 772                               "Forwarding altered incoming ICMPv6");
 773 }
 774 #endif
 775
 776 /* Handle relevant response ICMP messages - forward to the right
 777  * destination host.
 778  */
 779 static int handle_response_icmp(int af, struct sk_buff *skb,
 780                                 union nf_inet_addr *snet,
 781                                 __u8 protocol, struct ip_vs_conn *cp,
 782                                 struct ip_vs_protocol *pp,
 783                                 unsigned int offset, unsigned int ihl)
 784 {
 785         unsigned int verdict = NF_DROP;
 786
 787         if (IP_VS_FWD_METHOD(cp) != 0) {
 788                 pr_err("shouldn't reach here, because the box is on the "
 789                        "half connection in the tun/dr module.\n");
 790         }
 791
 792         /* Ensure the checksum is correct */
 793         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 794                 /* Failed checksum! */
 795                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
 796                               IP_VS_DBG_ADDR(af, snet));
 797                 goto out;
 798         }
 799
 800         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 801             IPPROTO_SCTP == protocol)
 802                 offset += 2 * sizeof(__u16);
 803         if (!skb_make_writable(skb, offset))
 804                 goto out;
 805
 806 #ifdef CONFIG_IP_VS_IPV6
 807         if (af == AF_INET6)
 808                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
 809         else
 810 #endif
 811                 ip_vs_nat_icmp(skb, pp, cp, 1);
 812
 813         if (ip_vs_route_me_harder(af, skb))
 814                 goto out;
 815
 816         /* do the statistics and put it back */
 817         ip_vs_out_stats(cp, skb);
 818
 819         skb->ipvs_property = 1;
 820         if (!(cp->flags & IP_VS_CONN_F_NFCT))
 821                 ip_vs_notrack(skb);
 822         else
 823                 ip_vs_update_conntrack(skb, cp, 0);
 824         verdict = NF_ACCEPT;
 825
 826 out:
 827         __ip_vs_conn_put(cp);
 828
 829         return verdict;
 830 }
 831
 832 /*
 833  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 834  *      Find any that might be relevant, check against existing connections.
 835  *      Currently handles error types - unreachable, quench, ttl exceeded.
 836  */
 837 static int ip_vs_out_icmp(struct sk_buff *skb, int *related,
 838                           unsigned int hooknum)
 839 {
 840         struct iphdr *iph;
 841         struct icmphdr  _icmph, *ic;
 842         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 843         struct ip_vs_iphdr ciph;
 844         struct ip_vs_conn *cp;
 845         struct ip_vs_protocol *pp;
 846         unsigned int offset, ihl;
 847         union nf_inet_addr snet;
 848
 849         *related = 1;
 850
 851         /* reassemble IP fragments */
 852         if (ip_is_fragment(ip_hdr(skb))) {
 853                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
 854                         return NF_STOLEN;
 855         }
 856
 857         iph = ip_hdr(skb);
 858         offset = ihl = iph->ihl * 4;
 859         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 860         if (ic == NULL)
 861                 return NF_DROP;
 862
 863         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
 864                   ic->type, ntohs(icmp_id(ic)),
 865                   &iph->saddr, &iph->daddr);
 866
 867         /*
 868          * Work through seeing if this is for us.
 869          * These checks are supposed to be in an order that means easy
 870          * things are checked first to speed up processing.... however
 871          * this means that some packets will manage to get a long way
 872          * down this stack and then be rejected, but that's life.
 873          */
 874         if ((ic->type != ICMP_DEST_UNREACH) &&
 875             (ic->type != ICMP_SOURCE_QUENCH) &&
 876             (ic->type != ICMP_TIME_EXCEEDED)) {
 877                 *related = 0;
 878                 return NF_ACCEPT;
 879         }
 880
 881         /* Now find the contained IP header */
 882         offset += sizeof(_icmph);
 883         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 884         if (cih == NULL)
 885                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 886
 887         pp = ip_vs_proto_get(cih->protocol);
 888         if (!pp)
 889                 return NF_ACCEPT;
 890
 891         /* Is the embedded protocol header present? */
 892         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 893                      pp->dont_defrag))
 894                 return NF_ACCEPT;
 895
 896         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
 897                       "Checking outgoing ICMP for");
 898
 899         ip_vs_fill_ip4hdr(cih, &ciph);
 900         ciph.len += offset;
 901         /* The embedded headers contain source and dest in reverse order */
 902         cp = pp->conn_out_get(AF_INET, skb, &ciph, 1);
 903         if (!cp)
 904                 return NF_ACCEPT;
 905
 906         snet.ip = iph->saddr;
 907         return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
 908                                     pp, ciph.len, ihl);
 909 }
 910
 911 #ifdef CONFIG_IP_VS_IPV6
 912 static int ip_vs_out_icmp_v6(struct sk_buff *skb, int *related,
 913                              unsigned int hooknum, struct ip_vs_iphdr *ipvsh)
 914 {
 915         struct icmp6hdr _icmph, *ic;
 916         struct ipv6hdr _ip6h, *ip6h; /* The ip header contained within ICMP */
 917         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
 918         struct ip_vs_conn *cp;
 919         struct ip_vs_protocol *pp;
 920         union nf_inet_addr snet;
 921         unsigned int writable;
 922
 923         *related = 1;
 924         ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph, ipvsh);
 925         if (ic == NULL)
 926                 return NF_DROP;
 927
 928         /*
 929          * Work through seeing if this is for us.
 930          * These checks are supposed to be in an order that means easy
 931          * things are checked first to speed up processing.... however
 932          * this means that some packets will manage to get a long way
 933          * down this stack and then be rejected, but that's life.
 934          */
 935         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
 936                 *related = 0;
 937                 return NF_ACCEPT;
 938         }
 939         /* Fragment header that is before ICMP header tells us that:
 940          * it's not an error message since they can't be fragmented.
 941          */
 942         if (ipvsh->flags & IP6_FH_F_FRAG)
 943                 return NF_DROP;
 944
 945         IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
 946                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
 947                   &ipvsh->saddr, &ipvsh->daddr);
 948
 949         /* Now find the contained IP header */
 950         ciph.len = ipvsh->len + sizeof(_icmph);
 951         ip6h = skb_header_pointer(skb, ciph.len, sizeof(_ip6h), &_ip6h);
 952         if (ip6h == NULL)
 953                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 954         ciph.saddr.in6 = ip6h->saddr; /* conn_out_get() handles reverse order */
 955         ciph.daddr.in6 = ip6h->daddr;
 956         /* skip possible IPv6 exthdrs of contained IPv6 packet */
 957         ciph.protocol = ipv6_find_hdr(skb, &ciph.len, -1, &ciph.fragoffs, NULL);
 958         if (ciph.protocol < 0)
 959                 return NF_ACCEPT; /* Contained IPv6 hdr looks wrong, ignore */
 960
 961         pp = ip_vs_proto_get(ciph.protocol);
 962         if (!pp)
 963                 return NF_ACCEPT;
 964
 965         /* The embedded headers contain source and dest in reverse order */
 966         cp = pp->conn_out_get(AF_INET6, skb, &ciph, 1);
 967         if (!cp)
 968                 return NF_ACCEPT;
 969
 970         snet.in6 = ciph.saddr.in6;
 971         writable = ciph.len;
 972         return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
 973                                     pp, writable, sizeof(struct ipv6hdr));
 974 }
 975 #endif
 976
 977 /*
 978  * Check if sctp chunc is ABORT chunk
 979  */
 980 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
 981 {
 982         sctp_chunkhdr_t *sch, schunk;
 983         sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
 984                         sizeof(schunk), &schunk);
 985         if (sch == NULL)
 986                 return 0;
 987         if (sch->type == SCTP_CID_ABORT)
 988                 return 1;
 989         return 0;
 990 }
 991
 992 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
 993 {
 994         struct tcphdr _tcph, *th;
 995
 996         th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
 997         if (th == NULL)
 998                 return 0;
 999         return th->rst;
1000 }
1001
1002 /* Handle response packets: rewrite addresses and send away...
1003  */
1004 static unsigned int
1005 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1006                 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph)
1007 {
1008         struct ip_vs_protocol *pp = pd->pp;
1009
1010         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Outgoing packet");
1011
1012         if (!skb_make_writable(skb, iph->len))
1013                 goto drop;
1014
1015         /* mangle the packet */
1016         if (pp->snat_handler && !pp->snat_handler(skb, pp, cp, iph))
1017                 goto drop;
1018
1019 #ifdef CONFIG_IP_VS_IPV6
1020         if (af == AF_INET6)
1021                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1022         else
1023 #endif
1024         {
1025                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1026                 ip_send_check(ip_hdr(skb));
1027         }
1028
1029         /*
1030          * nf_iterate does not expect change in the skb->dst->dev.
1031          * It looks like it is not fatal to enable this code for hooks
1032          * where our handlers are at the end of the chain list and
1033          * when all next handlers use skb->dst->dev and not outdev.
1034          * It will definitely route properly the inout NAT traffic
1035          * when multiple paths are used.
1036          */
1037
1038         /* For policy routing, packets originating from this
1039          * machine itself may be routed differently to packets
1040          * passing through.  We want this packet to be routed as
1041          * if it came from this machine itself.  So re-compute
1042          * the routing information.
1043          */
1044         if (ip_vs_route_me_harder(af, skb))
1045                 goto drop;
1046
1047         IP_VS_DBG_PKT(10, af, pp, skb, 0, "After SNAT");
1048
1049         ip_vs_out_stats(cp, skb);
1050         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1051         skb->ipvs_property = 1;
1052         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1053                 ip_vs_notrack(skb);
1054         else
1055                 ip_vs_update_conntrack(skb, cp, 0);
1056         ip_vs_conn_put(cp);
1057
1058         LeaveFunction(11);
1059         return NF_ACCEPT;
1060
1061 drop:
1062         ip_vs_conn_put(cp);
1063         kfree_skb(skb);
1064         LeaveFunction(11);
1065         return NF_STOLEN;
1066 }
1067
1068 /*
1069  *      Check if outgoing packet belongs to the established ip_vs_conn.
1070  */
1071 static unsigned int
1072 ip_vs_out(unsigned int hooknum, struct sk_buff *skb, int af)
1073 {
1074         struct net *net = NULL;
1075         struct ip_vs_iphdr iph;
1076         struct ip_vs_protocol *pp;
1077         struct ip_vs_proto_data *pd;
1078         struct ip_vs_conn *cp;
1079
1080         EnterFunction(11);
1081
1082         /* Already marked as IPVS request or reply? */
1083         if (skb->ipvs_property)
1084                 return NF_ACCEPT;
1085
1086         /* Bad... Do not break raw sockets */
1087         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1088                      af == AF_INET)) {
1089                 struct sock *sk = skb->sk;
1090                 struct inet_sock *inet = inet_sk(skb->sk);
1091
1092                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1093                         return NF_ACCEPT;
1094         }
1095
1096         if (unlikely(!skb_dst(skb)))
1097                 return NF_ACCEPT;
1098
1099         net = skb_net(skb);
1100         if (!net_ipvs(net)->enable)
1101                 return NF_ACCEPT;
1102
1103         ip_vs_fill_iph_skb(af, skb, &iph);
1104 #ifdef CONFIG_IP_VS_IPV6
1105         if (af == AF_INET6) {
1106                 if (!iph.fragoffs && skb_nfct_reasm(skb)) {
1107                         struct sk_buff *reasm = skb_nfct_reasm(skb);
1108                         /* Save fw mark for coming frags */
1109                         reasm->ipvs_property = 1;
1110                         reasm->mark = skb->mark;
1111                 }
1112                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1113                         int related;
1114                         int verdict = ip_vs_out_icmp_v6(skb, &related,
1115                                                         hooknum, &iph);
1116
1117                         if (related)
1118                                 return verdict;
1119                 }
1120         } else
1121 #endif
1122                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1123                         int related;
1124                         int verdict = ip_vs_out_icmp(skb, &related, hooknum);
1125
1126                         if (related)
1127                                 return verdict;
1128                 }
1129
1130         pd = ip_vs_proto_data_get(net, iph.protocol);
1131         if (unlikely(!pd))
1132                 return NF_ACCEPT;
1133         pp = pd->pp;
1134
1135         /* reassemble IP fragments */
1136 #ifdef CONFIG_IP_VS_IPV6
1137         if (af == AF_INET)
1138 #endif
1139                 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1140                         if (ip_vs_gather_frags(skb,
1141                                                ip_vs_defrag_user(hooknum)))
1142                                 return NF_STOLEN;
1143
1144                         ip_vs_fill_ip4hdr(skb_network_header(skb), &iph);
1145                 }
1146
1147         /*
1148          * Check if the packet belongs to an existing entry
1149          */
1150         cp = pp->conn_out_get(af, skb, &iph, 0);
1151
1152         if (likely(cp))
1153                 return handle_response(af, skb, pd, cp, &iph);
1154         if (sysctl_nat_icmp_send(net) &&
1155             (pp->protocol == IPPROTO_TCP ||
1156              pp->protocol == IPPROTO_UDP ||
1157              pp->protocol == IPPROTO_SCTP)) {
1158                 __be16 _ports[2], *pptr;
1159
1160                 pptr = frag_safe_skb_hp(skb, iph.len,
1161                                          sizeof(_ports), _ports, &iph);
1162                 if (pptr == NULL)
1163                         return NF_ACCEPT;       /* Not for me */
1164                 if (ip_vs_lookup_real_service(net, af, iph.protocol,
1165                                               &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         unsigned int verdict;
1224
1225         /* Disable BH in LOCAL_OUT until all places are fixed */
1226         local_bh_disable();
1227         verdict = ip_vs_out(hooknum, skb, AF_INET);
1228         local_bh_enable();
1229         return verdict;
1230 }
1231
1232 #ifdef CONFIG_IP_VS_IPV6
1233
1234 /*
1235  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1236  *      used only for VS/NAT.
1237  *      Check if packet is reply for established ip_vs_conn.
1238  */
1239 static unsigned int
1240 ip_vs_reply6(unsigned int hooknum, struct sk_buff *skb,
1241              const struct net_device *in, const struct net_device *out,
1242              int (*okfn)(struct sk_buff *))
1243 {
1244         return ip_vs_out(hooknum, skb, AF_INET6);
1245 }
1246
1247 /*
1248  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1249  *      Check if packet is reply for established ip_vs_conn.
1250  */
1251 static unsigned int
1252 ip_vs_local_reply6(unsigned int hooknum, struct sk_buff *skb,
1253                    const struct net_device *in, const struct net_device *out,
1254                    int (*okfn)(struct sk_buff *))
1255 {
1256         unsigned int verdict;
1257
1258         /* Disable BH in LOCAL_OUT until all places are fixed */
1259         local_bh_disable();
1260         verdict = ip_vs_out(hooknum, skb, AF_INET6);
1261         local_bh_enable();
1262         return verdict;
1263 }
1264
1265 #endif
1266
1267 /*
1268  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1269  *      Find any that might be relevant, check against existing connections,
1270  *      forward to the right destination host if relevant.
1271  *      Currently handles error types - unreachable, quench, ttl exceeded.
1272  */
1273 static int
1274 ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum)
1275 {
1276         struct net *net = NULL;
1277         struct iphdr *iph;
1278         struct icmphdr  _icmph, *ic;
1279         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1280         struct ip_vs_iphdr ciph;
1281         struct ip_vs_conn *cp;
1282         struct ip_vs_protocol *pp;
1283         struct ip_vs_proto_data *pd;
1284         unsigned int offset, offset2, ihl, verdict;
1285         bool ipip;
1286
1287         *related = 1;
1288
1289         /* reassemble IP fragments */
1290         if (ip_is_fragment(ip_hdr(skb))) {
1291                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
1292                         return NF_STOLEN;
1293         }
1294
1295         iph = ip_hdr(skb);
1296         offset = ihl = iph->ihl * 4;
1297         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1298         if (ic == NULL)
1299                 return NF_DROP;
1300
1301         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1302                   ic->type, ntohs(icmp_id(ic)),
1303                   &iph->saddr, &iph->daddr);
1304
1305         /*
1306          * Work through seeing if this is for us.
1307          * These checks are supposed to be in an order that means easy
1308          * things are checked first to speed up processing.... however
1309          * this means that some packets will manage to get a long way
1310          * down this stack and then be rejected, but that's life.
1311          */
1312         if ((ic->type != ICMP_DEST_UNREACH) &&
1313             (ic->type != ICMP_SOURCE_QUENCH) &&
1314             (ic->type != ICMP_TIME_EXCEEDED)) {
1315                 *related = 0;
1316                 return NF_ACCEPT;
1317         }
1318
1319         /* Now find the contained IP header */
1320         offset += sizeof(_icmph);
1321         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1322         if (cih == NULL)
1323                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1324
1325         net = skb_net(skb);
1326
1327         /* Special case for errors for IPIP packets */
1328         ipip = false;
1329         if (cih->protocol == IPPROTO_IPIP) {
1330                 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1331                         return NF_ACCEPT;
1332                 /* Error for our IPIP must arrive at LOCAL_IN */
1333                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1334                         return NF_ACCEPT;
1335                 offset += cih->ihl * 4;
1336                 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1337                 if (cih == NULL)
1338                         return NF_ACCEPT; /* The packet looks wrong, ignore */
1339                 ipip = true;
1340         }
1341
1342         pd = ip_vs_proto_data_get(net, cih->protocol);
1343         if (!pd)
1344                 return NF_ACCEPT;
1345         pp = pd->pp;
1346
1347         /* Is the embedded protocol header present? */
1348         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1349                      pp->dont_defrag))
1350                 return NF_ACCEPT;
1351
1352         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1353                       "Checking incoming ICMP for");
1354
1355         offset2 = offset;
1356         ip_vs_fill_ip4hdr(cih, &ciph);
1357         ciph.len += offset;
1358         offset = ciph.len;
1359         /* The embedded headers contain source and dest in reverse order.
1360          * For IPIP this is error for request, not for reply.
1361          */
1362         cp = pp->conn_in_get(AF_INET, skb, &ciph, ipip ? 0 : 1);
1363         if (!cp)
1364                 return NF_ACCEPT;
1365
1366         verdict = NF_DROP;
1367
1368         /* Ensure the checksum is correct */
1369         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1370                 /* Failed checksum! */
1371                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1372                           &iph->saddr);
1373                 goto out;
1374         }
1375
1376         if (ipip) {
1377                 __be32 info = ic->un.gateway;
1378
1379                 /* Update the MTU */
1380                 if (ic->type == ICMP_DEST_UNREACH &&
1381                     ic->code == ICMP_FRAG_NEEDED) {
1382                         struct ip_vs_dest *dest = cp->dest;
1383                         u32 mtu = ntohs(ic->un.frag.mtu);
1384
1385                         /* Strip outer IP and ICMP, go to IPIP header */
1386                         __skb_pull(skb, ihl + sizeof(_icmph));
1387                         offset2 -= ihl + sizeof(_icmph);
1388                         skb_reset_network_header(skb);
1389                         IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1390                                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1391                         rcu_read_lock();
1392                         ipv4_update_pmtu(skb, dev_net(skb->dev),
1393                                          mtu, 0, 0, 0, 0);
1394                         rcu_read_unlock();
1395                         /* Client uses PMTUD? */
1396                         if (!(cih->frag_off & htons(IP_DF)))
1397                                 goto ignore_ipip;
1398                         /* Prefer the resulting PMTU */
1399                         if (dest) {
1400                                 spin_lock(&dest->dst_lock);
1401                                 if (dest->dst_cache)
1402                                         mtu = dst_mtu(dest->dst_cache);
1403                                 spin_unlock(&dest->dst_lock);
1404                         }
1405                         if (mtu > 68 + sizeof(struct iphdr))
1406                                 mtu -= sizeof(struct iphdr);
1407                         info = htonl(mtu);
1408                 }
1409                 /* Strip outer IP, ICMP and IPIP, go to IP header of
1410                  * original request.
1411                  */
1412                 __skb_pull(skb, offset2);
1413                 skb_reset_network_header(skb);
1414                 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1415                         &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1416                         ic->type, ic->code, ntohl(info));
1417                 icmp_send(skb, ic->type, ic->code, info);
1418                 /* ICMP can be shorter but anyways, account it */
1419                 ip_vs_out_stats(cp, skb);
1420
1421 ignore_ipip:
1422                 consume_skb(skb);
1423                 verdict = NF_STOLEN;
1424                 goto out;
1425         }
1426
1427         /* do the statistics and put it back */
1428         ip_vs_in_stats(cp, skb);
1429         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol)
1430                 offset += 2 * sizeof(__u16);
1431         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1432
1433 out:
1434         __ip_vs_conn_put(cp);
1435
1436         return verdict;
1437 }
1438
1439 #ifdef CONFIG_IP_VS_IPV6
1440 static int ip_vs_in_icmp_v6(struct sk_buff *skb, int *related,
1441                             unsigned int hooknum, struct ip_vs_iphdr *iph)
1442 {
1443         struct net *net = NULL;
1444         struct ipv6hdr _ip6h, *ip6h;
1445         struct icmp6hdr _icmph, *ic;
1446         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1447         struct ip_vs_conn *cp;
1448         struct ip_vs_protocol *pp;
1449         struct ip_vs_proto_data *pd;
1450         unsigned int offs_ciph, writable, verdict;
1451
1452         *related = 1;
1453
1454         ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph, iph);
1455         if (ic == NULL)
1456                 return NF_DROP;
1457
1458         /*
1459          * Work through seeing if this is for us.
1460          * These checks are supposed to be in an order that means easy
1461          * things are checked first to speed up processing.... however
1462          * this means that some packets will manage to get a long way
1463          * down this stack and then be rejected, but that's life.
1464          */
1465         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1466                 *related = 0;
1467                 return NF_ACCEPT;
1468         }
1469         /* Fragment header that is before ICMP header tells us that:
1470          * it's not an error message since they can't be fragmented.
1471          */
1472         if (iph->flags & IP6_FH_F_FRAG)
1473                 return NF_DROP;
1474
1475         IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1476                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1477                   &iph->saddr, &iph->daddr);
1478
1479         /* Now find the contained IP header */
1480         ciph.len = iph->len + sizeof(_icmph);
1481         offs_ciph = ciph.len; /* Save ip header offset */
1482         ip6h = skb_header_pointer(skb, ciph.len, sizeof(_ip6h), &_ip6h);
1483         if (ip6h == NULL)
1484                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1485         ciph.saddr.in6 = ip6h->saddr; /* conn_in_get() handles reverse order */
1486         ciph.daddr.in6 = ip6h->daddr;
1487         /* skip possible IPv6 exthdrs of contained IPv6 packet */
1488         ciph.protocol = ipv6_find_hdr(skb, &ciph.len, -1, &ciph.fragoffs, NULL);
1489         if (ciph.protocol < 0)
1490                 return NF_ACCEPT; /* Contained IPv6 hdr looks wrong, ignore */
1491
1492         net = skb_net(skb);
1493         pd = ip_vs_proto_data_get(net, ciph.protocol);
1494         if (!pd)
1495                 return NF_ACCEPT;
1496         pp = pd->pp;
1497
1498         /* Cannot handle fragmented embedded protocol */
1499         if (ciph.fragoffs)
1500                 return NF_ACCEPT;
1501
1502         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offs_ciph,
1503                       "Checking incoming ICMPv6 for");
1504
1505         /* The embedded headers contain source and dest in reverse order
1506          * if not from localhost
1507          */
1508         cp = pp->conn_in_get(AF_INET6, skb, &ciph,
1509                              (hooknum == NF_INET_LOCAL_OUT) ? 0 : 1);
1510
1511         if (!cp)
1512                 return NF_ACCEPT;
1513         /* VS/TUN, VS/DR and LOCALNODE just let it go */
1514         if ((hooknum == NF_INET_LOCAL_OUT) &&
1515             (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1516                 __ip_vs_conn_put(cp);
1517                 return NF_ACCEPT;
1518         }
1519
1520         /* do the statistics and put it back */
1521         ip_vs_in_stats(cp, skb);
1522
1523         /* Need to mangle contained IPv6 header in ICMPv6 packet */
1524         writable = ciph.len;
1525         if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1526             IPPROTO_SCTP == ciph.protocol)
1527                 writable += 2 * sizeof(__u16); /* Also mangle ports */
1528
1529         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, writable, hooknum, &ciph);
1530
1531         __ip_vs_conn_put(cp);
1532
1533         return verdict;
1534 }
1535 #endif
1536
1537
1538 /*
1539  *      Check if it's for virtual services, look it up,
1540  *      and send it on its way...
1541  */
1542 static unsigned int
1543 ip_vs_in(unsigned int hooknum, struct sk_buff *skb, int af)
1544 {
1545         struct net *net;
1546         struct ip_vs_iphdr iph;
1547         struct ip_vs_protocol *pp;
1548         struct ip_vs_proto_data *pd;
1549         struct ip_vs_conn *cp;
1550         int ret, pkts;
1551         struct netns_ipvs *ipvs;
1552
1553         /* Already marked as IPVS request or reply? */
1554         if (skb->ipvs_property)
1555                 return NF_ACCEPT;
1556
1557         /*
1558          *      Big tappo:
1559          *      - remote client: only PACKET_HOST
1560          *      - route: used for struct net when skb->dev is unset
1561          */
1562         if (unlikely((skb->pkt_type != PACKET_HOST &&
1563                       hooknum != NF_INET_LOCAL_OUT) ||
1564                      !skb_dst(skb))) {
1565                 ip_vs_fill_iph_skb(af, skb, &iph);
1566                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1567                               " ignored in hook %u\n",
1568                               skb->pkt_type, iph.protocol,
1569                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1570                 return NF_ACCEPT;
1571         }
1572         /* ipvs enabled in this netns ? */
1573         net = skb_net(skb);
1574         if (!net_ipvs(net)->enable)
1575                 return NF_ACCEPT;
1576
1577         ip_vs_fill_iph_skb(af, skb, &iph);
1578
1579         /* Bad... Do not break raw sockets */
1580         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1581                      af == AF_INET)) {
1582                 struct sock *sk = skb->sk;
1583                 struct inet_sock *inet = inet_sk(skb->sk);
1584
1585                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1586                         return NF_ACCEPT;
1587         }
1588
1589 #ifdef CONFIG_IP_VS_IPV6
1590         if (af == AF_INET6) {
1591                 if (!iph.fragoffs && skb_nfct_reasm(skb)) {
1592                         struct sk_buff *reasm = skb_nfct_reasm(skb);
1593                         /* Save fw mark for coming frags. */
1594                         reasm->ipvs_property = 1;
1595                         reasm->mark = skb->mark;
1596                 }
1597                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1598                         int related;
1599                         int verdict = ip_vs_in_icmp_v6(skb, &related, hooknum,
1600                                                        &iph);
1601
1602                         if (related)
1603                                 return verdict;
1604                 }
1605         } else
1606 #endif
1607                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1608                         int related;
1609                         int verdict = ip_vs_in_icmp(skb, &related, hooknum);
1610
1611                         if (related)
1612                                 return verdict;
1613                 }
1614
1615         /* Protocol supported? */
1616         pd = ip_vs_proto_data_get(net, iph.protocol);
1617         if (unlikely(!pd))
1618                 return NF_ACCEPT;
1619         pp = pd->pp;
1620         /*
1621          * Check if the packet belongs to an existing connection entry
1622          */
1623         cp = pp->conn_in_get(af, skb, &iph, 0);
1624         if (unlikely(!cp) && !iph.fragoffs) {
1625                 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1626                  * replayed fragment zero will already have created the cp
1627                  */
1628                 int v;
1629
1630                 /* Schedule and create new connection entry into &cp */
1631                 if (!pp->conn_schedule(af, skb, pd, &v, &cp, &iph))
1632                         return v;
1633         }
1634
1635         if (unlikely(!cp)) {
1636                 /* sorry, all this trouble for a no-hit :) */
1637                 IP_VS_DBG_PKT(12, af, pp, skb, 0,
1638                               "ip_vs_in: packet continues traversal as normal");
1639                 if (iph.fragoffs && !skb_nfct_reasm(skb)) {
1640                         /* Fragment that couldn't be mapped to a conn entry
1641                          * and don't have any pointer to a reasm skb
1642                          * is missing module nf_defrag_ipv6
1643                          */
1644                         IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1645                         IP_VS_DBG_PKT(7, af, pp, skb, 0, "unhandled fragment");
1646                 }
1647                 return NF_ACCEPT;
1648         }
1649
1650         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Incoming packet");
1651         ipvs = net_ipvs(net);
1652         /* Check the server status */
1653         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1654                 /* the destination server is not available */
1655
1656                 if (sysctl_expire_nodest_conn(ipvs)) {
1657                         /* try to expire the connection immediately */
1658                         ip_vs_conn_expire_now(cp);
1659                 }
1660                 /* don't restart its timer, and silently
1661                    drop the packet. */
1662                 __ip_vs_conn_put(cp);
1663                 return NF_DROP;
1664         }
1665
1666         ip_vs_in_stats(cp, skb);
1667         ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1668         if (cp->packet_xmit)
1669                 ret = cp->packet_xmit(skb, cp, pp, &iph);
1670                 /* do not touch skb anymore */
1671         else {
1672                 IP_VS_DBG_RL("warning: packet_xmit is null");
1673                 ret = NF_ACCEPT;
1674         }
1675
1676         /* Increase its packet counter and check if it is needed
1677          * to be synchronized
1678          *
1679          * Sync connection if it is about to close to
1680          * encorage the standby servers to update the connections timeout
1681          *
1682          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1683          */
1684
1685         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
1686                 pkts = sysctl_sync_threshold(ipvs);
1687         else
1688                 pkts = atomic_add_return(1, &cp->in_pkts);
1689
1690         if (ipvs->sync_state & IP_VS_STATE_MASTER)
1691                 ip_vs_sync_conn(net, cp, pkts);
1692
1693         ip_vs_conn_put(cp);
1694         return ret;
1695 }
1696
1697 /*
1698  *      AF_INET handler in NF_INET_LOCAL_IN chain
1699  *      Schedule and forward packets from remote clients
1700  */
1701 static unsigned int
1702 ip_vs_remote_request4(unsigned int hooknum, struct sk_buff *skb,
1703                       const struct net_device *in,
1704                       const struct net_device *out,
1705                       int (*okfn)(struct sk_buff *))
1706 {
1707         return ip_vs_in(hooknum, skb, AF_INET);
1708 }
1709
1710 /*
1711  *      AF_INET handler in NF_INET_LOCAL_OUT chain
1712  *      Schedule and forward packets from local clients
1713  */
1714 static unsigned int
1715 ip_vs_local_request4(unsigned int hooknum, struct sk_buff *skb,
1716                      const struct net_device *in, const struct net_device *out,
1717                      int (*okfn)(struct sk_buff *))
1718 {
1719         unsigned int verdict;
1720
1721         /* Disable BH in LOCAL_OUT until all places are fixed */
1722         local_bh_disable();
1723         verdict = ip_vs_in(hooknum, skb, AF_INET);
1724         local_bh_enable();
1725         return verdict;
1726 }
1727
1728 #ifdef CONFIG_IP_VS_IPV6
1729
1730 /*
1731  * AF_INET6 fragment handling
1732  * Copy info from first fragment, to the rest of them.
1733  */
1734 static unsigned int
1735 ip_vs_preroute_frag6(unsigned int hooknum, struct sk_buff *skb,
1736                      const struct net_device *in,
1737                      const struct net_device *out,
1738                      int (*okfn)(struct sk_buff *))
1739 {
1740         struct sk_buff *reasm = skb_nfct_reasm(skb);
1741         struct net *net;
1742
1743         /* Skip if not a "replay" from nf_ct_frag6_output or first fragment.
1744          * ipvs_property is set when checking first fragment
1745          * in ip_vs_in() and ip_vs_out().
1746          */
1747         if (reasm)
1748                 IP_VS_DBG(2, "Fragment recv prop:%d\n", reasm->ipvs_property);
1749         if (!reasm || !reasm->ipvs_property)
1750                 return NF_ACCEPT;
1751
1752         net = skb_net(skb);
1753         if (!net_ipvs(net)->enable)
1754                 return NF_ACCEPT;
1755
1756         /* Copy stored fw mark, saved in ip_vs_{in,out} */
1757         skb->mark = reasm->mark;
1758
1759         return NF_ACCEPT;
1760 }
1761
1762 /*
1763  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
1764  *      Schedule and forward packets from remote clients
1765  */
1766 static unsigned int
1767 ip_vs_remote_request6(unsigned int hooknum, struct sk_buff *skb,
1768                       const struct net_device *in,
1769                       const struct net_device *out,
1770                       int (*okfn)(struct sk_buff *))
1771 {
1772         return ip_vs_in(hooknum, skb, AF_INET6);
1773 }
1774
1775 /*
1776  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
1777  *      Schedule and forward packets from local clients
1778  */
1779 static unsigned int
1780 ip_vs_local_request6(unsigned int hooknum, struct sk_buff *skb,
1781                      const struct net_device *in, const struct net_device *out,
1782                      int (*okfn)(struct sk_buff *))
1783 {
1784         unsigned int verdict;
1785
1786         /* Disable BH in LOCAL_OUT until all places are fixed */
1787         local_bh_disable();
1788         verdict = ip_vs_in(hooknum, skb, AF_INET6);
1789         local_bh_enable();
1790         return verdict;
1791 }
1792
1793 #endif
1794
1795
1796 /*
1797  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1798  *      related packets destined for 0.0.0.0/0.
1799  *      When fwmark-based virtual service is used, such as transparent
1800  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
1801  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1802  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1803  *      and send them to ip_vs_in_icmp.
1804  */
1805 static unsigned int
1806 ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff *skb,
1807                    const struct net_device *in, const struct net_device *out,
1808                    int (*okfn)(struct sk_buff *))
1809 {
1810         int r;
1811         struct net *net;
1812
1813         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1814                 return NF_ACCEPT;
1815
1816         /* ipvs enabled in this netns ? */
1817         net = skb_net(skb);
1818         if (!net_ipvs(net)->enable)
1819                 return NF_ACCEPT;
1820
1821         return ip_vs_in_icmp(skb, &r, hooknum);
1822 }
1823
1824 #ifdef CONFIG_IP_VS_IPV6
1825 static unsigned int
1826 ip_vs_forward_icmp_v6(unsigned int hooknum, struct sk_buff *skb,
1827                       const struct net_device *in, const struct net_device *out,
1828                       int (*okfn)(struct sk_buff *))
1829 {
1830         int r;
1831         struct net *net;
1832         struct ip_vs_iphdr iphdr;
1833
1834         ip_vs_fill_iph_skb(AF_INET6, skb, &iphdr);
1835         if (iphdr.protocol != IPPROTO_ICMPV6)
1836                 return NF_ACCEPT;
1837
1838         /* ipvs enabled in this netns ? */
1839         net = skb_net(skb);
1840         if (!net_ipvs(net)->enable)
1841                 return NF_ACCEPT;
1842
1843         return ip_vs_in_icmp_v6(skb, &r, hooknum, &iphdr);
1844 }
1845 #endif
1846
1847
1848 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
1849         /* After packet filtering, change source only for VS/NAT */
1850         {
1851                 .hook           = ip_vs_reply4,
1852                 .owner          = THIS_MODULE,
1853                 .pf             = NFPROTO_IPV4,
1854                 .hooknum        = NF_INET_LOCAL_IN,
1855                 .priority       = NF_IP_PRI_NAT_SRC - 2,
1856         },
1857         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1858          * or VS/NAT(change destination), so that filtering rules can be
1859          * applied to IPVS. */
1860         {
1861                 .hook           = ip_vs_remote_request4,
1862                 .owner          = THIS_MODULE,
1863                 .pf             = NFPROTO_IPV4,
1864                 .hooknum        = NF_INET_LOCAL_IN,
1865                 .priority       = NF_IP_PRI_NAT_SRC - 1,
1866         },
1867         /* Before ip_vs_in, change source only for VS/NAT */
1868         {
1869                 .hook           = ip_vs_local_reply4,
1870                 .owner          = THIS_MODULE,
1871                 .pf             = NFPROTO_IPV4,
1872                 .hooknum        = NF_INET_LOCAL_OUT,
1873                 .priority       = NF_IP_PRI_NAT_DST + 1,
1874         },
1875         /* After mangle, schedule and forward local requests */
1876         {
1877                 .hook           = ip_vs_local_request4,
1878                 .owner          = THIS_MODULE,
1879                 .pf             = NFPROTO_IPV4,
1880                 .hooknum        = NF_INET_LOCAL_OUT,
1881                 .priority       = NF_IP_PRI_NAT_DST + 2,
1882         },
1883         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1884          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1885         {
1886                 .hook           = ip_vs_forward_icmp,
1887                 .owner          = THIS_MODULE,
1888                 .pf             = NFPROTO_IPV4,
1889                 .hooknum        = NF_INET_FORWARD,
1890                 .priority       = 99,
1891         },
1892         /* After packet filtering, change source only for VS/NAT */
1893         {
1894                 .hook           = ip_vs_reply4,
1895                 .owner          = THIS_MODULE,
1896                 .pf             = NFPROTO_IPV4,
1897                 .hooknum        = NF_INET_FORWARD,
1898                 .priority       = 100,
1899         },
1900 #ifdef CONFIG_IP_VS_IPV6
1901         /* After mangle & nat fetch 2:nd fragment and following */
1902         {
1903                 .hook           = ip_vs_preroute_frag6,
1904                 .owner          = THIS_MODULE,
1905                 .pf             = NFPROTO_IPV6,
1906                 .hooknum        = NF_INET_PRE_ROUTING,
1907                 .priority       = NF_IP6_PRI_NAT_DST + 1,
1908         },
1909         /* After packet filtering, change source only for VS/NAT */
1910         {
1911                 .hook           = ip_vs_reply6,
1912                 .owner          = THIS_MODULE,
1913                 .pf             = NFPROTO_IPV6,
1914                 .hooknum        = NF_INET_LOCAL_IN,
1915                 .priority       = NF_IP6_PRI_NAT_SRC - 2,
1916         },
1917         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1918          * or VS/NAT(change destination), so that filtering rules can be
1919          * applied to IPVS. */
1920         {
1921                 .hook           = ip_vs_remote_request6,
1922                 .owner          = THIS_MODULE,
1923                 .pf             = NFPROTO_IPV6,
1924                 .hooknum        = NF_INET_LOCAL_IN,
1925                 .priority       = NF_IP6_PRI_NAT_SRC - 1,
1926         },
1927         /* Before ip_vs_in, change source only for VS/NAT */
1928         {
1929                 .hook           = ip_vs_local_reply6,
1930                 .owner          = THIS_MODULE,
1931                 .pf             = NFPROTO_IPV4,
1932                 .hooknum        = NF_INET_LOCAL_OUT,
1933                 .priority       = NF_IP6_PRI_NAT_DST + 1,
1934         },
1935         /* After mangle, schedule and forward local requests */
1936         {
1937                 .hook           = ip_vs_local_request6,
1938                 .owner          = THIS_MODULE,
1939                 .pf             = NFPROTO_IPV6,
1940                 .hooknum        = NF_INET_LOCAL_OUT,
1941                 .priority       = NF_IP6_PRI_NAT_DST + 2,
1942         },
1943         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1944          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1945         {
1946                 .hook           = ip_vs_forward_icmp_v6,
1947                 .owner          = THIS_MODULE,
1948                 .pf             = NFPROTO_IPV6,
1949                 .hooknum        = NF_INET_FORWARD,
1950                 .priority       = 99,
1951         },
1952         /* After packet filtering, change source only for VS/NAT */
1953         {
1954                 .hook           = ip_vs_reply6,
1955                 .owner          = THIS_MODULE,
1956                 .pf             = NFPROTO_IPV6,
1957                 .hooknum        = NF_INET_FORWARD,
1958                 .priority       = 100,
1959         },
1960 #endif
1961 };
1962 /*
1963  *      Initialize IP Virtual Server netns mem.
1964  */
1965 static int __net_init __ip_vs_init(struct net *net)
1966 {
1967         struct netns_ipvs *ipvs;
1968
1969         ipvs = net_generic(net, ip_vs_net_id);
1970         if (ipvs == NULL)
1971                 return -ENOMEM;
1972
1973         /* Hold the beast until a service is registerd */
1974         ipvs->enable = 0;
1975         ipvs->net = net;
1976         /* Counters used for creating unique names */
1977         ipvs->gen = atomic_read(&ipvs_netns_cnt);
1978         atomic_inc(&ipvs_netns_cnt);
1979         net->ipvs = ipvs;
1980
1981         if (ip_vs_estimator_net_init(net) < 0)
1982                 goto estimator_fail;
1983
1984         if (ip_vs_control_net_init(net) < 0)
1985                 goto control_fail;
1986
1987         if (ip_vs_protocol_net_init(net) < 0)
1988                 goto protocol_fail;
1989
1990         if (ip_vs_app_net_init(net) < 0)
1991                 goto app_fail;
1992
1993         if (ip_vs_conn_net_init(net) < 0)
1994                 goto conn_fail;
1995
1996         if (ip_vs_sync_net_init(net) < 0)
1997                 goto sync_fail;
1998
1999         printk(KERN_INFO "IPVS: Creating netns size=%zu id=%d\n",
2000                          sizeof(struct netns_ipvs), ipvs->gen);
2001         return 0;
2002 /*
2003  * Error handling
2004  */
2005
2006 sync_fail:
2007         ip_vs_conn_net_cleanup(net);
2008 conn_fail:
2009         ip_vs_app_net_cleanup(net);
2010 app_fail:
2011         ip_vs_protocol_net_cleanup(net);
2012 protocol_fail:
2013         ip_vs_control_net_cleanup(net);
2014 control_fail:
2015         ip_vs_estimator_net_cleanup(net);
2016 estimator_fail:
2017         net->ipvs = NULL;
2018         return -ENOMEM;
2019 }
2020
2021 static void __net_exit __ip_vs_cleanup(struct net *net)
2022 {
2023         ip_vs_service_net_cleanup(net); /* ip_vs_flush() with locks */
2024         ip_vs_conn_net_cleanup(net);
2025         ip_vs_app_net_cleanup(net);
2026         ip_vs_protocol_net_cleanup(net);
2027         ip_vs_control_net_cleanup(net);
2028         ip_vs_estimator_net_cleanup(net);
2029         IP_VS_DBG(2, "ipvs netns %d released\n", net_ipvs(net)->gen);
2030         net->ipvs = NULL;
2031 }
2032
2033 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
2034 {
2035         EnterFunction(2);
2036         net_ipvs(net)->enable = 0;      /* Disable packet reception */
2037         smp_wmb();
2038         ip_vs_sync_net_cleanup(net);
2039         LeaveFunction(2);
2040 }
2041
2042 static struct pernet_operations ipvs_core_ops = {
2043         .init = __ip_vs_init,
2044         .exit = __ip_vs_cleanup,
2045         .id   = &ip_vs_net_id,
2046         .size = sizeof(struct netns_ipvs),
2047 };
2048
2049 static struct pernet_operations ipvs_core_dev_ops = {
2050         .exit = __ip_vs_dev_cleanup,
2051 };
2052
2053 /*
2054  *      Initialize IP Virtual Server
2055  */
2056 static int __init ip_vs_init(void)
2057 {
2058         int ret;
2059
2060         ret = ip_vs_control_init();
2061         if (ret < 0) {
2062                 pr_err("can't setup control.\n");
2063                 goto exit;
2064         }
2065
2066         ip_vs_protocol_init();
2067
2068         ret = ip_vs_conn_init();
2069         if (ret < 0) {
2070                 pr_err("can't setup connection table.\n");
2071                 goto cleanup_protocol;
2072         }
2073
2074         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
2075         if (ret < 0)
2076                 goto cleanup_conn;
2077
2078         ret = register_pernet_device(&ipvs_core_dev_ops);
2079         if (ret < 0)
2080                 goto cleanup_sub;
2081
2082         ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2083         if (ret < 0) {
2084                 pr_err("can't register hooks.\n");
2085                 goto cleanup_dev;
2086         }
2087
2088         ret = ip_vs_register_nl_ioctl();
2089         if (ret < 0) {
2090                 pr_err("can't register netlink/ioctl.\n");
2091                 goto cleanup_hooks;
2092         }
2093
2094         pr_info("ipvs loaded.\n");
2095
2096         return ret;
2097
2098 cleanup_hooks:
2099         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2100 cleanup_dev:
2101         unregister_pernet_device(&ipvs_core_dev_ops);
2102 cleanup_sub:
2103         unregister_pernet_subsys(&ipvs_core_ops);
2104 cleanup_conn:
2105         ip_vs_conn_cleanup();
2106 cleanup_protocol:
2107         ip_vs_protocol_cleanup();
2108         ip_vs_control_cleanup();
2109 exit:
2110         return ret;
2111 }
2112
2113 static void __exit ip_vs_cleanup(void)
2114 {
2115         ip_vs_unregister_nl_ioctl();
2116         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2117         unregister_pernet_device(&ipvs_core_dev_ops);
2118         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
2119         ip_vs_conn_cleanup();
2120         ip_vs_protocol_cleanup();
2121         ip_vs_control_cleanup();
2122         pr_info("ipvs unloaded.\n");
2123 }
2124
2125 module_init(ip_vs_init);
2126 module_exit(ip_vs_cleanup);
2127 MODULE_LICENSE("GPL");