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 int ip_vs_net_id __read_mostly;
  73 #ifdef IP_VS_GENERIC_NETNS
  74 EXPORT_SYMBOL(ip_vs_net_id);
  75 #endif
  76 /* netns cnt used for uniqueness */
  77 static atomic_t ipvs_netns_cnt = ATOMIC_INIT(0);
  78
  79 /* ID used in ICMP lookups */
  80 #define icmp_id(icmph)          (((icmph)->un).echo.id)
  81 #define icmpv6_id(icmph)        (icmph->icmp6_dataun.u_echo.identifier)
  82
  83 const char *ip_vs_proto_name(unsigned int proto)
  84 {
  85         static char buf[20];
  86
  87         switch (proto) {
  88         case IPPROTO_IP:
  89                 return "IP";
  90         case IPPROTO_UDP:
  91                 return "UDP";
  92         case IPPROTO_TCP:
  93                 return "TCP";
  94         case IPPROTO_SCTP:
  95                 return "SCTP";
  96         case IPPROTO_ICMP:
  97                 return "ICMP";
  98 #ifdef CONFIG_IP_VS_IPV6
  99         case IPPROTO_ICMPV6:
 100                 return "ICMPv6";
 101 #endif
 102         default:
 103                 sprintf(buf, "IP_%d", proto);
 104                 return buf;
 105         }
 106 }
 107
 108 void ip_vs_init_hash_table(struct list_head *table, int rows)
 109 {
 110         while (--rows >= 0)
 111                 INIT_LIST_HEAD(&table[rows]);
 112 }
 113
 114 static inline void
 115 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 116 {
 117         struct ip_vs_dest *dest = cp->dest;
 118         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 119
 120         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 121                 struct ip_vs_cpu_stats *s;
 122
 123                 s = this_cpu_ptr(dest->stats.cpustats);
 124                 s->ustats.inpkts++;
 125                 u64_stats_update_begin(&s->syncp);
 126                 s->ustats.inbytes += skb->len;
 127                 u64_stats_update_end(&s->syncp);
 128
 129                 s = this_cpu_ptr(dest->svc->stats.cpustats);
 130                 s->ustats.inpkts++;
 131                 u64_stats_update_begin(&s->syncp);
 132                 s->ustats.inbytes += skb->len;
 133                 u64_stats_update_end(&s->syncp);
 134
 135                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 136                 s->ustats.inpkts++;
 137                 u64_stats_update_begin(&s->syncp);
 138                 s->ustats.inbytes += skb->len;
 139                 u64_stats_update_end(&s->syncp);
 140         }
 141 }
 142
 143
 144 static inline void
 145 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 146 {
 147         struct ip_vs_dest *dest = cp->dest;
 148         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 149
 150         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 151                 struct ip_vs_cpu_stats *s;
 152
 153                 s = this_cpu_ptr(dest->stats.cpustats);
 154                 s->ustats.outpkts++;
 155                 u64_stats_update_begin(&s->syncp);
 156                 s->ustats.outbytes += skb->len;
 157                 u64_stats_update_end(&s->syncp);
 158
 159                 s = this_cpu_ptr(dest->svc->stats.cpustats);
 160                 s->ustats.outpkts++;
 161                 u64_stats_update_begin(&s->syncp);
 162                 s->ustats.outbytes += skb->len;
 163                 u64_stats_update_end(&s->syncp);
 164
 165                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 166                 s->ustats.outpkts++;
 167                 u64_stats_update_begin(&s->syncp);
 168                 s->ustats.outbytes += skb->len;
 169                 u64_stats_update_end(&s->syncp);
 170         }
 171 }
 172
 173
 174 static inline void
 175 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
 176 {
 177         struct netns_ipvs *ipvs = net_ipvs(svc->net);
 178         struct ip_vs_cpu_stats *s;
 179
 180         s = this_cpu_ptr(cp->dest->stats.cpustats);
 181         s->ustats.conns++;
 182
 183         s = this_cpu_ptr(svc->stats.cpustats);
 184         s->ustats.conns++;
 185
 186         s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 187         s->ustats.conns++;
 188 }
 189
 190
 191 static inline void
 192 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
 193                 const struct sk_buff *skb,
 194                 struct ip_vs_proto_data *pd)
 195 {
 196         if (likely(pd->pp->state_transition))
 197                 pd->pp->state_transition(cp, direction, skb, pd);
 198 }
 199
 200 static inline int
 201 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
 202                               struct sk_buff *skb, int protocol,
 203                               const union nf_inet_addr *caddr, __be16 cport,
 204                               const union nf_inet_addr *vaddr, __be16 vport,
 205                               struct ip_vs_conn_param *p)
 206 {
 207         ip_vs_conn_fill_param(svc->net, svc->af, protocol, caddr, cport, vaddr,
 208                               vport, p);
 209         p->pe = svc->pe;
 210         if (p->pe && p->pe->fill_param)
 211                 return p->pe->fill_param(p, skb);
 212
 213         return 0;
 214 }
 215
 216 /*
 217  *  IPVS persistent scheduling function
 218  *  It creates a connection entry according to its template if exists,
 219  *  or selects a server and creates a connection entry plus a template.
 220  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
 221  *  Protocols supported: TCP, UDP
 222  */
 223 static struct ip_vs_conn *
 224 ip_vs_sched_persist(struct ip_vs_service *svc,
 225                     struct sk_buff *skb, __be16 src_port, __be16 dst_port,
 226                     int *ignored, struct ip_vs_iphdr *iph)
 227 {
 228         struct ip_vs_conn *cp = NULL;
 229         struct ip_vs_dest *dest;
 230         struct ip_vs_conn *ct;
 231         __be16 dport = 0;               /* destination port to forward */
 232         unsigned int flags;
 233         struct ip_vs_conn_param param;
 234         const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
 235         union nf_inet_addr snet;        /* source network of the client,
 236                                            after masking */
 237
 238         /* Mask saddr with the netmask to adjust template granularity */
 239 #ifdef CONFIG_IP_VS_IPV6
 240         if (svc->af == AF_INET6)
 241                 ipv6_addr_prefix(&snet.in6, &iph->saddr.in6, svc->netmask);
 242         else
 243 #endif
 244                 snet.ip = iph->saddr.ip & svc->netmask;
 245
 246         IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
 247                       "mnet %s\n",
 248                       IP_VS_DBG_ADDR(svc->af, &iph->saddr), ntohs(src_port),
 249                       IP_VS_DBG_ADDR(svc->af, &iph->daddr), ntohs(dst_port),
 250                       IP_VS_DBG_ADDR(svc->af, &snet));
 251
 252         /*
 253          * As far as we know, FTP is a very complicated network protocol, and
 254          * it uses control connection and data connections. For active FTP,
 255          * FTP server initialize data connection to the client, its source port
 256          * is often 20. For passive FTP, FTP server tells the clients the port
 257          * that it passively listens to,  and the client issues the data
 258          * connection. In the tunneling or direct routing mode, the load
 259          * balancer is on the client-to-server half of connection, the port
 260          * number is unknown to the load balancer. So, a conn template like
 261          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
 262          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
 263          * is created for other persistent services.
 264          */
 265         {
 266                 int protocol = iph->protocol;
 267                 const union nf_inet_addr *vaddr = &iph->daddr;
 268                 __be16 vport = 0;
 269
 270                 if (dst_port == svc->port) {
 271                         /* non-FTP template:
 272                          * <protocol, caddr, 0, vaddr, vport, daddr, dport>
 273                          * FTP template:
 274                          * <protocol, caddr, 0, vaddr, 0, daddr, 0>
 275                          */
 276                         if (svc->port != FTPPORT)
 277                                 vport = dst_port;
 278                 } else {
 279                         /* Note: persistent fwmark-based services and
 280                          * persistent port zero service are handled here.
 281                          * fwmark template:
 282                          * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
 283                          * port zero template:
 284                          * <protocol,caddr,0,vaddr,0,daddr,0>
 285                          */
 286                         if (svc->fwmark) {
 287                                 protocol = IPPROTO_IP;
 288                                 vaddr = &fwmark;
 289                         }
 290                 }
 291                 /* return *ignored = -1 so NF_DROP can be used */
 292                 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
 293                                                   vaddr, vport, &param) < 0) {
 294                         *ignored = -1;
 295                         return NULL;
 296                 }
 297         }
 298
 299         /* Check if a template already exists */
 300         ct = ip_vs_ct_in_get(&param);
 301         if (!ct || !ip_vs_check_template(ct)) {
 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                 dest = svc->scheduler->schedule(svc, skb);
 308                 if (!dest) {
 309                         IP_VS_DBG(1, "p-schedule: no dest found.\n");
 310                         kfree(param.pe_data);
 311                         *ignored = 0;
 312                         return NULL;
 313                 }
 314
 315                 if (dst_port == svc->port && svc->port != FTPPORT)
 316                         dport = dest->port;
 317
 318                 /* Create a template
 319                  * This adds param.pe_data to the template,
 320                  * and thus param.pe_data will be destroyed
 321                  * when the template expires */
 322                 ct = ip_vs_conn_new(&param, &dest->addr, dport,
 323                                     IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
 324                 if (ct == NULL) {
 325                         kfree(param.pe_data);
 326                         *ignored = -1;
 327                         return NULL;
 328                 }
 329
 330                 ct->timeout = svc->timeout;
 331         } else {
 332                 /* set destination with the found template */
 333                 dest = ct->dest;
 334                 kfree(param.pe_data);
 335         }
 336
 337         dport = dst_port;
 338         if (dport == svc->port && dest->port)
 339                 dport = dest->port;
 340
 341         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 342                  && iph->protocol == IPPROTO_UDP) ?
 343                 IP_VS_CONN_F_ONE_PACKET : 0;
 344
 345         /*
 346          *    Create a new connection according to the template
 347          */
 348         ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol, &iph->saddr,
 349                               src_port, &iph->daddr, dst_port, &param);
 350
 351         cp = ip_vs_conn_new(&param, &dest->addr, dport, flags, dest, skb->mark);
 352         if (cp == NULL) {
 353                 ip_vs_conn_put(ct);
 354                 *ignored = -1;
 355                 return NULL;
 356         }
 357
 358         /*
 359          *    Add its control
 360          */
 361         ip_vs_control_add(cp, ct);
 362         ip_vs_conn_put(ct);
 363
 364         ip_vs_conn_stats(cp, svc);
 365         return cp;
 366 }
 367
 368
 369 /*
 370  *  IPVS main scheduling function
 371  *  It selects a server according to the virtual service, and
 372  *  creates a connection entry.
 373  *  Protocols supported: TCP, UDP
 374  *
 375  *  Usage of *ignored
 376  *
 377  * 1 :   protocol tried to schedule (eg. on SYN), found svc but the
 378  *       svc/scheduler decides that this packet should be accepted with
 379  *       NF_ACCEPT because it must not be scheduled.
 380  *
 381  * 0 :   scheduler can not find destination, so try bypass or
 382  *       return ICMP and then NF_DROP (ip_vs_leave).
 383  *
 384  * -1 :  scheduler tried to schedule but fatal error occurred, eg.
 385  *       ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
 386  *       failure such as missing Call-ID, ENOMEM on skb_linearize
 387  *       or pe_data. In this case we should return NF_DROP without
 388  *       any attempts to send ICMP with ip_vs_leave.
 389  */
 390 struct ip_vs_conn *
 391 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
 392                struct ip_vs_proto_data *pd, int *ignored,
 393                struct ip_vs_iphdr *iph)
 394 {
 395         struct ip_vs_protocol *pp = pd->pp;
 396         struct ip_vs_conn *cp = NULL;
 397         struct ip_vs_dest *dest;
 398         __be16 _ports[2], *pptr;
 399         unsigned int flags;
 400
 401         *ignored = 1;
 402         /*
 403          * IPv6 frags, only the first hit here.
 404          */
 405         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 406         if (pptr == NULL)
 407                 return NULL;
 408
 409         /*
 410          * FTPDATA needs this check when using local real server.
 411          * Never schedule Active FTPDATA connections from real server.
 412          * For LVS-NAT they must be already created. For other methods
 413          * with persistence the connection is created on SYN+ACK.
 414          */
 415         if (pptr[0] == FTPDATA) {
 416                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 417                               "Not scheduling FTPDATA");
 418                 return NULL;
 419         }
 420
 421         /*
 422          *    Do not schedule replies from local real server.
 423          */
 424         if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
 425             (cp = pp->conn_in_get(svc->af, skb, iph, 1))) {
 426                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 427                               "Not scheduling reply for existing connection");
 428                 __ip_vs_conn_put(cp);
 429                 return NULL;
 430         }
 431
 432         /*
 433          *    Persistent service
 434          */
 435         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 436                 return ip_vs_sched_persist(svc, skb, pptr[0], pptr[1], ignored,
 437                                            iph);
 438
 439         *ignored = 0;
 440
 441         /*
 442          *    Non-persistent service
 443          */
 444         if (!svc->fwmark && pptr[1] != svc->port) {
 445                 if (!svc->port)
 446                         pr_err("Schedule: port zero only supported "
 447                                "in persistent services, "
 448                                "check your ipvs configuration\n");
 449                 return NULL;
 450         }
 451
 452         dest = svc->scheduler->schedule(svc, skb);
 453         if (dest == NULL) {
 454                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 455                 return NULL;
 456         }
 457
 458         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 459                  && iph->protocol == IPPROTO_UDP) ?
 460                 IP_VS_CONN_F_ONE_PACKET : 0;
 461
 462         /*
 463          *    Create a connection entry.
 464          */
 465         {
 466                 struct ip_vs_conn_param p;
 467
 468                 ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol,
 469                                       &iph->saddr, pptr[0], &iph->daddr,
 470                                       pptr[1], &p);
 471                 cp = ip_vs_conn_new(&p, &dest->addr,
 472                                     dest->port ? dest->port : pptr[1],
 473                                     flags, dest, skb->mark);
 474                 if (!cp) {
 475                         *ignored = -1;
 476                         return NULL;
 477                 }
 478         }
 479
 480         IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
 481                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
 482                       ip_vs_fwd_tag(cp),
 483                       IP_VS_DBG_ADDR(svc->af, &cp->caddr), ntohs(cp->cport),
 484                       IP_VS_DBG_ADDR(svc->af, &cp->vaddr), ntohs(cp->vport),
 485                       IP_VS_DBG_ADDR(svc->af, &cp->daddr), ntohs(cp->dport),
 486                       cp->flags, atomic_read(&cp->refcnt));
 487
 488         ip_vs_conn_stats(cp, svc);
 489         return cp;
 490 }
 491
 492
 493 /*
 494  *  Pass or drop the packet.
 495  *  Called by ip_vs_in, when the virtual service is available but
 496  *  no destination is available for a new connection.
 497  */
 498 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 499                 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
 500 {
 501         __be16 _ports[2], *pptr;
 502 #ifdef CONFIG_SYSCTL
 503         struct net *net;
 504         struct netns_ipvs *ipvs;
 505         int unicast;
 506 #endif
 507
 508         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 509         if (pptr == NULL) {
 510                 ip_vs_service_put(svc);
 511                 return NF_DROP;
 512         }
 513
 514 #ifdef CONFIG_SYSCTL
 515         net = skb_net(skb);
 516
 517 #ifdef CONFIG_IP_VS_IPV6
 518         if (svc->af == AF_INET6)
 519                 unicast = ipv6_addr_type(&iph->daddr.in6) & IPV6_ADDR_UNICAST;
 520         else
 521 #endif
 522                 unicast = (inet_addr_type(net, iph->daddr.ip) == RTN_UNICAST);
 523
 524         /* if it is fwmark-based service, the cache_bypass sysctl is up
 525            and the destination is a non-local unicast, then create
 526            a cache_bypass connection entry */
 527         ipvs = net_ipvs(net);
 528         if (ipvs->sysctl_cache_bypass && svc->fwmark && unicast) {
 529                 int ret;
 530                 struct ip_vs_conn *cp;
 531                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
 532                                       iph->protocol == IPPROTO_UDP) ?
 533                                       IP_VS_CONN_F_ONE_PACKET : 0;
 534                 union nf_inet_addr daddr =  { .all = { 0, 0, 0, 0 } };
 535
 536                 ip_vs_service_put(svc);
 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                 ip_vs_service_put(svc);
 576                 return NF_ACCEPT;
 577         }
 578
 579         ip_vs_service_put(svc);
 580
 581         /*
 582          * Notify the client that the destination is unreachable, and
 583          * release the socket buffer.
 584          * Since it is in IP layer, the TCP socket is not actually
 585          * created, the TCP RST packet cannot be sent, instead that
 586          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 587          */
 588 #ifdef CONFIG_IP_VS_IPV6
 589         if (svc->af == AF_INET6) {
 590                 if (!skb->dev) {
 591                         struct net *net = dev_net(skb_dst(skb)->dev);
 592
 593                         skb->dev = net->loopback_dev;
 594                 }
 595                 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 596         } else
 597 #endif
 598                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 599
 600         return NF_DROP;
 601 }
 602
 603 #ifdef CONFIG_SYSCTL
 604
 605 static int sysctl_snat_reroute(struct sk_buff *skb)
 606 {
 607         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 608         return ipvs->sysctl_snat_reroute;
 609 }
 610
 611 static int sysctl_nat_icmp_send(struct net *net)
 612 {
 613         struct netns_ipvs *ipvs = net_ipvs(net);
 614         return ipvs->sysctl_nat_icmp_send;
 615 }
 616
 617 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
 618 {
 619         return ipvs->sysctl_expire_nodest_conn;
 620 }
 621
 622 #else
 623
 624 static int sysctl_snat_reroute(struct sk_buff *skb) { return 0; }
 625 static int sysctl_nat_icmp_send(struct net *net) { return 0; }
 626 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
 627
 628 #endif
 629
 630 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 631 {
 632         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 633 }
 634
 635 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
 636 {
 637         if (NF_INET_LOCAL_IN == hooknum)
 638                 return IP_DEFRAG_VS_IN;
 639         if (NF_INET_FORWARD == hooknum)
 640                 return IP_DEFRAG_VS_FWD;
 641         return IP_DEFRAG_VS_OUT;
 642 }
 643
 644 static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user)
 645 {
 646         int err = ip_defrag(skb, user);
 647
 648         if (!err)
 649                 ip_send_check(ip_hdr(skb));
 650
 651         return err;
 652 }
 653
 654 static int ip_vs_route_me_harder(int af, struct sk_buff *skb)
 655 {
 656 #ifdef CONFIG_IP_VS_IPV6
 657         if (af == AF_INET6) {
 658                 if (sysctl_snat_reroute(skb) && ip6_route_me_harder(skb) != 0)
 659                         return 1;
 660         } else
 661 #endif
 662                 if ((sysctl_snat_reroute(skb) ||
 663                      skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
 664                     ip_route_me_harder(skb, RTN_LOCAL) != 0)
 665                         return 1;
 666
 667         return 0;
 668 }
 669
 670 /*
 671  * Packet has been made sufficiently writable in caller
 672  * - inout: 1=in->out, 0=out->in
 673  */
 674 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 675                     struct ip_vs_conn *cp, int inout)
 676 {
 677         struct iphdr *iph        = ip_hdr(skb);
 678         unsigned int icmp_offset = iph->ihl*4;
 679         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
 680                                                       icmp_offset);
 681         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 682
 683         if (inout) {
 684                 iph->saddr = cp->vaddr.ip;
 685                 ip_send_check(iph);
 686                 ciph->daddr = cp->vaddr.ip;
 687                 ip_send_check(ciph);
 688         } else {
 689                 iph->daddr = cp->daddr.ip;
 690                 ip_send_check(iph);
 691                 ciph->saddr = cp->daddr.ip;
 692                 ip_send_check(ciph);
 693         }
 694
 695         /* the TCP/UDP/SCTP port */
 696         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
 697             IPPROTO_SCTP == ciph->protocol) {
 698                 __be16 *ports = (void *)ciph + ciph->ihl*4;
 699
 700                 if (inout)
 701                         ports[1] = cp->vport;
 702                 else
 703                         ports[0] = cp->dport;
 704         }
 705
 706         /* And finally the ICMP checksum */
 707         icmph->checksum = 0;
 708         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 709         skb->ip_summed = CHECKSUM_UNNECESSARY;
 710
 711         if (inout)
 712                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 713                         "Forwarding altered outgoing ICMP");
 714         else
 715                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 716                         "Forwarding altered incoming ICMP");
 717 }
 718
 719 #ifdef CONFIG_IP_VS_IPV6
 720 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
 721                     struct ip_vs_conn *cp, int inout)
 722 {
 723         struct ipv6hdr *iph      = ipv6_hdr(skb);
 724         unsigned int icmp_offset = 0;
 725         unsigned int offs        = 0; /* header offset*/
 726         int protocol;
 727         struct icmp6hdr *icmph;
 728         struct ipv6hdr *ciph;
 729         unsigned short fragoffs;
 730
 731         ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
 732         icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
 733         offs = icmp_offset + sizeof(struct icmp6hdr);
 734         ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
 735
 736         protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
 737
 738         if (inout) {
 739                 iph->saddr = cp->vaddr.in6;
 740                 ciph->daddr = cp->vaddr.in6;
 741         } else {
 742                 iph->daddr = cp->daddr.in6;
 743                 ciph->saddr = cp->daddr.in6;
 744         }
 745
 746         /* the TCP/UDP/SCTP port */
 747         if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 748                           IPPROTO_SCTP == protocol)) {
 749                 __be16 *ports = (void *)(skb_network_header(skb) + offs);
 750
 751                 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
 752                               ntohs(inout ? ports[1] : ports[0]),
 753                               ntohs(inout ? cp->vport : cp->dport));
 754                 if (inout)
 755                         ports[1] = cp->vport;
 756                 else
 757                         ports[0] = cp->dport;
 758         }
 759
 760         /* And finally the ICMP checksum */
 761         icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
 762                                               skb->len - icmp_offset,
 763                                               IPPROTO_ICMPV6, 0);
 764         skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
 765         skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
 766         skb->ip_summed = CHECKSUM_PARTIAL;
 767
 768         if (inout)
 769                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 770                               (void *)ciph - (void *)iph,
 771                               "Forwarding altered outgoing ICMPv6");
 772         else
 773                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 774                               (void *)ciph - (void *)iph,
 775                               "Forwarding altered incoming ICMPv6");
 776 }
 777 #endif
 778
 779 /* Handle relevant response ICMP messages - forward to the right
 780  * destination host.
 781  */
 782 static int handle_response_icmp(int af, struct sk_buff *skb,
 783                                 union nf_inet_addr *snet,
 784                                 __u8 protocol, struct ip_vs_conn *cp,
 785                                 struct ip_vs_protocol *pp,
 786                                 unsigned int offset, unsigned int ihl)
 787 {
 788         unsigned int verdict = NF_DROP;
 789
 790         if (IP_VS_FWD_METHOD(cp) != 0) {
 791                 pr_err("shouldn't reach here, because the box is on the "
 792                        "half connection in the tun/dr module.\n");
 793         }
 794
 795         /* Ensure the checksum is correct */
 796         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 797                 /* Failed checksum! */
 798                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
 799                               IP_VS_DBG_ADDR(af, snet));
 800                 goto out;
 801         }
 802
 803         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 804             IPPROTO_SCTP == protocol)
 805                 offset += 2 * sizeof(__u16);
 806         if (!skb_make_writable(skb, offset))
 807                 goto out;
 808
 809 #ifdef CONFIG_IP_VS_IPV6
 810         if (af == AF_INET6)
 811                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
 812         else
 813 #endif
 814                 ip_vs_nat_icmp(skb, pp, cp, 1);
 815
 816         if (ip_vs_route_me_harder(af, skb))
 817                 goto out;
 818
 819         /* do the statistics and put it back */
 820         ip_vs_out_stats(cp, skb);
 821
 822         skb->ipvs_property = 1;
 823         if (!(cp->flags & IP_VS_CONN_F_NFCT))
 824                 ip_vs_notrack(skb);
 825         else
 826                 ip_vs_update_conntrack(skb, cp, 0);
 827         verdict = NF_ACCEPT;
 828
 829 out:
 830         __ip_vs_conn_put(cp);
 831
 832         return verdict;
 833 }
 834
 835 /*
 836  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 837  *      Find any that might be relevant, check against existing connections.
 838  *      Currently handles error types - unreachable, quench, ttl exceeded.
 839  */
 840 static int ip_vs_out_icmp(struct sk_buff *skb, int *related,
 841                           unsigned int hooknum)
 842 {
 843         struct iphdr *iph;
 844         struct icmphdr  _icmph, *ic;
 845         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 846         struct ip_vs_iphdr ciph;
 847         struct ip_vs_conn *cp;
 848         struct ip_vs_protocol *pp;
 849         unsigned int offset, ihl;
 850         union nf_inet_addr snet;
 851
 852         *related = 1;
 853
 854         /* reassemble IP fragments */
 855         if (ip_is_fragment(ip_hdr(skb))) {
 856                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
 857                         return NF_STOLEN;
 858         }
 859
 860         iph = ip_hdr(skb);
 861         offset = ihl = iph->ihl * 4;
 862         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 863         if (ic == NULL)
 864                 return NF_DROP;
 865
 866         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
 867                   ic->type, ntohs(icmp_id(ic)),
 868                   &iph->saddr, &iph->daddr);
 869
 870         /*
 871          * Work through seeing if this is for us.
 872          * These checks are supposed to be in an order that means easy
 873          * things are checked first to speed up processing.... however
 874          * this means that some packets will manage to get a long way
 875          * down this stack and then be rejected, but that's life.
 876          */
 877         if ((ic->type != ICMP_DEST_UNREACH) &&
 878             (ic->type != ICMP_SOURCE_QUENCH) &&
 879             (ic->type != ICMP_TIME_EXCEEDED)) {
 880                 *related = 0;
 881                 return NF_ACCEPT;
 882         }
 883
 884         /* Now find the contained IP header */
 885         offset += sizeof(_icmph);
 886         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 887         if (cih == NULL)
 888                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 889
 890         pp = ip_vs_proto_get(cih->protocol);
 891         if (!pp)
 892                 return NF_ACCEPT;
 893
 894         /* Is the embedded protocol header present? */
 895         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 896                      pp->dont_defrag))
 897                 return NF_ACCEPT;
 898
 899         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
 900                       "Checking outgoing ICMP for");
 901
 902         ip_vs_fill_ip4hdr(cih, &ciph);
 903         ciph.len += offset;
 904         /* The embedded headers contain source and dest in reverse order */
 905         cp = pp->conn_out_get(AF_INET, skb, &ciph, 1);
 906         if (!cp)
 907                 return NF_ACCEPT;
 908
 909         snet.ip = iph->saddr;
 910         return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
 911                                     pp, ciph.len, ihl);
 912 }
 913
 914 #ifdef CONFIG_IP_VS_IPV6
 915 static int ip_vs_out_icmp_v6(struct sk_buff *skb, int *related,
 916                              unsigned int hooknum, struct ip_vs_iphdr *ipvsh)
 917 {
 918         struct icmp6hdr _icmph, *ic;
 919         struct ipv6hdr _ip6h, *ip6h; /* The ip header contained within ICMP */
 920         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
 921         struct ip_vs_conn *cp;
 922         struct ip_vs_protocol *pp;
 923         union nf_inet_addr snet;
 924         unsigned int writable;
 925
 926         *related = 1;
 927         ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph, ipvsh);
 928         if (ic == NULL)
 929                 return NF_DROP;
 930
 931         /*
 932          * Work through seeing if this is for us.
 933          * These checks are supposed to be in an order that means easy
 934          * things are checked first to speed up processing.... however
 935          * this means that some packets will manage to get a long way
 936          * down this stack and then be rejected, but that's life.
 937          */
 938         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
 939                 *related = 0;
 940                 return NF_ACCEPT;
 941         }
 942         /* Fragment header that is before ICMP header tells us that:
 943          * it's not an error message since they can't be fragmented.
 944          */
 945         if (ipvsh->flags & IP6_FH_F_FRAG)
 946                 return NF_DROP;
 947
 948         IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
 949                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
 950                   &ipvsh->saddr, &ipvsh->daddr);
 951
 952         /* Now find the contained IP header */
 953         ciph.len = ipvsh->len + sizeof(_icmph);
 954         ip6h = skb_header_pointer(skb, ciph.len, sizeof(_ip6h), &_ip6h);
 955         if (ip6h == NULL)
 956                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 957         ciph.saddr.in6 = ip6h->saddr; /* conn_out_get() handles reverse order */
 958         ciph.daddr.in6 = ip6h->daddr;
 959         /* skip possible IPv6 exthdrs of contained IPv6 packet */
 960         ciph.protocol = ipv6_find_hdr(skb, &ciph.len, -1, &ciph.fragoffs, NULL);
 961         if (ciph.protocol < 0)
 962                 return NF_ACCEPT; /* Contained IPv6 hdr looks wrong, ignore */
 963
 964         pp = ip_vs_proto_get(ciph.protocol);
 965         if (!pp)
 966                 return NF_ACCEPT;
 967
 968         /* The embedded headers contain source and dest in reverse order */
 969         cp = pp->conn_out_get(AF_INET6, skb, &ciph, 1);
 970         if (!cp)
 971                 return NF_ACCEPT;
 972
 973         snet.in6 = ciph.saddr.in6;
 974         writable = ciph.len;
 975         return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
 976                                     pp, writable, sizeof(struct ipv6hdr));
 977 }
 978 #endif
 979
 980 /*
 981  * Check if sctp chunc is ABORT chunk
 982  */
 983 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
 984 {
 985         sctp_chunkhdr_t *sch, schunk;
 986         sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
 987                         sizeof(schunk), &schunk);
 988         if (sch == NULL)
 989                 return 0;
 990         if (sch->type == SCTP_CID_ABORT)
 991                 return 1;
 992         return 0;
 993 }
 994
 995 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
 996 {
 997         struct tcphdr _tcph, *th;
 998
 999         th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
1000         if (th == NULL)
1001                 return 0;
1002         return th->rst;
1003 }
1004
1005 /* Handle response packets: rewrite addresses and send away...
1006  */
1007 static unsigned int
1008 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1009                 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph)
1010 {
1011         struct ip_vs_protocol *pp = pd->pp;
1012
1013         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Outgoing packet");
1014
1015         if (!skb_make_writable(skb, iph->len))
1016                 goto drop;
1017
1018         /* mangle the packet */
1019         if (pp->snat_handler && !pp->snat_handler(skb, pp, cp, iph))
1020                 goto drop;
1021
1022 #ifdef CONFIG_IP_VS_IPV6
1023         if (af == AF_INET6)
1024                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1025         else
1026 #endif
1027         {
1028                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1029                 ip_send_check(ip_hdr(skb));
1030         }
1031
1032         /*
1033          * nf_iterate does not expect change in the skb->dst->dev.
1034          * It looks like it is not fatal to enable this code for hooks
1035          * where our handlers are at the end of the chain list and
1036          * when all next handlers use skb->dst->dev and not outdev.
1037          * It will definitely route properly the inout NAT traffic
1038          * when multiple paths are used.
1039          */
1040
1041         /* For policy routing, packets originating from this
1042          * machine itself may be routed differently to packets
1043          * passing through.  We want this packet to be routed as
1044          * if it came from this machine itself.  So re-compute
1045          * the routing information.
1046          */
1047         if (ip_vs_route_me_harder(af, skb))
1048                 goto drop;
1049
1050         IP_VS_DBG_PKT(10, af, pp, skb, 0, "After SNAT");
1051
1052         ip_vs_out_stats(cp, skb);
1053         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1054         skb->ipvs_property = 1;
1055         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1056                 ip_vs_notrack(skb);
1057         else
1058                 ip_vs_update_conntrack(skb, cp, 0);
1059         ip_vs_conn_put(cp);
1060
1061         LeaveFunction(11);
1062         return NF_ACCEPT;
1063
1064 drop:
1065         ip_vs_conn_put(cp);
1066         kfree_skb(skb);
1067         LeaveFunction(11);
1068         return NF_STOLEN;
1069 }
1070
1071 /*
1072  *      Check if outgoing packet belongs to the established ip_vs_conn.
1073  */
1074 static unsigned int
1075 ip_vs_out(unsigned int hooknum, struct sk_buff *skb, int af)
1076 {
1077         struct net *net = NULL;
1078         struct ip_vs_iphdr iph;
1079         struct ip_vs_protocol *pp;
1080         struct ip_vs_proto_data *pd;
1081         struct ip_vs_conn *cp;
1082
1083         EnterFunction(11);
1084
1085         /* Already marked as IPVS request or reply? */
1086         if (skb->ipvs_property)
1087                 return NF_ACCEPT;
1088
1089         /* Bad... Do not break raw sockets */
1090         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1091                      af == AF_INET)) {
1092                 struct sock *sk = skb->sk;
1093                 struct inet_sock *inet = inet_sk(skb->sk);
1094
1095                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1096                         return NF_ACCEPT;
1097         }
1098
1099         if (unlikely(!skb_dst(skb)))
1100                 return NF_ACCEPT;
1101
1102         net = skb_net(skb);
1103         if (!net_ipvs(net)->enable)
1104                 return NF_ACCEPT;
1105
1106         ip_vs_fill_iph_skb(af, skb, &iph);
1107 #ifdef CONFIG_IP_VS_IPV6
1108         if (af == AF_INET6) {
1109                 if (!iph.fragoffs && skb_nfct_reasm(skb)) {
1110                         struct sk_buff *reasm = skb_nfct_reasm(skb);
1111                         /* Save fw mark for coming frags */
1112                         reasm->ipvs_property = 1;
1113                         reasm->mark = skb->mark;
1114                 }
1115                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1116                         int related;
1117                         int verdict = ip_vs_out_icmp_v6(skb, &related,
1118                                                         hooknum, &iph);
1119
1120                         if (related)
1121                                 return verdict;
1122                 }
1123         } else
1124 #endif
1125                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1126                         int related;
1127                         int verdict = ip_vs_out_icmp(skb, &related, hooknum);
1128
1129                         if (related)
1130                                 return verdict;
1131                 }
1132
1133         pd = ip_vs_proto_data_get(net, iph.protocol);
1134         if (unlikely(!pd))
1135                 return NF_ACCEPT;
1136         pp = pd->pp;
1137
1138         /* reassemble IP fragments */
1139 #ifdef CONFIG_IP_VS_IPV6
1140         if (af == AF_INET)
1141 #endif
1142                 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1143                         if (ip_vs_gather_frags(skb,
1144                                                ip_vs_defrag_user(hooknum)))
1145                                 return NF_STOLEN;
1146
1147                         ip_vs_fill_ip4hdr(skb_network_header(skb), &iph);
1148                 }
1149
1150         /*
1151          * Check if the packet belongs to an existing entry
1152          */
1153         cp = pp->conn_out_get(af, skb, &iph, 0);
1154
1155         if (likely(cp))
1156                 return handle_response(af, skb, pd, cp, &iph);
1157         if (sysctl_nat_icmp_send(net) &&
1158             (pp->protocol == IPPROTO_TCP ||
1159              pp->protocol == IPPROTO_UDP ||
1160              pp->protocol == IPPROTO_SCTP)) {
1161                 __be16 _ports[2], *pptr;
1162
1163                 pptr = frag_safe_skb_hp(skb, iph.len,
1164                                          sizeof(_ports), _ports, &iph);
1165                 if (pptr == NULL)
1166                         return NF_ACCEPT;       /* Not for me */
1167                 if (ip_vs_lookup_real_service(net, af, iph.protocol,
1168                                               &iph.saddr,
1169                                               pptr[0])) {
1170                         /*
1171                          * Notify the real server: there is no
1172                          * existing entry if it is not RST
1173                          * packet or not TCP packet.
1174                          */
1175                         if ((iph.protocol != IPPROTO_TCP &&
1176                              iph.protocol != IPPROTO_SCTP)
1177                              || ((iph.protocol == IPPROTO_TCP
1178                                   && !is_tcp_reset(skb, iph.len))
1179                                  || (iph.protocol == IPPROTO_SCTP
1180                                         && !is_sctp_abort(skb,
1181                                                 iph.len)))) {
1182 #ifdef CONFIG_IP_VS_IPV6
1183                                 if (af == AF_INET6) {
1184                                         struct net *net =
1185                                                 dev_net(skb_dst(skb)->dev);
1186
1187                                         if (!skb->dev)
1188                                                 skb->dev = net->loopback_dev;
1189                                         icmpv6_send(skb,
1190                                                     ICMPV6_DEST_UNREACH,
1191                                                     ICMPV6_PORT_UNREACH,
1192                                                     0);
1193                                 } else
1194 #endif
1195                                         icmp_send(skb,
1196                                                   ICMP_DEST_UNREACH,
1197                                                   ICMP_PORT_UNREACH, 0);
1198                                 return NF_DROP;
1199                         }
1200                 }
1201         }
1202         IP_VS_DBG_PKT(12, af, pp, skb, 0,
1203                       "ip_vs_out: packet continues traversal as normal");
1204         return NF_ACCEPT;
1205 }
1206
1207 /*
1208  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1209  *      used only for VS/NAT.
1210  *      Check if packet is reply for established ip_vs_conn.
1211  */
1212 static unsigned int
1213 ip_vs_reply4(unsigned int hooknum, struct sk_buff *skb,
1214              const struct net_device *in, const struct net_device *out,
1215              int (*okfn)(struct sk_buff *))
1216 {
1217         return ip_vs_out(hooknum, skb, AF_INET);
1218 }
1219
1220 /*
1221  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1222  *      Check if packet is reply for established ip_vs_conn.
1223  */
1224 static unsigned int
1225 ip_vs_local_reply4(unsigned int hooknum, struct sk_buff *skb,
1226                    const struct net_device *in, const struct net_device *out,
1227                    int (*okfn)(struct sk_buff *))
1228 {
1229         unsigned int verdict;
1230
1231         /* Disable BH in LOCAL_OUT until all places are fixed */
1232         local_bh_disable();
1233         verdict = ip_vs_out(hooknum, skb, AF_INET);
1234         local_bh_enable();
1235         return verdict;
1236 }
1237
1238 #ifdef CONFIG_IP_VS_IPV6
1239
1240 /*
1241  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1242  *      used only for VS/NAT.
1243  *      Check if packet is reply for established ip_vs_conn.
1244  */
1245 static unsigned int
1246 ip_vs_reply6(unsigned int hooknum, struct sk_buff *skb,
1247              const struct net_device *in, const struct net_device *out,
1248              int (*okfn)(struct sk_buff *))
1249 {
1250         return ip_vs_out(hooknum, skb, AF_INET6);
1251 }
1252
1253 /*
1254  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1255  *      Check if packet is reply for established ip_vs_conn.
1256  */
1257 static unsigned int
1258 ip_vs_local_reply6(unsigned int hooknum, struct sk_buff *skb,
1259                    const struct net_device *in, const struct net_device *out,
1260                    int (*okfn)(struct sk_buff *))
1261 {
1262         unsigned int verdict;
1263
1264         /* Disable BH in LOCAL_OUT until all places are fixed */
1265         local_bh_disable();
1266         verdict = ip_vs_out(hooknum, skb, AF_INET6);
1267         local_bh_enable();
1268         return verdict;
1269 }
1270
1271 #endif
1272
1273 /*
1274  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1275  *      Find any that might be relevant, check against existing connections,
1276  *      forward to the right destination host if relevant.
1277  *      Currently handles error types - unreachable, quench, ttl exceeded.
1278  */
1279 static int
1280 ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum)
1281 {
1282         struct net *net = NULL;
1283         struct iphdr *iph;
1284         struct icmphdr  _icmph, *ic;
1285         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1286         struct ip_vs_iphdr ciph;
1287         struct ip_vs_conn *cp;
1288         struct ip_vs_protocol *pp;
1289         struct ip_vs_proto_data *pd;
1290         unsigned int offset, offset2, ihl, verdict;
1291         bool ipip;
1292
1293         *related = 1;
1294
1295         /* reassemble IP fragments */
1296         if (ip_is_fragment(ip_hdr(skb))) {
1297                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
1298                         return NF_STOLEN;
1299         }
1300
1301         iph = ip_hdr(skb);
1302         offset = ihl = iph->ihl * 4;
1303         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1304         if (ic == NULL)
1305                 return NF_DROP;
1306
1307         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1308                   ic->type, ntohs(icmp_id(ic)),
1309                   &iph->saddr, &iph->daddr);
1310
1311         /*
1312          * Work through seeing if this is for us.
1313          * These checks are supposed to be in an order that means easy
1314          * things are checked first to speed up processing.... however
1315          * this means that some packets will manage to get a long way
1316          * down this stack and then be rejected, but that's life.
1317          */
1318         if ((ic->type != ICMP_DEST_UNREACH) &&
1319             (ic->type != ICMP_SOURCE_QUENCH) &&
1320             (ic->type != ICMP_TIME_EXCEEDED)) {
1321                 *related = 0;
1322                 return NF_ACCEPT;
1323         }
1324
1325         /* Now find the contained IP header */
1326         offset += sizeof(_icmph);
1327         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1328         if (cih == NULL)
1329                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1330
1331         net = skb_net(skb);
1332
1333         /* Special case for errors for IPIP packets */
1334         ipip = false;
1335         if (cih->protocol == IPPROTO_IPIP) {
1336                 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1337                         return NF_ACCEPT;
1338                 /* Error for our IPIP must arrive at LOCAL_IN */
1339                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1340                         return NF_ACCEPT;
1341                 offset += cih->ihl * 4;
1342                 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1343                 if (cih == NULL)
1344                         return NF_ACCEPT; /* The packet looks wrong, ignore */
1345                 ipip = true;
1346         }
1347
1348         pd = ip_vs_proto_data_get(net, cih->protocol);
1349         if (!pd)
1350                 return NF_ACCEPT;
1351         pp = pd->pp;
1352
1353         /* Is the embedded protocol header present? */
1354         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1355                      pp->dont_defrag))
1356                 return NF_ACCEPT;
1357
1358         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1359                       "Checking incoming ICMP for");
1360
1361         offset2 = offset;
1362         ip_vs_fill_ip4hdr(cih, &ciph);
1363         ciph.len += offset;
1364         offset = ciph.len;
1365         /* The embedded headers contain source and dest in reverse order.
1366          * For IPIP this is error for request, not for reply.
1367          */
1368         cp = pp->conn_in_get(AF_INET, skb, &ciph, ipip ? 0 : 1);
1369         if (!cp)
1370                 return NF_ACCEPT;
1371
1372         verdict = NF_DROP;
1373
1374         /* Ensure the checksum is correct */
1375         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1376                 /* Failed checksum! */
1377                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1378                           &iph->saddr);
1379                 goto out;
1380         }
1381
1382         if (ipip) {
1383                 __be32 info = ic->un.gateway;
1384
1385                 /* Update the MTU */
1386                 if (ic->type == ICMP_DEST_UNREACH &&
1387                     ic->code == ICMP_FRAG_NEEDED) {
1388                         struct ip_vs_dest *dest = cp->dest;
1389                         u32 mtu = ntohs(ic->un.frag.mtu);
1390
1391                         /* Strip outer IP and ICMP, go to IPIP header */
1392                         __skb_pull(skb, ihl + sizeof(_icmph));
1393                         offset2 -= ihl + sizeof(_icmph);
1394                         skb_reset_network_header(skb);
1395                         IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1396                                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1397                         rcu_read_lock();
1398                         ipv4_update_pmtu(skb, dev_net(skb->dev),
1399                                          mtu, 0, 0, 0, 0);
1400                         rcu_read_unlock();
1401                         /* Client uses PMTUD? */
1402                         if (!(cih->frag_off & htons(IP_DF)))
1403                                 goto ignore_ipip;
1404                         /* Prefer the resulting PMTU */
1405                         if (dest) {
1406                                 spin_lock(&dest->dst_lock);
1407                                 if (dest->dst_cache)
1408                                         mtu = dst_mtu(dest->dst_cache);
1409                                 spin_unlock(&dest->dst_lock);
1410                         }
1411                         if (mtu > 68 + sizeof(struct iphdr))
1412                                 mtu -= sizeof(struct iphdr);
1413                         info = htonl(mtu);
1414                 }
1415                 /* Strip outer IP, ICMP and IPIP, go to IP header of
1416                  * original request.
1417                  */
1418                 __skb_pull(skb, offset2);
1419                 skb_reset_network_header(skb);
1420                 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1421                         &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1422                         ic->type, ic->code, ntohl(info));
1423                 icmp_send(skb, ic->type, ic->code, info);
1424                 /* ICMP can be shorter but anyways, account it */
1425                 ip_vs_out_stats(cp, skb);
1426
1427 ignore_ipip:
1428                 consume_skb(skb);
1429                 verdict = NF_STOLEN;
1430                 goto out;
1431         }
1432
1433         /* do the statistics and put it back */
1434         ip_vs_in_stats(cp, skb);
1435         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol)
1436                 offset += 2 * sizeof(__u16);
1437         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1438
1439 out:
1440         __ip_vs_conn_put(cp);
1441
1442         return verdict;
1443 }
1444
1445 #ifdef CONFIG_IP_VS_IPV6
1446 static int ip_vs_in_icmp_v6(struct sk_buff *skb, int *related,
1447                             unsigned int hooknum, struct ip_vs_iphdr *iph)
1448 {
1449         struct net *net = NULL;
1450         struct ipv6hdr _ip6h, *ip6h;
1451         struct icmp6hdr _icmph, *ic;
1452         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1453         struct ip_vs_conn *cp;
1454         struct ip_vs_protocol *pp;
1455         struct ip_vs_proto_data *pd;
1456         unsigned int offs_ciph, writable, verdict;
1457
1458         *related = 1;
1459
1460         ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph, iph);
1461         if (ic == NULL)
1462                 return NF_DROP;
1463
1464         /*
1465          * Work through seeing if this is for us.
1466          * These checks are supposed to be in an order that means easy
1467          * things are checked first to speed up processing.... however
1468          * this means that some packets will manage to get a long way
1469          * down this stack and then be rejected, but that's life.
1470          */
1471         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1472                 *related = 0;
1473                 return NF_ACCEPT;
1474         }
1475         /* Fragment header that is before ICMP header tells us that:
1476          * it's not an error message since they can't be fragmented.
1477          */
1478         if (iph->flags & IP6_FH_F_FRAG)
1479                 return NF_DROP;
1480
1481         IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1482                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1483                   &iph->saddr, &iph->daddr);
1484
1485         /* Now find the contained IP header */
1486         ciph.len = iph->len + sizeof(_icmph);
1487         offs_ciph = ciph.len; /* Save ip header offset */
1488         ip6h = skb_header_pointer(skb, ciph.len, sizeof(_ip6h), &_ip6h);
1489         if (ip6h == NULL)
1490                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1491         ciph.saddr.in6 = ip6h->saddr; /* conn_in_get() handles reverse order */
1492         ciph.daddr.in6 = ip6h->daddr;
1493         /* skip possible IPv6 exthdrs of contained IPv6 packet */
1494         ciph.protocol = ipv6_find_hdr(skb, &ciph.len, -1, &ciph.fragoffs, NULL);
1495         if (ciph.protocol < 0)
1496                 return NF_ACCEPT; /* Contained IPv6 hdr looks wrong, ignore */
1497
1498         net = skb_net(skb);
1499         pd = ip_vs_proto_data_get(net, ciph.protocol);
1500         if (!pd)
1501                 return NF_ACCEPT;
1502         pp = pd->pp;
1503
1504         /* Cannot handle fragmented embedded protocol */
1505         if (ciph.fragoffs)
1506                 return NF_ACCEPT;
1507
1508         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offs_ciph,
1509                       "Checking incoming ICMPv6 for");
1510
1511         /* The embedded headers contain source and dest in reverse order
1512          * if not from localhost
1513          */
1514         cp = pp->conn_in_get(AF_INET6, skb, &ciph,
1515                              (hooknum == NF_INET_LOCAL_OUT) ? 0 : 1);
1516
1517         if (!cp)
1518                 return NF_ACCEPT;
1519         /* VS/TUN, VS/DR and LOCALNODE just let it go */
1520         if ((hooknum == NF_INET_LOCAL_OUT) &&
1521             (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1522                 __ip_vs_conn_put(cp);
1523                 return NF_ACCEPT;
1524         }
1525
1526         /* do the statistics and put it back */
1527         ip_vs_in_stats(cp, skb);
1528
1529         /* Need to mangle contained IPv6 header in ICMPv6 packet */
1530         writable = ciph.len;
1531         if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1532             IPPROTO_SCTP == ciph.protocol)
1533                 writable += 2 * sizeof(__u16); /* Also mangle ports */
1534
1535         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, writable, hooknum, &ciph);
1536
1537         __ip_vs_conn_put(cp);
1538
1539         return verdict;
1540 }
1541 #endif
1542
1543
1544 /*
1545  *      Check if it's for virtual services, look it up,
1546  *      and send it on its way...
1547  */
1548 static unsigned int
1549 ip_vs_in(unsigned int hooknum, struct sk_buff *skb, int af)
1550 {
1551         struct net *net;
1552         struct ip_vs_iphdr iph;
1553         struct ip_vs_protocol *pp;
1554         struct ip_vs_proto_data *pd;
1555         struct ip_vs_conn *cp;
1556         int ret, pkts;
1557         struct netns_ipvs *ipvs;
1558
1559         /* Already marked as IPVS request or reply? */
1560         if (skb->ipvs_property)
1561                 return NF_ACCEPT;
1562
1563         /*
1564          *      Big tappo:
1565          *      - remote client: only PACKET_HOST
1566          *      - route: used for struct net when skb->dev is unset
1567          */
1568         if (unlikely((skb->pkt_type != PACKET_HOST &&
1569                       hooknum != NF_INET_LOCAL_OUT) ||
1570                      !skb_dst(skb))) {
1571                 ip_vs_fill_iph_skb(af, skb, &iph);
1572                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1573                               " ignored in hook %u\n",
1574                               skb->pkt_type, iph.protocol,
1575                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1576                 return NF_ACCEPT;
1577         }
1578         /* ipvs enabled in this netns ? */
1579         net = skb_net(skb);
1580         if (!net_ipvs(net)->enable)
1581                 return NF_ACCEPT;
1582
1583         ip_vs_fill_iph_skb(af, skb, &iph);
1584
1585         /* Bad... Do not break raw sockets */
1586         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1587                      af == AF_INET)) {
1588                 struct sock *sk = skb->sk;
1589                 struct inet_sock *inet = inet_sk(skb->sk);
1590
1591                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1592                         return NF_ACCEPT;
1593         }
1594
1595 #ifdef CONFIG_IP_VS_IPV6
1596         if (af == AF_INET6) {
1597                 if (!iph.fragoffs && skb_nfct_reasm(skb)) {
1598                         struct sk_buff *reasm = skb_nfct_reasm(skb);
1599                         /* Save fw mark for coming frags. */
1600                         reasm->ipvs_property = 1;
1601                         reasm->mark = skb->mark;
1602                 }
1603                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1604                         int related;
1605                         int verdict = ip_vs_in_icmp_v6(skb, &related, hooknum,
1606                                                        &iph);
1607
1608                         if (related)
1609                                 return verdict;
1610                 }
1611         } else
1612 #endif
1613                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1614                         int related;
1615                         int verdict = ip_vs_in_icmp(skb, &related, hooknum);
1616
1617                         if (related)
1618                                 return verdict;
1619                 }
1620
1621         /* Protocol supported? */
1622         pd = ip_vs_proto_data_get(net, iph.protocol);
1623         if (unlikely(!pd))
1624                 return NF_ACCEPT;
1625         pp = pd->pp;
1626         /*
1627          * Check if the packet belongs to an existing connection entry
1628          */
1629         cp = pp->conn_in_get(af, skb, &iph, 0);
1630         if (unlikely(!cp) && !iph.fragoffs) {
1631                 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1632                  * replayed fragment zero will already have created the cp
1633                  */
1634                 int v;
1635
1636                 /* Schedule and create new connection entry into &cp */
1637                 if (!pp->conn_schedule(af, skb, pd, &v, &cp, &iph))
1638                         return v;
1639         }
1640
1641         if (unlikely(!cp)) {
1642                 /* sorry, all this trouble for a no-hit :) */
1643                 IP_VS_DBG_PKT(12, af, pp, skb, 0,
1644                               "ip_vs_in: packet continues traversal as normal");
1645                 if (iph.fragoffs && !skb_nfct_reasm(skb)) {
1646                         /* Fragment that couldn't be mapped to a conn entry
1647                          * and don't have any pointer to a reasm skb
1648                          * is missing module nf_defrag_ipv6
1649                          */
1650                         IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1651                         IP_VS_DBG_PKT(7, af, pp, skb, 0, "unhandled fragment");
1652                 }
1653                 return NF_ACCEPT;
1654         }
1655
1656         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Incoming packet");
1657         ipvs = net_ipvs(net);
1658         /* Check the server status */
1659         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1660                 /* the destination server is not available */
1661
1662                 if (sysctl_expire_nodest_conn(ipvs)) {
1663                         /* try to expire the connection immediately */
1664                         ip_vs_conn_expire_now(cp);
1665                 }
1666                 /* don't restart its timer, and silently
1667                    drop the packet. */
1668                 __ip_vs_conn_put(cp);
1669                 return NF_DROP;
1670         }
1671
1672         ip_vs_in_stats(cp, skb);
1673         ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1674         if (cp->packet_xmit)
1675                 ret = cp->packet_xmit(skb, cp, pp, &iph);
1676                 /* do not touch skb anymore */
1677         else {
1678                 IP_VS_DBG_RL("warning: packet_xmit is null");
1679                 ret = NF_ACCEPT;
1680         }
1681
1682         /* Increase its packet counter and check if it is needed
1683          * to be synchronized
1684          *
1685          * Sync connection if it is about to close to
1686          * encorage the standby servers to update the connections timeout
1687          *
1688          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1689          */
1690
1691         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
1692                 pkts = sysctl_sync_threshold(ipvs);
1693         else
1694                 pkts = atomic_add_return(1, &cp->in_pkts);
1695
1696         if (ipvs->sync_state & IP_VS_STATE_MASTER)
1697                 ip_vs_sync_conn(net, cp, pkts);
1698
1699         ip_vs_conn_put(cp);
1700         return ret;
1701 }
1702
1703 /*
1704  *      AF_INET handler in NF_INET_LOCAL_IN chain
1705  *      Schedule and forward packets from remote clients
1706  */
1707 static unsigned int
1708 ip_vs_remote_request4(unsigned int hooknum, struct sk_buff *skb,
1709                       const struct net_device *in,
1710                       const struct net_device *out,
1711                       int (*okfn)(struct sk_buff *))
1712 {
1713         return ip_vs_in(hooknum, skb, AF_INET);
1714 }
1715
1716 /*
1717  *      AF_INET handler in NF_INET_LOCAL_OUT chain
1718  *      Schedule and forward packets from local clients
1719  */
1720 static unsigned int
1721 ip_vs_local_request4(unsigned int hooknum, struct sk_buff *skb,
1722                      const struct net_device *in, const struct net_device *out,
1723                      int (*okfn)(struct sk_buff *))
1724 {
1725         unsigned int verdict;
1726
1727         /* Disable BH in LOCAL_OUT until all places are fixed */
1728         local_bh_disable();
1729         verdict = ip_vs_in(hooknum, skb, AF_INET);
1730         local_bh_enable();
1731         return verdict;
1732 }
1733
1734 #ifdef CONFIG_IP_VS_IPV6
1735
1736 /*
1737  * AF_INET6 fragment handling
1738  * Copy info from first fragment, to the rest of them.
1739  */
1740 static unsigned int
1741 ip_vs_preroute_frag6(unsigned int hooknum, struct sk_buff *skb,
1742                      const struct net_device *in,
1743                      const struct net_device *out,
1744                      int (*okfn)(struct sk_buff *))
1745 {
1746         struct sk_buff *reasm = skb_nfct_reasm(skb);
1747         struct net *net;
1748
1749         /* Skip if not a "replay" from nf_ct_frag6_output or first fragment.
1750          * ipvs_property is set when checking first fragment
1751          * in ip_vs_in() and ip_vs_out().
1752          */
1753         if (reasm)
1754                 IP_VS_DBG(2, "Fragment recv prop:%d\n", reasm->ipvs_property);
1755         if (!reasm || !reasm->ipvs_property)
1756                 return NF_ACCEPT;
1757
1758         net = skb_net(skb);
1759         if (!net_ipvs(net)->enable)
1760                 return NF_ACCEPT;
1761
1762         /* Copy stored fw mark, saved in ip_vs_{in,out} */
1763         skb->mark = reasm->mark;
1764
1765         return NF_ACCEPT;
1766 }
1767
1768 /*
1769  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
1770  *      Schedule and forward packets from remote clients
1771  */
1772 static unsigned int
1773 ip_vs_remote_request6(unsigned int hooknum, struct sk_buff *skb,
1774                       const struct net_device *in,
1775                       const struct net_device *out,
1776                       int (*okfn)(struct sk_buff *))
1777 {
1778         return ip_vs_in(hooknum, skb, AF_INET6);
1779 }
1780
1781 /*
1782  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
1783  *      Schedule and forward packets from local clients
1784  */
1785 static unsigned int
1786 ip_vs_local_request6(unsigned int hooknum, struct sk_buff *skb,
1787                      const struct net_device *in, const struct net_device *out,
1788                      int (*okfn)(struct sk_buff *))
1789 {
1790         unsigned int verdict;
1791
1792         /* Disable BH in LOCAL_OUT until all places are fixed */
1793         local_bh_disable();
1794         verdict = ip_vs_in(hooknum, skb, AF_INET6);
1795         local_bh_enable();
1796         return verdict;
1797 }
1798
1799 #endif
1800
1801
1802 /*
1803  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1804  *      related packets destined for 0.0.0.0/0.
1805  *      When fwmark-based virtual service is used, such as transparent
1806  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
1807  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1808  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1809  *      and send them to ip_vs_in_icmp.
1810  */
1811 static unsigned int
1812 ip_vs_forward_icmp(unsigned int hooknum, struct sk_buff *skb,
1813                    const struct net_device *in, const struct net_device *out,
1814                    int (*okfn)(struct sk_buff *))
1815 {
1816         int r;
1817         struct net *net;
1818
1819         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1820                 return NF_ACCEPT;
1821
1822         /* ipvs enabled in this netns ? */
1823         net = skb_net(skb);
1824         if (!net_ipvs(net)->enable)
1825                 return NF_ACCEPT;
1826
1827         return ip_vs_in_icmp(skb, &r, hooknum);
1828 }
1829
1830 #ifdef CONFIG_IP_VS_IPV6
1831 static unsigned int
1832 ip_vs_forward_icmp_v6(unsigned int hooknum, struct sk_buff *skb,
1833                       const struct net_device *in, const struct net_device *out,
1834                       int (*okfn)(struct sk_buff *))
1835 {
1836         int r;
1837         struct net *net;
1838         struct ip_vs_iphdr iphdr;
1839
1840         ip_vs_fill_iph_skb(AF_INET6, skb, &iphdr);
1841         if (iphdr.protocol != IPPROTO_ICMPV6)
1842                 return NF_ACCEPT;
1843
1844         /* ipvs enabled in this netns ? */
1845         net = skb_net(skb);
1846         if (!net_ipvs(net)->enable)
1847                 return NF_ACCEPT;
1848
1849         return ip_vs_in_icmp_v6(skb, &r, hooknum, &iphdr);
1850 }
1851 #endif
1852
1853
1854 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
1855         /* After packet filtering, change source only for VS/NAT */
1856         {
1857                 .hook           = ip_vs_reply4,
1858                 .owner          = THIS_MODULE,
1859                 .pf             = NFPROTO_IPV4,
1860                 .hooknum        = NF_INET_LOCAL_IN,
1861                 .priority       = NF_IP_PRI_NAT_SRC - 2,
1862         },
1863         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1864          * or VS/NAT(change destination), so that filtering rules can be
1865          * applied to IPVS. */
1866         {
1867                 .hook           = ip_vs_remote_request4,
1868                 .owner          = THIS_MODULE,
1869                 .pf             = NFPROTO_IPV4,
1870                 .hooknum        = NF_INET_LOCAL_IN,
1871                 .priority       = NF_IP_PRI_NAT_SRC - 1,
1872         },
1873         /* Before ip_vs_in, change source only for VS/NAT */
1874         {
1875                 .hook           = ip_vs_local_reply4,
1876                 .owner          = THIS_MODULE,
1877                 .pf             = NFPROTO_IPV4,
1878                 .hooknum        = NF_INET_LOCAL_OUT,
1879                 .priority       = NF_IP_PRI_NAT_DST + 1,
1880         },
1881         /* After mangle, schedule and forward local requests */
1882         {
1883                 .hook           = ip_vs_local_request4,
1884                 .owner          = THIS_MODULE,
1885                 .pf             = NFPROTO_IPV4,
1886                 .hooknum        = NF_INET_LOCAL_OUT,
1887                 .priority       = NF_IP_PRI_NAT_DST + 2,
1888         },
1889         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1890          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1891         {
1892                 .hook           = ip_vs_forward_icmp,
1893                 .owner          = THIS_MODULE,
1894                 .pf             = NFPROTO_IPV4,
1895                 .hooknum        = NF_INET_FORWARD,
1896                 .priority       = 99,
1897         },
1898         /* After packet filtering, change source only for VS/NAT */
1899         {
1900                 .hook           = ip_vs_reply4,
1901                 .owner          = THIS_MODULE,
1902                 .pf             = NFPROTO_IPV4,
1903                 .hooknum        = NF_INET_FORWARD,
1904                 .priority       = 100,
1905         },
1906 #ifdef CONFIG_IP_VS_IPV6
1907         /* After mangle & nat fetch 2:nd fragment and following */
1908         {
1909                 .hook           = ip_vs_preroute_frag6,
1910                 .owner          = THIS_MODULE,
1911                 .pf             = NFPROTO_IPV6,
1912                 .hooknum        = NF_INET_PRE_ROUTING,
1913                 .priority       = NF_IP6_PRI_NAT_DST + 1,
1914         },
1915         /* After packet filtering, change source only for VS/NAT */
1916         {
1917                 .hook           = ip_vs_reply6,
1918                 .owner          = THIS_MODULE,
1919                 .pf             = NFPROTO_IPV6,
1920                 .hooknum        = NF_INET_LOCAL_IN,
1921                 .priority       = NF_IP6_PRI_NAT_SRC - 2,
1922         },
1923         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1924          * or VS/NAT(change destination), so that filtering rules can be
1925          * applied to IPVS. */
1926         {
1927                 .hook           = ip_vs_remote_request6,
1928                 .owner          = THIS_MODULE,
1929                 .pf             = NFPROTO_IPV6,
1930                 .hooknum        = NF_INET_LOCAL_IN,
1931                 .priority       = NF_IP6_PRI_NAT_SRC - 1,
1932         },
1933         /* Before ip_vs_in, change source only for VS/NAT */
1934         {
1935                 .hook           = ip_vs_local_reply6,
1936                 .owner          = THIS_MODULE,
1937                 .pf             = NFPROTO_IPV4,
1938                 .hooknum        = NF_INET_LOCAL_OUT,
1939                 .priority       = NF_IP6_PRI_NAT_DST + 1,
1940         },
1941         /* After mangle, schedule and forward local requests */
1942         {
1943                 .hook           = ip_vs_local_request6,
1944                 .owner          = THIS_MODULE,
1945                 .pf             = NFPROTO_IPV6,
1946                 .hooknum        = NF_INET_LOCAL_OUT,
1947                 .priority       = NF_IP6_PRI_NAT_DST + 2,
1948         },
1949         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1950          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1951         {
1952                 .hook           = ip_vs_forward_icmp_v6,
1953                 .owner          = THIS_MODULE,
1954                 .pf             = NFPROTO_IPV6,
1955                 .hooknum        = NF_INET_FORWARD,
1956                 .priority       = 99,
1957         },
1958         /* After packet filtering, change source only for VS/NAT */
1959         {
1960                 .hook           = ip_vs_reply6,
1961                 .owner          = THIS_MODULE,
1962                 .pf             = NFPROTO_IPV6,
1963                 .hooknum        = NF_INET_FORWARD,
1964                 .priority       = 100,
1965         },
1966 #endif
1967 };
1968 /*
1969  *      Initialize IP Virtual Server netns mem.
1970  */
1971 static int __net_init __ip_vs_init(struct net *net)
1972 {
1973         struct netns_ipvs *ipvs;
1974
1975         ipvs = net_generic(net, ip_vs_net_id);
1976         if (ipvs == NULL)
1977                 return -ENOMEM;
1978
1979         /* Hold the beast until a service is registerd */
1980         ipvs->enable = 0;
1981         ipvs->net = net;
1982         /* Counters used for creating unique names */
1983         ipvs->gen = atomic_read(&ipvs_netns_cnt);
1984         atomic_inc(&ipvs_netns_cnt);
1985         net->ipvs = ipvs;
1986
1987         if (ip_vs_estimator_net_init(net) < 0)
1988                 goto estimator_fail;
1989
1990         if (ip_vs_control_net_init(net) < 0)
1991                 goto control_fail;
1992
1993         if (ip_vs_protocol_net_init(net) < 0)
1994                 goto protocol_fail;
1995
1996         if (ip_vs_app_net_init(net) < 0)
1997                 goto app_fail;
1998
1999         if (ip_vs_conn_net_init(net) < 0)
2000                 goto conn_fail;
2001
2002         if (ip_vs_sync_net_init(net) < 0)
2003                 goto sync_fail;
2004
2005         printk(KERN_INFO "IPVS: Creating netns size=%zu id=%d\n",
2006                          sizeof(struct netns_ipvs), ipvs->gen);
2007         return 0;
2008 /*
2009  * Error handling
2010  */
2011
2012 sync_fail:
2013         ip_vs_conn_net_cleanup(net);
2014 conn_fail:
2015         ip_vs_app_net_cleanup(net);
2016 app_fail:
2017         ip_vs_protocol_net_cleanup(net);
2018 protocol_fail:
2019         ip_vs_control_net_cleanup(net);
2020 control_fail:
2021         ip_vs_estimator_net_cleanup(net);
2022 estimator_fail:
2023         net->ipvs = NULL;
2024         return -ENOMEM;
2025 }
2026
2027 static void __net_exit __ip_vs_cleanup(struct net *net)
2028 {
2029         ip_vs_service_net_cleanup(net); /* ip_vs_flush() with locks */
2030         ip_vs_conn_net_cleanup(net);
2031         ip_vs_app_net_cleanup(net);
2032         ip_vs_protocol_net_cleanup(net);
2033         ip_vs_control_net_cleanup(net);
2034         ip_vs_estimator_net_cleanup(net);
2035         IP_VS_DBG(2, "ipvs netns %d released\n", net_ipvs(net)->gen);
2036         net->ipvs = NULL;
2037 }
2038
2039 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
2040 {
2041         EnterFunction(2);
2042         net_ipvs(net)->enable = 0;      /* Disable packet reception */
2043         smp_wmb();
2044         ip_vs_sync_net_cleanup(net);
2045         LeaveFunction(2);
2046 }
2047
2048 static struct pernet_operations ipvs_core_ops = {
2049         .init = __ip_vs_init,
2050         .exit = __ip_vs_cleanup,
2051         .id   = &ip_vs_net_id,
2052         .size = sizeof(struct netns_ipvs),
2053 };
2054
2055 static struct pernet_operations ipvs_core_dev_ops = {
2056         .exit = __ip_vs_dev_cleanup,
2057 };
2058
2059 /*
2060  *      Initialize IP Virtual Server
2061  */
2062 static int __init ip_vs_init(void)
2063 {
2064         int ret;
2065
2066         ret = ip_vs_control_init();
2067         if (ret < 0) {
2068                 pr_err("can't setup control.\n");
2069                 goto exit;
2070         }
2071
2072         ip_vs_protocol_init();
2073
2074         ret = ip_vs_conn_init();
2075         if (ret < 0) {
2076                 pr_err("can't setup connection table.\n");
2077                 goto cleanup_protocol;
2078         }
2079
2080         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
2081         if (ret < 0)
2082                 goto cleanup_conn;
2083
2084         ret = register_pernet_device(&ipvs_core_dev_ops);
2085         if (ret < 0)
2086                 goto cleanup_sub;
2087
2088         ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2089         if (ret < 0) {
2090                 pr_err("can't register hooks.\n");
2091                 goto cleanup_dev;
2092         }
2093
2094         ret = ip_vs_register_nl_ioctl();
2095         if (ret < 0) {
2096                 pr_err("can't register netlink/ioctl.\n");
2097                 goto cleanup_hooks;
2098         }
2099
2100         pr_info("ipvs loaded.\n");
2101
2102         return ret;
2103
2104 cleanup_hooks:
2105         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2106 cleanup_dev:
2107         unregister_pernet_device(&ipvs_core_dev_ops);
2108 cleanup_sub:
2109         unregister_pernet_subsys(&ipvs_core_ops);
2110 cleanup_conn:
2111         ip_vs_conn_cleanup();
2112 cleanup_protocol:
2113         ip_vs_protocol_cleanup();
2114         ip_vs_control_cleanup();
2115 exit:
2116         return ret;
2117 }
2118
2119 static void __exit ip_vs_cleanup(void)
2120 {
2121         ip_vs_unregister_nl_ioctl();
2122         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2123         unregister_pernet_device(&ipvs_core_dev_ops);
2124         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
2125         ip_vs_conn_cleanup();
2126         ip_vs_protocol_cleanup();
2127         ip_vs_control_cleanup();
2128         pr_info("ipvs unloaded.\n");
2129 }
2130
2131 module_init(ip_vs_init);
2132 module_exit(ip_vs_cleanup);
2133 MODULE_LICENSE("GPL");