net/netfilter/ipvs/ip_vs_core.c

   1 /*
   2  * IPVS         An implementation of the IP virtual server support for the
   3  *              LINUX operating system.  IPVS is now implemented as a module
   4  *              over the Netfilter framework. IPVS can be used to build a
   5  *              high-performance and highly available server based on a
   6  *              cluster of servers.
   7  *
   8  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
   9  *              Peter Kese <peter.kese@ijs.si>
  10  *              Julian Anastasov <ja@ssi.bg>
  11  *
  12  *              This program is free software; you can redistribute it and/or
  13  *              modify it under the terms of the GNU General Public License
  14  *              as published by the Free Software Foundation; either version
  15  *              2 of the License, or (at your option) any later version.
  16  *
  17  * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
  18  * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
  19  * and others.
  20  *
  21  * Changes:
  22  *      Paul `Rusty' Russell            properly handle non-linear skbs
  23  *      Harald Welte                    don't use nfcache
  24  *
  25  */
  26
  27 #define KMSG_COMPONENT "IPVS"
  28 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  29
  30 #include <linux/module.h>
  31 #include <linux/kernel.h>
  32 #include <linux/ip.h>
  33 #include <linux/tcp.h>
  34 #include <linux/sctp.h>
  35 #include <linux/icmp.h>
  36 #include <linux/slab.h>
  37
  38 #include <net/ip.h>
  39 #include <net/tcp.h>
  40 #include <net/udp.h>
  41 #include <net/icmp.h>                   /* for icmp_send */
  42 #include <net/route.h>
  43 #include <net/ip6_checksum.h>
  44 #include <net/netns/generic.h>          /* net_generic() */
  45
  46 #include <linux/netfilter.h>
  47 #include <linux/netfilter_ipv4.h>
  48
  49 #ifdef CONFIG_IP_VS_IPV6
  50 #include <net/ipv6.h>
  51 #include <linux/netfilter_ipv6.h>
  52 #include <net/ip6_route.h>
  53 #endif
  54
  55 #include <net/ip_vs.h>
  56
  57
  58 EXPORT_SYMBOL(register_ip_vs_scheduler);
  59 EXPORT_SYMBOL(unregister_ip_vs_scheduler);
  60 EXPORT_SYMBOL(ip_vs_proto_name);
  61 EXPORT_SYMBOL(ip_vs_conn_new);
  62 EXPORT_SYMBOL(ip_vs_conn_in_get);
  63 EXPORT_SYMBOL(ip_vs_conn_out_get);
  64 #ifdef CONFIG_IP_VS_PROTO_TCP
  65 EXPORT_SYMBOL(ip_vs_tcp_conn_listen);
  66 #endif
  67 EXPORT_SYMBOL(ip_vs_conn_put);
  68 #ifdef CONFIG_IP_VS_DEBUG
  69 EXPORT_SYMBOL(ip_vs_get_debug_level);
  70 #endif
  71
  72 static int ip_vs_net_id __read_mostly;
  73 /* netns cnt used for uniqueness */
  74 static atomic_t ipvs_netns_cnt = ATOMIC_INIT(0);
  75
  76 /* ID used in ICMP lookups */
  77 #define icmp_id(icmph)          (((icmph)->un).echo.id)
  78 #define icmpv6_id(icmph)        (icmph->icmp6_dataun.u_echo.identifier)
  79
  80 const char *ip_vs_proto_name(unsigned int proto)
  81 {
  82         static char buf[20];
  83
  84         switch (proto) {
  85         case IPPROTO_IP:
  86                 return "IP";
  87         case IPPROTO_UDP:
  88                 return "UDP";
  89         case IPPROTO_TCP:
  90                 return "TCP";
  91         case IPPROTO_SCTP:
  92                 return "SCTP";
  93         case IPPROTO_ICMP:
  94                 return "ICMP";
  95 #ifdef CONFIG_IP_VS_IPV6
  96         case IPPROTO_ICMPV6:
  97                 return "ICMPv6";
  98 #endif
  99         default:
 100                 sprintf(buf, "IP_%u", proto);
 101                 return buf;
 102         }
 103 }
 104
 105 void ip_vs_init_hash_table(struct list_head *table, int rows)
 106 {
 107         while (--rows >= 0)
 108                 INIT_LIST_HEAD(&table[rows]);
 109 }
 110
 111 static inline void
 112 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 113 {
 114         struct ip_vs_dest *dest = cp->dest;
 115         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 116
 117         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 118                 struct ip_vs_cpu_stats *s;
 119                 struct ip_vs_service *svc;
 120
 121                 s = this_cpu_ptr(dest->stats.cpustats);
 122                 u64_stats_update_begin(&s->syncp);
 123                 s->cnt.inpkts++;
 124                 s->cnt.inbytes += skb->len;
 125                 u64_stats_update_end(&s->syncp);
 126
 127                 rcu_read_lock();
 128                 svc = rcu_dereference(dest->svc);
 129                 s = this_cpu_ptr(svc->stats.cpustats);
 130                 u64_stats_update_begin(&s->syncp);
 131                 s->cnt.inpkts++;
 132                 s->cnt.inbytes += skb->len;
 133                 u64_stats_update_end(&s->syncp);
 134                 rcu_read_unlock();
 135
 136                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 137                 u64_stats_update_begin(&s->syncp);
 138                 s->cnt.inpkts++;
 139                 s->cnt.inbytes += skb->len;
 140                 u64_stats_update_end(&s->syncp);
 141         }
 142 }
 143
 144
 145 static inline void
 146 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 147 {
 148         struct ip_vs_dest *dest = cp->dest;
 149         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 150
 151         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 152                 struct ip_vs_cpu_stats *s;
 153                 struct ip_vs_service *svc;
 154
 155                 s = this_cpu_ptr(dest->stats.cpustats);
 156                 u64_stats_update_begin(&s->syncp);
 157                 s->cnt.outpkts++;
 158                 s->cnt.outbytes += skb->len;
 159                 u64_stats_update_end(&s->syncp);
 160
 161                 rcu_read_lock();
 162                 svc = rcu_dereference(dest->svc);
 163                 s = this_cpu_ptr(svc->stats.cpustats);
 164                 u64_stats_update_begin(&s->syncp);
 165                 s->cnt.outpkts++;
 166                 s->cnt.outbytes += skb->len;
 167                 u64_stats_update_end(&s->syncp);
 168                 rcu_read_unlock();
 169
 170                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 171                 u64_stats_update_begin(&s->syncp);
 172                 s->cnt.outpkts++;
 173                 s->cnt.outbytes += skb->len;
 174                 u64_stats_update_end(&s->syncp);
 175         }
 176 }
 177
 178
 179 static inline void
 180 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
 181 {
 182         struct netns_ipvs *ipvs = net_ipvs(svc->net);
 183         struct ip_vs_cpu_stats *s;
 184
 185         s = this_cpu_ptr(cp->dest->stats.cpustats);
 186         u64_stats_update_begin(&s->syncp);
 187         s->cnt.conns++;
 188         u64_stats_update_end(&s->syncp);
 189
 190         s = this_cpu_ptr(svc->stats.cpustats);
 191         u64_stats_update_begin(&s->syncp);
 192         s->cnt.conns++;
 193         u64_stats_update_end(&s->syncp);
 194
 195         s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 196         u64_stats_update_begin(&s->syncp);
 197         s->cnt.conns++;
 198         u64_stats_update_end(&s->syncp);
 199 }
 200
 201
 202 static inline void
 203 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
 204                 const struct sk_buff *skb,
 205                 struct ip_vs_proto_data *pd)
 206 {
 207         if (likely(pd->pp->state_transition))
 208                 pd->pp->state_transition(cp, direction, skb, pd);
 209 }
 210
 211 static inline int
 212 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
 213                               struct sk_buff *skb, int protocol,
 214                               const union nf_inet_addr *caddr, __be16 cport,
 215                               const union nf_inet_addr *vaddr, __be16 vport,
 216                               struct ip_vs_conn_param *p)
 217 {
 218         ip_vs_conn_fill_param(svc->net, svc->af, protocol, caddr, cport, vaddr,
 219                               vport, p);
 220         p->pe = rcu_dereference(svc->pe);
 221         if (p->pe && p->pe->fill_param)
 222                 return p->pe->fill_param(p, skb);
 223
 224         return 0;
 225 }
 226
 227 /*
 228  *  IPVS persistent scheduling function
 229  *  It creates a connection entry according to its template if exists,
 230  *  or selects a server and creates a connection entry plus a template.
 231  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
 232  *  Protocols supported: TCP, UDP
 233  */
 234 static struct ip_vs_conn *
 235 ip_vs_sched_persist(struct ip_vs_service *svc,
 236                     struct sk_buff *skb, __be16 src_port, __be16 dst_port,
 237                     int *ignored, struct ip_vs_iphdr *iph)
 238 {
 239         struct ip_vs_conn *cp = NULL;
 240         struct ip_vs_dest *dest;
 241         struct ip_vs_conn *ct;
 242         __be16 dport = 0;               /* destination port to forward */
 243         unsigned int flags;
 244         struct ip_vs_conn_param param;
 245         const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
 246         union nf_inet_addr snet;        /* source network of the client,
 247                                            after masking */
 248
 249         /* Mask saddr with the netmask to adjust template granularity */
 250 #ifdef CONFIG_IP_VS_IPV6
 251         if (svc->af == AF_INET6)
 252                 ipv6_addr_prefix(&snet.in6, &iph->saddr.in6,
 253                                  (__force __u32) svc->netmask);
 254         else
 255 #endif
 256                 snet.ip = iph->saddr.ip & svc->netmask;
 257
 258         IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
 259                       "mnet %s\n",
 260                       IP_VS_DBG_ADDR(svc->af, &iph->saddr), ntohs(src_port),
 261                       IP_VS_DBG_ADDR(svc->af, &iph->daddr), ntohs(dst_port),
 262                       IP_VS_DBG_ADDR(svc->af, &snet));
 263
 264         /*
 265          * As far as we know, FTP is a very complicated network protocol, and
 266          * it uses control connection and data connections. For active FTP,
 267          * FTP server initialize data connection to the client, its source port
 268          * is often 20. For passive FTP, FTP server tells the clients the port
 269          * that it passively listens to,  and the client issues the data
 270          * connection. In the tunneling or direct routing mode, the load
 271          * balancer is on the client-to-server half of connection, the port
 272          * number is unknown to the load balancer. So, a conn template like
 273          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
 274          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
 275          * is created for other persistent services.
 276          */
 277         {
 278                 int protocol = iph->protocol;
 279                 const union nf_inet_addr *vaddr = &iph->daddr;
 280                 __be16 vport = 0;
 281
 282                 if (dst_port == svc->port) {
 283                         /* non-FTP template:
 284                          * <protocol, caddr, 0, vaddr, vport, daddr, dport>
 285                          * FTP template:
 286                          * <protocol, caddr, 0, vaddr, 0, daddr, 0>
 287                          */
 288                         if (svc->port != FTPPORT)
 289                                 vport = dst_port;
 290                 } else {
 291                         /* Note: persistent fwmark-based services and
 292                          * persistent port zero service are handled here.
 293                          * fwmark template:
 294                          * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
 295                          * port zero template:
 296                          * <protocol,caddr,0,vaddr,0,daddr,0>
 297                          */
 298                         if (svc->fwmark) {
 299                                 protocol = IPPROTO_IP;
 300                                 vaddr = &fwmark;
 301                         }
 302                 }
 303                 /* return *ignored = -1 so NF_DROP can be used */
 304                 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
 305                                                   vaddr, vport, &param) < 0) {
 306                         *ignored = -1;
 307                         return NULL;
 308                 }
 309         }
 310
 311         /* Check if a template already exists */
 312         ct = ip_vs_ct_in_get(&param);
 313         if (!ct || !ip_vs_check_template(ct)) {
 314                 struct ip_vs_scheduler *sched;
 315
 316                 /*
 317                  * No template found or the dest of the connection
 318                  * template is not available.
 319                  * return *ignored=0 i.e. ICMP and NF_DROP
 320                  */
 321                 sched = rcu_dereference(svc->scheduler);
 322                 if (sched) {
 323                         /* read svc->sched_data after svc->scheduler */
 324                         smp_rmb();
 325                         dest = sched->schedule(svc, skb, iph);
 326                 } else {
 327                         dest = NULL;
 328                 }
 329                 if (!dest) {
 330                         IP_VS_DBG(1, "p-schedule: no dest found.\n");
 331                         kfree(param.pe_data);
 332                         *ignored = 0;
 333                         return NULL;
 334                 }
 335
 336                 if (dst_port == svc->port && svc->port != FTPPORT)
 337                         dport = dest->port;
 338
 339                 /* Create a template
 340                  * This adds param.pe_data to the template,
 341                  * and thus param.pe_data will be destroyed
 342                  * when the template expires */
 343                 ct = ip_vs_conn_new(&param, dest->af, &dest->addr, dport,
 344                                     IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
 345                 if (ct == NULL) {
 346                         kfree(param.pe_data);
 347                         *ignored = -1;
 348                         return NULL;
 349                 }
 350
 351                 ct->timeout = svc->timeout;
 352         } else {
 353                 /* set destination with the found template */
 354                 dest = ct->dest;
 355                 kfree(param.pe_data);
 356         }
 357
 358         dport = dst_port;
 359         if (dport == svc->port && dest->port)
 360                 dport = dest->port;
 361
 362         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 363                  && iph->protocol == IPPROTO_UDP) ?
 364                 IP_VS_CONN_F_ONE_PACKET : 0;
 365
 366         /*
 367          *    Create a new connection according to the template
 368          */
 369         ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol, &iph->saddr,
 370                               src_port, &iph->daddr, dst_port, &param);
 371
 372         cp = ip_vs_conn_new(&param, dest->af, &dest->addr, dport, flags, dest,
 373                             skb->mark);
 374         if (cp == NULL) {
 375                 ip_vs_conn_put(ct);
 376                 *ignored = -1;
 377                 return NULL;
 378         }
 379
 380         /*
 381          *    Add its control
 382          */
 383         ip_vs_control_add(cp, ct);
 384         ip_vs_conn_put(ct);
 385
 386         ip_vs_conn_stats(cp, svc);
 387         return cp;
 388 }
 389
 390
 391 /*
 392  *  IPVS main scheduling function
 393  *  It selects a server according to the virtual service, and
 394  *  creates a connection entry.
 395  *  Protocols supported: TCP, UDP
 396  *
 397  *  Usage of *ignored
 398  *
 399  * 1 :   protocol tried to schedule (eg. on SYN), found svc but the
 400  *       svc/scheduler decides that this packet should be accepted with
 401  *       NF_ACCEPT because it must not be scheduled.
 402  *
 403  * 0 :   scheduler can not find destination, so try bypass or
 404  *       return ICMP and then NF_DROP (ip_vs_leave).
 405  *
 406  * -1 :  scheduler tried to schedule but fatal error occurred, eg.
 407  *       ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
 408  *       failure such as missing Call-ID, ENOMEM on skb_linearize
 409  *       or pe_data. In this case we should return NF_DROP without
 410  *       any attempts to send ICMP with ip_vs_leave.
 411  */
 412 struct ip_vs_conn *
 413 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
 414                struct ip_vs_proto_data *pd, int *ignored,
 415                struct ip_vs_iphdr *iph)
 416 {
 417         struct ip_vs_protocol *pp = pd->pp;
 418         struct ip_vs_conn *cp = NULL;
 419         struct ip_vs_scheduler *sched;
 420         struct ip_vs_dest *dest;
 421         __be16 _ports[2], *pptr;
 422         unsigned int flags;
 423
 424         *ignored = 1;
 425         /*
 426          * IPv6 frags, only the first hit here.
 427          */
 428         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 429         if (pptr == NULL)
 430                 return NULL;
 431
 432         /*
 433          * FTPDATA needs this check when using local real server.
 434          * Never schedule Active FTPDATA connections from real server.
 435          * For LVS-NAT they must be already created. For other methods
 436          * with persistence the connection is created on SYN+ACK.
 437          */
 438         if (pptr[0] == FTPDATA) {
 439                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 440                               "Not scheduling FTPDATA");
 441                 return NULL;
 442         }
 443
 444         /*
 445          *    Do not schedule replies from local real server.
 446          */
 447         if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
 448             (cp = pp->conn_in_get(svc->af, skb, iph, 1))) {
 449                 IP_VS_DBG_PKT(12, svc->af, pp, skb, 0,
 450                               "Not scheduling reply for existing connection");
 451                 __ip_vs_conn_put(cp);
 452                 return NULL;
 453         }
 454
 455         /*
 456          *    Persistent service
 457          */
 458         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 459                 return ip_vs_sched_persist(svc, skb, pptr[0], pptr[1], ignored,
 460                                            iph);
 461
 462         *ignored = 0;
 463
 464         /*
 465          *    Non-persistent service
 466          */
 467         if (!svc->fwmark && pptr[1] != svc->port) {
 468                 if (!svc->port)
 469                         pr_err("Schedule: port zero only supported "
 470                                "in persistent services, "
 471                                "check your ipvs configuration\n");
 472                 return NULL;
 473         }
 474
 475         sched = rcu_dereference(svc->scheduler);
 476         if (sched) {
 477                 /* read svc->sched_data after svc->scheduler */
 478                 smp_rmb();
 479                 dest = sched->schedule(svc, skb, iph);
 480         } else {
 481                 dest = NULL;
 482         }
 483         if (dest == NULL) {
 484                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 485                 return NULL;
 486         }
 487
 488         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 489                  && iph->protocol == IPPROTO_UDP) ?
 490                 IP_VS_CONN_F_ONE_PACKET : 0;
 491
 492         /*
 493          *    Create a connection entry.
 494          */
 495         {
 496                 struct ip_vs_conn_param p;
 497
 498                 ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol,
 499                                       &iph->saddr, pptr[0], &iph->daddr,
 500                                       pptr[1], &p);
 501                 cp = ip_vs_conn_new(&p, dest->af, &dest->addr,
 502                                     dest->port ? dest->port : pptr[1],
 503                                     flags, dest, skb->mark);
 504                 if (!cp) {
 505                         *ignored = -1;
 506                         return NULL;
 507                 }
 508         }
 509
 510         IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
 511                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
 512                       ip_vs_fwd_tag(cp),
 513                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 514                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 515                       IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
 516                       cp->flags, atomic_read(&cp->refcnt));
 517
 518         ip_vs_conn_stats(cp, svc);
 519         return cp;
 520 }
 521
 522
 523 /*
 524  *  Pass or drop the packet.
 525  *  Called by ip_vs_in, when the virtual service is available but
 526  *  no destination is available for a new connection.
 527  */
 528 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 529                 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
 530 {
 531         __be16 _ports[2], *pptr;
 532 #ifdef CONFIG_SYSCTL
 533         struct net *net;
 534         struct netns_ipvs *ipvs;
 535         int unicast;
 536 #endif
 537
 538         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 539         if (pptr == NULL) {
 540                 return NF_DROP;
 541         }
 542
 543 #ifdef CONFIG_SYSCTL
 544         net = skb_net(skb);
 545
 546 #ifdef CONFIG_IP_VS_IPV6
 547         if (svc->af == AF_INET6)
 548                 unicast = ipv6_addr_type(&iph->daddr.in6) & IPV6_ADDR_UNICAST;
 549         else
 550 #endif
 551                 unicast = (inet_addr_type(net, iph->daddr.ip) == RTN_UNICAST);
 552
 553         /* if it is fwmark-based service, the cache_bypass sysctl is up
 554            and the destination is a non-local unicast, then create
 555            a cache_bypass connection entry */
 556         ipvs = net_ipvs(net);
 557         if (ipvs->sysctl_cache_bypass && svc->fwmark && unicast) {
 558                 int ret;
 559                 struct ip_vs_conn *cp;
 560                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
 561                                       iph->protocol == IPPROTO_UDP) ?
 562                                       IP_VS_CONN_F_ONE_PACKET : 0;
 563                 union nf_inet_addr daddr =  { .all = { 0, 0, 0, 0 } };
 564
 565                 /* create a new connection entry */
 566                 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
 567                 {
 568                         struct ip_vs_conn_param p;
 569                         ip_vs_conn_fill_param(svc->net, svc->af, iph->protocol,
 570                                               &iph->saddr, pptr[0],
 571                                               &iph->daddr, pptr[1], &p);
 572                         cp = ip_vs_conn_new(&p, svc->af, &daddr, 0,
 573                                             IP_VS_CONN_F_BYPASS | flags,
 574                                             NULL, skb->mark);
 575                         if (!cp)
 576                                 return NF_DROP;
 577                 }
 578
 579                 /* statistics */
 580                 ip_vs_in_stats(cp, skb);
 581
 582                 /* set state */
 583                 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
 584
 585                 /* transmit the first SYN packet */
 586                 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
 587                 /* do not touch skb anymore */
 588
 589                 atomic_inc(&cp->in_pkts);
 590                 ip_vs_conn_put(cp);
 591                 return ret;
 592         }
 593 #endif
 594
 595         /*
 596          * When the virtual ftp service is presented, packets destined
 597          * for other services on the VIP may get here (except services
 598          * listed in the ipvs table), pass the packets, because it is
 599          * not ipvs job to decide to drop the packets.
 600          */
 601         if ((svc->port == FTPPORT) && (pptr[1] != FTPPORT))
 602                 return NF_ACCEPT;
 603
 604         /*
 605          * Notify the client that the destination is unreachable, and
 606          * release the socket buffer.
 607          * Since it is in IP layer, the TCP socket is not actually
 608          * created, the TCP RST packet cannot be sent, instead that
 609          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 610          */
 611 #ifdef CONFIG_IP_VS_IPV6
 612         if (svc->af == AF_INET6) {
 613                 if (!skb->dev) {
 614                         struct net *net_ = dev_net(skb_dst(skb)->dev);
 615
 616                         skb->dev = net_->loopback_dev;
 617                 }
 618                 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 619         } else
 620 #endif
 621                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 622
 623         return NF_DROP;
 624 }
 625
 626 #ifdef CONFIG_SYSCTL
 627
 628 static int sysctl_snat_reroute(struct sk_buff *skb)
 629 {
 630         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 631         return ipvs->sysctl_snat_reroute;
 632 }
 633
 634 static int sysctl_nat_icmp_send(struct net *net)
 635 {
 636         struct netns_ipvs *ipvs = net_ipvs(net);
 637         return ipvs->sysctl_nat_icmp_send;
 638 }
 639
 640 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
 641 {
 642         return ipvs->sysctl_expire_nodest_conn;
 643 }
 644
 645 #else
 646
 647 static int sysctl_snat_reroute(struct sk_buff *skb) { return 0; }
 648 static int sysctl_nat_icmp_send(struct net *net) { return 0; }
 649 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
 650
 651 #endif
 652
 653 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 654 {
 655         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 656 }
 657
 658 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
 659 {
 660         if (NF_INET_LOCAL_IN == hooknum)
 661                 return IP_DEFRAG_VS_IN;
 662         if (NF_INET_FORWARD == hooknum)
 663                 return IP_DEFRAG_VS_FWD;
 664         return IP_DEFRAG_VS_OUT;
 665 }
 666
 667 static inline int ip_vs_gather_frags(struct sk_buff *skb, u_int32_t user)
 668 {
 669         int err;
 670
 671         local_bh_disable();
 672         err = ip_defrag(skb, user);
 673         local_bh_enable();
 674         if (!err)
 675                 ip_send_check(ip_hdr(skb));
 676
 677         return err;
 678 }
 679
 680 static int ip_vs_route_me_harder(int af, struct sk_buff *skb,
 681                                  unsigned int hooknum)
 682 {
 683         if (!sysctl_snat_reroute(skb))
 684                 return 0;
 685         /* Reroute replies only to remote clients (FORWARD and LOCAL_OUT) */
 686         if (NF_INET_LOCAL_IN == hooknum)
 687                 return 0;
 688 #ifdef CONFIG_IP_VS_IPV6
 689         if (af == AF_INET6) {
 690                 struct dst_entry *dst = skb_dst(skb);
 691
 692                 if (dst->dev && !(dst->dev->flags & IFF_LOOPBACK) &&
 693                     ip6_route_me_harder(skb) != 0)
 694                         return 1;
 695         } else
 696 #endif
 697                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
 698                     ip_route_me_harder(skb, RTN_LOCAL) != 0)
 699                         return 1;
 700
 701         return 0;
 702 }
 703
 704 /*
 705  * Packet has been made sufficiently writable in caller
 706  * - inout: 1=in->out, 0=out->in
 707  */
 708 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 709                     struct ip_vs_conn *cp, int inout)
 710 {
 711         struct iphdr *iph        = ip_hdr(skb);
 712         unsigned int icmp_offset = iph->ihl*4;
 713         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
 714                                                       icmp_offset);
 715         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 716
 717         if (inout) {
 718                 iph->saddr = cp->vaddr.ip;
 719                 ip_send_check(iph);
 720                 ciph->daddr = cp->vaddr.ip;
 721                 ip_send_check(ciph);
 722         } else {
 723                 iph->daddr = cp->daddr.ip;
 724                 ip_send_check(iph);
 725                 ciph->saddr = cp->daddr.ip;
 726                 ip_send_check(ciph);
 727         }
 728
 729         /* the TCP/UDP/SCTP port */
 730         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
 731             IPPROTO_SCTP == ciph->protocol) {
 732                 __be16 *ports = (void *)ciph + ciph->ihl*4;
 733
 734                 if (inout)
 735                         ports[1] = cp->vport;
 736                 else
 737                         ports[0] = cp->dport;
 738         }
 739
 740         /* And finally the ICMP checksum */
 741         icmph->checksum = 0;
 742         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 743         skb->ip_summed = CHECKSUM_UNNECESSARY;
 744
 745         if (inout)
 746                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 747                         "Forwarding altered outgoing ICMP");
 748         else
 749                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 750                         "Forwarding altered incoming ICMP");
 751 }
 752
 753 #ifdef CONFIG_IP_VS_IPV6
 754 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
 755                     struct ip_vs_conn *cp, int inout)
 756 {
 757         struct ipv6hdr *iph      = ipv6_hdr(skb);
 758         unsigned int icmp_offset = 0;
 759         unsigned int offs        = 0; /* header offset*/
 760         int protocol;
 761         struct icmp6hdr *icmph;
 762         struct ipv6hdr *ciph;
 763         unsigned short fragoffs;
 764
 765         ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
 766         icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
 767         offs = icmp_offset + sizeof(struct icmp6hdr);
 768         ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
 769
 770         protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
 771
 772         if (inout) {
 773                 iph->saddr = cp->vaddr.in6;
 774                 ciph->daddr = cp->vaddr.in6;
 775         } else {
 776                 iph->daddr = cp->daddr.in6;
 777                 ciph->saddr = cp->daddr.in6;
 778         }
 779
 780         /* the TCP/UDP/SCTP port */
 781         if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 782                           IPPROTO_SCTP == protocol)) {
 783                 __be16 *ports = (void *)(skb_network_header(skb) + offs);
 784
 785                 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
 786                               ntohs(inout ? ports[1] : ports[0]),
 787                               ntohs(inout ? cp->vport : cp->dport));
 788                 if (inout)
 789                         ports[1] = cp->vport;
 790                 else
 791                         ports[0] = cp->dport;
 792         }
 793
 794         /* And finally the ICMP checksum */
 795         icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
 796                                               skb->len - icmp_offset,
 797                                               IPPROTO_ICMPV6, 0);
 798         skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
 799         skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
 800         skb->ip_summed = CHECKSUM_PARTIAL;
 801
 802         if (inout)
 803                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 804                               (void *)ciph - (void *)iph,
 805                               "Forwarding altered outgoing ICMPv6");
 806         else
 807                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 808                               (void *)ciph - (void *)iph,
 809                               "Forwarding altered incoming ICMPv6");
 810 }
 811 #endif
 812
 813 /* Handle relevant response ICMP messages - forward to the right
 814  * destination host.
 815  */
 816 static int handle_response_icmp(int af, struct sk_buff *skb,
 817                                 union nf_inet_addr *snet,
 818                                 __u8 protocol, struct ip_vs_conn *cp,
 819                                 struct ip_vs_protocol *pp,
 820                                 unsigned int offset, unsigned int ihl,
 821                                 unsigned int hooknum)
 822 {
 823         unsigned int verdict = NF_DROP;
 824
 825         if (IP_VS_FWD_METHOD(cp) != 0) {
 826                 pr_err("shouldn't reach here, because the box is on the "
 827                        "half connection in the tun/dr module.\n");
 828         }
 829
 830         /* Ensure the checksum is correct */
 831         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 832                 /* Failed checksum! */
 833                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
 834                               IP_VS_DBG_ADDR(af, snet));
 835                 goto out;
 836         }
 837
 838         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 839             IPPROTO_SCTP == protocol)
 840                 offset += 2 * sizeof(__u16);
 841         if (!skb_make_writable(skb, offset))
 842                 goto out;
 843
 844 #ifdef CONFIG_IP_VS_IPV6
 845         if (af == AF_INET6)
 846                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
 847         else
 848 #endif
 849                 ip_vs_nat_icmp(skb, pp, cp, 1);
 850
 851         if (ip_vs_route_me_harder(af, skb, hooknum))
 852                 goto out;
 853
 854         /* do the statistics and put it back */
 855         ip_vs_out_stats(cp, skb);
 856
 857         skb->ipvs_property = 1;
 858         if (!(cp->flags & IP_VS_CONN_F_NFCT))
 859                 ip_vs_notrack(skb);
 860         else
 861                 ip_vs_update_conntrack(skb, cp, 0);
 862         verdict = NF_ACCEPT;
 863
 864 out:
 865         __ip_vs_conn_put(cp);
 866
 867         return verdict;
 868 }
 869
 870 /*
 871  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 872  *      Find any that might be relevant, check against existing connections.
 873  *      Currently handles error types - unreachable, quench, ttl exceeded.
 874  */
 875 static int ip_vs_out_icmp(struct sk_buff *skb, int *related,
 876                           unsigned int hooknum)
 877 {
 878         struct iphdr *iph;
 879         struct icmphdr  _icmph, *ic;
 880         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 881         struct ip_vs_iphdr ciph;
 882         struct ip_vs_conn *cp;
 883         struct ip_vs_protocol *pp;
 884         unsigned int offset, ihl;
 885         union nf_inet_addr snet;
 886
 887         *related = 1;
 888
 889         /* reassemble IP fragments */
 890         if (ip_is_fragment(ip_hdr(skb))) {
 891                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
 892                         return NF_STOLEN;
 893         }
 894
 895         iph = ip_hdr(skb);
 896         offset = ihl = iph->ihl * 4;
 897         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 898         if (ic == NULL)
 899                 return NF_DROP;
 900
 901         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
 902                   ic->type, ntohs(icmp_id(ic)),
 903                   &iph->saddr, &iph->daddr);
 904
 905         /*
 906          * Work through seeing if this is for us.
 907          * These checks are supposed to be in an order that means easy
 908          * things are checked first to speed up processing.... however
 909          * this means that some packets will manage to get a long way
 910          * down this stack and then be rejected, but that's life.
 911          */
 912         if ((ic->type != ICMP_DEST_UNREACH) &&
 913             (ic->type != ICMP_SOURCE_QUENCH) &&
 914             (ic->type != ICMP_TIME_EXCEEDED)) {
 915                 *related = 0;
 916                 return NF_ACCEPT;
 917         }
 918
 919         /* Now find the contained IP header */
 920         offset += sizeof(_icmph);
 921         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 922         if (cih == NULL)
 923                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 924
 925         pp = ip_vs_proto_get(cih->protocol);
 926         if (!pp)
 927                 return NF_ACCEPT;
 928
 929         /* Is the embedded protocol header present? */
 930         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 931                      pp->dont_defrag))
 932                 return NF_ACCEPT;
 933
 934         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
 935                       "Checking outgoing ICMP for");
 936
 937         ip_vs_fill_ip4hdr(cih, &ciph);
 938         ciph.len += offset;
 939         /* The embedded headers contain source and dest in reverse order */
 940         cp = pp->conn_out_get(AF_INET, skb, &ciph, 1);
 941         if (!cp)
 942                 return NF_ACCEPT;
 943
 944         snet.ip = iph->saddr;
 945         return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
 946                                     pp, ciph.len, ihl, hooknum);
 947 }
 948
 949 #ifdef CONFIG_IP_VS_IPV6
 950 static int ip_vs_out_icmp_v6(struct sk_buff *skb, int *related,
 951                              unsigned int hooknum, struct ip_vs_iphdr *ipvsh)
 952 {
 953         struct icmp6hdr _icmph, *ic;
 954         struct ipv6hdr _ip6h, *ip6h; /* The ip header contained within ICMP */
 955         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
 956         struct ip_vs_conn *cp;
 957         struct ip_vs_protocol *pp;
 958         union nf_inet_addr snet;
 959         unsigned int writable;
 960
 961         *related = 1;
 962         ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph, ipvsh);
 963         if (ic == NULL)
 964                 return NF_DROP;
 965
 966         /*
 967          * Work through seeing if this is for us.
 968          * These checks are supposed to be in an order that means easy
 969          * things are checked first to speed up processing.... however
 970          * this means that some packets will manage to get a long way
 971          * down this stack and then be rejected, but that's life.
 972          */
 973         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
 974                 *related = 0;
 975                 return NF_ACCEPT;
 976         }
 977         /* Fragment header that is before ICMP header tells us that:
 978          * it's not an error message since they can't be fragmented.
 979          */
 980         if (ipvsh->flags & IP6_FH_F_FRAG)
 981                 return NF_DROP;
 982
 983         IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
 984                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
 985                   &ipvsh->saddr, &ipvsh->daddr);
 986
 987         /* Now find the contained IP header */
 988         ciph.len = ipvsh->len + sizeof(_icmph);
 989         ip6h = skb_header_pointer(skb, ciph.len, sizeof(_ip6h), &_ip6h);
 990         if (ip6h == NULL)
 991                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 992         ciph.saddr.in6 = ip6h->saddr; /* conn_out_get() handles reverse order */
 993         ciph.daddr.in6 = ip6h->daddr;
 994         /* skip possible IPv6 exthdrs of contained IPv6 packet */
 995         ciph.protocol = ipv6_find_hdr(skb, &ciph.len, -1, &ciph.fragoffs, NULL);
 996         if (ciph.protocol < 0)
 997                 return NF_ACCEPT; /* Contained IPv6 hdr looks wrong, ignore */
 998
 999         pp = ip_vs_proto_get(ciph.protocol);
1000         if (!pp)
1001                 return NF_ACCEPT;
1002
1003         /* The embedded headers contain source and dest in reverse order */
1004         cp = pp->conn_out_get(AF_INET6, skb, &ciph, 1);
1005         if (!cp)
1006                 return NF_ACCEPT;
1007
1008         snet.in6 = ciph.saddr.in6;
1009         writable = ciph.len;
1010         return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
1011                                     pp, writable, sizeof(struct ipv6hdr),
1012                                     hooknum);
1013 }
1014 #endif
1015
1016 /*
1017  * Check if sctp chunc is ABORT chunk
1018  */
1019 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
1020 {
1021         sctp_chunkhdr_t *sch, schunk;
1022         sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
1023                         sizeof(schunk), &schunk);
1024         if (sch == NULL)
1025                 return 0;
1026         if (sch->type == SCTP_CID_ABORT)
1027                 return 1;
1028         return 0;
1029 }
1030
1031 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
1032 {
1033         struct tcphdr _tcph, *th;
1034
1035         th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
1036         if (th == NULL)
1037                 return 0;
1038         return th->rst;
1039 }
1040
1041 static inline bool is_new_conn(const struct sk_buff *skb,
1042                                struct ip_vs_iphdr *iph)
1043 {
1044         switch (iph->protocol) {
1045         case IPPROTO_TCP: {
1046                 struct tcphdr _tcph, *th;
1047
1048                 th = skb_header_pointer(skb, iph->len, sizeof(_tcph), &_tcph);
1049                 if (th == NULL)
1050                         return false;
1051                 return th->syn;
1052         }
1053         case IPPROTO_SCTP: {
1054                 sctp_chunkhdr_t *sch, schunk;
1055
1056                 sch = skb_header_pointer(skb, iph->len + sizeof(sctp_sctphdr_t),
1057                                          sizeof(schunk), &schunk);
1058                 if (sch == NULL)
1059                         return false;
1060                 return sch->type == SCTP_CID_INIT;
1061         }
1062         default:
1063                 return false;
1064         }
1065 }
1066
1067 static inline bool is_new_conn_expected(const struct ip_vs_conn *cp,
1068                                         int conn_reuse_mode)
1069 {
1070         /* Controlled (FTP DATA or persistence)? */
1071         if (cp->control)
1072                 return false;
1073
1074         switch (cp->protocol) {
1075         case IPPROTO_TCP:
1076                 return (cp->state == IP_VS_TCP_S_TIME_WAIT) ||
1077                         ((conn_reuse_mode & 2) &&
1078                          (cp->state == IP_VS_TCP_S_FIN_WAIT) &&
1079                          (cp->flags & IP_VS_CONN_F_NOOUTPUT));
1080         case IPPROTO_SCTP:
1081                 return cp->state == IP_VS_SCTP_S_CLOSED;
1082         default:
1083                 return false;
1084         }
1085 }
1086
1087 /* Handle response packets: rewrite addresses and send away...
1088  */
1089 static unsigned int
1090 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1091                 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph,
1092                 unsigned int hooknum)
1093 {
1094         struct ip_vs_protocol *pp = pd->pp;
1095
1096         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Outgoing packet");
1097
1098         if (!skb_make_writable(skb, iph->len))
1099                 goto drop;
1100
1101         /* mangle the packet */
1102         if (pp->snat_handler && !pp->snat_handler(skb, pp, cp, iph))
1103                 goto drop;
1104
1105 #ifdef CONFIG_IP_VS_IPV6
1106         if (af == AF_INET6)
1107                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1108         else
1109 #endif
1110         {
1111                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1112                 ip_send_check(ip_hdr(skb));
1113         }
1114
1115         /*
1116          * nf_iterate does not expect change in the skb->dst->dev.
1117          * It looks like it is not fatal to enable this code for hooks
1118          * where our handlers are at the end of the chain list and
1119          * when all next handlers use skb->dst->dev and not outdev.
1120          * It will definitely route properly the inout NAT traffic
1121          * when multiple paths are used.
1122          */
1123
1124         /* For policy routing, packets originating from this
1125          * machine itself may be routed differently to packets
1126          * passing through.  We want this packet to be routed as
1127          * if it came from this machine itself.  So re-compute
1128          * the routing information.
1129          */
1130         if (ip_vs_route_me_harder(af, skb, hooknum))
1131                 goto drop;
1132
1133         IP_VS_DBG_PKT(10, af, pp, skb, 0, "After SNAT");
1134
1135         ip_vs_out_stats(cp, skb);
1136         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1137         skb->ipvs_property = 1;
1138         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1139                 ip_vs_notrack(skb);
1140         else
1141                 ip_vs_update_conntrack(skb, cp, 0);
1142         ip_vs_conn_put(cp);
1143
1144         LeaveFunction(11);
1145         return NF_ACCEPT;
1146
1147 drop:
1148         ip_vs_conn_put(cp);
1149         kfree_skb(skb);
1150         LeaveFunction(11);
1151         return NF_STOLEN;
1152 }
1153
1154 /*
1155  *      Check if outgoing packet belongs to the established ip_vs_conn.
1156  */
1157 static unsigned int
1158 ip_vs_out(unsigned int hooknum, struct sk_buff *skb, int af)
1159 {
1160         struct net *net = NULL;
1161         struct ip_vs_iphdr iph;
1162         struct ip_vs_protocol *pp;
1163         struct ip_vs_proto_data *pd;
1164         struct ip_vs_conn *cp;
1165
1166         EnterFunction(11);
1167
1168         /* Already marked as IPVS request or reply? */
1169         if (skb->ipvs_property)
1170                 return NF_ACCEPT;
1171
1172         /* Bad... Do not break raw sockets */
1173         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1174                      af == AF_INET)) {
1175                 struct sock *sk = skb->sk;
1176                 struct inet_sock *inet = inet_sk(skb->sk);
1177
1178                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1179                         return NF_ACCEPT;
1180         }
1181
1182         if (unlikely(!skb_dst(skb)))
1183                 return NF_ACCEPT;
1184
1185         net = skb_net(skb);
1186         if (!net_ipvs(net)->enable)
1187                 return NF_ACCEPT;
1188
1189         ip_vs_fill_iph_skb(af, skb, &iph);
1190 #ifdef CONFIG_IP_VS_IPV6
1191         if (af == AF_INET6) {
1192                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1193                         int related;
1194                         int verdict = ip_vs_out_icmp_v6(skb, &related,
1195                                                         hooknum, &iph);
1196
1197                         if (related)
1198                                 return verdict;
1199                 }
1200         } else
1201 #endif
1202                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1203                         int related;
1204                         int verdict = ip_vs_out_icmp(skb, &related, hooknum);
1205
1206                         if (related)
1207                                 return verdict;
1208                 }
1209
1210         pd = ip_vs_proto_data_get(net, iph.protocol);
1211         if (unlikely(!pd))
1212                 return NF_ACCEPT;
1213         pp = pd->pp;
1214
1215         /* reassemble IP fragments */
1216 #ifdef CONFIG_IP_VS_IPV6
1217         if (af == AF_INET)
1218 #endif
1219                 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1220                         if (ip_vs_gather_frags(skb,
1221                                                ip_vs_defrag_user(hooknum)))
1222                                 return NF_STOLEN;
1223
1224                         ip_vs_fill_ip4hdr(skb_network_header(skb), &iph);
1225                 }
1226
1227         /*
1228          * Check if the packet belongs to an existing entry
1229          */
1230         cp = pp->conn_out_get(af, skb, &iph, 0);
1231
1232         if (likely(cp))
1233                 return handle_response(af, skb, pd, cp, &iph, hooknum);
1234         if (sysctl_nat_icmp_send(net) &&
1235             (pp->protocol == IPPROTO_TCP ||
1236              pp->protocol == IPPROTO_UDP ||
1237              pp->protocol == IPPROTO_SCTP)) {
1238                 __be16 _ports[2], *pptr;
1239
1240                 pptr = frag_safe_skb_hp(skb, iph.len,
1241                                          sizeof(_ports), _ports, &iph);
1242                 if (pptr == NULL)
1243                         return NF_ACCEPT;       /* Not for me */
1244                 if (ip_vs_has_real_service(net, af, iph.protocol, &iph.saddr,
1245                                            pptr[0])) {
1246                         /*
1247                          * Notify the real server: there is no
1248                          * existing entry if it is not RST
1249                          * packet or not TCP packet.
1250                          */
1251                         if ((iph.protocol != IPPROTO_TCP &&
1252                              iph.protocol != IPPROTO_SCTP)
1253                              || ((iph.protocol == IPPROTO_TCP
1254                                   && !is_tcp_reset(skb, iph.len))
1255                                  || (iph.protocol == IPPROTO_SCTP
1256                                         && !is_sctp_abort(skb,
1257                                                 iph.len)))) {
1258 #ifdef CONFIG_IP_VS_IPV6
1259                                 if (af == AF_INET6) {
1260                                         if (!skb->dev)
1261                                                 skb->dev = net->loopback_dev;
1262                                         icmpv6_send(skb,
1263                                                     ICMPV6_DEST_UNREACH,
1264                                                     ICMPV6_PORT_UNREACH,
1265                                                     0);
1266                                 } else
1267 #endif
1268                                         icmp_send(skb,
1269                                                   ICMP_DEST_UNREACH,
1270                                                   ICMP_PORT_UNREACH, 0);
1271                                 return NF_DROP;
1272                         }
1273                 }
1274         }
1275         IP_VS_DBG_PKT(12, af, pp, skb, 0,
1276                       "ip_vs_out: packet continues traversal as normal");
1277         return NF_ACCEPT;
1278 }
1279
1280 /*
1281  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1282  *      used only for VS/NAT.
1283  *      Check if packet is reply for established ip_vs_conn.
1284  */
1285 static unsigned int
1286 ip_vs_reply4(const struct nf_hook_ops *ops, struct sk_buff *skb,
1287              const struct nf_hook_state *state)
1288 {
1289         return ip_vs_out(ops->hooknum, skb, AF_INET);
1290 }
1291
1292 /*
1293  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1294  *      Check if packet is reply for established ip_vs_conn.
1295  */
1296 static unsigned int
1297 ip_vs_local_reply4(const struct nf_hook_ops *ops, struct sk_buff *skb,
1298                    const struct nf_hook_state *state)
1299 {
1300         return ip_vs_out(ops->hooknum, skb, AF_INET);
1301 }
1302
1303 #ifdef CONFIG_IP_VS_IPV6
1304
1305 /*
1306  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1307  *      used only for VS/NAT.
1308  *      Check if packet is reply for established ip_vs_conn.
1309  */
1310 static unsigned int
1311 ip_vs_reply6(const struct nf_hook_ops *ops, struct sk_buff *skb,
1312              const struct nf_hook_state *state)
1313 {
1314         return ip_vs_out(ops->hooknum, skb, AF_INET6);
1315 }
1316
1317 /*
1318  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1319  *      Check if packet is reply for established ip_vs_conn.
1320  */
1321 static unsigned int
1322 ip_vs_local_reply6(const struct nf_hook_ops *ops, struct sk_buff *skb,
1323                    const struct nf_hook_state *state)
1324 {
1325         return ip_vs_out(ops->hooknum, skb, AF_INET6);
1326 }
1327
1328 #endif
1329
1330 /*
1331  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1332  *      Find any that might be relevant, check against existing connections,
1333  *      forward to the right destination host if relevant.
1334  *      Currently handles error types - unreachable, quench, ttl exceeded.
1335  */
1336 static int
1337 ip_vs_in_icmp(struct sk_buff *skb, int *related, unsigned int hooknum)
1338 {
1339         struct net *net = NULL;
1340         struct iphdr *iph;
1341         struct icmphdr  _icmph, *ic;
1342         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1343         struct ip_vs_iphdr ciph;
1344         struct ip_vs_conn *cp;
1345         struct ip_vs_protocol *pp;
1346         struct ip_vs_proto_data *pd;
1347         unsigned int offset, offset2, ihl, verdict;
1348         bool ipip;
1349
1350         *related = 1;
1351
1352         /* reassemble IP fragments */
1353         if (ip_is_fragment(ip_hdr(skb))) {
1354                 if (ip_vs_gather_frags(skb, ip_vs_defrag_user(hooknum)))
1355                         return NF_STOLEN;
1356         }
1357
1358         iph = ip_hdr(skb);
1359         offset = ihl = iph->ihl * 4;
1360         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1361         if (ic == NULL)
1362                 return NF_DROP;
1363
1364         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1365                   ic->type, ntohs(icmp_id(ic)),
1366                   &iph->saddr, &iph->daddr);
1367
1368         /*
1369          * Work through seeing if this is for us.
1370          * These checks are supposed to be in an order that means easy
1371          * things are checked first to speed up processing.... however
1372          * this means that some packets will manage to get a long way
1373          * down this stack and then be rejected, but that's life.
1374          */
1375         if ((ic->type != ICMP_DEST_UNREACH) &&
1376             (ic->type != ICMP_SOURCE_QUENCH) &&
1377             (ic->type != ICMP_TIME_EXCEEDED)) {
1378                 *related = 0;
1379                 return NF_ACCEPT;
1380         }
1381
1382         /* Now find the contained IP header */
1383         offset += sizeof(_icmph);
1384         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1385         if (cih == NULL)
1386                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1387
1388         net = skb_net(skb);
1389
1390         /* Special case for errors for IPIP packets */
1391         ipip = false;
1392         if (cih->protocol == IPPROTO_IPIP) {
1393                 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1394                         return NF_ACCEPT;
1395                 /* Error for our IPIP must arrive at LOCAL_IN */
1396                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1397                         return NF_ACCEPT;
1398                 offset += cih->ihl * 4;
1399                 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1400                 if (cih == NULL)
1401                         return NF_ACCEPT; /* The packet looks wrong, ignore */
1402                 ipip = true;
1403         }
1404
1405         pd = ip_vs_proto_data_get(net, cih->protocol);
1406         if (!pd)
1407                 return NF_ACCEPT;
1408         pp = pd->pp;
1409
1410         /* Is the embedded protocol header present? */
1411         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1412                      pp->dont_defrag))
1413                 return NF_ACCEPT;
1414
1415         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1416                       "Checking incoming ICMP for");
1417
1418         offset2 = offset;
1419         ip_vs_fill_ip4hdr(cih, &ciph);
1420         ciph.len += offset;
1421         offset = ciph.len;
1422         /* The embedded headers contain source and dest in reverse order.
1423          * For IPIP this is error for request, not for reply.
1424          */
1425         cp = pp->conn_in_get(AF_INET, skb, &ciph, ipip ? 0 : 1);
1426         if (!cp)
1427                 return NF_ACCEPT;
1428
1429         verdict = NF_DROP;
1430
1431         /* Ensure the checksum is correct */
1432         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1433                 /* Failed checksum! */
1434                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1435                           &iph->saddr);
1436                 goto out;
1437         }
1438
1439         if (ipip) {
1440                 __be32 info = ic->un.gateway;
1441                 __u8 type = ic->type;
1442                 __u8 code = ic->code;
1443
1444                 /* Update the MTU */
1445                 if (ic->type == ICMP_DEST_UNREACH &&
1446                     ic->code == ICMP_FRAG_NEEDED) {
1447                         struct ip_vs_dest *dest = cp->dest;
1448                         u32 mtu = ntohs(ic->un.frag.mtu);
1449                         __be16 frag_off = cih->frag_off;
1450
1451                         /* Strip outer IP and ICMP, go to IPIP header */
1452                         if (pskb_pull(skb, ihl + sizeof(_icmph)) == NULL)
1453                                 goto ignore_ipip;
1454                         offset2 -= ihl + sizeof(_icmph);
1455                         skb_reset_network_header(skb);
1456                         IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1457                                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1458                         ipv4_update_pmtu(skb, dev_net(skb->dev),
1459                                          mtu, 0, 0, 0, 0);
1460                         /* Client uses PMTUD? */
1461                         if (!(frag_off & htons(IP_DF)))
1462                                 goto ignore_ipip;
1463                         /* Prefer the resulting PMTU */
1464                         if (dest) {
1465                                 struct ip_vs_dest_dst *dest_dst;
1466
1467                                 rcu_read_lock();
1468                                 dest_dst = rcu_dereference(dest->dest_dst);
1469                                 if (dest_dst)
1470                                         mtu = dst_mtu(dest_dst->dst_cache);
1471                                 rcu_read_unlock();
1472                         }
1473                         if (mtu > 68 + sizeof(struct iphdr))
1474                                 mtu -= sizeof(struct iphdr);
1475                         info = htonl(mtu);
1476                 }
1477                 /* Strip outer IP, ICMP and IPIP, go to IP header of
1478                  * original request.
1479                  */
1480                 if (pskb_pull(skb, offset2) == NULL)
1481                         goto ignore_ipip;
1482                 skb_reset_network_header(skb);
1483                 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1484                         &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1485                         type, code, ntohl(info));
1486                 icmp_send(skb, type, code, info);
1487                 /* ICMP can be shorter but anyways, account it */
1488                 ip_vs_out_stats(cp, skb);
1489
1490 ignore_ipip:
1491                 consume_skb(skb);
1492                 verdict = NF_STOLEN;
1493                 goto out;
1494         }
1495
1496         /* do the statistics and put it back */
1497         ip_vs_in_stats(cp, skb);
1498         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol ||
1499             IPPROTO_SCTP == cih->protocol)
1500                 offset += 2 * sizeof(__u16);
1501         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1502
1503 out:
1504         __ip_vs_conn_put(cp);
1505
1506         return verdict;
1507 }
1508
1509 #ifdef CONFIG_IP_VS_IPV6
1510 static int ip_vs_in_icmp_v6(struct sk_buff *skb, int *related,
1511                             unsigned int hooknum, struct ip_vs_iphdr *iph)
1512 {
1513         struct net *net = NULL;
1514         struct ipv6hdr _ip6h, *ip6h;
1515         struct icmp6hdr _icmph, *ic;
1516         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1517         struct ip_vs_conn *cp;
1518         struct ip_vs_protocol *pp;
1519         struct ip_vs_proto_data *pd;
1520         unsigned int offs_ciph, writable, verdict;
1521
1522         *related = 1;
1523
1524         ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph, iph);
1525         if (ic == NULL)
1526                 return NF_DROP;
1527
1528         /*
1529          * Work through seeing if this is for us.
1530          * These checks are supposed to be in an order that means easy
1531          * things are checked first to speed up processing.... however
1532          * this means that some packets will manage to get a long way
1533          * down this stack and then be rejected, but that's life.
1534          */
1535         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1536                 *related = 0;
1537                 return NF_ACCEPT;
1538         }
1539         /* Fragment header that is before ICMP header tells us that:
1540          * it's not an error message since they can't be fragmented.
1541          */
1542         if (iph->flags & IP6_FH_F_FRAG)
1543                 return NF_DROP;
1544
1545         IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1546                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1547                   &iph->saddr, &iph->daddr);
1548
1549         /* Now find the contained IP header */
1550         ciph.len = iph->len + sizeof(_icmph);
1551         offs_ciph = ciph.len; /* Save ip header offset */
1552         ip6h = skb_header_pointer(skb, ciph.len, sizeof(_ip6h), &_ip6h);
1553         if (ip6h == NULL)
1554                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1555         ciph.saddr.in6 = ip6h->saddr; /* conn_in_get() handles reverse order */
1556         ciph.daddr.in6 = ip6h->daddr;
1557         /* skip possible IPv6 exthdrs of contained IPv6 packet */
1558         ciph.protocol = ipv6_find_hdr(skb, &ciph.len, -1, &ciph.fragoffs, NULL);
1559         if (ciph.protocol < 0)
1560                 return NF_ACCEPT; /* Contained IPv6 hdr looks wrong, ignore */
1561
1562         net = skb_net(skb);
1563         pd = ip_vs_proto_data_get(net, ciph.protocol);
1564         if (!pd)
1565                 return NF_ACCEPT;
1566         pp = pd->pp;
1567
1568         /* Cannot handle fragmented embedded protocol */
1569         if (ciph.fragoffs)
1570                 return NF_ACCEPT;
1571
1572         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offs_ciph,
1573                       "Checking incoming ICMPv6 for");
1574
1575         /* The embedded headers contain source and dest in reverse order
1576          * if not from localhost
1577          */
1578         cp = pp->conn_in_get(AF_INET6, skb, &ciph,
1579                              (hooknum == NF_INET_LOCAL_OUT) ? 0 : 1);
1580
1581         if (!cp)
1582                 return NF_ACCEPT;
1583         /* VS/TUN, VS/DR and LOCALNODE just let it go */
1584         if ((hooknum == NF_INET_LOCAL_OUT) &&
1585             (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1586                 __ip_vs_conn_put(cp);
1587                 return NF_ACCEPT;
1588         }
1589
1590         /* do the statistics and put it back */
1591         ip_vs_in_stats(cp, skb);
1592
1593         /* Need to mangle contained IPv6 header in ICMPv6 packet */
1594         writable = ciph.len;
1595         if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1596             IPPROTO_SCTP == ciph.protocol)
1597                 writable += 2 * sizeof(__u16); /* Also mangle ports */
1598
1599         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, writable, hooknum, &ciph);
1600
1601         __ip_vs_conn_put(cp);
1602
1603         return verdict;
1604 }
1605 #endif
1606
1607
1608 /*
1609  *      Check if it's for virtual services, look it up,
1610  *      and send it on its way...
1611  */
1612 static unsigned int
1613 ip_vs_in(unsigned int hooknum, struct sk_buff *skb, int af)
1614 {
1615         struct net *net;
1616         struct ip_vs_iphdr iph;
1617         struct ip_vs_protocol *pp;
1618         struct ip_vs_proto_data *pd;
1619         struct ip_vs_conn *cp;
1620         int ret, pkts;
1621         struct netns_ipvs *ipvs;
1622         int conn_reuse_mode;
1623
1624         /* Already marked as IPVS request or reply? */
1625         if (skb->ipvs_property)
1626                 return NF_ACCEPT;
1627
1628         /*
1629          *      Big tappo:
1630          *      - remote client: only PACKET_HOST
1631          *      - route: used for struct net when skb->dev is unset
1632          */
1633         if (unlikely((skb->pkt_type != PACKET_HOST &&
1634                       hooknum != NF_INET_LOCAL_OUT) ||
1635                      !skb_dst(skb))) {
1636                 ip_vs_fill_iph_skb(af, skb, &iph);
1637                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1638                               " ignored in hook %u\n",
1639                               skb->pkt_type, iph.protocol,
1640                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1641                 return NF_ACCEPT;
1642         }
1643         /* ipvs enabled in this netns ? */
1644         net = skb_net(skb);
1645         ipvs = net_ipvs(net);
1646         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1647                 return NF_ACCEPT;
1648
1649         ip_vs_fill_iph_skb(af, skb, &iph);
1650
1651         /* Bad... Do not break raw sockets */
1652         if (unlikely(skb->sk != NULL && hooknum == NF_INET_LOCAL_OUT &&
1653                      af == AF_INET)) {
1654                 struct sock *sk = skb->sk;
1655                 struct inet_sock *inet = inet_sk(skb->sk);
1656
1657                 if (inet && sk->sk_family == PF_INET && inet->nodefrag)
1658                         return NF_ACCEPT;
1659         }
1660
1661 #ifdef CONFIG_IP_VS_IPV6
1662         if (af == AF_INET6) {
1663                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1664                         int related;
1665                         int verdict = ip_vs_in_icmp_v6(skb, &related, hooknum,
1666                                                        &iph);
1667
1668                         if (related)
1669                                 return verdict;
1670                 }
1671         } else
1672 #endif
1673                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1674                         int related;
1675                         int verdict = ip_vs_in_icmp(skb, &related, hooknum);
1676
1677                         if (related)
1678                                 return verdict;
1679                 }
1680
1681         /* Protocol supported? */
1682         pd = ip_vs_proto_data_get(net, iph.protocol);
1683         if (unlikely(!pd))
1684                 return NF_ACCEPT;
1685         pp = pd->pp;
1686         /*
1687          * Check if the packet belongs to an existing connection entry
1688          */
1689         cp = pp->conn_in_get(af, skb, &iph, 0);
1690
1691         conn_reuse_mode = sysctl_conn_reuse_mode(ipvs);
1692         if (conn_reuse_mode && !iph.fragoffs &&
1693             is_new_conn(skb, &iph) && cp &&
1694             ((unlikely(sysctl_expire_nodest_conn(ipvs)) && cp->dest &&
1695               unlikely(!atomic_read(&cp->dest->weight))) ||
1696              unlikely(is_new_conn_expected(cp, conn_reuse_mode)))) {
1697                 if (!atomic_read(&cp->n_control))
1698                         ip_vs_conn_expire_now(cp);
1699                 __ip_vs_conn_put(cp);
1700                 cp = NULL;
1701         }
1702
1703         if (unlikely(!cp) && !iph.fragoffs) {
1704                 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1705                  * replayed fragment zero will already have created the cp
1706                  */
1707                 int v;
1708
1709                 /* Schedule and create new connection entry into &cp */
1710                 if (!pp->conn_schedule(af, skb, pd, &v, &cp, &iph))
1711                         return v;
1712         }
1713
1714         if (unlikely(!cp)) {
1715                 /* sorry, all this trouble for a no-hit :) */
1716                 IP_VS_DBG_PKT(12, af, pp, skb, 0,
1717                               "ip_vs_in: packet continues traversal as normal");
1718                 if (iph.fragoffs) {
1719                         /* Fragment that couldn't be mapped to a conn entry
1720                          * is missing module nf_defrag_ipv6
1721                          */
1722                         IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1723                         IP_VS_DBG_PKT(7, af, pp, skb, 0, "unhandled fragment");
1724                 }
1725                 return NF_ACCEPT;
1726         }
1727
1728         IP_VS_DBG_PKT(11, af, pp, skb, 0, "Incoming packet");
1729         /* Check the server status */
1730         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1731                 /* the destination server is not available */
1732
1733                 if (sysctl_expire_nodest_conn(ipvs)) {
1734                         /* try to expire the connection immediately */
1735                         ip_vs_conn_expire_now(cp);
1736                 }
1737                 /* don't restart its timer, and silently
1738                    drop the packet. */
1739                 __ip_vs_conn_put(cp);
1740                 return NF_DROP;
1741         }
1742
1743         ip_vs_in_stats(cp, skb);
1744         ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1745         if (cp->packet_xmit)
1746                 ret = cp->packet_xmit(skb, cp, pp, &iph);
1747                 /* do not touch skb anymore */
1748         else {
1749                 IP_VS_DBG_RL("warning: packet_xmit is null");
1750                 ret = NF_ACCEPT;
1751         }
1752
1753         /* Increase its packet counter and check if it is needed
1754          * to be synchronized
1755          *
1756          * Sync connection if it is about to close to
1757          * encorage the standby servers to update the connections timeout
1758          *
1759          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1760          */
1761
1762         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
1763                 pkts = sysctl_sync_threshold(ipvs);
1764         else
1765                 pkts = atomic_add_return(1, &cp->in_pkts);
1766
1767         if (ipvs->sync_state & IP_VS_STATE_MASTER)
1768                 ip_vs_sync_conn(net, cp, pkts);
1769
1770         ip_vs_conn_put(cp);
1771         return ret;
1772 }
1773
1774 /*
1775  *      AF_INET handler in NF_INET_LOCAL_IN chain
1776  *      Schedule and forward packets from remote clients
1777  */
1778 static unsigned int
1779 ip_vs_remote_request4(const struct nf_hook_ops *ops, struct sk_buff *skb,
1780                       const struct nf_hook_state *state)
1781 {
1782         return ip_vs_in(ops->hooknum, skb, AF_INET);
1783 }
1784
1785 /*
1786  *      AF_INET handler in NF_INET_LOCAL_OUT chain
1787  *      Schedule and forward packets from local clients
1788  */
1789 static unsigned int
1790 ip_vs_local_request4(const struct nf_hook_ops *ops, struct sk_buff *skb,
1791                      const struct nf_hook_state *state)
1792 {
1793         return ip_vs_in(ops->hooknum, skb, AF_INET);
1794 }
1795
1796 #ifdef CONFIG_IP_VS_IPV6
1797
1798 /*
1799  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
1800  *      Schedule and forward packets from remote clients
1801  */
1802 static unsigned int
1803 ip_vs_remote_request6(const struct nf_hook_ops *ops, struct sk_buff *skb,
1804                       const struct nf_hook_state *state)
1805 {
1806         return ip_vs_in(ops->hooknum, skb, AF_INET6);
1807 }
1808
1809 /*
1810  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
1811  *      Schedule and forward packets from local clients
1812  */
1813 static unsigned int
1814 ip_vs_local_request6(const struct nf_hook_ops *ops, struct sk_buff *skb,
1815                      const struct nf_hook_state *state)
1816 {
1817         return ip_vs_in(ops->hooknum, skb, AF_INET6);
1818 }
1819
1820 #endif
1821
1822
1823 /*
1824  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1825  *      related packets destined for 0.0.0.0/0.
1826  *      When fwmark-based virtual service is used, such as transparent
1827  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
1828  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1829  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1830  *      and send them to ip_vs_in_icmp.
1831  */
1832 static unsigned int
1833 ip_vs_forward_icmp(const struct nf_hook_ops *ops, struct sk_buff *skb,
1834                    const struct nf_hook_state *state)
1835 {
1836         int r;
1837         struct net *net;
1838         struct netns_ipvs *ipvs;
1839
1840         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1841                 return NF_ACCEPT;
1842
1843         /* ipvs enabled in this netns ? */
1844         net = skb_net(skb);
1845         ipvs = net_ipvs(net);
1846         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1847                 return NF_ACCEPT;
1848
1849         return ip_vs_in_icmp(skb, &r, ops->hooknum);
1850 }
1851
1852 #ifdef CONFIG_IP_VS_IPV6
1853 static unsigned int
1854 ip_vs_forward_icmp_v6(const struct nf_hook_ops *ops, struct sk_buff *skb,
1855                       const struct nf_hook_state *state)
1856 {
1857         int r;
1858         struct net *net;
1859         struct netns_ipvs *ipvs;
1860         struct ip_vs_iphdr iphdr;
1861
1862         ip_vs_fill_iph_skb(AF_INET6, skb, &iphdr);
1863         if (iphdr.protocol != IPPROTO_ICMPV6)
1864                 return NF_ACCEPT;
1865
1866         /* ipvs enabled in this netns ? */
1867         net = skb_net(skb);
1868         ipvs = net_ipvs(net);
1869         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1870                 return NF_ACCEPT;
1871
1872         return ip_vs_in_icmp_v6(skb, &r, ops->hooknum, &iphdr);
1873 }
1874 #endif
1875
1876
1877 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
1878         /* After packet filtering, change source only for VS/NAT */
1879         {
1880                 .hook           = ip_vs_reply4,
1881                 .owner          = THIS_MODULE,
1882                 .pf             = NFPROTO_IPV4,
1883                 .hooknum        = NF_INET_LOCAL_IN,
1884                 .priority       = NF_IP_PRI_NAT_SRC - 2,
1885         },
1886         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1887          * or VS/NAT(change destination), so that filtering rules can be
1888          * applied to IPVS. */
1889         {
1890                 .hook           = ip_vs_remote_request4,
1891                 .owner          = THIS_MODULE,
1892                 .pf             = NFPROTO_IPV4,
1893                 .hooknum        = NF_INET_LOCAL_IN,
1894                 .priority       = NF_IP_PRI_NAT_SRC - 1,
1895         },
1896         /* Before ip_vs_in, change source only for VS/NAT */
1897         {
1898                 .hook           = ip_vs_local_reply4,
1899                 .owner          = THIS_MODULE,
1900                 .pf             = NFPROTO_IPV4,
1901                 .hooknum        = NF_INET_LOCAL_OUT,
1902                 .priority       = NF_IP_PRI_NAT_DST + 1,
1903         },
1904         /* After mangle, schedule and forward local requests */
1905         {
1906                 .hook           = ip_vs_local_request4,
1907                 .owner          = THIS_MODULE,
1908                 .pf             = NFPROTO_IPV4,
1909                 .hooknum        = NF_INET_LOCAL_OUT,
1910                 .priority       = NF_IP_PRI_NAT_DST + 2,
1911         },
1912         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1913          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1914         {
1915                 .hook           = ip_vs_forward_icmp,
1916                 .owner          = THIS_MODULE,
1917                 .pf             = NFPROTO_IPV4,
1918                 .hooknum        = NF_INET_FORWARD,
1919                 .priority       = 99,
1920         },
1921         /* After packet filtering, change source only for VS/NAT */
1922         {
1923                 .hook           = ip_vs_reply4,
1924                 .owner          = THIS_MODULE,
1925                 .pf             = NFPROTO_IPV4,
1926                 .hooknum        = NF_INET_FORWARD,
1927                 .priority       = 100,
1928         },
1929 #ifdef CONFIG_IP_VS_IPV6
1930         /* After packet filtering, change source only for VS/NAT */
1931         {
1932                 .hook           = ip_vs_reply6,
1933                 .owner          = THIS_MODULE,
1934                 .pf             = NFPROTO_IPV6,
1935                 .hooknum        = NF_INET_LOCAL_IN,
1936                 .priority       = NF_IP6_PRI_NAT_SRC - 2,
1937         },
1938         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1939          * or VS/NAT(change destination), so that filtering rules can be
1940          * applied to IPVS. */
1941         {
1942                 .hook           = ip_vs_remote_request6,
1943                 .owner          = THIS_MODULE,
1944                 .pf             = NFPROTO_IPV6,
1945                 .hooknum        = NF_INET_LOCAL_IN,
1946                 .priority       = NF_IP6_PRI_NAT_SRC - 1,
1947         },
1948         /* Before ip_vs_in, change source only for VS/NAT */
1949         {
1950                 .hook           = ip_vs_local_reply6,
1951                 .owner          = THIS_MODULE,
1952                 .pf             = NFPROTO_IPV6,
1953                 .hooknum        = NF_INET_LOCAL_OUT,
1954                 .priority       = NF_IP6_PRI_NAT_DST + 1,
1955         },
1956         /* After mangle, schedule and forward local requests */
1957         {
1958                 .hook           = ip_vs_local_request6,
1959                 .owner          = THIS_MODULE,
1960                 .pf             = NFPROTO_IPV6,
1961                 .hooknum        = NF_INET_LOCAL_OUT,
1962                 .priority       = NF_IP6_PRI_NAT_DST + 2,
1963         },
1964         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1965          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1966         {
1967                 .hook           = ip_vs_forward_icmp_v6,
1968                 .owner          = THIS_MODULE,
1969                 .pf             = NFPROTO_IPV6,
1970                 .hooknum        = NF_INET_FORWARD,
1971                 .priority       = 99,
1972         },
1973         /* After packet filtering, change source only for VS/NAT */
1974         {
1975                 .hook           = ip_vs_reply6,
1976                 .owner          = THIS_MODULE,
1977                 .pf             = NFPROTO_IPV6,
1978                 .hooknum        = NF_INET_FORWARD,
1979                 .priority       = 100,
1980         },
1981 #endif
1982 };
1983 /*
1984  *      Initialize IP Virtual Server netns mem.
1985  */
1986 static int __net_init __ip_vs_init(struct net *net)
1987 {
1988         struct netns_ipvs *ipvs;
1989
1990         ipvs = net_generic(net, ip_vs_net_id);
1991         if (ipvs == NULL)
1992                 return -ENOMEM;
1993
1994         /* Hold the beast until a service is registerd */
1995         ipvs->enable = 0;
1996         ipvs->net = net;
1997         /* Counters used for creating unique names */
1998         ipvs->gen = atomic_read(&ipvs_netns_cnt);
1999         atomic_inc(&ipvs_netns_cnt);
2000         net->ipvs = ipvs;
2001
2002         if (ip_vs_estimator_net_init(net) < 0)
2003                 goto estimator_fail;
2004
2005         if (ip_vs_control_net_init(net) < 0)
2006                 goto control_fail;
2007
2008         if (ip_vs_protocol_net_init(net) < 0)
2009                 goto protocol_fail;
2010
2011         if (ip_vs_app_net_init(net) < 0)
2012                 goto app_fail;
2013
2014         if (ip_vs_conn_net_init(net) < 0)
2015                 goto conn_fail;
2016
2017         if (ip_vs_sync_net_init(net) < 0)
2018                 goto sync_fail;
2019
2020         printk(KERN_INFO "IPVS: Creating netns size=%zu id=%d\n",
2021                          sizeof(struct netns_ipvs), ipvs->gen);
2022         return 0;
2023 /*
2024  * Error handling
2025  */
2026
2027 sync_fail:
2028         ip_vs_conn_net_cleanup(net);
2029 conn_fail:
2030         ip_vs_app_net_cleanup(net);
2031 app_fail:
2032         ip_vs_protocol_net_cleanup(net);
2033 protocol_fail:
2034         ip_vs_control_net_cleanup(net);
2035 control_fail:
2036         ip_vs_estimator_net_cleanup(net);
2037 estimator_fail:
2038         net->ipvs = NULL;
2039         return -ENOMEM;
2040 }
2041
2042 static void __net_exit __ip_vs_cleanup(struct net *net)
2043 {
2044         ip_vs_service_net_cleanup(net); /* ip_vs_flush() with locks */
2045         ip_vs_conn_net_cleanup(net);
2046         ip_vs_app_net_cleanup(net);
2047         ip_vs_protocol_net_cleanup(net);
2048         ip_vs_control_net_cleanup(net);
2049         ip_vs_estimator_net_cleanup(net);
2050         IP_VS_DBG(2, "ipvs netns %d released\n", net_ipvs(net)->gen);
2051         net->ipvs = NULL;
2052 }
2053
2054 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
2055 {
2056         EnterFunction(2);
2057         net_ipvs(net)->enable = 0;      /* Disable packet reception */
2058         smp_wmb();
2059         ip_vs_sync_net_cleanup(net);
2060         LeaveFunction(2);
2061 }
2062
2063 static struct pernet_operations ipvs_core_ops = {
2064         .init = __ip_vs_init,
2065         .exit = __ip_vs_cleanup,
2066         .id   = &ip_vs_net_id,
2067         .size = sizeof(struct netns_ipvs),
2068 };
2069
2070 static struct pernet_operations ipvs_core_dev_ops = {
2071         .exit = __ip_vs_dev_cleanup,
2072 };
2073
2074 /*
2075  *      Initialize IP Virtual Server
2076  */
2077 static int __init ip_vs_init(void)
2078 {
2079         int ret;
2080
2081         ret = ip_vs_control_init();
2082         if (ret < 0) {
2083                 pr_err("can't setup control.\n");
2084                 goto exit;
2085         }
2086
2087         ip_vs_protocol_init();
2088
2089         ret = ip_vs_conn_init();
2090         if (ret < 0) {
2091                 pr_err("can't setup connection table.\n");
2092                 goto cleanup_protocol;
2093         }
2094
2095         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
2096         if (ret < 0)
2097                 goto cleanup_conn;
2098
2099         ret = register_pernet_device(&ipvs_core_dev_ops);
2100         if (ret < 0)
2101                 goto cleanup_sub;
2102
2103         ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2104         if (ret < 0) {
2105                 pr_err("can't register hooks.\n");
2106                 goto cleanup_dev;
2107         }
2108
2109         ret = ip_vs_register_nl_ioctl();
2110         if (ret < 0) {
2111                 pr_err("can't register netlink/ioctl.\n");
2112                 goto cleanup_hooks;
2113         }
2114
2115         pr_info("ipvs loaded.\n");
2116
2117         return ret;
2118
2119 cleanup_hooks:
2120         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2121 cleanup_dev:
2122         unregister_pernet_device(&ipvs_core_dev_ops);
2123 cleanup_sub:
2124         unregister_pernet_subsys(&ipvs_core_ops);
2125 cleanup_conn:
2126         ip_vs_conn_cleanup();
2127 cleanup_protocol:
2128         ip_vs_protocol_cleanup();
2129         ip_vs_control_cleanup();
2130 exit:
2131         return ret;
2132 }
2133
2134 static void __exit ip_vs_cleanup(void)
2135 {
2136         ip_vs_unregister_nl_ioctl();
2137         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2138         unregister_pernet_device(&ipvs_core_dev_ops);
2139         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
2140         ip_vs_conn_cleanup();
2141         ip_vs_protocol_cleanup();
2142         ip_vs_control_cleanup();
2143         pr_info("ipvs unloaded.\n");
2144 }
2145
2146 module_init(ip_vs_init);
2147 module_exit(ip_vs_cleanup);
2148 MODULE_LICENSE("GPL");