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 EXPORT_SYMBOL(ip_vs_new_conn_out);
  72
  73 static int ip_vs_net_id __read_mostly;
  74 /* netns cnt used for uniqueness */
  75 static atomic_t ipvs_netns_cnt = ATOMIC_INIT(0);
  76
  77 /* ID used in ICMP lookups */
  78 #define icmp_id(icmph)          (((icmph)->un).echo.id)
  79 #define icmpv6_id(icmph)        (icmph->icmp6_dataun.u_echo.identifier)
  80
  81 const char *ip_vs_proto_name(unsigned int proto)
  82 {
  83         static char buf[20];
  84
  85         switch (proto) {
  86         case IPPROTO_IP:
  87                 return "IP";
  88         case IPPROTO_UDP:
  89                 return "UDP";
  90         case IPPROTO_TCP:
  91                 return "TCP";
  92         case IPPROTO_SCTP:
  93                 return "SCTP";
  94         case IPPROTO_ICMP:
  95                 return "ICMP";
  96 #ifdef CONFIG_IP_VS_IPV6
  97         case IPPROTO_ICMPV6:
  98                 return "ICMPv6";
  99 #endif
 100         default:
 101                 sprintf(buf, "IP_%u", proto);
 102                 return buf;
 103         }
 104 }
 105
 106 void ip_vs_init_hash_table(struct list_head *table, int rows)
 107 {
 108         while (--rows >= 0)
 109                 INIT_LIST_HEAD(&table[rows]);
 110 }
 111
 112 static inline void
 113 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 114 {
 115         struct ip_vs_dest *dest = cp->dest;
 116         struct netns_ipvs *ipvs = cp->ipvs;
 117
 118         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 119                 struct ip_vs_cpu_stats *s;
 120                 struct ip_vs_service *svc;
 121
 122                 s = this_cpu_ptr(dest->stats.cpustats);
 123                 u64_stats_update_begin(&s->syncp);
 124                 s->cnt.inpkts++;
 125                 s->cnt.inbytes += skb->len;
 126                 u64_stats_update_end(&s->syncp);
 127
 128                 rcu_read_lock();
 129                 svc = rcu_dereference(dest->svc);
 130                 s = this_cpu_ptr(svc->stats.cpustats);
 131                 u64_stats_update_begin(&s->syncp);
 132                 s->cnt.inpkts++;
 133                 s->cnt.inbytes += skb->len;
 134                 u64_stats_update_end(&s->syncp);
 135                 rcu_read_unlock();
 136
 137                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 138                 u64_stats_update_begin(&s->syncp);
 139                 s->cnt.inpkts++;
 140                 s->cnt.inbytes += skb->len;
 141                 u64_stats_update_end(&s->syncp);
 142         }
 143 }
 144
 145
 146 static inline void
 147 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 148 {
 149         struct ip_vs_dest *dest = cp->dest;
 150         struct netns_ipvs *ipvs = cp->ipvs;
 151
 152         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 153                 struct ip_vs_cpu_stats *s;
 154                 struct ip_vs_service *svc;
 155
 156                 s = this_cpu_ptr(dest->stats.cpustats);
 157                 u64_stats_update_begin(&s->syncp);
 158                 s->cnt.outpkts++;
 159                 s->cnt.outbytes += skb->len;
 160                 u64_stats_update_end(&s->syncp);
 161
 162                 rcu_read_lock();
 163                 svc = rcu_dereference(dest->svc);
 164                 s = this_cpu_ptr(svc->stats.cpustats);
 165                 u64_stats_update_begin(&s->syncp);
 166                 s->cnt.outpkts++;
 167                 s->cnt.outbytes += skb->len;
 168                 u64_stats_update_end(&s->syncp);
 169                 rcu_read_unlock();
 170
 171                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 172                 u64_stats_update_begin(&s->syncp);
 173                 s->cnt.outpkts++;
 174                 s->cnt.outbytes += skb->len;
 175                 u64_stats_update_end(&s->syncp);
 176         }
 177 }
 178
 179
 180 static inline void
 181 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
 182 {
 183         struct netns_ipvs *ipvs = svc->ipvs;
 184         struct ip_vs_cpu_stats *s;
 185
 186         s = this_cpu_ptr(cp->dest->stats.cpustats);
 187         u64_stats_update_begin(&s->syncp);
 188         s->cnt.conns++;
 189         u64_stats_update_end(&s->syncp);
 190
 191         s = this_cpu_ptr(svc->stats.cpustats);
 192         u64_stats_update_begin(&s->syncp);
 193         s->cnt.conns++;
 194         u64_stats_update_end(&s->syncp);
 195
 196         s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 197         u64_stats_update_begin(&s->syncp);
 198         s->cnt.conns++;
 199         u64_stats_update_end(&s->syncp);
 200 }
 201
 202
 203 static inline void
 204 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
 205                 const struct sk_buff *skb,
 206                 struct ip_vs_proto_data *pd)
 207 {
 208         if (likely(pd->pp->state_transition))
 209                 pd->pp->state_transition(cp, direction, skb, pd);
 210 }
 211
 212 static inline int
 213 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
 214                               struct sk_buff *skb, int protocol,
 215                               const union nf_inet_addr *caddr, __be16 cport,
 216                               const union nf_inet_addr *vaddr, __be16 vport,
 217                               struct ip_vs_conn_param *p)
 218 {
 219         ip_vs_conn_fill_param(svc->ipvs, svc->af, protocol, caddr, cport, vaddr,
 220                               vport, p);
 221         p->pe = rcu_dereference(svc->pe);
 222         if (p->pe && p->pe->fill_param)
 223                 return p->pe->fill_param(p, skb);
 224
 225         return 0;
 226 }
 227
 228 /*
 229  *  IPVS persistent scheduling function
 230  *  It creates a connection entry according to its template if exists,
 231  *  or selects a server and creates a connection entry plus a template.
 232  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
 233  *  Protocols supported: TCP, UDP
 234  */
 235 static struct ip_vs_conn *
 236 ip_vs_sched_persist(struct ip_vs_service *svc,
 237                     struct sk_buff *skb, __be16 src_port, __be16 dst_port,
 238                     int *ignored, struct ip_vs_iphdr *iph)
 239 {
 240         struct ip_vs_conn *cp = NULL;
 241         struct ip_vs_dest *dest;
 242         struct ip_vs_conn *ct;
 243         __be16 dport = 0;               /* destination port to forward */
 244         unsigned int flags;
 245         struct ip_vs_conn_param param;
 246         const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
 247         union nf_inet_addr snet;        /* source network of the client,
 248                                            after masking */
 249         const union nf_inet_addr *src_addr, *dst_addr;
 250
 251         if (likely(!ip_vs_iph_inverse(iph))) {
 252                 src_addr = &iph->saddr;
 253                 dst_addr = &iph->daddr;
 254         } else {
 255                 src_addr = &iph->daddr;
 256                 dst_addr = &iph->saddr;
 257         }
 258
 259
 260         /* Mask saddr with the netmask to adjust template granularity */
 261 #ifdef CONFIG_IP_VS_IPV6
 262         if (svc->af == AF_INET6)
 263                 ipv6_addr_prefix(&snet.in6, &src_addr->in6,
 264                                  (__force __u32) svc->netmask);
 265         else
 266 #endif
 267                 snet.ip = src_addr->ip & svc->netmask;
 268
 269         IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
 270                       "mnet %s\n",
 271                       IP_VS_DBG_ADDR(svc->af, src_addr), ntohs(src_port),
 272                       IP_VS_DBG_ADDR(svc->af, dst_addr), ntohs(dst_port),
 273                       IP_VS_DBG_ADDR(svc->af, &snet));
 274
 275         /*
 276          * As far as we know, FTP is a very complicated network protocol, and
 277          * it uses control connection and data connections. For active FTP,
 278          * FTP server initialize data connection to the client, its source port
 279          * is often 20. For passive FTP, FTP server tells the clients the port
 280          * that it passively listens to,  and the client issues the data
 281          * connection. In the tunneling or direct routing mode, the load
 282          * balancer is on the client-to-server half of connection, the port
 283          * number is unknown to the load balancer. So, a conn template like
 284          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
 285          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
 286          * is created for other persistent services.
 287          */
 288         {
 289                 int protocol = iph->protocol;
 290                 const union nf_inet_addr *vaddr = dst_addr;
 291                 __be16 vport = 0;
 292
 293                 if (dst_port == svc->port) {
 294                         /* non-FTP template:
 295                          * <protocol, caddr, 0, vaddr, vport, daddr, dport>
 296                          * FTP template:
 297                          * <protocol, caddr, 0, vaddr, 0, daddr, 0>
 298                          */
 299                         if (svc->port != FTPPORT)
 300                                 vport = dst_port;
 301                 } else {
 302                         /* Note: persistent fwmark-based services and
 303                          * persistent port zero service are handled here.
 304                          * fwmark template:
 305                          * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
 306                          * port zero template:
 307                          * <protocol,caddr,0,vaddr,0,daddr,0>
 308                          */
 309                         if (svc->fwmark) {
 310                                 protocol = IPPROTO_IP;
 311                                 vaddr = &fwmark;
 312                         }
 313                 }
 314                 /* return *ignored = -1 so NF_DROP can be used */
 315                 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
 316                                                   vaddr, vport, &param) < 0) {
 317                         *ignored = -1;
 318                         return NULL;
 319                 }
 320         }
 321
 322         /* Check if a template already exists */
 323         ct = ip_vs_ct_in_get(&param);
 324         if (!ct || !ip_vs_check_template(ct)) {
 325                 struct ip_vs_scheduler *sched;
 326
 327                 /*
 328                  * No template found or the dest of the connection
 329                  * template is not available.
 330                  * return *ignored=0 i.e. ICMP and NF_DROP
 331                  */
 332                 sched = rcu_dereference(svc->scheduler);
 333                 if (sched) {
 334                         /* read svc->sched_data after svc->scheduler */
 335                         smp_rmb();
 336                         dest = sched->schedule(svc, skb, iph);
 337                 } else {
 338                         dest = NULL;
 339                 }
 340                 if (!dest) {
 341                         IP_VS_DBG(1, "p-schedule: no dest found.\n");
 342                         kfree(param.pe_data);
 343                         *ignored = 0;
 344                         return NULL;
 345                 }
 346
 347                 if (dst_port == svc->port && svc->port != FTPPORT)
 348                         dport = dest->port;
 349
 350                 /* Create a template
 351                  * This adds param.pe_data to the template,
 352                  * and thus param.pe_data will be destroyed
 353                  * when the template expires */
 354                 ct = ip_vs_conn_new(&param, dest->af, &dest->addr, dport,
 355                                     IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
 356                 if (ct == NULL) {
 357                         kfree(param.pe_data);
 358                         *ignored = -1;
 359                         return NULL;
 360                 }
 361
 362                 ct->timeout = svc->timeout;
 363         } else {
 364                 /* set destination with the found template */
 365                 dest = ct->dest;
 366                 kfree(param.pe_data);
 367         }
 368
 369         dport = dst_port;
 370         if (dport == svc->port && dest->port)
 371                 dport = dest->port;
 372
 373         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 374                  && iph->protocol == IPPROTO_UDP) ?
 375                 IP_VS_CONN_F_ONE_PACKET : 0;
 376
 377         /*
 378          *    Create a new connection according to the template
 379          */
 380         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol, src_addr,
 381                               src_port, dst_addr, dst_port, &param);
 382
 383         cp = ip_vs_conn_new(&param, dest->af, &dest->addr, dport, flags, dest,
 384                             skb->mark);
 385         if (cp == NULL) {
 386                 ip_vs_conn_put(ct);
 387                 *ignored = -1;
 388                 return NULL;
 389         }
 390
 391         /*
 392          *    Add its control
 393          */
 394         ip_vs_control_add(cp, ct);
 395         ip_vs_conn_put(ct);
 396
 397         ip_vs_conn_stats(cp, svc);
 398         return cp;
 399 }
 400
 401
 402 /*
 403  *  IPVS main scheduling function
 404  *  It selects a server according to the virtual service, and
 405  *  creates a connection entry.
 406  *  Protocols supported: TCP, UDP
 407  *
 408  *  Usage of *ignored
 409  *
 410  * 1 :   protocol tried to schedule (eg. on SYN), found svc but the
 411  *       svc/scheduler decides that this packet should be accepted with
 412  *       NF_ACCEPT because it must not be scheduled.
 413  *
 414  * 0 :   scheduler can not find destination, so try bypass or
 415  *       return ICMP and then NF_DROP (ip_vs_leave).
 416  *
 417  * -1 :  scheduler tried to schedule but fatal error occurred, eg.
 418  *       ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
 419  *       failure such as missing Call-ID, ENOMEM on skb_linearize
 420  *       or pe_data. In this case we should return NF_DROP without
 421  *       any attempts to send ICMP with ip_vs_leave.
 422  */
 423 struct ip_vs_conn *
 424 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
 425                struct ip_vs_proto_data *pd, int *ignored,
 426                struct ip_vs_iphdr *iph)
 427 {
 428         struct ip_vs_protocol *pp = pd->pp;
 429         struct ip_vs_conn *cp = NULL;
 430         struct ip_vs_scheduler *sched;
 431         struct ip_vs_dest *dest;
 432         __be16 _ports[2], *pptr, cport, vport;
 433         const void *caddr, *vaddr;
 434         unsigned int flags;
 435
 436         *ignored = 1;
 437         /*
 438          * IPv6 frags, only the first hit here.
 439          */
 440         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 441         if (pptr == NULL)
 442                 return NULL;
 443
 444         if (likely(!ip_vs_iph_inverse(iph))) {
 445                 cport = pptr[0];
 446                 caddr = &iph->saddr;
 447                 vport = pptr[1];
 448                 vaddr = &iph->daddr;
 449         } else {
 450                 cport = pptr[1];
 451                 caddr = &iph->daddr;
 452                 vport = pptr[0];
 453                 vaddr = &iph->saddr;
 454         }
 455
 456         /*
 457          * FTPDATA needs this check when using local real server.
 458          * Never schedule Active FTPDATA connections from real server.
 459          * For LVS-NAT they must be already created. For other methods
 460          * with persistence the connection is created on SYN+ACK.
 461          */
 462         if (cport == FTPDATA) {
 463                 IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
 464                               "Not scheduling FTPDATA");
 465                 return NULL;
 466         }
 467
 468         /*
 469          *    Do not schedule replies from local real server.
 470          */
 471         if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK)) {
 472                 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
 473                 cp = pp->conn_in_get(svc->ipvs, svc->af, skb, iph);
 474                 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
 475
 476                 if (cp) {
 477                         IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
 478                                       "Not scheduling reply for existing"
 479                                       " connection");
 480                         __ip_vs_conn_put(cp);
 481                         return NULL;
 482                 }
 483         }
 484
 485         /*
 486          *    Persistent service
 487          */
 488         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 489                 return ip_vs_sched_persist(svc, skb, cport, vport, ignored,
 490                                            iph);
 491
 492         *ignored = 0;
 493
 494         /*
 495          *    Non-persistent service
 496          */
 497         if (!svc->fwmark && vport != svc->port) {
 498                 if (!svc->port)
 499                         pr_err("Schedule: port zero only supported "
 500                                "in persistent services, "
 501                                "check your ipvs configuration\n");
 502                 return NULL;
 503         }
 504
 505         sched = rcu_dereference(svc->scheduler);
 506         if (sched) {
 507                 /* read svc->sched_data after svc->scheduler */
 508                 smp_rmb();
 509                 dest = sched->schedule(svc, skb, iph);
 510         } else {
 511                 dest = NULL;
 512         }
 513         if (dest == NULL) {
 514                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 515                 return NULL;
 516         }
 517
 518         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 519                  && iph->protocol == IPPROTO_UDP) ?
 520                 IP_VS_CONN_F_ONE_PACKET : 0;
 521
 522         /*
 523          *    Create a connection entry.
 524          */
 525         {
 526                 struct ip_vs_conn_param p;
 527
 528                 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
 529                                       caddr, cport, vaddr, vport, &p);
 530                 cp = ip_vs_conn_new(&p, dest->af, &dest->addr,
 531                                     dest->port ? dest->port : vport,
 532                                     flags, dest, skb->mark);
 533                 if (!cp) {
 534                         *ignored = -1;
 535                         return NULL;
 536                 }
 537         }
 538
 539         IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
 540                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
 541                       ip_vs_fwd_tag(cp),
 542                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 543                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 544                       IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
 545                       cp->flags, atomic_read(&cp->refcnt));
 546
 547         ip_vs_conn_stats(cp, svc);
 548         return cp;
 549 }
 550
 551 static inline int ip_vs_addr_is_unicast(struct net *net, int af,
 552                                         union nf_inet_addr *addr)
 553 {
 554 #ifdef CONFIG_IP_VS_IPV6
 555         if (af == AF_INET6)
 556                 return ipv6_addr_type(&addr->in6) & IPV6_ADDR_UNICAST;
 557 #endif
 558         return (inet_addr_type(net, addr->ip) == RTN_UNICAST);
 559 }
 560
 561 /*
 562  *  Pass or drop the packet.
 563  *  Called by ip_vs_in, when the virtual service is available but
 564  *  no destination is available for a new connection.
 565  */
 566 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 567                 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
 568 {
 569         __be16 _ports[2], *pptr, dport;
 570         struct netns_ipvs *ipvs = svc->ipvs;
 571         struct net *net = ipvs->net;
 572
 573         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 574         if (!pptr)
 575                 return NF_DROP;
 576         dport = likely(!ip_vs_iph_inverse(iph)) ? pptr[1] : pptr[0];
 577
 578         /* if it is fwmark-based service, the cache_bypass sysctl is up
 579            and the destination is a non-local unicast, then create
 580            a cache_bypass connection entry */
 581         if (sysctl_cache_bypass(ipvs) && svc->fwmark &&
 582             !(iph->hdr_flags & (IP_VS_HDR_INVERSE | IP_VS_HDR_ICMP)) &&
 583             ip_vs_addr_is_unicast(net, svc->af, &iph->daddr)) {
 584                 int ret;
 585                 struct ip_vs_conn *cp;
 586                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
 587                                       iph->protocol == IPPROTO_UDP) ?
 588                                       IP_VS_CONN_F_ONE_PACKET : 0;
 589                 union nf_inet_addr daddr =  { .all = { 0, 0, 0, 0 } };
 590
 591                 /* create a new connection entry */
 592                 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
 593                 {
 594                         struct ip_vs_conn_param p;
 595                         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
 596                                               &iph->saddr, pptr[0],
 597                                               &iph->daddr, pptr[1], &p);
 598                         cp = ip_vs_conn_new(&p, svc->af, &daddr, 0,
 599                                             IP_VS_CONN_F_BYPASS | flags,
 600                                             NULL, skb->mark);
 601                         if (!cp)
 602                                 return NF_DROP;
 603                 }
 604
 605                 /* statistics */
 606                 ip_vs_in_stats(cp, skb);
 607
 608                 /* set state */
 609                 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
 610
 611                 /* transmit the first SYN packet */
 612                 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
 613                 /* do not touch skb anymore */
 614
 615                 if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) && cp->control)
 616                         atomic_inc(&cp->control->in_pkts);
 617                 else
 618                         atomic_inc(&cp->in_pkts);
 619                 ip_vs_conn_put(cp);
 620                 return ret;
 621         }
 622
 623         /*
 624          * When the virtual ftp service is presented, packets destined
 625          * for other services on the VIP may get here (except services
 626          * listed in the ipvs table), pass the packets, because it is
 627          * not ipvs job to decide to drop the packets.
 628          */
 629         if (svc->port == FTPPORT && dport != FTPPORT)
 630                 return NF_ACCEPT;
 631
 632         if (unlikely(ip_vs_iph_icmp(iph)))
 633                 return NF_DROP;
 634
 635         /*
 636          * Notify the client that the destination is unreachable, and
 637          * release the socket buffer.
 638          * Since it is in IP layer, the TCP socket is not actually
 639          * created, the TCP RST packet cannot be sent, instead that
 640          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 641          */
 642 #ifdef CONFIG_IP_VS_IPV6
 643         if (svc->af == AF_INET6) {
 644                 if (!skb->dev)
 645                         skb->dev = net->loopback_dev;
 646                 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 647         } else
 648 #endif
 649                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 650
 651         return NF_DROP;
 652 }
 653
 654 #ifdef CONFIG_SYSCTL
 655
 656 static int sysctl_snat_reroute(struct netns_ipvs *ipvs)
 657 {
 658         return ipvs->sysctl_snat_reroute;
 659 }
 660
 661 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs)
 662 {
 663         return ipvs->sysctl_nat_icmp_send;
 664 }
 665
 666 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
 667 {
 668         return ipvs->sysctl_expire_nodest_conn;
 669 }
 670
 671 #else
 672
 673 static int sysctl_snat_reroute(struct netns_ipvs *ipvs) { return 0; }
 674 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs) { return 0; }
 675 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
 676
 677 #endif
 678
 679 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 680 {
 681         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 682 }
 683
 684 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
 685 {
 686         if (NF_INET_LOCAL_IN == hooknum)
 687                 return IP_DEFRAG_VS_IN;
 688         if (NF_INET_FORWARD == hooknum)
 689                 return IP_DEFRAG_VS_FWD;
 690         return IP_DEFRAG_VS_OUT;
 691 }
 692
 693 static inline int ip_vs_gather_frags(struct netns_ipvs *ipvs,
 694                                      struct sk_buff *skb, u_int32_t user)
 695 {
 696         int err;
 697
 698         local_bh_disable();
 699         err = ip_defrag(ipvs->net, skb, user);
 700         local_bh_enable();
 701         if (!err)
 702                 ip_send_check(ip_hdr(skb));
 703
 704         return err;
 705 }
 706
 707 static int ip_vs_route_me_harder(struct netns_ipvs *ipvs, int af,
 708                                  struct sk_buff *skb, unsigned int hooknum)
 709 {
 710         if (!sysctl_snat_reroute(ipvs))
 711                 return 0;
 712         /* Reroute replies only to remote clients (FORWARD and LOCAL_OUT) */
 713         if (NF_INET_LOCAL_IN == hooknum)
 714                 return 0;
 715 #ifdef CONFIG_IP_VS_IPV6
 716         if (af == AF_INET6) {
 717                 struct dst_entry *dst = skb_dst(skb);
 718
 719                 if (dst->dev && !(dst->dev->flags & IFF_LOOPBACK) &&
 720                     ip6_route_me_harder(ipvs->net, skb) != 0)
 721                         return 1;
 722         } else
 723 #endif
 724                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
 725                     ip_route_me_harder(ipvs->net, skb, RTN_LOCAL) != 0)
 726                         return 1;
 727
 728         return 0;
 729 }
 730
 731 /*
 732  * Packet has been made sufficiently writable in caller
 733  * - inout: 1=in->out, 0=out->in
 734  */
 735 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 736                     struct ip_vs_conn *cp, int inout)
 737 {
 738         struct iphdr *iph        = ip_hdr(skb);
 739         unsigned int icmp_offset = iph->ihl*4;
 740         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
 741                                                       icmp_offset);
 742         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 743
 744         if (inout) {
 745                 iph->saddr = cp->vaddr.ip;
 746                 ip_send_check(iph);
 747                 ciph->daddr = cp->vaddr.ip;
 748                 ip_send_check(ciph);
 749         } else {
 750                 iph->daddr = cp->daddr.ip;
 751                 ip_send_check(iph);
 752                 ciph->saddr = cp->daddr.ip;
 753                 ip_send_check(ciph);
 754         }
 755
 756         /* the TCP/UDP/SCTP port */
 757         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
 758             IPPROTO_SCTP == ciph->protocol) {
 759                 __be16 *ports = (void *)ciph + ciph->ihl*4;
 760
 761                 if (inout)
 762                         ports[1] = cp->vport;
 763                 else
 764                         ports[0] = cp->dport;
 765         }
 766
 767         /* And finally the ICMP checksum */
 768         icmph->checksum = 0;
 769         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 770         skb->ip_summed = CHECKSUM_UNNECESSARY;
 771
 772         if (inout)
 773                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 774                         "Forwarding altered outgoing ICMP");
 775         else
 776                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 777                         "Forwarding altered incoming ICMP");
 778 }
 779
 780 #ifdef CONFIG_IP_VS_IPV6
 781 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
 782                     struct ip_vs_conn *cp, int inout)
 783 {
 784         struct ipv6hdr *iph      = ipv6_hdr(skb);
 785         unsigned int icmp_offset = 0;
 786         unsigned int offs        = 0; /* header offset*/
 787         int protocol;
 788         struct icmp6hdr *icmph;
 789         struct ipv6hdr *ciph;
 790         unsigned short fragoffs;
 791
 792         ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
 793         icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
 794         offs = icmp_offset + sizeof(struct icmp6hdr);
 795         ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
 796
 797         protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
 798
 799         if (inout) {
 800                 iph->saddr = cp->vaddr.in6;
 801                 ciph->daddr = cp->vaddr.in6;
 802         } else {
 803                 iph->daddr = cp->daddr.in6;
 804                 ciph->saddr = cp->daddr.in6;
 805         }
 806
 807         /* the TCP/UDP/SCTP port */
 808         if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 809                           IPPROTO_SCTP == protocol)) {
 810                 __be16 *ports = (void *)(skb_network_header(skb) + offs);
 811
 812                 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
 813                               ntohs(inout ? ports[1] : ports[0]),
 814                               ntohs(inout ? cp->vport : cp->dport));
 815                 if (inout)
 816                         ports[1] = cp->vport;
 817                 else
 818                         ports[0] = cp->dport;
 819         }
 820
 821         /* And finally the ICMP checksum */
 822         icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
 823                                               skb->len - icmp_offset,
 824                                               IPPROTO_ICMPV6, 0);
 825         skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
 826         skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
 827         skb->ip_summed = CHECKSUM_PARTIAL;
 828
 829         if (inout)
 830                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 831                               (void *)ciph - (void *)iph,
 832                               "Forwarding altered outgoing ICMPv6");
 833         else
 834                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 835                               (void *)ciph - (void *)iph,
 836                               "Forwarding altered incoming ICMPv6");
 837 }
 838 #endif
 839
 840 /* Handle relevant response ICMP messages - forward to the right
 841  * destination host.
 842  */
 843 static int handle_response_icmp(int af, struct sk_buff *skb,
 844                                 union nf_inet_addr *snet,
 845                                 __u8 protocol, struct ip_vs_conn *cp,
 846                                 struct ip_vs_protocol *pp,
 847                                 unsigned int offset, unsigned int ihl,
 848                                 unsigned int hooknum)
 849 {
 850         unsigned int verdict = NF_DROP;
 851
 852         if (IP_VS_FWD_METHOD(cp) != 0) {
 853                 pr_err("shouldn't reach here, because the box is on the "
 854                        "half connection in the tun/dr module.\n");
 855         }
 856
 857         /* Ensure the checksum is correct */
 858         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 859                 /* Failed checksum! */
 860                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
 861                               IP_VS_DBG_ADDR(af, snet));
 862                 goto out;
 863         }
 864
 865         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 866             IPPROTO_SCTP == protocol)
 867                 offset += 2 * sizeof(__u16);
 868         if (!skb_make_writable(skb, offset))
 869                 goto out;
 870
 871 #ifdef CONFIG_IP_VS_IPV6
 872         if (af == AF_INET6)
 873                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
 874         else
 875 #endif
 876                 ip_vs_nat_icmp(skb, pp, cp, 1);
 877
 878         if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
 879                 goto out;
 880
 881         /* do the statistics and put it back */
 882         ip_vs_out_stats(cp, skb);
 883
 884         skb->ipvs_property = 1;
 885         if (!(cp->flags & IP_VS_CONN_F_NFCT))
 886                 ip_vs_notrack(skb);
 887         else
 888                 ip_vs_update_conntrack(skb, cp, 0);
 889         verdict = NF_ACCEPT;
 890
 891 out:
 892         __ip_vs_conn_put(cp);
 893
 894         return verdict;
 895 }
 896
 897 /*
 898  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 899  *      Find any that might be relevant, check against existing connections.
 900  *      Currently handles error types - unreachable, quench, ttl exceeded.
 901  */
 902 static int ip_vs_out_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb,
 903                           int *related, unsigned int hooknum)
 904 {
 905         struct iphdr *iph;
 906         struct icmphdr  _icmph, *ic;
 907         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 908         struct ip_vs_iphdr ciph;
 909         struct ip_vs_conn *cp;
 910         struct ip_vs_protocol *pp;
 911         unsigned int offset, ihl;
 912         union nf_inet_addr snet;
 913
 914         *related = 1;
 915
 916         /* reassemble IP fragments */
 917         if (ip_is_fragment(ip_hdr(skb))) {
 918                 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
 919                         return NF_STOLEN;
 920         }
 921
 922         iph = ip_hdr(skb);
 923         offset = ihl = iph->ihl * 4;
 924         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 925         if (ic == NULL)
 926                 return NF_DROP;
 927
 928         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
 929                   ic->type, ntohs(icmp_id(ic)),
 930                   &iph->saddr, &iph->daddr);
 931
 932         /*
 933          * Work through seeing if this is for us.
 934          * These checks are supposed to be in an order that means easy
 935          * things are checked first to speed up processing.... however
 936          * this means that some packets will manage to get a long way
 937          * down this stack and then be rejected, but that's life.
 938          */
 939         if ((ic->type != ICMP_DEST_UNREACH) &&
 940             (ic->type != ICMP_SOURCE_QUENCH) &&
 941             (ic->type != ICMP_TIME_EXCEEDED)) {
 942                 *related = 0;
 943                 return NF_ACCEPT;
 944         }
 945
 946         /* Now find the contained IP header */
 947         offset += sizeof(_icmph);
 948         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 949         if (cih == NULL)
 950                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 951
 952         pp = ip_vs_proto_get(cih->protocol);
 953         if (!pp)
 954                 return NF_ACCEPT;
 955
 956         /* Is the embedded protocol header present? */
 957         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 958                      pp->dont_defrag))
 959                 return NF_ACCEPT;
 960
 961         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
 962                       "Checking outgoing ICMP for");
 963
 964         ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, true, &ciph);
 965
 966         /* The embedded headers contain source and dest in reverse order */
 967         cp = pp->conn_out_get(ipvs, AF_INET, skb, &ciph);
 968         if (!cp)
 969                 return NF_ACCEPT;
 970
 971         snet.ip = iph->saddr;
 972         return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
 973                                     pp, ciph.len, ihl, hooknum);
 974 }
 975
 976 #ifdef CONFIG_IP_VS_IPV6
 977 static int ip_vs_out_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
 978                              int *related,  unsigned int hooknum,
 979                              struct ip_vs_iphdr *ipvsh)
 980 {
 981         struct icmp6hdr _icmph, *ic;
 982         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
 983         struct ip_vs_conn *cp;
 984         struct ip_vs_protocol *pp;
 985         union nf_inet_addr snet;
 986         unsigned int offset;
 987
 988         *related = 1;
 989         ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph, ipvsh);
 990         if (ic == NULL)
 991                 return NF_DROP;
 992
 993         /*
 994          * Work through seeing if this is for us.
 995          * These checks are supposed to be in an order that means easy
 996          * things are checked first to speed up processing.... however
 997          * this means that some packets will manage to get a long way
 998          * down this stack and then be rejected, but that's life.
 999          */
1000         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1001                 *related = 0;
1002                 return NF_ACCEPT;
1003         }
1004         /* Fragment header that is before ICMP header tells us that:
1005          * it's not an error message since they can't be fragmented.
1006          */
1007         if (ipvsh->flags & IP6_FH_F_FRAG)
1008                 return NF_DROP;
1009
1010         IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1011                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1012                   &ipvsh->saddr, &ipvsh->daddr);
1013
1014         if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, ipvsh->len + sizeof(_icmph),
1015                                      true, &ciph))
1016                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1017
1018         pp = ip_vs_proto_get(ciph.protocol);
1019         if (!pp)
1020                 return NF_ACCEPT;
1021
1022         /* The embedded headers contain source and dest in reverse order */
1023         cp = pp->conn_out_get(ipvs, AF_INET6, skb, &ciph);
1024         if (!cp)
1025                 return NF_ACCEPT;
1026
1027         snet.in6 = ciph.saddr.in6;
1028         offset = ciph.len;
1029         return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
1030                                     pp, offset, sizeof(struct ipv6hdr),
1031                                     hooknum);
1032 }
1033 #endif
1034
1035 /*
1036  * Check if sctp chunc is ABORT chunk
1037  */
1038 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
1039 {
1040         sctp_chunkhdr_t *sch, schunk;
1041         sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
1042                         sizeof(schunk), &schunk);
1043         if (sch == NULL)
1044                 return 0;
1045         if (sch->type == SCTP_CID_ABORT)
1046                 return 1;
1047         return 0;
1048 }
1049
1050 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
1051 {
1052         struct tcphdr _tcph, *th;
1053
1054         th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
1055         if (th == NULL)
1056                 return 0;
1057         return th->rst;
1058 }
1059
1060 static inline bool is_new_conn(const struct sk_buff *skb,
1061                                struct ip_vs_iphdr *iph)
1062 {
1063         switch (iph->protocol) {
1064         case IPPROTO_TCP: {
1065                 struct tcphdr _tcph, *th;
1066
1067                 th = skb_header_pointer(skb, iph->len, sizeof(_tcph), &_tcph);
1068                 if (th == NULL)
1069                         return false;
1070                 return th->syn;
1071         }
1072         case IPPROTO_SCTP: {
1073                 sctp_chunkhdr_t *sch, schunk;
1074
1075                 sch = skb_header_pointer(skb, iph->len + sizeof(sctp_sctphdr_t),
1076                                          sizeof(schunk), &schunk);
1077                 if (sch == NULL)
1078                         return false;
1079                 return sch->type == SCTP_CID_INIT;
1080         }
1081         default:
1082                 return false;
1083         }
1084 }
1085
1086 static inline bool is_new_conn_expected(const struct ip_vs_conn *cp,
1087                                         int conn_reuse_mode)
1088 {
1089         /* Controlled (FTP DATA or persistence)? */
1090         if (cp->control)
1091                 return false;
1092
1093         switch (cp->protocol) {
1094         case IPPROTO_TCP:
1095                 return (cp->state == IP_VS_TCP_S_TIME_WAIT) ||
1096                        (cp->state == IP_VS_TCP_S_CLOSE) ||
1097                         ((conn_reuse_mode & 2) &&
1098                          (cp->state == IP_VS_TCP_S_FIN_WAIT) &&
1099                          (cp->flags & IP_VS_CONN_F_NOOUTPUT));
1100         case IPPROTO_SCTP:
1101                 return cp->state == IP_VS_SCTP_S_CLOSED;
1102         default:
1103                 return false;
1104         }
1105 }
1106
1107 /* Generic function to create new connections for outgoing RS packets
1108  *
1109  * Pre-requisites for successful connection creation:
1110  * 1) Virtual Service is NOT fwmark based:
1111  *    In fwmark-VS actual vaddr and vport are unknown to IPVS
1112  * 2) Real Server and Virtual Service were NOT configured without port:
1113  *    This is to allow match of different VS to the same RS ip-addr
1114  */
1115 struct ip_vs_conn *ip_vs_new_conn_out(struct ip_vs_service *svc,
1116                                       struct ip_vs_dest *dest,
1117                                       struct sk_buff *skb,
1118                                       const struct ip_vs_iphdr *iph,
1119                                       __be16 dport,
1120                                       __be16 cport)
1121 {
1122         struct ip_vs_conn_param param;
1123         struct ip_vs_conn *ct = NULL, *cp = NULL;
1124         const union nf_inet_addr *vaddr, *daddr, *caddr;
1125         union nf_inet_addr snet;
1126         __be16 vport;
1127         unsigned int flags;
1128
1129         EnterFunction(12);
1130         vaddr = &svc->addr;
1131         vport = svc->port;
1132         daddr = &iph->saddr;
1133         caddr = &iph->daddr;
1134
1135         /* check pre-requisites are satisfied */
1136         if (svc->fwmark)
1137                 return NULL;
1138         if (!vport || !dport)
1139                 return NULL;
1140
1141         /* for persistent service first create connection template */
1142         if (svc->flags & IP_VS_SVC_F_PERSISTENT) {
1143                 /* apply netmask the same way ingress-side does */
1144 #ifdef CONFIG_IP_VS_IPV6
1145                 if (svc->af == AF_INET6)
1146                         ipv6_addr_prefix(&snet.in6, &caddr->in6,
1147                                          (__force __u32)svc->netmask);
1148                 else
1149 #endif
1150                         snet.ip = caddr->ip & svc->netmask;
1151                 /* fill params and create template if not existent */
1152                 if (ip_vs_conn_fill_param_persist(svc, skb, iph->protocol,
1153                                                   &snet, 0, vaddr,
1154                                                   vport, &param) < 0)
1155                         return NULL;
1156                 ct = ip_vs_ct_in_get(&param);
1157                 if (!ct) {
1158                         ct = ip_vs_conn_new(&param, dest->af, daddr, dport,
1159                                             IP_VS_CONN_F_TEMPLATE, dest, 0);
1160                         if (!ct) {
1161                                 kfree(param.pe_data);
1162                                 return NULL;
1163                         }
1164                         ct->timeout = svc->timeout;
1165                 } else {
1166                         kfree(param.pe_data);
1167                 }
1168         }
1169
1170         /* connection flags */
1171         flags = ((svc->flags & IP_VS_SVC_F_ONEPACKET) &&
1172                  iph->protocol == IPPROTO_UDP) ? IP_VS_CONN_F_ONE_PACKET : 0;
1173         /* create connection */
1174         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
1175                               caddr, cport, vaddr, vport, &param);
1176         cp = ip_vs_conn_new(&param, dest->af, daddr, dport, flags, dest, 0);
1177         if (!cp) {
1178                 if (ct)
1179                         ip_vs_conn_put(ct);
1180                 return NULL;
1181         }
1182         if (ct) {
1183                 ip_vs_control_add(cp, ct);
1184                 ip_vs_conn_put(ct);
1185         }
1186         ip_vs_conn_stats(cp, svc);
1187
1188         /* return connection (will be used to handle outgoing packet) */
1189         IP_VS_DBG_BUF(6, "New connection RS-initiated:%c c:%s:%u v:%s:%u "
1190                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
1191                       ip_vs_fwd_tag(cp),
1192                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
1193                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
1194                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
1195                       cp->flags, atomic_read(&cp->refcnt));
1196         LeaveFunction(12);
1197         return cp;
1198 }
1199
1200 /* Handle outgoing packets which are considered requests initiated by
1201  * real servers, so that subsequent responses from external client can be
1202  * routed to the right real server.
1203  * Used also for outgoing responses in OPS mode.
1204  *
1205  * Connection management is handled by persistent-engine specific callback.
1206  */
1207 static struct ip_vs_conn *__ip_vs_rs_conn_out(unsigned int hooknum,
1208                                               struct netns_ipvs *ipvs,
1209                                               int af, struct sk_buff *skb,
1210                                               const struct ip_vs_iphdr *iph)
1211 {
1212         struct ip_vs_dest *dest;
1213         struct ip_vs_conn *cp = NULL;
1214         __be16 _ports[2], *pptr;
1215
1216         if (hooknum == NF_INET_LOCAL_IN)
1217                 return NULL;
1218
1219         pptr = frag_safe_skb_hp(skb, iph->len,
1220                                 sizeof(_ports), _ports, iph);
1221         if (!pptr)
1222                 return NULL;
1223
1224         rcu_read_lock();
1225         dest = ip_vs_find_real_service(ipvs, af, iph->protocol,
1226                                        &iph->saddr, pptr[0]);
1227         if (dest) {
1228                 struct ip_vs_service *svc;
1229                 struct ip_vs_pe *pe;
1230
1231                 svc = rcu_dereference(dest->svc);
1232                 if (svc) {
1233                         pe = rcu_dereference(svc->pe);
1234                         if (pe && pe->conn_out)
1235                                 cp = pe->conn_out(svc, dest, skb, iph,
1236                                                   pptr[0], pptr[1]);
1237                 }
1238         }
1239         rcu_read_unlock();
1240
1241         return cp;
1242 }
1243
1244 /* Handle response packets: rewrite addresses and send away...
1245  */
1246 static unsigned int
1247 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1248                 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph,
1249                 unsigned int hooknum)
1250 {
1251         struct ip_vs_protocol *pp = pd->pp;
1252
1253         IP_VS_DBG_PKT(11, af, pp, skb, iph->off, "Outgoing packet");
1254
1255         if (!skb_make_writable(skb, iph->len))
1256                 goto drop;
1257
1258         /* mangle the packet */
1259         if (pp->snat_handler && !pp->snat_handler(skb, pp, cp, iph))
1260                 goto drop;
1261
1262 #ifdef CONFIG_IP_VS_IPV6
1263         if (af == AF_INET6)
1264                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1265         else
1266 #endif
1267         {
1268                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1269                 ip_send_check(ip_hdr(skb));
1270         }
1271
1272         /*
1273          * nf_iterate does not expect change in the skb->dst->dev.
1274          * It looks like it is not fatal to enable this code for hooks
1275          * where our handlers are at the end of the chain list and
1276          * when all next handlers use skb->dst->dev and not outdev.
1277          * It will definitely route properly the inout NAT traffic
1278          * when multiple paths are used.
1279          */
1280
1281         /* For policy routing, packets originating from this
1282          * machine itself may be routed differently to packets
1283          * passing through.  We want this packet to be routed as
1284          * if it came from this machine itself.  So re-compute
1285          * the routing information.
1286          */
1287         if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
1288                 goto drop;
1289
1290         IP_VS_DBG_PKT(10, af, pp, skb, iph->off, "After SNAT");
1291
1292         ip_vs_out_stats(cp, skb);
1293         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1294         skb->ipvs_property = 1;
1295         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1296                 ip_vs_notrack(skb);
1297         else
1298                 ip_vs_update_conntrack(skb, cp, 0);
1299         ip_vs_conn_put(cp);
1300
1301         LeaveFunction(11);
1302         return NF_ACCEPT;
1303
1304 drop:
1305         ip_vs_conn_put(cp);
1306         kfree_skb(skb);
1307         LeaveFunction(11);
1308         return NF_STOLEN;
1309 }
1310
1311 /*
1312  *      Check if outgoing packet belongs to the established ip_vs_conn.
1313  */
1314 static unsigned int
1315 ip_vs_out(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1316 {
1317         struct ip_vs_iphdr iph;
1318         struct ip_vs_protocol *pp;
1319         struct ip_vs_proto_data *pd;
1320         struct ip_vs_conn *cp;
1321         struct sock *sk;
1322
1323         EnterFunction(11);
1324
1325         /* Already marked as IPVS request or reply? */
1326         if (skb->ipvs_property)
1327                 return NF_ACCEPT;
1328
1329         sk = skb_to_full_sk(skb);
1330         /* Bad... Do not break raw sockets */
1331         if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
1332                      af == AF_INET)) {
1333
1334                 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
1335                         return NF_ACCEPT;
1336         }
1337
1338         if (unlikely(!skb_dst(skb)))
1339                 return NF_ACCEPT;
1340
1341         if (!ipvs->enable)
1342                 return NF_ACCEPT;
1343
1344         ip_vs_fill_iph_skb(af, skb, false, &iph);
1345 #ifdef CONFIG_IP_VS_IPV6
1346         if (af == AF_INET6) {
1347                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1348                         int related;
1349                         int verdict = ip_vs_out_icmp_v6(ipvs, skb, &related,
1350                                                         hooknum, &iph);
1351
1352                         if (related)
1353                                 return verdict;
1354                 }
1355         } else
1356 #endif
1357                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1358                         int related;
1359                         int verdict = ip_vs_out_icmp(ipvs, skb, &related, hooknum);
1360
1361                         if (related)
1362                                 return verdict;
1363                 }
1364
1365         pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1366         if (unlikely(!pd))
1367                 return NF_ACCEPT;
1368         pp = pd->pp;
1369
1370         /* reassemble IP fragments */
1371 #ifdef CONFIG_IP_VS_IPV6
1372         if (af == AF_INET)
1373 #endif
1374                 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1375                         if (ip_vs_gather_frags(ipvs, skb,
1376                                                ip_vs_defrag_user(hooknum)))
1377                                 return NF_STOLEN;
1378
1379                         ip_vs_fill_iph_skb(AF_INET, skb, false, &iph);
1380                 }
1381
1382         /*
1383          * Check if the packet belongs to an existing entry
1384          */
1385         cp = pp->conn_out_get(ipvs, af, skb, &iph);
1386
1387         if (likely(cp))
1388                 return handle_response(af, skb, pd, cp, &iph, hooknum);
1389
1390         /* Check for real-server-started requests */
1391         if (atomic_read(&ipvs->conn_out_counter)) {
1392                 /* Currently only for UDP:
1393                  * connection oriented protocols typically use
1394                  * ephemeral ports for outgoing connections, so
1395                  * related incoming responses would not match any VS
1396                  */
1397                 if (pp->protocol == IPPROTO_UDP) {
1398                         cp = __ip_vs_rs_conn_out(hooknum, ipvs, af, skb, &iph);
1399                         if (likely(cp))
1400                                 return handle_response(af, skb, pd, cp, &iph,
1401                                                        hooknum);
1402                 }
1403         }
1404
1405         if (sysctl_nat_icmp_send(ipvs) &&
1406             (pp->protocol == IPPROTO_TCP ||
1407              pp->protocol == IPPROTO_UDP ||
1408              pp->protocol == IPPROTO_SCTP)) {
1409                 __be16 _ports[2], *pptr;
1410
1411                 pptr = frag_safe_skb_hp(skb, iph.len,
1412                                          sizeof(_ports), _ports, &iph);
1413                 if (pptr == NULL)
1414                         return NF_ACCEPT;       /* Not for me */
1415                 if (ip_vs_has_real_service(ipvs, af, iph.protocol, &iph.saddr,
1416                                            pptr[0])) {
1417                         /*
1418                          * Notify the real server: there is no
1419                          * existing entry if it is not RST
1420                          * packet or not TCP packet.
1421                          */
1422                         if ((iph.protocol != IPPROTO_TCP &&
1423                              iph.protocol != IPPROTO_SCTP)
1424                              || ((iph.protocol == IPPROTO_TCP
1425                                   && !is_tcp_reset(skb, iph.len))
1426                                  || (iph.protocol == IPPROTO_SCTP
1427                                         && !is_sctp_abort(skb,
1428                                                 iph.len)))) {
1429 #ifdef CONFIG_IP_VS_IPV6
1430                                 if (af == AF_INET6) {
1431                                         if (!skb->dev)
1432                                                 skb->dev = ipvs->net->loopback_dev;
1433                                         icmpv6_send(skb,
1434                                                     ICMPV6_DEST_UNREACH,
1435                                                     ICMPV6_PORT_UNREACH,
1436                                                     0);
1437                                 } else
1438 #endif
1439                                         icmp_send(skb,
1440                                                   ICMP_DEST_UNREACH,
1441                                                   ICMP_PORT_UNREACH, 0);
1442                                 return NF_DROP;
1443                         }
1444                 }
1445         }
1446         IP_VS_DBG_PKT(12, af, pp, skb, iph.off,
1447                       "ip_vs_out: packet continues traversal as normal");
1448         return NF_ACCEPT;
1449 }
1450
1451 /*
1452  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1453  *      used only for VS/NAT.
1454  *      Check if packet is reply for established ip_vs_conn.
1455  */
1456 static unsigned int
1457 ip_vs_reply4(void *priv, struct sk_buff *skb,
1458              const struct nf_hook_state *state)
1459 {
1460         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1461 }
1462
1463 /*
1464  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1465  *      Check if packet is reply for established ip_vs_conn.
1466  */
1467 static unsigned int
1468 ip_vs_local_reply4(void *priv, struct sk_buff *skb,
1469                    const struct nf_hook_state *state)
1470 {
1471         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1472 }
1473
1474 #ifdef CONFIG_IP_VS_IPV6
1475
1476 /*
1477  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1478  *      used only for VS/NAT.
1479  *      Check if packet is reply for established ip_vs_conn.
1480  */
1481 static unsigned int
1482 ip_vs_reply6(void *priv, struct sk_buff *skb,
1483              const struct nf_hook_state *state)
1484 {
1485         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1486 }
1487
1488 /*
1489  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1490  *      Check if packet is reply for established ip_vs_conn.
1491  */
1492 static unsigned int
1493 ip_vs_local_reply6(void *priv, struct sk_buff *skb,
1494                    const struct nf_hook_state *state)
1495 {
1496         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1497 }
1498
1499 #endif
1500
1501 static unsigned int
1502 ip_vs_try_to_schedule(struct netns_ipvs *ipvs, int af, struct sk_buff *skb,
1503                       struct ip_vs_proto_data *pd,
1504                       int *verdict, struct ip_vs_conn **cpp,
1505                       struct ip_vs_iphdr *iph)
1506 {
1507         struct ip_vs_protocol *pp = pd->pp;
1508
1509         if (!iph->fragoffs) {
1510                 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1511                  * replayed fragment zero will already have created the cp
1512                  */
1513
1514                 /* Schedule and create new connection entry into cpp */
1515                 if (!pp->conn_schedule(ipvs, af, skb, pd, verdict, cpp, iph))
1516                         return 0;
1517         }
1518
1519         if (unlikely(!*cpp)) {
1520                 /* sorry, all this trouble for a no-hit :) */
1521                 IP_VS_DBG_PKT(12, af, pp, skb, iph->off,
1522                               "ip_vs_in: packet continues traversal as normal");
1523                 if (iph->fragoffs) {
1524                         /* Fragment that couldn't be mapped to a conn entry
1525                          * is missing module nf_defrag_ipv6
1526                          */
1527                         IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1528                         IP_VS_DBG_PKT(7, af, pp, skb, iph->off,
1529                                       "unhandled fragment");
1530                 }
1531                 *verdict = NF_ACCEPT;
1532                 return 0;
1533         }
1534
1535         return 1;
1536 }
1537
1538 /*
1539  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1540  *      Find any that might be relevant, check against existing connections,
1541  *      forward to the right destination host if relevant.
1542  *      Currently handles error types - unreachable, quench, ttl exceeded.
1543  */
1544 static int
1545 ip_vs_in_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb, int *related,
1546               unsigned int hooknum)
1547 {
1548         struct iphdr *iph;
1549         struct icmphdr  _icmph, *ic;
1550         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1551         struct ip_vs_iphdr ciph;
1552         struct ip_vs_conn *cp;
1553         struct ip_vs_protocol *pp;
1554         struct ip_vs_proto_data *pd;
1555         unsigned int offset, offset2, ihl, verdict;
1556         bool ipip, new_cp = false;
1557
1558         *related = 1;
1559
1560         /* reassemble IP fragments */
1561         if (ip_is_fragment(ip_hdr(skb))) {
1562                 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
1563                         return NF_STOLEN;
1564         }
1565
1566         iph = ip_hdr(skb);
1567         offset = ihl = iph->ihl * 4;
1568         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1569         if (ic == NULL)
1570                 return NF_DROP;
1571
1572         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1573                   ic->type, ntohs(icmp_id(ic)),
1574                   &iph->saddr, &iph->daddr);
1575
1576         /*
1577          * Work through seeing if this is for us.
1578          * These checks are supposed to be in an order that means easy
1579          * things are checked first to speed up processing.... however
1580          * this means that some packets will manage to get a long way
1581          * down this stack and then be rejected, but that's life.
1582          */
1583         if ((ic->type != ICMP_DEST_UNREACH) &&
1584             (ic->type != ICMP_SOURCE_QUENCH) &&
1585             (ic->type != ICMP_TIME_EXCEEDED)) {
1586                 *related = 0;
1587                 return NF_ACCEPT;
1588         }
1589
1590         /* Now find the contained IP header */
1591         offset += sizeof(_icmph);
1592         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1593         if (cih == NULL)
1594                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1595
1596         /* Special case for errors for IPIP packets */
1597         ipip = false;
1598         if (cih->protocol == IPPROTO_IPIP) {
1599                 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1600                         return NF_ACCEPT;
1601                 /* Error for our IPIP must arrive at LOCAL_IN */
1602                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1603                         return NF_ACCEPT;
1604                 offset += cih->ihl * 4;
1605                 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1606                 if (cih == NULL)
1607                         return NF_ACCEPT; /* The packet looks wrong, ignore */
1608                 ipip = true;
1609         }
1610
1611         pd = ip_vs_proto_data_get(ipvs, cih->protocol);
1612         if (!pd)
1613                 return NF_ACCEPT;
1614         pp = pd->pp;
1615
1616         /* Is the embedded protocol header present? */
1617         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1618                      pp->dont_defrag))
1619                 return NF_ACCEPT;
1620
1621         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1622                       "Checking incoming ICMP for");
1623
1624         offset2 = offset;
1625         ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, !ipip, &ciph);
1626         offset = ciph.len;
1627
1628         /* The embedded headers contain source and dest in reverse order.
1629          * For IPIP this is error for request, not for reply.
1630          */
1631         cp = pp->conn_in_get(ipvs, AF_INET, skb, &ciph);
1632
1633         if (!cp) {
1634                 int v;
1635
1636                 if (!sysctl_schedule_icmp(ipvs))
1637                         return NF_ACCEPT;
1638
1639                 if (!ip_vs_try_to_schedule(ipvs, AF_INET, skb, pd, &v, &cp, &ciph))
1640                         return v;
1641                 new_cp = true;
1642         }
1643
1644         verdict = NF_DROP;
1645
1646         /* Ensure the checksum is correct */
1647         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1648                 /* Failed checksum! */
1649                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1650                           &iph->saddr);
1651                 goto out;
1652         }
1653
1654         if (ipip) {
1655                 __be32 info = ic->un.gateway;
1656                 __u8 type = ic->type;
1657                 __u8 code = ic->code;
1658
1659                 /* Update the MTU */
1660                 if (ic->type == ICMP_DEST_UNREACH &&
1661                     ic->code == ICMP_FRAG_NEEDED) {
1662                         struct ip_vs_dest *dest = cp->dest;
1663                         u32 mtu = ntohs(ic->un.frag.mtu);
1664                         __be16 frag_off = cih->frag_off;
1665
1666                         /* Strip outer IP and ICMP, go to IPIP header */
1667                         if (pskb_pull(skb, ihl + sizeof(_icmph)) == NULL)
1668                                 goto ignore_ipip;
1669                         offset2 -= ihl + sizeof(_icmph);
1670                         skb_reset_network_header(skb);
1671                         IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1672                                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1673                         ipv4_update_pmtu(skb, ipvs->net,
1674                                          mtu, 0, 0, 0, 0);
1675                         /* Client uses PMTUD? */
1676                         if (!(frag_off & htons(IP_DF)))
1677                                 goto ignore_ipip;
1678                         /* Prefer the resulting PMTU */
1679                         if (dest) {
1680                                 struct ip_vs_dest_dst *dest_dst;
1681
1682                                 rcu_read_lock();
1683                                 dest_dst = rcu_dereference(dest->dest_dst);
1684                                 if (dest_dst)
1685                                         mtu = dst_mtu(dest_dst->dst_cache);
1686                                 rcu_read_unlock();
1687                         }
1688                         if (mtu > 68 + sizeof(struct iphdr))
1689                                 mtu -= sizeof(struct iphdr);
1690                         info = htonl(mtu);
1691                 }
1692                 /* Strip outer IP, ICMP and IPIP, go to IP header of
1693                  * original request.
1694                  */
1695                 if (pskb_pull(skb, offset2) == NULL)
1696                         goto ignore_ipip;
1697                 skb_reset_network_header(skb);
1698                 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1699                         &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1700                         type, code, ntohl(info));
1701                 icmp_send(skb, type, code, info);
1702                 /* ICMP can be shorter but anyways, account it */
1703                 ip_vs_out_stats(cp, skb);
1704
1705 ignore_ipip:
1706                 consume_skb(skb);
1707                 verdict = NF_STOLEN;
1708                 goto out;
1709         }
1710
1711         /* do the statistics and put it back */
1712         ip_vs_in_stats(cp, skb);
1713         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol ||
1714             IPPROTO_SCTP == cih->protocol)
1715                 offset += 2 * sizeof(__u16);
1716         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1717
1718 out:
1719         if (likely(!new_cp))
1720                 __ip_vs_conn_put(cp);
1721         else
1722                 ip_vs_conn_put(cp);
1723
1724         return verdict;
1725 }
1726
1727 #ifdef CONFIG_IP_VS_IPV6
1728 static int ip_vs_in_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
1729                             int *related, unsigned int hooknum,
1730                             struct ip_vs_iphdr *iph)
1731 {
1732         struct icmp6hdr _icmph, *ic;
1733         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1734         struct ip_vs_conn *cp;
1735         struct ip_vs_protocol *pp;
1736         struct ip_vs_proto_data *pd;
1737         unsigned int offset, verdict;
1738         bool new_cp = false;
1739
1740         *related = 1;
1741
1742         ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph, iph);
1743         if (ic == NULL)
1744                 return NF_DROP;
1745
1746         /*
1747          * Work through seeing if this is for us.
1748          * These checks are supposed to be in an order that means easy
1749          * things are checked first to speed up processing.... however
1750          * this means that some packets will manage to get a long way
1751          * down this stack and then be rejected, but that's life.
1752          */
1753         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1754                 *related = 0;
1755                 return NF_ACCEPT;
1756         }
1757         /* Fragment header that is before ICMP header tells us that:
1758          * it's not an error message since they can't be fragmented.
1759          */
1760         if (iph->flags & IP6_FH_F_FRAG)
1761                 return NF_DROP;
1762
1763         IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1764                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1765                   &iph->saddr, &iph->daddr);
1766
1767         offset = iph->len + sizeof(_icmph);
1768         if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, offset, true, &ciph))
1769                 return NF_ACCEPT;
1770
1771         pd = ip_vs_proto_data_get(ipvs, ciph.protocol);
1772         if (!pd)
1773                 return NF_ACCEPT;
1774         pp = pd->pp;
1775
1776         /* Cannot handle fragmented embedded protocol */
1777         if (ciph.fragoffs)
1778                 return NF_ACCEPT;
1779
1780         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
1781                       "Checking incoming ICMPv6 for");
1782
1783         /* The embedded headers contain source and dest in reverse order
1784          * if not from localhost
1785          */
1786         cp = pp->conn_in_get(ipvs, AF_INET6, skb, &ciph);
1787
1788         if (!cp) {
1789                 int v;
1790
1791                 if (!sysctl_schedule_icmp(ipvs))
1792                         return NF_ACCEPT;
1793
1794                 if (!ip_vs_try_to_schedule(ipvs, AF_INET6, skb, pd, &v, &cp, &ciph))
1795                         return v;
1796
1797                 new_cp = true;
1798         }
1799
1800         /* VS/TUN, VS/DR and LOCALNODE just let it go */
1801         if ((hooknum == NF_INET_LOCAL_OUT) &&
1802             (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1803                 verdict = NF_ACCEPT;
1804                 goto out;
1805         }
1806
1807         /* do the statistics and put it back */
1808         ip_vs_in_stats(cp, skb);
1809
1810         /* Need to mangle contained IPv6 header in ICMPv6 packet */
1811         offset = ciph.len;
1812         if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1813             IPPROTO_SCTP == ciph.protocol)
1814                 offset += 2 * sizeof(__u16); /* Also mangle ports */
1815
1816         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset, hooknum, &ciph);
1817
1818 out:
1819         if (likely(!new_cp))
1820                 __ip_vs_conn_put(cp);
1821         else
1822                 ip_vs_conn_put(cp);
1823
1824         return verdict;
1825 }
1826 #endif
1827
1828
1829 /*
1830  *      Check if it's for virtual services, look it up,
1831  *      and send it on its way...
1832  */
1833 static unsigned int
1834 ip_vs_in(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1835 {
1836         struct ip_vs_iphdr iph;
1837         struct ip_vs_protocol *pp;
1838         struct ip_vs_proto_data *pd;
1839         struct ip_vs_conn *cp;
1840         int ret, pkts;
1841         int conn_reuse_mode;
1842         struct sock *sk;
1843
1844         /* Already marked as IPVS request or reply? */
1845         if (skb->ipvs_property)
1846                 return NF_ACCEPT;
1847
1848         /*
1849          *      Big tappo:
1850          *      - remote client: only PACKET_HOST
1851          *      - route: used for struct net when skb->dev is unset
1852          */
1853         if (unlikely((skb->pkt_type != PACKET_HOST &&
1854                       hooknum != NF_INET_LOCAL_OUT) ||
1855                      !skb_dst(skb))) {
1856                 ip_vs_fill_iph_skb(af, skb, false, &iph);
1857                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1858                               " ignored in hook %u\n",
1859                               skb->pkt_type, iph.protocol,
1860                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1861                 return NF_ACCEPT;
1862         }
1863         /* ipvs enabled in this netns ? */
1864         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1865                 return NF_ACCEPT;
1866
1867         ip_vs_fill_iph_skb(af, skb, false, &iph);
1868
1869         /* Bad... Do not break raw sockets */
1870         sk = skb_to_full_sk(skb);
1871         if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
1872                      af == AF_INET)) {
1873
1874                 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
1875                         return NF_ACCEPT;
1876         }
1877
1878 #ifdef CONFIG_IP_VS_IPV6
1879         if (af == AF_INET6) {
1880                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1881                         int related;
1882                         int verdict = ip_vs_in_icmp_v6(ipvs, skb, &related,
1883                                                        hooknum, &iph);
1884
1885                         if (related)
1886                                 return verdict;
1887                 }
1888         } else
1889 #endif
1890                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1891                         int related;
1892                         int verdict = ip_vs_in_icmp(ipvs, skb, &related,
1893                                                     hooknum);
1894
1895                         if (related)
1896                                 return verdict;
1897                 }
1898
1899         /* Protocol supported? */
1900         pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1901         if (unlikely(!pd)) {
1902                 /* The only way we'll see this packet again is if it's
1903                  * encapsulated, so mark it with ipvs_property=1 so we
1904                  * skip it if we're ignoring tunneled packets
1905                  */
1906                 if (sysctl_ignore_tunneled(ipvs))
1907                         skb->ipvs_property = 1;
1908
1909                 return NF_ACCEPT;
1910         }
1911         pp = pd->pp;
1912         /*
1913          * Check if the packet belongs to an existing connection entry
1914          */
1915         cp = pp->conn_in_get(ipvs, af, skb, &iph);
1916
1917         conn_reuse_mode = sysctl_conn_reuse_mode(ipvs);
1918         if (conn_reuse_mode && !iph.fragoffs && is_new_conn(skb, &iph) && cp) {
1919                 bool uses_ct = false, resched = false;
1920
1921                 if (unlikely(sysctl_expire_nodest_conn(ipvs)) && cp->dest &&
1922                     unlikely(!atomic_read(&cp->dest->weight))) {
1923                         resched = true;
1924                         uses_ct = ip_vs_conn_uses_conntrack(cp, skb);
1925                 } else if (is_new_conn_expected(cp, conn_reuse_mode)) {
1926                         uses_ct = ip_vs_conn_uses_conntrack(cp, skb);
1927                         if (!atomic_read(&cp->n_control)) {
1928                                 resched = true;
1929                         } else {
1930                                 /* Do not reschedule controlling connection
1931                                  * that uses conntrack while it is still
1932                                  * referenced by controlled connection(s).
1933                                  */
1934                                 resched = !uses_ct;
1935                         }
1936                 }
1937
1938                 if (resched) {
1939                         if (!atomic_read(&cp->n_control))
1940                                 ip_vs_conn_expire_now(cp);
1941                         __ip_vs_conn_put(cp);
1942                         if (uses_ct)
1943                                 return NF_DROP;
1944                         cp = NULL;
1945                 }
1946         }
1947
1948         if (unlikely(!cp)) {
1949                 int v;
1950
1951                 if (!ip_vs_try_to_schedule(ipvs, af, skb, pd, &v, &cp, &iph))
1952                         return v;
1953         }
1954
1955         IP_VS_DBG_PKT(11, af, pp, skb, iph.off, "Incoming packet");
1956
1957         /* Check the server status */
1958         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1959                 /* the destination server is not available */
1960
1961                 if (sysctl_expire_nodest_conn(ipvs)) {
1962                         /* try to expire the connection immediately */
1963                         ip_vs_conn_expire_now(cp);
1964                 }
1965                 /* don't restart its timer, and silently
1966                    drop the packet. */
1967                 __ip_vs_conn_put(cp);
1968                 return NF_DROP;
1969         }
1970
1971         ip_vs_in_stats(cp, skb);
1972         ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1973         if (cp->packet_xmit)
1974                 ret = cp->packet_xmit(skb, cp, pp, &iph);
1975                 /* do not touch skb anymore */
1976         else {
1977                 IP_VS_DBG_RL("warning: packet_xmit is null");
1978                 ret = NF_ACCEPT;
1979         }
1980
1981         /* Increase its packet counter and check if it is needed
1982          * to be synchronized
1983          *
1984          * Sync connection if it is about to close to
1985          * encorage the standby servers to update the connections timeout
1986          *
1987          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1988          */
1989
1990         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
1991                 pkts = sysctl_sync_threshold(ipvs);
1992         else
1993                 pkts = atomic_add_return(1, &cp->in_pkts);
1994
1995         if (ipvs->sync_state & IP_VS_STATE_MASTER)
1996                 ip_vs_sync_conn(ipvs, cp, pkts);
1997         else if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) && cp->control)
1998                 /* increment is done inside ip_vs_sync_conn too */
1999                 atomic_inc(&cp->control->in_pkts);
2000
2001         ip_vs_conn_put(cp);
2002         return ret;
2003 }
2004
2005 /*
2006  *      AF_INET handler in NF_INET_LOCAL_IN chain
2007  *      Schedule and forward packets from remote clients
2008  */
2009 static unsigned int
2010 ip_vs_remote_request4(void *priv, struct sk_buff *skb,
2011                       const struct nf_hook_state *state)
2012 {
2013         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
2014 }
2015
2016 /*
2017  *      AF_INET handler in NF_INET_LOCAL_OUT chain
2018  *      Schedule and forward packets from local clients
2019  */
2020 static unsigned int
2021 ip_vs_local_request4(void *priv, struct sk_buff *skb,
2022                      const struct nf_hook_state *state)
2023 {
2024         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
2025 }
2026
2027 #ifdef CONFIG_IP_VS_IPV6
2028
2029 /*
2030  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
2031  *      Schedule and forward packets from remote clients
2032  */
2033 static unsigned int
2034 ip_vs_remote_request6(void *priv, struct sk_buff *skb,
2035                       const struct nf_hook_state *state)
2036 {
2037         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
2038 }
2039
2040 /*
2041  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
2042  *      Schedule and forward packets from local clients
2043  */
2044 static unsigned int
2045 ip_vs_local_request6(void *priv, struct sk_buff *skb,
2046                      const struct nf_hook_state *state)
2047 {
2048         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
2049 }
2050
2051 #endif
2052
2053
2054 /*
2055  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
2056  *      related packets destined for 0.0.0.0/0.
2057  *      When fwmark-based virtual service is used, such as transparent
2058  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
2059  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
2060  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
2061  *      and send them to ip_vs_in_icmp.
2062  */
2063 static unsigned int
2064 ip_vs_forward_icmp(void *priv, struct sk_buff *skb,
2065                    const struct nf_hook_state *state)
2066 {
2067         int r;
2068         struct netns_ipvs *ipvs = net_ipvs(state->net);
2069
2070         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
2071                 return NF_ACCEPT;
2072
2073         /* ipvs enabled in this netns ? */
2074         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
2075                 return NF_ACCEPT;
2076
2077         return ip_vs_in_icmp(ipvs, skb, &r, state->hook);
2078 }
2079
2080 #ifdef CONFIG_IP_VS_IPV6
2081 static unsigned int
2082 ip_vs_forward_icmp_v6(void *priv, struct sk_buff *skb,
2083                       const struct nf_hook_state *state)
2084 {
2085         int r;
2086         struct netns_ipvs *ipvs = net_ipvs(state->net);
2087         struct ip_vs_iphdr iphdr;
2088
2089         ip_vs_fill_iph_skb(AF_INET6, skb, false, &iphdr);
2090         if (iphdr.protocol != IPPROTO_ICMPV6)
2091                 return NF_ACCEPT;
2092
2093         /* ipvs enabled in this netns ? */
2094         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
2095                 return NF_ACCEPT;
2096
2097         return ip_vs_in_icmp_v6(ipvs, skb, &r, state->hook, &iphdr);
2098 }
2099 #endif
2100
2101
2102 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
2103         /* After packet filtering, change source only for VS/NAT */
2104         {
2105                 .hook           = ip_vs_reply4,
2106                 .pf             = NFPROTO_IPV4,
2107                 .hooknum        = NF_INET_LOCAL_IN,
2108                 .priority       = NF_IP_PRI_NAT_SRC - 2,
2109         },
2110         /* After packet filtering, forward packet through VS/DR, VS/TUN,
2111          * or VS/NAT(change destination), so that filtering rules can be
2112          * applied to IPVS. */
2113         {
2114                 .hook           = ip_vs_remote_request4,
2115                 .pf             = NFPROTO_IPV4,
2116                 .hooknum        = NF_INET_LOCAL_IN,
2117                 .priority       = NF_IP_PRI_NAT_SRC - 1,
2118         },
2119         /* Before ip_vs_in, change source only for VS/NAT */
2120         {
2121                 .hook           = ip_vs_local_reply4,
2122                 .pf             = NFPROTO_IPV4,
2123                 .hooknum        = NF_INET_LOCAL_OUT,
2124                 .priority       = NF_IP_PRI_NAT_DST + 1,
2125         },
2126         /* After mangle, schedule and forward local requests */
2127         {
2128                 .hook           = ip_vs_local_request4,
2129                 .pf             = NFPROTO_IPV4,
2130                 .hooknum        = NF_INET_LOCAL_OUT,
2131                 .priority       = NF_IP_PRI_NAT_DST + 2,
2132         },
2133         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2134          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2135         {
2136                 .hook           = ip_vs_forward_icmp,
2137                 .pf             = NFPROTO_IPV4,
2138                 .hooknum        = NF_INET_FORWARD,
2139                 .priority       = 99,
2140         },
2141         /* After packet filtering, change source only for VS/NAT */
2142         {
2143                 .hook           = ip_vs_reply4,
2144                 .pf             = NFPROTO_IPV4,
2145                 .hooknum        = NF_INET_FORWARD,
2146                 .priority       = 100,
2147         },
2148 #ifdef CONFIG_IP_VS_IPV6
2149         /* After packet filtering, change source only for VS/NAT */
2150         {
2151                 .hook           = ip_vs_reply6,
2152                 .pf             = NFPROTO_IPV6,
2153                 .hooknum        = NF_INET_LOCAL_IN,
2154                 .priority       = NF_IP6_PRI_NAT_SRC - 2,
2155         },
2156         /* After packet filtering, forward packet through VS/DR, VS/TUN,
2157          * or VS/NAT(change destination), so that filtering rules can be
2158          * applied to IPVS. */
2159         {
2160                 .hook           = ip_vs_remote_request6,
2161                 .pf             = NFPROTO_IPV6,
2162                 .hooknum        = NF_INET_LOCAL_IN,
2163                 .priority       = NF_IP6_PRI_NAT_SRC - 1,
2164         },
2165         /* Before ip_vs_in, change source only for VS/NAT */
2166         {
2167                 .hook           = ip_vs_local_reply6,
2168                 .pf             = NFPROTO_IPV6,
2169                 .hooknum        = NF_INET_LOCAL_OUT,
2170                 .priority       = NF_IP6_PRI_NAT_DST + 1,
2171         },
2172         /* After mangle, schedule and forward local requests */
2173         {
2174                 .hook           = ip_vs_local_request6,
2175                 .pf             = NFPROTO_IPV6,
2176                 .hooknum        = NF_INET_LOCAL_OUT,
2177                 .priority       = NF_IP6_PRI_NAT_DST + 2,
2178         },
2179         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2180          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2181         {
2182                 .hook           = ip_vs_forward_icmp_v6,
2183                 .pf             = NFPROTO_IPV6,
2184                 .hooknum        = NF_INET_FORWARD,
2185                 .priority       = 99,
2186         },
2187         /* After packet filtering, change source only for VS/NAT */
2188         {
2189                 .hook           = ip_vs_reply6,
2190                 .pf             = NFPROTO_IPV6,
2191                 .hooknum        = NF_INET_FORWARD,
2192                 .priority       = 100,
2193         },
2194 #endif
2195 };
2196 /*
2197  *      Initialize IP Virtual Server netns mem.
2198  */
2199 static int __net_init __ip_vs_init(struct net *net)
2200 {
2201         struct netns_ipvs *ipvs;
2202
2203         ipvs = net_generic(net, ip_vs_net_id);
2204         if (ipvs == NULL)
2205                 return -ENOMEM;
2206
2207         /* Hold the beast until a service is registerd */
2208         ipvs->enable = 0;
2209         ipvs->net = net;
2210         /* Counters used for creating unique names */
2211         ipvs->gen = atomic_read(&ipvs_netns_cnt);
2212         atomic_inc(&ipvs_netns_cnt);
2213         net->ipvs = ipvs;
2214
2215         if (ip_vs_estimator_net_init(ipvs) < 0)
2216                 goto estimator_fail;
2217
2218         if (ip_vs_control_net_init(ipvs) < 0)
2219                 goto control_fail;
2220
2221         if (ip_vs_protocol_net_init(ipvs) < 0)
2222                 goto protocol_fail;
2223
2224         if (ip_vs_app_net_init(ipvs) < 0)
2225                 goto app_fail;
2226
2227         if (ip_vs_conn_net_init(ipvs) < 0)
2228                 goto conn_fail;
2229
2230         if (ip_vs_sync_net_init(ipvs) < 0)
2231                 goto sync_fail;
2232
2233         printk(KERN_INFO "IPVS: Creating netns size=%zu id=%d\n",
2234                          sizeof(struct netns_ipvs), ipvs->gen);
2235         return 0;
2236 /*
2237  * Error handling
2238  */
2239
2240 sync_fail:
2241         ip_vs_conn_net_cleanup(ipvs);
2242 conn_fail:
2243         ip_vs_app_net_cleanup(ipvs);
2244 app_fail:
2245         ip_vs_protocol_net_cleanup(ipvs);
2246 protocol_fail:
2247         ip_vs_control_net_cleanup(ipvs);
2248 control_fail:
2249         ip_vs_estimator_net_cleanup(ipvs);
2250 estimator_fail:
2251         net->ipvs = NULL;
2252         return -ENOMEM;
2253 }
2254
2255 static void __net_exit __ip_vs_cleanup(struct net *net)
2256 {
2257         struct netns_ipvs *ipvs = net_ipvs(net);
2258
2259         ip_vs_service_net_cleanup(ipvs);        /* ip_vs_flush() with locks */
2260         ip_vs_conn_net_cleanup(ipvs);
2261         ip_vs_app_net_cleanup(ipvs);
2262         ip_vs_protocol_net_cleanup(ipvs);
2263         ip_vs_control_net_cleanup(ipvs);
2264         ip_vs_estimator_net_cleanup(ipvs);
2265         IP_VS_DBG(2, "ipvs netns %d released\n", ipvs->gen);
2266         net->ipvs = NULL;
2267 }
2268
2269 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
2270 {
2271         struct netns_ipvs *ipvs = net_ipvs(net);
2272         EnterFunction(2);
2273         ipvs->enable = 0;       /* Disable packet reception */
2274         smp_wmb();
2275         ip_vs_sync_net_cleanup(ipvs);
2276         LeaveFunction(2);
2277 }
2278
2279 static struct pernet_operations ipvs_core_ops = {
2280         .init = __ip_vs_init,
2281         .exit = __ip_vs_cleanup,
2282         .id   = &ip_vs_net_id,
2283         .size = sizeof(struct netns_ipvs),
2284 };
2285
2286 static struct pernet_operations ipvs_core_dev_ops = {
2287         .exit = __ip_vs_dev_cleanup,
2288 };
2289
2290 /*
2291  *      Initialize IP Virtual Server
2292  */
2293 static int __init ip_vs_init(void)
2294 {
2295         int ret;
2296
2297         ret = ip_vs_control_init();
2298         if (ret < 0) {
2299                 pr_err("can't setup control.\n");
2300                 goto exit;
2301         }
2302
2303         ip_vs_protocol_init();
2304
2305         ret = ip_vs_conn_init();
2306         if (ret < 0) {
2307                 pr_err("can't setup connection table.\n");
2308                 goto cleanup_protocol;
2309         }
2310
2311         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
2312         if (ret < 0)
2313                 goto cleanup_conn;
2314
2315         ret = register_pernet_device(&ipvs_core_dev_ops);
2316         if (ret < 0)
2317                 goto cleanup_sub;
2318
2319         ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2320         if (ret < 0) {
2321                 pr_err("can't register hooks.\n");
2322                 goto cleanup_dev;
2323         }
2324
2325         ret = ip_vs_register_nl_ioctl();
2326         if (ret < 0) {
2327                 pr_err("can't register netlink/ioctl.\n");
2328                 goto cleanup_hooks;
2329         }
2330
2331         pr_info("ipvs loaded.\n");
2332
2333         return ret;
2334
2335 cleanup_hooks:
2336         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2337 cleanup_dev:
2338         unregister_pernet_device(&ipvs_core_dev_ops);
2339 cleanup_sub:
2340         unregister_pernet_subsys(&ipvs_core_ops);
2341 cleanup_conn:
2342         ip_vs_conn_cleanup();
2343 cleanup_protocol:
2344         ip_vs_protocol_cleanup();
2345         ip_vs_control_cleanup();
2346 exit:
2347         return ret;
2348 }
2349
2350 static void __exit ip_vs_cleanup(void)
2351 {
2352         ip_vs_unregister_nl_ioctl();
2353         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2354         unregister_pernet_device(&ipvs_core_dev_ops);
2355         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
2356         ip_vs_conn_cleanup();
2357         ip_vs_protocol_cleanup();
2358         ip_vs_control_cleanup();
2359         pr_info("ipvs unloaded.\n");
2360 }
2361
2362 module_init(ip_vs_init);
2363 module_exit(ip_vs_cleanup);
2364 MODULE_LICENSE("GPL");