net/netfilter/ipvs/ip_vs_core.c

   1 /*
   2  * IPVS         An implementation of the IP virtual server support for the
   3  *              LINUX operating system.  IPVS is now implemented as a module
   4  *              over the Netfilter framework. IPVS can be used to build a
   5  *              high-performance and highly available server based on a
   6  *              cluster of servers.
   7  *
   8  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
   9  *              Peter Kese <peter.kese@ijs.si>
  10  *              Julian Anastasov <ja@ssi.bg>
  11  *
  12  *              This program is free software; you can redistribute it and/or
  13  *              modify it under the terms of the GNU General Public License
  14  *              as published by the Free Software Foundation; either version
  15  *              2 of the License, or (at your option) any later version.
  16  *
  17  * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
  18  * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
  19  * and others.
  20  *
  21  * Changes:
  22  *      Paul `Rusty' Russell            properly handle non-linear skbs
  23  *      Harald Welte                    don't use nfcache
  24  *
  25  */
  26
  27 #define KMSG_COMPONENT "IPVS"
  28 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  29
  30 #include <linux/module.h>
  31 #include <linux/kernel.h>
  32 #include <linux/ip.h>
  33 #include <linux/tcp.h>
  34 #include <linux/sctp.h>
  35 #include <linux/icmp.h>
  36 #include <linux/slab.h>
  37
  38 #include <net/ip.h>
  39 #include <net/tcp.h>
  40 #include <net/udp.h>
  41 #include <net/icmp.h>                   /* for icmp_send */
  42 #include <net/route.h>
  43 #include <net/ip6_checksum.h>
  44 #include <net/netns/generic.h>          /* net_generic() */
  45
  46 #include <linux/netfilter.h>
  47 #include <linux/netfilter_ipv4.h>
  48
  49 #ifdef CONFIG_IP_VS_IPV6
  50 #include <net/ipv6.h>
  51 #include <linux/netfilter_ipv6.h>
  52 #include <net/ip6_route.h>
  53 #endif
  54
  55 #include <net/ip_vs.h>
  56
  57
  58 EXPORT_SYMBOL(register_ip_vs_scheduler);
  59 EXPORT_SYMBOL(unregister_ip_vs_scheduler);
  60 EXPORT_SYMBOL(ip_vs_proto_name);
  61 EXPORT_SYMBOL(ip_vs_conn_new);
  62 EXPORT_SYMBOL(ip_vs_conn_in_get);
  63 EXPORT_SYMBOL(ip_vs_conn_out_get);
  64 #ifdef CONFIG_IP_VS_PROTO_TCP
  65 EXPORT_SYMBOL(ip_vs_tcp_conn_listen);
  66 #endif
  67 EXPORT_SYMBOL(ip_vs_conn_put);
  68 #ifdef CONFIG_IP_VS_DEBUG
  69 EXPORT_SYMBOL(ip_vs_get_debug_level);
  70 #endif
  71
  72 static int ip_vs_net_id __read_mostly;
  73 /* netns cnt used for uniqueness */
  74 static atomic_t ipvs_netns_cnt = ATOMIC_INIT(0);
  75
  76 /* ID used in ICMP lookups */
  77 #define icmp_id(icmph)          (((icmph)->un).echo.id)
  78 #define icmpv6_id(icmph)        (icmph->icmp6_dataun.u_echo.identifier)
  79
  80 const char *ip_vs_proto_name(unsigned int proto)
  81 {
  82         static char buf[20];
  83
  84         switch (proto) {
  85         case IPPROTO_IP:
  86                 return "IP";
  87         case IPPROTO_UDP:
  88                 return "UDP";
  89         case IPPROTO_TCP:
  90                 return "TCP";
  91         case IPPROTO_SCTP:
  92                 return "SCTP";
  93         case IPPROTO_ICMP:
  94                 return "ICMP";
  95 #ifdef CONFIG_IP_VS_IPV6
  96         case IPPROTO_ICMPV6:
  97                 return "ICMPv6";
  98 #endif
  99         default:
 100                 sprintf(buf, "IP_%u", proto);
 101                 return buf;
 102         }
 103 }
 104
 105 void ip_vs_init_hash_table(struct list_head *table, int rows)
 106 {
 107         while (--rows >= 0)
 108                 INIT_LIST_HEAD(&table[rows]);
 109 }
 110
 111 static inline void
 112 ip_vs_in_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 113 {
 114         struct ip_vs_dest *dest = cp->dest;
 115         struct netns_ipvs *ipvs = cp->ipvs;
 116
 117         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 118                 struct ip_vs_cpu_stats *s;
 119                 struct ip_vs_service *svc;
 120
 121                 s = this_cpu_ptr(dest->stats.cpustats);
 122                 u64_stats_update_begin(&s->syncp);
 123                 s->cnt.inpkts++;
 124                 s->cnt.inbytes += skb->len;
 125                 u64_stats_update_end(&s->syncp);
 126
 127                 rcu_read_lock();
 128                 svc = rcu_dereference(dest->svc);
 129                 s = this_cpu_ptr(svc->stats.cpustats);
 130                 u64_stats_update_begin(&s->syncp);
 131                 s->cnt.inpkts++;
 132                 s->cnt.inbytes += skb->len;
 133                 u64_stats_update_end(&s->syncp);
 134                 rcu_read_unlock();
 135
 136                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 137                 u64_stats_update_begin(&s->syncp);
 138                 s->cnt.inpkts++;
 139                 s->cnt.inbytes += skb->len;
 140                 u64_stats_update_end(&s->syncp);
 141         }
 142 }
 143
 144
 145 static inline void
 146 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 147 {
 148         struct ip_vs_dest *dest = cp->dest;
 149         struct netns_ipvs *ipvs = cp->ipvs;
 150
 151         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 152                 struct ip_vs_cpu_stats *s;
 153                 struct ip_vs_service *svc;
 154
 155                 s = this_cpu_ptr(dest->stats.cpustats);
 156                 u64_stats_update_begin(&s->syncp);
 157                 s->cnt.outpkts++;
 158                 s->cnt.outbytes += skb->len;
 159                 u64_stats_update_end(&s->syncp);
 160
 161                 rcu_read_lock();
 162                 svc = rcu_dereference(dest->svc);
 163                 s = this_cpu_ptr(svc->stats.cpustats);
 164                 u64_stats_update_begin(&s->syncp);
 165                 s->cnt.outpkts++;
 166                 s->cnt.outbytes += skb->len;
 167                 u64_stats_update_end(&s->syncp);
 168                 rcu_read_unlock();
 169
 170                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 171                 u64_stats_update_begin(&s->syncp);
 172                 s->cnt.outpkts++;
 173                 s->cnt.outbytes += skb->len;
 174                 u64_stats_update_end(&s->syncp);
 175         }
 176 }
 177
 178
 179 static inline void
 180 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
 181 {
 182         struct netns_ipvs *ipvs = svc->ipvs;
 183         struct ip_vs_cpu_stats *s;
 184
 185         s = this_cpu_ptr(cp->dest->stats.cpustats);
 186         u64_stats_update_begin(&s->syncp);
 187         s->cnt.conns++;
 188         u64_stats_update_end(&s->syncp);
 189
 190         s = this_cpu_ptr(svc->stats.cpustats);
 191         u64_stats_update_begin(&s->syncp);
 192         s->cnt.conns++;
 193         u64_stats_update_end(&s->syncp);
 194
 195         s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 196         u64_stats_update_begin(&s->syncp);
 197         s->cnt.conns++;
 198         u64_stats_update_end(&s->syncp);
 199 }
 200
 201
 202 static inline void
 203 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
 204                 const struct sk_buff *skb,
 205                 struct ip_vs_proto_data *pd)
 206 {
 207         if (likely(pd->pp->state_transition))
 208                 pd->pp->state_transition(cp, direction, skb, pd);
 209 }
 210
 211 static inline int
 212 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
 213                               struct sk_buff *skb, int protocol,
 214                               const union nf_inet_addr *caddr, __be16 cport,
 215                               const union nf_inet_addr *vaddr, __be16 vport,
 216                               struct ip_vs_conn_param *p)
 217 {
 218         ip_vs_conn_fill_param(svc->ipvs, svc->af, protocol, caddr, cport, vaddr,
 219                               vport, p);
 220         p->pe = rcu_dereference(svc->pe);
 221         if (p->pe && p->pe->fill_param)
 222                 return p->pe->fill_param(p, skb);
 223
 224         return 0;
 225 }
 226
 227 /*
 228  *  IPVS persistent scheduling function
 229  *  It creates a connection entry according to its template if exists,
 230  *  or selects a server and creates a connection entry plus a template.
 231  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
 232  *  Protocols supported: TCP, UDP
 233  */
 234 static struct ip_vs_conn *
 235 ip_vs_sched_persist(struct ip_vs_service *svc,
 236                     struct sk_buff *skb, __be16 src_port, __be16 dst_port,
 237                     int *ignored, struct ip_vs_iphdr *iph)
 238 {
 239         struct ip_vs_conn *cp = NULL;
 240         struct ip_vs_dest *dest;
 241         struct ip_vs_conn *ct;
 242         __be16 dport = 0;               /* destination port to forward */
 243         unsigned int flags;
 244         struct ip_vs_conn_param param;
 245         const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
 246         union nf_inet_addr snet;        /* source network of the client,
 247                                            after masking */
 248         const union nf_inet_addr *src_addr, *dst_addr;
 249
 250         if (likely(!ip_vs_iph_inverse(iph))) {
 251                 src_addr = &iph->saddr;
 252                 dst_addr = &iph->daddr;
 253         } else {
 254                 src_addr = &iph->daddr;
 255                 dst_addr = &iph->saddr;
 256         }
 257
 258
 259         /* Mask saddr with the netmask to adjust template granularity */
 260 #ifdef CONFIG_IP_VS_IPV6
 261         if (svc->af == AF_INET6)
 262                 ipv6_addr_prefix(&snet.in6, &src_addr->in6,
 263                                  (__force __u32) svc->netmask);
 264         else
 265 #endif
 266                 snet.ip = src_addr->ip & svc->netmask;
 267
 268         IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
 269                       "mnet %s\n",
 270                       IP_VS_DBG_ADDR(svc->af, src_addr), ntohs(src_port),
 271                       IP_VS_DBG_ADDR(svc->af, dst_addr), ntohs(dst_port),
 272                       IP_VS_DBG_ADDR(svc->af, &snet));
 273
 274         /*
 275          * As far as we know, FTP is a very complicated network protocol, and
 276          * it uses control connection and data connections. For active FTP,
 277          * FTP server initialize data connection to the client, its source port
 278          * is often 20. For passive FTP, FTP server tells the clients the port
 279          * that it passively listens to,  and the client issues the data
 280          * connection. In the tunneling or direct routing mode, the load
 281          * balancer is on the client-to-server half of connection, the port
 282          * number is unknown to the load balancer. So, a conn template like
 283          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
 284          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
 285          * is created for other persistent services.
 286          */
 287         {
 288                 int protocol = iph->protocol;
 289                 const union nf_inet_addr *vaddr = dst_addr;
 290                 __be16 vport = 0;
 291
 292                 if (dst_port == svc->port) {
 293                         /* non-FTP template:
 294                          * <protocol, caddr, 0, vaddr, vport, daddr, dport>
 295                          * FTP template:
 296                          * <protocol, caddr, 0, vaddr, 0, daddr, 0>
 297                          */
 298                         if (svc->port != FTPPORT)
 299                                 vport = dst_port;
 300                 } else {
 301                         /* Note: persistent fwmark-based services and
 302                          * persistent port zero service are handled here.
 303                          * fwmark template:
 304                          * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
 305                          * port zero template:
 306                          * <protocol,caddr,0,vaddr,0,daddr,0>
 307                          */
 308                         if (svc->fwmark) {
 309                                 protocol = IPPROTO_IP;
 310                                 vaddr = &fwmark;
 311                         }
 312                 }
 313                 /* return *ignored = -1 so NF_DROP can be used */
 314                 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
 315                                                   vaddr, vport, &param) < 0) {
 316                         *ignored = -1;
 317                         return NULL;
 318                 }
 319         }
 320
 321         /* Check if a template already exists */
 322         ct = ip_vs_ct_in_get(&param);
 323         if (!ct || !ip_vs_check_template(ct)) {
 324                 struct ip_vs_scheduler *sched;
 325
 326                 /*
 327                  * No template found or the dest of the connection
 328                  * template is not available.
 329                  * return *ignored=0 i.e. ICMP and NF_DROP
 330                  */
 331                 sched = rcu_dereference(svc->scheduler);
 332                 if (sched) {
 333                         /* read svc->sched_data after svc->scheduler */
 334                         smp_rmb();
 335                         dest = sched->schedule(svc, skb, iph);
 336                 } else {
 337                         dest = NULL;
 338                 }
 339                 if (!dest) {
 340                         IP_VS_DBG(1, "p-schedule: no dest found.\n");
 341                         kfree(param.pe_data);
 342                         *ignored = 0;
 343                         return NULL;
 344                 }
 345
 346                 if (dst_port == svc->port && svc->port != FTPPORT)
 347                         dport = dest->port;
 348
 349                 /* Create a template
 350                  * This adds param.pe_data to the template,
 351                  * and thus param.pe_data will be destroyed
 352                  * when the template expires */
 353                 ct = ip_vs_conn_new(&param, dest->af, &dest->addr, dport,
 354                                     IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
 355                 if (ct == NULL) {
 356                         kfree(param.pe_data);
 357                         *ignored = -1;
 358                         return NULL;
 359                 }
 360
 361                 ct->timeout = svc->timeout;
 362         } else {
 363                 /* set destination with the found template */
 364                 dest = ct->dest;
 365                 kfree(param.pe_data);
 366         }
 367
 368         dport = dst_port;
 369         if (dport == svc->port && dest->port)
 370                 dport = dest->port;
 371
 372         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 373                  && iph->protocol == IPPROTO_UDP) ?
 374                 IP_VS_CONN_F_ONE_PACKET : 0;
 375
 376         /*
 377          *    Create a new connection according to the template
 378          */
 379         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol, src_addr,
 380                               src_port, dst_addr, dst_port, &param);
 381
 382         cp = ip_vs_conn_new(&param, dest->af, &dest->addr, dport, flags, dest,
 383                             skb->mark);
 384         if (cp == NULL) {
 385                 ip_vs_conn_put(ct);
 386                 *ignored = -1;
 387                 return NULL;
 388         }
 389
 390         /*
 391          *    Add its control
 392          */
 393         ip_vs_control_add(cp, ct);
 394         ip_vs_conn_put(ct);
 395
 396         ip_vs_conn_stats(cp, svc);
 397         return cp;
 398 }
 399
 400
 401 /*
 402  *  IPVS main scheduling function
 403  *  It selects a server according to the virtual service, and
 404  *  creates a connection entry.
 405  *  Protocols supported: TCP, UDP
 406  *
 407  *  Usage of *ignored
 408  *
 409  * 1 :   protocol tried to schedule (eg. on SYN), found svc but the
 410  *       svc/scheduler decides that this packet should be accepted with
 411  *       NF_ACCEPT because it must not be scheduled.
 412  *
 413  * 0 :   scheduler can not find destination, so try bypass or
 414  *       return ICMP and then NF_DROP (ip_vs_leave).
 415  *
 416  * -1 :  scheduler tried to schedule but fatal error occurred, eg.
 417  *       ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
 418  *       failure such as missing Call-ID, ENOMEM on skb_linearize
 419  *       or pe_data. In this case we should return NF_DROP without
 420  *       any attempts to send ICMP with ip_vs_leave.
 421  */
 422 struct ip_vs_conn *
 423 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
 424                struct ip_vs_proto_data *pd, int *ignored,
 425                struct ip_vs_iphdr *iph)
 426 {
 427         struct ip_vs_protocol *pp = pd->pp;
 428         struct ip_vs_conn *cp = NULL;
 429         struct ip_vs_scheduler *sched;
 430         struct ip_vs_dest *dest;
 431         __be16 _ports[2], *pptr, cport, vport;
 432         const void *caddr, *vaddr;
 433         unsigned int flags;
 434
 435         *ignored = 1;
 436         /*
 437          * IPv6 frags, only the first hit here.
 438          */
 439         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 440         if (pptr == NULL)
 441                 return NULL;
 442
 443         if (likely(!ip_vs_iph_inverse(iph))) {
 444                 cport = pptr[0];
 445                 caddr = &iph->saddr;
 446                 vport = pptr[1];
 447                 vaddr = &iph->daddr;
 448         } else {
 449                 cport = pptr[1];
 450                 caddr = &iph->daddr;
 451                 vport = pptr[0];
 452                 vaddr = &iph->saddr;
 453         }
 454
 455         /*
 456          * FTPDATA needs this check when using local real server.
 457          * Never schedule Active FTPDATA connections from real server.
 458          * For LVS-NAT they must be already created. For other methods
 459          * with persistence the connection is created on SYN+ACK.
 460          */
 461         if (cport == FTPDATA) {
 462                 IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
 463                               "Not scheduling FTPDATA");
 464                 return NULL;
 465         }
 466
 467         /*
 468          *    Do not schedule replies from local real server.
 469          */
 470         if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK)) {
 471                 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
 472                 cp = pp->conn_in_get(svc->ipvs, svc->af, skb, iph);
 473                 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
 474
 475                 if (cp) {
 476                         IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
 477                                       "Not scheduling reply for existing"
 478                                       " connection");
 479                         __ip_vs_conn_put(cp);
 480                         return NULL;
 481                 }
 482         }
 483
 484         /*
 485          *    Persistent service
 486          */
 487         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 488                 return ip_vs_sched_persist(svc, skb, cport, vport, ignored,
 489                                            iph);
 490
 491         *ignored = 0;
 492
 493         /*
 494          *    Non-persistent service
 495          */
 496         if (!svc->fwmark && vport != svc->port) {
 497                 if (!svc->port)
 498                         pr_err("Schedule: port zero only supported "
 499                                "in persistent services, "
 500                                "check your ipvs configuration\n");
 501                 return NULL;
 502         }
 503
 504         sched = rcu_dereference(svc->scheduler);
 505         if (sched) {
 506                 /* read svc->sched_data after svc->scheduler */
 507                 smp_rmb();
 508                 dest = sched->schedule(svc, skb, iph);
 509         } else {
 510                 dest = NULL;
 511         }
 512         if (dest == NULL) {
 513                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 514                 return NULL;
 515         }
 516
 517         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 518                  && iph->protocol == IPPROTO_UDP) ?
 519                 IP_VS_CONN_F_ONE_PACKET : 0;
 520
 521         /*
 522          *    Create a connection entry.
 523          */
 524         {
 525                 struct ip_vs_conn_param p;
 526
 527                 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
 528                                       caddr, cport, vaddr, vport, &p);
 529                 cp = ip_vs_conn_new(&p, dest->af, &dest->addr,
 530                                     dest->port ? dest->port : vport,
 531                                     flags, dest, skb->mark);
 532                 if (!cp) {
 533                         *ignored = -1;
 534                         return NULL;
 535                 }
 536         }
 537
 538         IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
 539                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
 540                       ip_vs_fwd_tag(cp),
 541                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 542                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 543                       IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
 544                       cp->flags, atomic_read(&cp->refcnt));
 545
 546         ip_vs_conn_stats(cp, svc);
 547         return cp;
 548 }
 549
 550 static inline int ip_vs_addr_is_unicast(struct net *net, int af,
 551                                         union nf_inet_addr *addr)
 552 {
 553 #ifdef CONFIG_IP_VS_IPV6
 554         if (af == AF_INET6)
 555                 return ipv6_addr_type(&addr->in6) & IPV6_ADDR_UNICAST;
 556 #endif
 557         return (inet_addr_type(net, addr->ip) == RTN_UNICAST);
 558 }
 559
 560 /*
 561  *  Pass or drop the packet.
 562  *  Called by ip_vs_in, when the virtual service is available but
 563  *  no destination is available for a new connection.
 564  */
 565 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 566                 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
 567 {
 568         __be16 _ports[2], *pptr, dport;
 569         struct netns_ipvs *ipvs = svc->ipvs;
 570         struct net *net = ipvs->net;
 571
 572         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 573         if (!pptr)
 574                 return NF_DROP;
 575         dport = likely(!ip_vs_iph_inverse(iph)) ? pptr[1] : pptr[0];
 576
 577         /* if it is fwmark-based service, the cache_bypass sysctl is up
 578            and the destination is a non-local unicast, then create
 579            a cache_bypass connection entry */
 580         if (sysctl_cache_bypass(ipvs) && svc->fwmark &&
 581             !(iph->hdr_flags & (IP_VS_HDR_INVERSE | IP_VS_HDR_ICMP)) &&
 582             ip_vs_addr_is_unicast(net, svc->af, &iph->daddr)) {
 583                 int ret;
 584                 struct ip_vs_conn *cp;
 585                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
 586                                       iph->protocol == IPPROTO_UDP) ?
 587                                       IP_VS_CONN_F_ONE_PACKET : 0;
 588                 union nf_inet_addr daddr =  { .all = { 0, 0, 0, 0 } };
 589
 590                 /* create a new connection entry */
 591                 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
 592                 {
 593                         struct ip_vs_conn_param p;
 594                         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
 595                                               &iph->saddr, pptr[0],
 596                                               &iph->daddr, pptr[1], &p);
 597                         cp = ip_vs_conn_new(&p, svc->af, &daddr, 0,
 598                                             IP_VS_CONN_F_BYPASS | flags,
 599                                             NULL, skb->mark);
 600                         if (!cp)
 601                                 return NF_DROP;
 602                 }
 603
 604                 /* statistics */
 605                 ip_vs_in_stats(cp, skb);
 606
 607                 /* set state */
 608                 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
 609
 610                 /* transmit the first SYN packet */
 611                 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
 612                 /* do not touch skb anymore */
 613
 614                 atomic_inc(&cp->in_pkts);
 615                 ip_vs_conn_put(cp);
 616                 return ret;
 617         }
 618
 619         /*
 620          * When the virtual ftp service is presented, packets destined
 621          * for other services on the VIP may get here (except services
 622          * listed in the ipvs table), pass the packets, because it is
 623          * not ipvs job to decide to drop the packets.
 624          */
 625         if (svc->port == FTPPORT && dport != FTPPORT)
 626                 return NF_ACCEPT;
 627
 628         if (unlikely(ip_vs_iph_icmp(iph)))
 629                 return NF_DROP;
 630
 631         /*
 632          * Notify the client that the destination is unreachable, and
 633          * release the socket buffer.
 634          * Since it is in IP layer, the TCP socket is not actually
 635          * created, the TCP RST packet cannot be sent, instead that
 636          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 637          */
 638 #ifdef CONFIG_IP_VS_IPV6
 639         if (svc->af == AF_INET6) {
 640                 if (!skb->dev)
 641                         skb->dev = net->loopback_dev;
 642                 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 643         } else
 644 #endif
 645                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 646
 647         return NF_DROP;
 648 }
 649
 650 #ifdef CONFIG_SYSCTL
 651
 652 static int sysctl_snat_reroute(struct netns_ipvs *ipvs)
 653 {
 654         return ipvs->sysctl_snat_reroute;
 655 }
 656
 657 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs)
 658 {
 659         return ipvs->sysctl_nat_icmp_send;
 660 }
 661
 662 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
 663 {
 664         return ipvs->sysctl_expire_nodest_conn;
 665 }
 666
 667 #else
 668
 669 static int sysctl_snat_reroute(struct netns_ipvs *ipvs) { return 0; }
 670 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs) { return 0; }
 671 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
 672
 673 #endif
 674
 675 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 676 {
 677         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 678 }
 679
 680 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
 681 {
 682         if (NF_INET_LOCAL_IN == hooknum)
 683                 return IP_DEFRAG_VS_IN;
 684         if (NF_INET_FORWARD == hooknum)
 685                 return IP_DEFRAG_VS_FWD;
 686         return IP_DEFRAG_VS_OUT;
 687 }
 688
 689 static inline int ip_vs_gather_frags(struct netns_ipvs *ipvs,
 690                                      struct sk_buff *skb, u_int32_t user)
 691 {
 692         int err;
 693
 694         local_bh_disable();
 695         err = ip_defrag(ipvs->net, skb, user);
 696         local_bh_enable();
 697         if (!err)
 698                 ip_send_check(ip_hdr(skb));
 699
 700         return err;
 701 }
 702
 703 static int ip_vs_route_me_harder(struct netns_ipvs *ipvs, int af,
 704                                  struct sk_buff *skb, unsigned int hooknum)
 705 {
 706         if (!sysctl_snat_reroute(ipvs))
 707                 return 0;
 708         /* Reroute replies only to remote clients (FORWARD and LOCAL_OUT) */
 709         if (NF_INET_LOCAL_IN == hooknum)
 710                 return 0;
 711 #ifdef CONFIG_IP_VS_IPV6
 712         if (af == AF_INET6) {
 713                 struct dst_entry *dst = skb_dst(skb);
 714
 715                 if (dst->dev && !(dst->dev->flags & IFF_LOOPBACK) &&
 716                     ip6_route_me_harder(ipvs->net, skb) != 0)
 717                         return 1;
 718         } else
 719 #endif
 720                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
 721                     ip_route_me_harder(ipvs->net, skb, RTN_LOCAL) != 0)
 722                         return 1;
 723
 724         return 0;
 725 }
 726
 727 /*
 728  * Packet has been made sufficiently writable in caller
 729  * - inout: 1=in->out, 0=out->in
 730  */
 731 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 732                     struct ip_vs_conn *cp, int inout)
 733 {
 734         struct iphdr *iph        = ip_hdr(skb);
 735         unsigned int icmp_offset = iph->ihl*4;
 736         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
 737                                                       icmp_offset);
 738         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 739
 740         if (inout) {
 741                 iph->saddr = cp->vaddr.ip;
 742                 ip_send_check(iph);
 743                 ciph->daddr = cp->vaddr.ip;
 744                 ip_send_check(ciph);
 745         } else {
 746                 iph->daddr = cp->daddr.ip;
 747                 ip_send_check(iph);
 748                 ciph->saddr = cp->daddr.ip;
 749                 ip_send_check(ciph);
 750         }
 751
 752         /* the TCP/UDP/SCTP port */
 753         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
 754             IPPROTO_SCTP == ciph->protocol) {
 755                 __be16 *ports = (void *)ciph + ciph->ihl*4;
 756
 757                 if (inout)
 758                         ports[1] = cp->vport;
 759                 else
 760                         ports[0] = cp->dport;
 761         }
 762
 763         /* And finally the ICMP checksum */
 764         icmph->checksum = 0;
 765         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 766         skb->ip_summed = CHECKSUM_UNNECESSARY;
 767
 768         if (inout)
 769                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 770                         "Forwarding altered outgoing ICMP");
 771         else
 772                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 773                         "Forwarding altered incoming ICMP");
 774 }
 775
 776 #ifdef CONFIG_IP_VS_IPV6
 777 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
 778                     struct ip_vs_conn *cp, int inout)
 779 {
 780         struct ipv6hdr *iph      = ipv6_hdr(skb);
 781         unsigned int icmp_offset = 0;
 782         unsigned int offs        = 0; /* header offset*/
 783         int protocol;
 784         struct icmp6hdr *icmph;
 785         struct ipv6hdr *ciph;
 786         unsigned short fragoffs;
 787
 788         ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
 789         icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
 790         offs = icmp_offset + sizeof(struct icmp6hdr);
 791         ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
 792
 793         protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
 794
 795         if (inout) {
 796                 iph->saddr = cp->vaddr.in6;
 797                 ciph->daddr = cp->vaddr.in6;
 798         } else {
 799                 iph->daddr = cp->daddr.in6;
 800                 ciph->saddr = cp->daddr.in6;
 801         }
 802
 803         /* the TCP/UDP/SCTP port */
 804         if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 805                           IPPROTO_SCTP == protocol)) {
 806                 __be16 *ports = (void *)(skb_network_header(skb) + offs);
 807
 808                 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
 809                               ntohs(inout ? ports[1] : ports[0]),
 810                               ntohs(inout ? cp->vport : cp->dport));
 811                 if (inout)
 812                         ports[1] = cp->vport;
 813                 else
 814                         ports[0] = cp->dport;
 815         }
 816
 817         /* And finally the ICMP checksum */
 818         icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
 819                                               skb->len - icmp_offset,
 820                                               IPPROTO_ICMPV6, 0);
 821         skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
 822         skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
 823         skb->ip_summed = CHECKSUM_PARTIAL;
 824
 825         if (inout)
 826                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 827                               (void *)ciph - (void *)iph,
 828                               "Forwarding altered outgoing ICMPv6");
 829         else
 830                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 831                               (void *)ciph - (void *)iph,
 832                               "Forwarding altered incoming ICMPv6");
 833 }
 834 #endif
 835
 836 /* Handle relevant response ICMP messages - forward to the right
 837  * destination host.
 838  */
 839 static int handle_response_icmp(int af, struct sk_buff *skb,
 840                                 union nf_inet_addr *snet,
 841                                 __u8 protocol, struct ip_vs_conn *cp,
 842                                 struct ip_vs_protocol *pp,
 843                                 unsigned int offset, unsigned int ihl,
 844                                 unsigned int hooknum)
 845 {
 846         unsigned int verdict = NF_DROP;
 847
 848         if (IP_VS_FWD_METHOD(cp) != 0) {
 849                 pr_err("shouldn't reach here, because the box is on the "
 850                        "half connection in the tun/dr module.\n");
 851         }
 852
 853         /* Ensure the checksum is correct */
 854         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 855                 /* Failed checksum! */
 856                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
 857                               IP_VS_DBG_ADDR(af, snet));
 858                 goto out;
 859         }
 860
 861         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 862             IPPROTO_SCTP == protocol)
 863                 offset += 2 * sizeof(__u16);
 864         if (!skb_make_writable(skb, offset))
 865                 goto out;
 866
 867 #ifdef CONFIG_IP_VS_IPV6
 868         if (af == AF_INET6)
 869                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
 870         else
 871 #endif
 872                 ip_vs_nat_icmp(skb, pp, cp, 1);
 873
 874         if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
 875                 goto out;
 876
 877         /* do the statistics and put it back */
 878         ip_vs_out_stats(cp, skb);
 879
 880         skb->ipvs_property = 1;
 881         if (!(cp->flags & IP_VS_CONN_F_NFCT))
 882                 ip_vs_notrack(skb);
 883         else
 884                 ip_vs_update_conntrack(skb, cp, 0);
 885         verdict = NF_ACCEPT;
 886
 887 out:
 888         __ip_vs_conn_put(cp);
 889
 890         return verdict;
 891 }
 892
 893 /*
 894  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 895  *      Find any that might be relevant, check against existing connections.
 896  *      Currently handles error types - unreachable, quench, ttl exceeded.
 897  */
 898 static int ip_vs_out_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb,
 899                           int *related, unsigned int hooknum)
 900 {
 901         struct iphdr *iph;
 902         struct icmphdr  _icmph, *ic;
 903         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 904         struct ip_vs_iphdr ciph;
 905         struct ip_vs_conn *cp;
 906         struct ip_vs_protocol *pp;
 907         unsigned int offset, ihl;
 908         union nf_inet_addr snet;
 909
 910         *related = 1;
 911
 912         /* reassemble IP fragments */
 913         if (ip_is_fragment(ip_hdr(skb))) {
 914                 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
 915                         return NF_STOLEN;
 916         }
 917
 918         iph = ip_hdr(skb);
 919         offset = ihl = iph->ihl * 4;
 920         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 921         if (ic == NULL)
 922                 return NF_DROP;
 923
 924         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
 925                   ic->type, ntohs(icmp_id(ic)),
 926                   &iph->saddr, &iph->daddr);
 927
 928         /*
 929          * Work through seeing if this is for us.
 930          * These checks are supposed to be in an order that means easy
 931          * things are checked first to speed up processing.... however
 932          * this means that some packets will manage to get a long way
 933          * down this stack and then be rejected, but that's life.
 934          */
 935         if ((ic->type != ICMP_DEST_UNREACH) &&
 936             (ic->type != ICMP_SOURCE_QUENCH) &&
 937             (ic->type != ICMP_TIME_EXCEEDED)) {
 938                 *related = 0;
 939                 return NF_ACCEPT;
 940         }
 941
 942         /* Now find the contained IP header */
 943         offset += sizeof(_icmph);
 944         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 945         if (cih == NULL)
 946                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 947
 948         pp = ip_vs_proto_get(cih->protocol);
 949         if (!pp)
 950                 return NF_ACCEPT;
 951
 952         /* Is the embedded protocol header present? */
 953         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 954                      pp->dont_defrag))
 955                 return NF_ACCEPT;
 956
 957         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
 958                       "Checking outgoing ICMP for");
 959
 960         ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, true, &ciph);
 961
 962         /* The embedded headers contain source and dest in reverse order */
 963         cp = pp->conn_out_get(ipvs, AF_INET, skb, &ciph);
 964         if (!cp)
 965                 return NF_ACCEPT;
 966
 967         snet.ip = iph->saddr;
 968         return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
 969                                     pp, ciph.len, ihl, hooknum);
 970 }
 971
 972 #ifdef CONFIG_IP_VS_IPV6
 973 static int ip_vs_out_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
 974                              int *related,  unsigned int hooknum,
 975                              struct ip_vs_iphdr *ipvsh)
 976 {
 977         struct icmp6hdr _icmph, *ic;
 978         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
 979         struct ip_vs_conn *cp;
 980         struct ip_vs_protocol *pp;
 981         union nf_inet_addr snet;
 982         unsigned int offset;
 983
 984         *related = 1;
 985         ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph, ipvsh);
 986         if (ic == NULL)
 987                 return NF_DROP;
 988
 989         /*
 990          * Work through seeing if this is for us.
 991          * These checks are supposed to be in an order that means easy
 992          * things are checked first to speed up processing.... however
 993          * this means that some packets will manage to get a long way
 994          * down this stack and then be rejected, but that's life.
 995          */
 996         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
 997                 *related = 0;
 998                 return NF_ACCEPT;
 999         }
1000         /* Fragment header that is before ICMP header tells us that:
1001          * it's not an error message since they can't be fragmented.
1002          */
1003         if (ipvsh->flags & IP6_FH_F_FRAG)
1004                 return NF_DROP;
1005
1006         IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1007                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1008                   &ipvsh->saddr, &ipvsh->daddr);
1009
1010         if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, ipvsh->len + sizeof(_icmph),
1011                                      true, &ciph))
1012                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1013
1014         pp = ip_vs_proto_get(ciph.protocol);
1015         if (!pp)
1016                 return NF_ACCEPT;
1017
1018         /* The embedded headers contain source and dest in reverse order */
1019         cp = pp->conn_out_get(ipvs, AF_INET6, skb, &ciph);
1020         if (!cp)
1021                 return NF_ACCEPT;
1022
1023         snet.in6 = ciph.saddr.in6;
1024         offset = ciph.len;
1025         return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
1026                                     pp, offset, sizeof(struct ipv6hdr),
1027                                     hooknum);
1028 }
1029 #endif
1030
1031 /*
1032  * Check if sctp chunc is ABORT chunk
1033  */
1034 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
1035 {
1036         sctp_chunkhdr_t *sch, schunk;
1037         sch = skb_header_pointer(skb, nh_len + sizeof(sctp_sctphdr_t),
1038                         sizeof(schunk), &schunk);
1039         if (sch == NULL)
1040                 return 0;
1041         if (sch->type == SCTP_CID_ABORT)
1042                 return 1;
1043         return 0;
1044 }
1045
1046 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
1047 {
1048         struct tcphdr _tcph, *th;
1049
1050         th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
1051         if (th == NULL)
1052                 return 0;
1053         return th->rst;
1054 }
1055
1056 static inline bool is_new_conn(const struct sk_buff *skb,
1057                                struct ip_vs_iphdr *iph)
1058 {
1059         switch (iph->protocol) {
1060         case IPPROTO_TCP: {
1061                 struct tcphdr _tcph, *th;
1062
1063                 th = skb_header_pointer(skb, iph->len, sizeof(_tcph), &_tcph);
1064                 if (th == NULL)
1065                         return false;
1066                 return th->syn;
1067         }
1068         case IPPROTO_SCTP: {
1069                 sctp_chunkhdr_t *sch, schunk;
1070
1071                 sch = skb_header_pointer(skb, iph->len + sizeof(sctp_sctphdr_t),
1072                                          sizeof(schunk), &schunk);
1073                 if (sch == NULL)
1074                         return false;
1075                 return sch->type == SCTP_CID_INIT;
1076         }
1077         default:
1078                 return false;
1079         }
1080 }
1081
1082 static inline bool is_new_conn_expected(const struct ip_vs_conn *cp,
1083                                         int conn_reuse_mode)
1084 {
1085         /* Controlled (FTP DATA or persistence)? */
1086         if (cp->control)
1087                 return false;
1088
1089         switch (cp->protocol) {
1090         case IPPROTO_TCP:
1091                 return (cp->state == IP_VS_TCP_S_TIME_WAIT) ||
1092                        (cp->state == IP_VS_TCP_S_CLOSE) ||
1093                         ((conn_reuse_mode & 2) &&
1094                          (cp->state == IP_VS_TCP_S_FIN_WAIT) &&
1095                          (cp->flags & IP_VS_CONN_F_NOOUTPUT));
1096         case IPPROTO_SCTP:
1097                 return cp->state == IP_VS_SCTP_S_CLOSED;
1098         default:
1099                 return false;
1100         }
1101 }
1102
1103 /* Handle response packets: rewrite addresses and send away...
1104  */
1105 static unsigned int
1106 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1107                 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph,
1108                 unsigned int hooknum)
1109 {
1110         struct ip_vs_protocol *pp = pd->pp;
1111
1112         IP_VS_DBG_PKT(11, af, pp, skb, iph->off, "Outgoing packet");
1113
1114         if (!skb_make_writable(skb, iph->len))
1115                 goto drop;
1116
1117         /* mangle the packet */
1118         if (pp->snat_handler && !pp->snat_handler(skb, pp, cp, iph))
1119                 goto drop;
1120
1121 #ifdef CONFIG_IP_VS_IPV6
1122         if (af == AF_INET6)
1123                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1124         else
1125 #endif
1126         {
1127                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1128                 ip_send_check(ip_hdr(skb));
1129         }
1130
1131         /*
1132          * nf_iterate does not expect change in the skb->dst->dev.
1133          * It looks like it is not fatal to enable this code for hooks
1134          * where our handlers are at the end of the chain list and
1135          * when all next handlers use skb->dst->dev and not outdev.
1136          * It will definitely route properly the inout NAT traffic
1137          * when multiple paths are used.
1138          */
1139
1140         /* For policy routing, packets originating from this
1141          * machine itself may be routed differently to packets
1142          * passing through.  We want this packet to be routed as
1143          * if it came from this machine itself.  So re-compute
1144          * the routing information.
1145          */
1146         if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
1147                 goto drop;
1148
1149         IP_VS_DBG_PKT(10, af, pp, skb, iph->off, "After SNAT");
1150
1151         ip_vs_out_stats(cp, skb);
1152         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1153         skb->ipvs_property = 1;
1154         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1155                 ip_vs_notrack(skb);
1156         else
1157                 ip_vs_update_conntrack(skb, cp, 0);
1158         ip_vs_conn_put(cp);
1159
1160         LeaveFunction(11);
1161         return NF_ACCEPT;
1162
1163 drop:
1164         ip_vs_conn_put(cp);
1165         kfree_skb(skb);
1166         LeaveFunction(11);
1167         return NF_STOLEN;
1168 }
1169
1170 /*
1171  *      Check if outgoing packet belongs to the established ip_vs_conn.
1172  */
1173 static unsigned int
1174 ip_vs_out(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1175 {
1176         struct ip_vs_iphdr iph;
1177         struct ip_vs_protocol *pp;
1178         struct ip_vs_proto_data *pd;
1179         struct ip_vs_conn *cp;
1180         struct sock *sk;
1181
1182         EnterFunction(11);
1183
1184         /* Already marked as IPVS request or reply? */
1185         if (skb->ipvs_property)
1186                 return NF_ACCEPT;
1187
1188         sk = skb_to_full_sk(skb);
1189         /* Bad... Do not break raw sockets */
1190         if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
1191                      af == AF_INET)) {
1192
1193                 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
1194                         return NF_ACCEPT;
1195         }
1196
1197         if (unlikely(!skb_dst(skb)))
1198                 return NF_ACCEPT;
1199
1200         if (!ipvs->enable)
1201                 return NF_ACCEPT;
1202
1203         ip_vs_fill_iph_skb(af, skb, false, &iph);
1204 #ifdef CONFIG_IP_VS_IPV6
1205         if (af == AF_INET6) {
1206                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1207                         int related;
1208                         int verdict = ip_vs_out_icmp_v6(ipvs, skb, &related,
1209                                                         hooknum, &iph);
1210
1211                         if (related)
1212                                 return verdict;
1213                 }
1214         } else
1215 #endif
1216                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1217                         int related;
1218                         int verdict = ip_vs_out_icmp(ipvs, skb, &related, hooknum);
1219
1220                         if (related)
1221                                 return verdict;
1222                 }
1223
1224         pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1225         if (unlikely(!pd))
1226                 return NF_ACCEPT;
1227         pp = pd->pp;
1228
1229         /* reassemble IP fragments */
1230 #ifdef CONFIG_IP_VS_IPV6
1231         if (af == AF_INET)
1232 #endif
1233                 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1234                         if (ip_vs_gather_frags(ipvs, skb,
1235                                                ip_vs_defrag_user(hooknum)))
1236                                 return NF_STOLEN;
1237
1238                         ip_vs_fill_iph_skb(AF_INET, skb, false, &iph);
1239                 }
1240
1241         /*
1242          * Check if the packet belongs to an existing entry
1243          */
1244         cp = pp->conn_out_get(ipvs, af, skb, &iph);
1245
1246         if (likely(cp))
1247                 return handle_response(af, skb, pd, cp, &iph, hooknum);
1248         if (sysctl_nat_icmp_send(ipvs) &&
1249             (pp->protocol == IPPROTO_TCP ||
1250              pp->protocol == IPPROTO_UDP ||
1251              pp->protocol == IPPROTO_SCTP)) {
1252                 __be16 _ports[2], *pptr;
1253
1254                 pptr = frag_safe_skb_hp(skb, iph.len,
1255                                          sizeof(_ports), _ports, &iph);
1256                 if (pptr == NULL)
1257                         return NF_ACCEPT;       /* Not for me */
1258                 if (ip_vs_has_real_service(ipvs, af, iph.protocol, &iph.saddr,
1259                                            pptr[0])) {
1260                         /*
1261                          * Notify the real server: there is no
1262                          * existing entry if it is not RST
1263                          * packet or not TCP packet.
1264                          */
1265                         if ((iph.protocol != IPPROTO_TCP &&
1266                              iph.protocol != IPPROTO_SCTP)
1267                              || ((iph.protocol == IPPROTO_TCP
1268                                   && !is_tcp_reset(skb, iph.len))
1269                                  || (iph.protocol == IPPROTO_SCTP
1270                                         && !is_sctp_abort(skb,
1271                                                 iph.len)))) {
1272 #ifdef CONFIG_IP_VS_IPV6
1273                                 if (af == AF_INET6) {
1274                                         if (!skb->dev)
1275                                                 skb->dev = ipvs->net->loopback_dev;
1276                                         icmpv6_send(skb,
1277                                                     ICMPV6_DEST_UNREACH,
1278                                                     ICMPV6_PORT_UNREACH,
1279                                                     0);
1280                                 } else
1281 #endif
1282                                         icmp_send(skb,
1283                                                   ICMP_DEST_UNREACH,
1284                                                   ICMP_PORT_UNREACH, 0);
1285                                 return NF_DROP;
1286                         }
1287                 }
1288         }
1289         IP_VS_DBG_PKT(12, af, pp, skb, iph.off,
1290                       "ip_vs_out: packet continues traversal as normal");
1291         return NF_ACCEPT;
1292 }
1293
1294 /*
1295  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1296  *      used only for VS/NAT.
1297  *      Check if packet is reply for established ip_vs_conn.
1298  */
1299 static unsigned int
1300 ip_vs_reply4(void *priv, struct sk_buff *skb,
1301              const struct nf_hook_state *state)
1302 {
1303         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1304 }
1305
1306 /*
1307  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1308  *      Check if packet is reply for established ip_vs_conn.
1309  */
1310 static unsigned int
1311 ip_vs_local_reply4(void *priv, struct sk_buff *skb,
1312                    const struct nf_hook_state *state)
1313 {
1314         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1315 }
1316
1317 #ifdef CONFIG_IP_VS_IPV6
1318
1319 /*
1320  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1321  *      used only for VS/NAT.
1322  *      Check if packet is reply for established ip_vs_conn.
1323  */
1324 static unsigned int
1325 ip_vs_reply6(void *priv, struct sk_buff *skb,
1326              const struct nf_hook_state *state)
1327 {
1328         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1329 }
1330
1331 /*
1332  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1333  *      Check if packet is reply for established ip_vs_conn.
1334  */
1335 static unsigned int
1336 ip_vs_local_reply6(void *priv, struct sk_buff *skb,
1337                    const struct nf_hook_state *state)
1338 {
1339         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1340 }
1341
1342 #endif
1343
1344 static unsigned int
1345 ip_vs_try_to_schedule(struct netns_ipvs *ipvs, int af, struct sk_buff *skb,
1346                       struct ip_vs_proto_data *pd,
1347                       int *verdict, struct ip_vs_conn **cpp,
1348                       struct ip_vs_iphdr *iph)
1349 {
1350         struct ip_vs_protocol *pp = pd->pp;
1351
1352         if (!iph->fragoffs) {
1353                 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1354                  * replayed fragment zero will already have created the cp
1355                  */
1356
1357                 /* Schedule and create new connection entry into cpp */
1358                 if (!pp->conn_schedule(ipvs, af, skb, pd, verdict, cpp, iph))
1359                         return 0;
1360         }
1361
1362         if (unlikely(!*cpp)) {
1363                 /* sorry, all this trouble for a no-hit :) */
1364                 IP_VS_DBG_PKT(12, af, pp, skb, iph->off,
1365                               "ip_vs_in: packet continues traversal as normal");
1366                 if (iph->fragoffs) {
1367                         /* Fragment that couldn't be mapped to a conn entry
1368                          * is missing module nf_defrag_ipv6
1369                          */
1370                         IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1371                         IP_VS_DBG_PKT(7, af, pp, skb, iph->off,
1372                                       "unhandled fragment");
1373                 }
1374                 *verdict = NF_ACCEPT;
1375                 return 0;
1376         }
1377
1378         return 1;
1379 }
1380
1381 /*
1382  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1383  *      Find any that might be relevant, check against existing connections,
1384  *      forward to the right destination host if relevant.
1385  *      Currently handles error types - unreachable, quench, ttl exceeded.
1386  */
1387 static int
1388 ip_vs_in_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb, int *related,
1389               unsigned int hooknum)
1390 {
1391         struct iphdr *iph;
1392         struct icmphdr  _icmph, *ic;
1393         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1394         struct ip_vs_iphdr ciph;
1395         struct ip_vs_conn *cp;
1396         struct ip_vs_protocol *pp;
1397         struct ip_vs_proto_data *pd;
1398         unsigned int offset, offset2, ihl, verdict;
1399         bool ipip, new_cp = false;
1400
1401         *related = 1;
1402
1403         /* reassemble IP fragments */
1404         if (ip_is_fragment(ip_hdr(skb))) {
1405                 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
1406                         return NF_STOLEN;
1407         }
1408
1409         iph = ip_hdr(skb);
1410         offset = ihl = iph->ihl * 4;
1411         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1412         if (ic == NULL)
1413                 return NF_DROP;
1414
1415         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1416                   ic->type, ntohs(icmp_id(ic)),
1417                   &iph->saddr, &iph->daddr);
1418
1419         /*
1420          * Work through seeing if this is for us.
1421          * These checks are supposed to be in an order that means easy
1422          * things are checked first to speed up processing.... however
1423          * this means that some packets will manage to get a long way
1424          * down this stack and then be rejected, but that's life.
1425          */
1426         if ((ic->type != ICMP_DEST_UNREACH) &&
1427             (ic->type != ICMP_SOURCE_QUENCH) &&
1428             (ic->type != ICMP_TIME_EXCEEDED)) {
1429                 *related = 0;
1430                 return NF_ACCEPT;
1431         }
1432
1433         /* Now find the contained IP header */
1434         offset += sizeof(_icmph);
1435         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1436         if (cih == NULL)
1437                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1438
1439         /* Special case for errors for IPIP packets */
1440         ipip = false;
1441         if (cih->protocol == IPPROTO_IPIP) {
1442                 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1443                         return NF_ACCEPT;
1444                 /* Error for our IPIP must arrive at LOCAL_IN */
1445                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1446                         return NF_ACCEPT;
1447                 offset += cih->ihl * 4;
1448                 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1449                 if (cih == NULL)
1450                         return NF_ACCEPT; /* The packet looks wrong, ignore */
1451                 ipip = true;
1452         }
1453
1454         pd = ip_vs_proto_data_get(ipvs, cih->protocol);
1455         if (!pd)
1456                 return NF_ACCEPT;
1457         pp = pd->pp;
1458
1459         /* Is the embedded protocol header present? */
1460         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1461                      pp->dont_defrag))
1462                 return NF_ACCEPT;
1463
1464         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1465                       "Checking incoming ICMP for");
1466
1467         offset2 = offset;
1468         ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, !ipip, &ciph);
1469         offset = ciph.len;
1470
1471         /* The embedded headers contain source and dest in reverse order.
1472          * For IPIP this is error for request, not for reply.
1473          */
1474         cp = pp->conn_in_get(ipvs, AF_INET, skb, &ciph);
1475
1476         if (!cp) {
1477                 int v;
1478
1479                 if (!sysctl_schedule_icmp(ipvs))
1480                         return NF_ACCEPT;
1481
1482                 if (!ip_vs_try_to_schedule(ipvs, AF_INET, skb, pd, &v, &cp, &ciph))
1483                         return v;
1484                 new_cp = true;
1485         }
1486
1487         verdict = NF_DROP;
1488
1489         /* Ensure the checksum is correct */
1490         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1491                 /* Failed checksum! */
1492                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1493                           &iph->saddr);
1494                 goto out;
1495         }
1496
1497         if (ipip) {
1498                 __be32 info = ic->un.gateway;
1499                 __u8 type = ic->type;
1500                 __u8 code = ic->code;
1501
1502                 /* Update the MTU */
1503                 if (ic->type == ICMP_DEST_UNREACH &&
1504                     ic->code == ICMP_FRAG_NEEDED) {
1505                         struct ip_vs_dest *dest = cp->dest;
1506                         u32 mtu = ntohs(ic->un.frag.mtu);
1507                         __be16 frag_off = cih->frag_off;
1508
1509                         /* Strip outer IP and ICMP, go to IPIP header */
1510                         if (pskb_pull(skb, ihl + sizeof(_icmph)) == NULL)
1511                                 goto ignore_ipip;
1512                         offset2 -= ihl + sizeof(_icmph);
1513                         skb_reset_network_header(skb);
1514                         IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1515                                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1516                         ipv4_update_pmtu(skb, ipvs->net,
1517                                          mtu, 0, 0, 0, 0);
1518                         /* Client uses PMTUD? */
1519                         if (!(frag_off & htons(IP_DF)))
1520                                 goto ignore_ipip;
1521                         /* Prefer the resulting PMTU */
1522                         if (dest) {
1523                                 struct ip_vs_dest_dst *dest_dst;
1524
1525                                 rcu_read_lock();
1526                                 dest_dst = rcu_dereference(dest->dest_dst);
1527                                 if (dest_dst)
1528                                         mtu = dst_mtu(dest_dst->dst_cache);
1529                                 rcu_read_unlock();
1530                         }
1531                         if (mtu > 68 + sizeof(struct iphdr))
1532                                 mtu -= sizeof(struct iphdr);
1533                         info = htonl(mtu);
1534                 }
1535                 /* Strip outer IP, ICMP and IPIP, go to IP header of
1536                  * original request.
1537                  */
1538                 if (pskb_pull(skb, offset2) == NULL)
1539                         goto ignore_ipip;
1540                 skb_reset_network_header(skb);
1541                 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1542                         &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1543                         type, code, ntohl(info));
1544                 icmp_send(skb, type, code, info);
1545                 /* ICMP can be shorter but anyways, account it */
1546                 ip_vs_out_stats(cp, skb);
1547
1548 ignore_ipip:
1549                 consume_skb(skb);
1550                 verdict = NF_STOLEN;
1551                 goto out;
1552         }
1553
1554         /* do the statistics and put it back */
1555         ip_vs_in_stats(cp, skb);
1556         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol ||
1557             IPPROTO_SCTP == cih->protocol)
1558                 offset += 2 * sizeof(__u16);
1559         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1560
1561 out:
1562         if (likely(!new_cp))
1563                 __ip_vs_conn_put(cp);
1564         else
1565                 ip_vs_conn_put(cp);
1566
1567         return verdict;
1568 }
1569
1570 #ifdef CONFIG_IP_VS_IPV6
1571 static int ip_vs_in_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
1572                             int *related, unsigned int hooknum,
1573                             struct ip_vs_iphdr *iph)
1574 {
1575         struct icmp6hdr _icmph, *ic;
1576         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1577         struct ip_vs_conn *cp;
1578         struct ip_vs_protocol *pp;
1579         struct ip_vs_proto_data *pd;
1580         unsigned int offset, verdict;
1581         bool new_cp = false;
1582
1583         *related = 1;
1584
1585         ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph, iph);
1586         if (ic == NULL)
1587                 return NF_DROP;
1588
1589         /*
1590          * Work through seeing if this is for us.
1591          * These checks are supposed to be in an order that means easy
1592          * things are checked first to speed up processing.... however
1593          * this means that some packets will manage to get a long way
1594          * down this stack and then be rejected, but that's life.
1595          */
1596         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1597                 *related = 0;
1598                 return NF_ACCEPT;
1599         }
1600         /* Fragment header that is before ICMP header tells us that:
1601          * it's not an error message since they can't be fragmented.
1602          */
1603         if (iph->flags & IP6_FH_F_FRAG)
1604                 return NF_DROP;
1605
1606         IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1607                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1608                   &iph->saddr, &iph->daddr);
1609
1610         offset = iph->len + sizeof(_icmph);
1611         if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, offset, true, &ciph))
1612                 return NF_ACCEPT;
1613
1614         pd = ip_vs_proto_data_get(ipvs, ciph.protocol);
1615         if (!pd)
1616                 return NF_ACCEPT;
1617         pp = pd->pp;
1618
1619         /* Cannot handle fragmented embedded protocol */
1620         if (ciph.fragoffs)
1621                 return NF_ACCEPT;
1622
1623         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
1624                       "Checking incoming ICMPv6 for");
1625
1626         /* The embedded headers contain source and dest in reverse order
1627          * if not from localhost
1628          */
1629         cp = pp->conn_in_get(ipvs, AF_INET6, skb, &ciph);
1630
1631         if (!cp) {
1632                 int v;
1633
1634                 if (!sysctl_schedule_icmp(ipvs))
1635                         return NF_ACCEPT;
1636
1637                 if (!ip_vs_try_to_schedule(ipvs, AF_INET6, skb, pd, &v, &cp, &ciph))
1638                         return v;
1639
1640                 new_cp = true;
1641         }
1642
1643         /* VS/TUN, VS/DR and LOCALNODE just let it go */
1644         if ((hooknum == NF_INET_LOCAL_OUT) &&
1645             (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1646                 verdict = NF_ACCEPT;
1647                 goto out;
1648         }
1649
1650         /* do the statistics and put it back */
1651         ip_vs_in_stats(cp, skb);
1652
1653         /* Need to mangle contained IPv6 header in ICMPv6 packet */
1654         offset = ciph.len;
1655         if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1656             IPPROTO_SCTP == ciph.protocol)
1657                 offset += 2 * sizeof(__u16); /* Also mangle ports */
1658
1659         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset, hooknum, &ciph);
1660
1661 out:
1662         if (likely(!new_cp))
1663                 __ip_vs_conn_put(cp);
1664         else
1665                 ip_vs_conn_put(cp);
1666
1667         return verdict;
1668 }
1669 #endif
1670
1671
1672 /*
1673  *      Check if it's for virtual services, look it up,
1674  *      and send it on its way...
1675  */
1676 static unsigned int
1677 ip_vs_in(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1678 {
1679         struct ip_vs_iphdr iph;
1680         struct ip_vs_protocol *pp;
1681         struct ip_vs_proto_data *pd;
1682         struct ip_vs_conn *cp;
1683         int ret, pkts;
1684         int conn_reuse_mode;
1685         struct sock *sk;
1686
1687         /* Already marked as IPVS request or reply? */
1688         if (skb->ipvs_property)
1689                 return NF_ACCEPT;
1690
1691         /*
1692          *      Big tappo:
1693          *      - remote client: only PACKET_HOST
1694          *      - route: used for struct net when skb->dev is unset
1695          */
1696         if (unlikely((skb->pkt_type != PACKET_HOST &&
1697                       hooknum != NF_INET_LOCAL_OUT) ||
1698                      !skb_dst(skb))) {
1699                 ip_vs_fill_iph_skb(af, skb, false, &iph);
1700                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1701                               " ignored in hook %u\n",
1702                               skb->pkt_type, iph.protocol,
1703                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1704                 return NF_ACCEPT;
1705         }
1706         /* ipvs enabled in this netns ? */
1707         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1708                 return NF_ACCEPT;
1709
1710         ip_vs_fill_iph_skb(af, skb, false, &iph);
1711
1712         /* Bad... Do not break raw sockets */
1713         sk = skb_to_full_sk(skb);
1714         if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
1715                      af == AF_INET)) {
1716
1717                 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
1718                         return NF_ACCEPT;
1719         }
1720
1721 #ifdef CONFIG_IP_VS_IPV6
1722         if (af == AF_INET6) {
1723                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1724                         int related;
1725                         int verdict = ip_vs_in_icmp_v6(ipvs, skb, &related,
1726                                                        hooknum, &iph);
1727
1728                         if (related)
1729                                 return verdict;
1730                 }
1731         } else
1732 #endif
1733                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1734                         int related;
1735                         int verdict = ip_vs_in_icmp(ipvs, skb, &related,
1736                                                     hooknum);
1737
1738                         if (related)
1739                                 return verdict;
1740                 }
1741
1742         /* Protocol supported? */
1743         pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1744         if (unlikely(!pd)) {
1745                 /* The only way we'll see this packet again is if it's
1746                  * encapsulated, so mark it with ipvs_property=1 so we
1747                  * skip it if we're ignoring tunneled packets
1748                  */
1749                 if (sysctl_ignore_tunneled(ipvs))
1750                         skb->ipvs_property = 1;
1751
1752                 return NF_ACCEPT;
1753         }
1754         pp = pd->pp;
1755         /*
1756          * Check if the packet belongs to an existing connection entry
1757          */
1758         cp = pp->conn_in_get(ipvs, af, skb, &iph);
1759
1760         conn_reuse_mode = sysctl_conn_reuse_mode(ipvs);
1761         if (conn_reuse_mode && !iph.fragoffs && is_new_conn(skb, &iph) && cp) {
1762                 bool uses_ct = false, resched = false;
1763
1764                 if (unlikely(sysctl_expire_nodest_conn(ipvs)) && cp->dest &&
1765                     unlikely(!atomic_read(&cp->dest->weight))) {
1766                         resched = true;
1767                         uses_ct = ip_vs_conn_uses_conntrack(cp, skb);
1768                 } else if (is_new_conn_expected(cp, conn_reuse_mode)) {
1769                         uses_ct = ip_vs_conn_uses_conntrack(cp, skb);
1770                         if (!atomic_read(&cp->n_control)) {
1771                                 resched = true;
1772                         } else {
1773                                 /* Do not reschedule controlling connection
1774                                  * that uses conntrack while it is still
1775                                  * referenced by controlled connection(s).
1776                                  */
1777                                 resched = !uses_ct;
1778                         }
1779                 }
1780
1781                 if (resched) {
1782                         if (!atomic_read(&cp->n_control))
1783                                 ip_vs_conn_expire_now(cp);
1784                         __ip_vs_conn_put(cp);
1785                         if (uses_ct)
1786                                 return NF_DROP;
1787                         cp = NULL;
1788                 }
1789         }
1790
1791         if (unlikely(!cp)) {
1792                 int v;
1793
1794                 if (!ip_vs_try_to_schedule(ipvs, af, skb, pd, &v, &cp, &iph))
1795                         return v;
1796         }
1797
1798         IP_VS_DBG_PKT(11, af, pp, skb, iph.off, "Incoming packet");
1799
1800         /* Check the server status */
1801         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1802                 /* the destination server is not available */
1803
1804                 if (sysctl_expire_nodest_conn(ipvs)) {
1805                         /* try to expire the connection immediately */
1806                         ip_vs_conn_expire_now(cp);
1807                 }
1808                 /* don't restart its timer, and silently
1809                    drop the packet. */
1810                 __ip_vs_conn_put(cp);
1811                 return NF_DROP;
1812         }
1813
1814         ip_vs_in_stats(cp, skb);
1815         ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1816         if (cp->packet_xmit)
1817                 ret = cp->packet_xmit(skb, cp, pp, &iph);
1818                 /* do not touch skb anymore */
1819         else {
1820                 IP_VS_DBG_RL("warning: packet_xmit is null");
1821                 ret = NF_ACCEPT;
1822         }
1823
1824         /* Increase its packet counter and check if it is needed
1825          * to be synchronized
1826          *
1827          * Sync connection if it is about to close to
1828          * encorage the standby servers to update the connections timeout
1829          *
1830          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1831          */
1832
1833         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
1834                 pkts = sysctl_sync_threshold(ipvs);
1835         else
1836                 pkts = atomic_add_return(1, &cp->in_pkts);
1837
1838         if (ipvs->sync_state & IP_VS_STATE_MASTER)
1839                 ip_vs_sync_conn(ipvs, cp, pkts);
1840
1841         ip_vs_conn_put(cp);
1842         return ret;
1843 }
1844
1845 /*
1846  *      AF_INET handler in NF_INET_LOCAL_IN chain
1847  *      Schedule and forward packets from remote clients
1848  */
1849 static unsigned int
1850 ip_vs_remote_request4(void *priv, struct sk_buff *skb,
1851                       const struct nf_hook_state *state)
1852 {
1853         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
1854 }
1855
1856 /*
1857  *      AF_INET handler in NF_INET_LOCAL_OUT chain
1858  *      Schedule and forward packets from local clients
1859  */
1860 static unsigned int
1861 ip_vs_local_request4(void *priv, struct sk_buff *skb,
1862                      const struct nf_hook_state *state)
1863 {
1864         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
1865 }
1866
1867 #ifdef CONFIG_IP_VS_IPV6
1868
1869 /*
1870  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
1871  *      Schedule and forward packets from remote clients
1872  */
1873 static unsigned int
1874 ip_vs_remote_request6(void *priv, struct sk_buff *skb,
1875                       const struct nf_hook_state *state)
1876 {
1877         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
1878 }
1879
1880 /*
1881  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
1882  *      Schedule and forward packets from local clients
1883  */
1884 static unsigned int
1885 ip_vs_local_request6(void *priv, struct sk_buff *skb,
1886                      const struct nf_hook_state *state)
1887 {
1888         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
1889 }
1890
1891 #endif
1892
1893
1894 /*
1895  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
1896  *      related packets destined for 0.0.0.0/0.
1897  *      When fwmark-based virtual service is used, such as transparent
1898  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
1899  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
1900  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
1901  *      and send them to ip_vs_in_icmp.
1902  */
1903 static unsigned int
1904 ip_vs_forward_icmp(void *priv, struct sk_buff *skb,
1905                    const struct nf_hook_state *state)
1906 {
1907         int r;
1908         struct netns_ipvs *ipvs = net_ipvs(state->net);
1909
1910         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
1911                 return NF_ACCEPT;
1912
1913         /* ipvs enabled in this netns ? */
1914         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1915                 return NF_ACCEPT;
1916
1917         return ip_vs_in_icmp(ipvs, skb, &r, state->hook);
1918 }
1919
1920 #ifdef CONFIG_IP_VS_IPV6
1921 static unsigned int
1922 ip_vs_forward_icmp_v6(void *priv, struct sk_buff *skb,
1923                       const struct nf_hook_state *state)
1924 {
1925         int r;
1926         struct netns_ipvs *ipvs = net_ipvs(state->net);
1927         struct ip_vs_iphdr iphdr;
1928
1929         ip_vs_fill_iph_skb(AF_INET6, skb, false, &iphdr);
1930         if (iphdr.protocol != IPPROTO_ICMPV6)
1931                 return NF_ACCEPT;
1932
1933         /* ipvs enabled in this netns ? */
1934         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1935                 return NF_ACCEPT;
1936
1937         return ip_vs_in_icmp_v6(ipvs, skb, &r, state->hook, &iphdr);
1938 }
1939 #endif
1940
1941
1942 static struct nf_hook_ops ip_vs_ops[] __read_mostly = {
1943         /* After packet filtering, change source only for VS/NAT */
1944         {
1945                 .hook           = ip_vs_reply4,
1946                 .pf             = NFPROTO_IPV4,
1947                 .hooknum        = NF_INET_LOCAL_IN,
1948                 .priority       = NF_IP_PRI_NAT_SRC - 2,
1949         },
1950         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1951          * or VS/NAT(change destination), so that filtering rules can be
1952          * applied to IPVS. */
1953         {
1954                 .hook           = ip_vs_remote_request4,
1955                 .pf             = NFPROTO_IPV4,
1956                 .hooknum        = NF_INET_LOCAL_IN,
1957                 .priority       = NF_IP_PRI_NAT_SRC - 1,
1958         },
1959         /* Before ip_vs_in, change source only for VS/NAT */
1960         {
1961                 .hook           = ip_vs_local_reply4,
1962                 .pf             = NFPROTO_IPV4,
1963                 .hooknum        = NF_INET_LOCAL_OUT,
1964                 .priority       = NF_IP_PRI_NAT_DST + 1,
1965         },
1966         /* After mangle, schedule and forward local requests */
1967         {
1968                 .hook           = ip_vs_local_request4,
1969                 .pf             = NFPROTO_IPV4,
1970                 .hooknum        = NF_INET_LOCAL_OUT,
1971                 .priority       = NF_IP_PRI_NAT_DST + 2,
1972         },
1973         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
1974          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
1975         {
1976                 .hook           = ip_vs_forward_icmp,
1977                 .pf             = NFPROTO_IPV4,
1978                 .hooknum        = NF_INET_FORWARD,
1979                 .priority       = 99,
1980         },
1981         /* After packet filtering, change source only for VS/NAT */
1982         {
1983                 .hook           = ip_vs_reply4,
1984                 .pf             = NFPROTO_IPV4,
1985                 .hooknum        = NF_INET_FORWARD,
1986                 .priority       = 100,
1987         },
1988 #ifdef CONFIG_IP_VS_IPV6
1989         /* After packet filtering, change source only for VS/NAT */
1990         {
1991                 .hook           = ip_vs_reply6,
1992                 .pf             = NFPROTO_IPV6,
1993                 .hooknum        = NF_INET_LOCAL_IN,
1994                 .priority       = NF_IP6_PRI_NAT_SRC - 2,
1995         },
1996         /* After packet filtering, forward packet through VS/DR, VS/TUN,
1997          * or VS/NAT(change destination), so that filtering rules can be
1998          * applied to IPVS. */
1999         {
2000                 .hook           = ip_vs_remote_request6,
2001                 .pf             = NFPROTO_IPV6,
2002                 .hooknum        = NF_INET_LOCAL_IN,
2003                 .priority       = NF_IP6_PRI_NAT_SRC - 1,
2004         },
2005         /* Before ip_vs_in, change source only for VS/NAT */
2006         {
2007                 .hook           = ip_vs_local_reply6,
2008                 .pf             = NFPROTO_IPV6,
2009                 .hooknum        = NF_INET_LOCAL_OUT,
2010                 .priority       = NF_IP6_PRI_NAT_DST + 1,
2011         },
2012         /* After mangle, schedule and forward local requests */
2013         {
2014                 .hook           = ip_vs_local_request6,
2015                 .pf             = NFPROTO_IPV6,
2016                 .hooknum        = NF_INET_LOCAL_OUT,
2017                 .priority       = NF_IP6_PRI_NAT_DST + 2,
2018         },
2019         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2020          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2021         {
2022                 .hook           = ip_vs_forward_icmp_v6,
2023                 .pf             = NFPROTO_IPV6,
2024                 .hooknum        = NF_INET_FORWARD,
2025                 .priority       = 99,
2026         },
2027         /* After packet filtering, change source only for VS/NAT */
2028         {
2029                 .hook           = ip_vs_reply6,
2030                 .pf             = NFPROTO_IPV6,
2031                 .hooknum        = NF_INET_FORWARD,
2032                 .priority       = 100,
2033         },
2034 #endif
2035 };
2036 /*
2037  *      Initialize IP Virtual Server netns mem.
2038  */
2039 static int __net_init __ip_vs_init(struct net *net)
2040 {
2041         struct netns_ipvs *ipvs;
2042
2043         ipvs = net_generic(net, ip_vs_net_id);
2044         if (ipvs == NULL)
2045                 return -ENOMEM;
2046
2047         /* Hold the beast until a service is registerd */
2048         ipvs->enable = 0;
2049         ipvs->net = net;
2050         /* Counters used for creating unique names */
2051         ipvs->gen = atomic_read(&ipvs_netns_cnt);
2052         atomic_inc(&ipvs_netns_cnt);
2053         net->ipvs = ipvs;
2054
2055         if (ip_vs_estimator_net_init(ipvs) < 0)
2056                 goto estimator_fail;
2057
2058         if (ip_vs_control_net_init(ipvs) < 0)
2059                 goto control_fail;
2060
2061         if (ip_vs_protocol_net_init(ipvs) < 0)
2062                 goto protocol_fail;
2063
2064         if (ip_vs_app_net_init(ipvs) < 0)
2065                 goto app_fail;
2066
2067         if (ip_vs_conn_net_init(ipvs) < 0)
2068                 goto conn_fail;
2069
2070         if (ip_vs_sync_net_init(ipvs) < 0)
2071                 goto sync_fail;
2072
2073         printk(KERN_INFO "IPVS: Creating netns size=%zu id=%d\n",
2074                          sizeof(struct netns_ipvs), ipvs->gen);
2075         return 0;
2076 /*
2077  * Error handling
2078  */
2079
2080 sync_fail:
2081         ip_vs_conn_net_cleanup(ipvs);
2082 conn_fail:
2083         ip_vs_app_net_cleanup(ipvs);
2084 app_fail:
2085         ip_vs_protocol_net_cleanup(ipvs);
2086 protocol_fail:
2087         ip_vs_control_net_cleanup(ipvs);
2088 control_fail:
2089         ip_vs_estimator_net_cleanup(ipvs);
2090 estimator_fail:
2091         net->ipvs = NULL;
2092         return -ENOMEM;
2093 }
2094
2095 static void __net_exit __ip_vs_cleanup(struct net *net)
2096 {
2097         struct netns_ipvs *ipvs = net_ipvs(net);
2098
2099         ip_vs_service_net_cleanup(ipvs);        /* ip_vs_flush() with locks */
2100         ip_vs_conn_net_cleanup(ipvs);
2101         ip_vs_app_net_cleanup(ipvs);
2102         ip_vs_protocol_net_cleanup(ipvs);
2103         ip_vs_control_net_cleanup(ipvs);
2104         ip_vs_estimator_net_cleanup(ipvs);
2105         IP_VS_DBG(2, "ipvs netns %d released\n", ipvs->gen);
2106         net->ipvs = NULL;
2107 }
2108
2109 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
2110 {
2111         struct netns_ipvs *ipvs = net_ipvs(net);
2112         EnterFunction(2);
2113         ipvs->enable = 0;       /* Disable packet reception */
2114         smp_wmb();
2115         ip_vs_sync_net_cleanup(ipvs);
2116         LeaveFunction(2);
2117 }
2118
2119 static struct pernet_operations ipvs_core_ops = {
2120         .init = __ip_vs_init,
2121         .exit = __ip_vs_cleanup,
2122         .id   = &ip_vs_net_id,
2123         .size = sizeof(struct netns_ipvs),
2124 };
2125
2126 static struct pernet_operations ipvs_core_dev_ops = {
2127         .exit = __ip_vs_dev_cleanup,
2128 };
2129
2130 /*
2131  *      Initialize IP Virtual Server
2132  */
2133 static int __init ip_vs_init(void)
2134 {
2135         int ret;
2136
2137         ret = ip_vs_control_init();
2138         if (ret < 0) {
2139                 pr_err("can't setup control.\n");
2140                 goto exit;
2141         }
2142
2143         ip_vs_protocol_init();
2144
2145         ret = ip_vs_conn_init();
2146         if (ret < 0) {
2147                 pr_err("can't setup connection table.\n");
2148                 goto cleanup_protocol;
2149         }
2150
2151         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
2152         if (ret < 0)
2153                 goto cleanup_conn;
2154
2155         ret = register_pernet_device(&ipvs_core_dev_ops);
2156         if (ret < 0)
2157                 goto cleanup_sub;
2158
2159         ret = nf_register_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2160         if (ret < 0) {
2161                 pr_err("can't register hooks.\n");
2162                 goto cleanup_dev;
2163         }
2164
2165         ret = ip_vs_register_nl_ioctl();
2166         if (ret < 0) {
2167                 pr_err("can't register netlink/ioctl.\n");
2168                 goto cleanup_hooks;
2169         }
2170
2171         pr_info("ipvs loaded.\n");
2172
2173         return ret;
2174
2175 cleanup_hooks:
2176         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2177 cleanup_dev:
2178         unregister_pernet_device(&ipvs_core_dev_ops);
2179 cleanup_sub:
2180         unregister_pernet_subsys(&ipvs_core_ops);
2181 cleanup_conn:
2182         ip_vs_conn_cleanup();
2183 cleanup_protocol:
2184         ip_vs_protocol_cleanup();
2185         ip_vs_control_cleanup();
2186 exit:
2187         return ret;
2188 }
2189
2190 static void __exit ip_vs_cleanup(void)
2191 {
2192         ip_vs_unregister_nl_ioctl();
2193         nf_unregister_hooks(ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2194         unregister_pernet_device(&ipvs_core_dev_ops);
2195         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
2196         ip_vs_conn_cleanup();
2197         ip_vs_protocol_cleanup();
2198         ip_vs_control_cleanup();
2199         pr_info("ipvs unloaded.\n");
2200 }
2201
2202 module_init(ip_vs_init);
2203 module_exit(ip_vs_cleanup);
2204 MODULE_LICENSE("GPL");