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 unsigned 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                 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
 135                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 136                 u64_stats_update_begin(&s->syncp);
 137                 s->cnt.inpkts++;
 138                 s->cnt.inbytes += skb->len;
 139                 u64_stats_update_end(&s->syncp);
 140         }
 141 }
 142
 143
 144 static inline void
 145 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 146 {
 147         struct ip_vs_dest *dest = cp->dest;
 148         struct netns_ipvs *ipvs = cp->ipvs;
 149
 150         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 151                 struct ip_vs_cpu_stats *s;
 152                 struct ip_vs_service *svc;
 153
 154                 s = this_cpu_ptr(dest->stats.cpustats);
 155                 u64_stats_update_begin(&s->syncp);
 156                 s->cnt.outpkts++;
 157                 s->cnt.outbytes += skb->len;
 158                 u64_stats_update_end(&s->syncp);
 159
 160                 svc = rcu_dereference(dest->svc);
 161                 s = this_cpu_ptr(svc->stats.cpustats);
 162                 u64_stats_update_begin(&s->syncp);
 163                 s->cnt.outpkts++;
 164                 s->cnt.outbytes += skb->len;
 165                 u64_stats_update_end(&s->syncp);
 166
 167                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 168                 u64_stats_update_begin(&s->syncp);
 169                 s->cnt.outpkts++;
 170                 s->cnt.outbytes += skb->len;
 171                 u64_stats_update_end(&s->syncp);
 172         }
 173 }
 174
 175
 176 static inline void
 177 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
 178 {
 179         struct netns_ipvs *ipvs = svc->ipvs;
 180         struct ip_vs_cpu_stats *s;
 181
 182         s = this_cpu_ptr(cp->dest->stats.cpustats);
 183         u64_stats_update_begin(&s->syncp);
 184         s->cnt.conns++;
 185         u64_stats_update_end(&s->syncp);
 186
 187         s = this_cpu_ptr(svc->stats.cpustats);
 188         u64_stats_update_begin(&s->syncp);
 189         s->cnt.conns++;
 190         u64_stats_update_end(&s->syncp);
 191
 192         s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 193         u64_stats_update_begin(&s->syncp);
 194         s->cnt.conns++;
 195         u64_stats_update_end(&s->syncp);
 196 }
 197
 198
 199 static inline void
 200 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
 201                 const struct sk_buff *skb,
 202                 struct ip_vs_proto_data *pd)
 203 {
 204         if (likely(pd->pp->state_transition))
 205                 pd->pp->state_transition(cp, direction, skb, pd);
 206 }
 207
 208 static inline int
 209 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
 210                               struct sk_buff *skb, int protocol,
 211                               const union nf_inet_addr *caddr, __be16 cport,
 212                               const union nf_inet_addr *vaddr, __be16 vport,
 213                               struct ip_vs_conn_param *p)
 214 {
 215         ip_vs_conn_fill_param(svc->ipvs, svc->af, protocol, caddr, cport, vaddr,
 216                               vport, p);
 217         p->pe = rcu_dereference(svc->pe);
 218         if (p->pe && p->pe->fill_param)
 219                 return p->pe->fill_param(p, skb);
 220
 221         return 0;
 222 }
 223
 224 /*
 225  *  IPVS persistent scheduling function
 226  *  It creates a connection entry according to its template if exists,
 227  *  or selects a server and creates a connection entry plus a template.
 228  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
 229  *  Protocols supported: TCP, UDP
 230  */
 231 static struct ip_vs_conn *
 232 ip_vs_sched_persist(struct ip_vs_service *svc,
 233                     struct sk_buff *skb, __be16 src_port, __be16 dst_port,
 234                     int *ignored, struct ip_vs_iphdr *iph)
 235 {
 236         struct ip_vs_conn *cp = NULL;
 237         struct ip_vs_dest *dest;
 238         struct ip_vs_conn *ct;
 239         __be16 dport = 0;               /* destination port to forward */
 240         unsigned int flags;
 241         struct ip_vs_conn_param param;
 242         const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
 243         union nf_inet_addr snet;        /* source network of the client,
 244                                            after masking */
 245         const union nf_inet_addr *src_addr, *dst_addr;
 246
 247         if (likely(!ip_vs_iph_inverse(iph))) {
 248                 src_addr = &iph->saddr;
 249                 dst_addr = &iph->daddr;
 250         } else {
 251                 src_addr = &iph->daddr;
 252                 dst_addr = &iph->saddr;
 253         }
 254
 255
 256         /* Mask saddr with the netmask to adjust template granularity */
 257 #ifdef CONFIG_IP_VS_IPV6
 258         if (svc->af == AF_INET6)
 259                 ipv6_addr_prefix(&snet.in6, &src_addr->in6,
 260                                  (__force __u32) svc->netmask);
 261         else
 262 #endif
 263                 snet.ip = src_addr->ip & svc->netmask;
 264
 265         IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
 266                       "mnet %s\n",
 267                       IP_VS_DBG_ADDR(svc->af, src_addr), ntohs(src_port),
 268                       IP_VS_DBG_ADDR(svc->af, dst_addr), ntohs(dst_port),
 269                       IP_VS_DBG_ADDR(svc->af, &snet));
 270
 271         /*
 272          * As far as we know, FTP is a very complicated network protocol, and
 273          * it uses control connection and data connections. For active FTP,
 274          * FTP server initialize data connection to the client, its source port
 275          * is often 20. For passive FTP, FTP server tells the clients the port
 276          * that it passively listens to,  and the client issues the data
 277          * connection. In the tunneling or direct routing mode, the load
 278          * balancer is on the client-to-server half of connection, the port
 279          * number is unknown to the load balancer. So, a conn template like
 280          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
 281          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
 282          * is created for other persistent services.
 283          */
 284         {
 285                 int protocol = iph->protocol;
 286                 const union nf_inet_addr *vaddr = dst_addr;
 287                 __be16 vport = 0;
 288
 289                 if (dst_port == svc->port) {
 290                         /* non-FTP template:
 291                          * <protocol, caddr, 0, vaddr, vport, daddr, dport>
 292                          * FTP template:
 293                          * <protocol, caddr, 0, vaddr, 0, daddr, 0>
 294                          */
 295                         if (svc->port != FTPPORT)
 296                                 vport = dst_port;
 297                 } else {
 298                         /* Note: persistent fwmark-based services and
 299                          * persistent port zero service are handled here.
 300                          * fwmark template:
 301                          * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
 302                          * port zero template:
 303                          * <protocol,caddr,0,vaddr,0,daddr,0>
 304                          */
 305                         if (svc->fwmark) {
 306                                 protocol = IPPROTO_IP;
 307                                 vaddr = &fwmark;
 308                         }
 309                 }
 310                 /* return *ignored = -1 so NF_DROP can be used */
 311                 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
 312                                                   vaddr, vport, &param) < 0) {
 313                         *ignored = -1;
 314                         return NULL;
 315                 }
 316         }
 317
 318         /* Check if a template already exists */
 319         ct = ip_vs_ct_in_get(&param);
 320         if (!ct || !ip_vs_check_template(ct, NULL)) {
 321                 struct ip_vs_scheduler *sched;
 322
 323                 /*
 324                  * No template found or the dest of the connection
 325                  * template is not available.
 326                  * return *ignored=0 i.e. ICMP and NF_DROP
 327                  */
 328                 sched = rcu_dereference(svc->scheduler);
 329                 if (sched) {
 330                         /* read svc->sched_data after svc->scheduler */
 331                         smp_rmb();
 332                         dest = sched->schedule(svc, skb, iph);
 333                 } else {
 334                         dest = NULL;
 335                 }
 336                 if (!dest) {
 337                         IP_VS_DBG(1, "p-schedule: no dest found.\n");
 338                         kfree(param.pe_data);
 339                         *ignored = 0;
 340                         return NULL;
 341                 }
 342
 343                 if (dst_port == svc->port && svc->port != FTPPORT)
 344                         dport = dest->port;
 345
 346                 /* Create a template
 347                  * This adds param.pe_data to the template,
 348                  * and thus param.pe_data will be destroyed
 349                  * when the template expires */
 350                 ct = ip_vs_conn_new(&param, dest->af, &dest->addr, dport,
 351                                     IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
 352                 if (ct == NULL) {
 353                         kfree(param.pe_data);
 354                         *ignored = -1;
 355                         return NULL;
 356                 }
 357
 358                 ct->timeout = svc->timeout;
 359         } else {
 360                 /* set destination with the found template */
 361                 dest = ct->dest;
 362                 kfree(param.pe_data);
 363         }
 364
 365         dport = dst_port;
 366         if (dport == svc->port && dest->port)
 367                 dport = dest->port;
 368
 369         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 370                  && iph->protocol == IPPROTO_UDP) ?
 371                 IP_VS_CONN_F_ONE_PACKET : 0;
 372
 373         /*
 374          *    Create a new connection according to the template
 375          */
 376         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol, src_addr,
 377                               src_port, dst_addr, dst_port, &param);
 378
 379         cp = ip_vs_conn_new(&param, dest->af, &dest->addr, dport, flags, dest,
 380                             skb->mark);
 381         if (cp == NULL) {
 382                 ip_vs_conn_put(ct);
 383                 *ignored = -1;
 384                 return NULL;
 385         }
 386
 387         /*
 388          *    Add its control
 389          */
 390         ip_vs_control_add(cp, ct);
 391         ip_vs_conn_put(ct);
 392
 393         ip_vs_conn_stats(cp, svc);
 394         return cp;
 395 }
 396
 397
 398 /*
 399  *  IPVS main scheduling function
 400  *  It selects a server according to the virtual service, and
 401  *  creates a connection entry.
 402  *  Protocols supported: TCP, UDP
 403  *
 404  *  Usage of *ignored
 405  *
 406  * 1 :   protocol tried to schedule (eg. on SYN), found svc but the
 407  *       svc/scheduler decides that this packet should be accepted with
 408  *       NF_ACCEPT because it must not be scheduled.
 409  *
 410  * 0 :   scheduler can not find destination, so try bypass or
 411  *       return ICMP and then NF_DROP (ip_vs_leave).
 412  *
 413  * -1 :  scheduler tried to schedule but fatal error occurred, eg.
 414  *       ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
 415  *       failure such as missing Call-ID, ENOMEM on skb_linearize
 416  *       or pe_data. In this case we should return NF_DROP without
 417  *       any attempts to send ICMP with ip_vs_leave.
 418  */
 419 struct ip_vs_conn *
 420 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
 421                struct ip_vs_proto_data *pd, int *ignored,
 422                struct ip_vs_iphdr *iph)
 423 {
 424         struct ip_vs_protocol *pp = pd->pp;
 425         struct ip_vs_conn *cp = NULL;
 426         struct ip_vs_scheduler *sched;
 427         struct ip_vs_dest *dest;
 428         __be16 _ports[2], *pptr, cport, vport;
 429         const void *caddr, *vaddr;
 430         unsigned int flags;
 431
 432         *ignored = 1;
 433         /*
 434          * IPv6 frags, only the first hit here.
 435          */
 436         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 437         if (pptr == NULL)
 438                 return NULL;
 439
 440         if (likely(!ip_vs_iph_inverse(iph))) {
 441                 cport = pptr[0];
 442                 caddr = &iph->saddr;
 443                 vport = pptr[1];
 444                 vaddr = &iph->daddr;
 445         } else {
 446                 cport = pptr[1];
 447                 caddr = &iph->daddr;
 448                 vport = pptr[0];
 449                 vaddr = &iph->saddr;
 450         }
 451
 452         /*
 453          * FTPDATA needs this check when using local real server.
 454          * Never schedule Active FTPDATA connections from real server.
 455          * For LVS-NAT they must be already created. For other methods
 456          * with persistence the connection is created on SYN+ACK.
 457          */
 458         if (cport == FTPDATA) {
 459                 IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
 460                               "Not scheduling FTPDATA");
 461                 return NULL;
 462         }
 463
 464         /*
 465          *    Do not schedule replies from local real server.
 466          */
 467         if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK)) {
 468                 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
 469                 cp = pp->conn_in_get(svc->ipvs, svc->af, skb, iph);
 470                 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
 471
 472                 if (cp) {
 473                         IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
 474                                       "Not scheduling reply for existing"
 475                                       " connection");
 476                         __ip_vs_conn_put(cp);
 477                         return NULL;
 478                 }
 479         }
 480
 481         /*
 482          *    Persistent service
 483          */
 484         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 485                 return ip_vs_sched_persist(svc, skb, cport, vport, ignored,
 486                                            iph);
 487
 488         *ignored = 0;
 489
 490         /*
 491          *    Non-persistent service
 492          */
 493         if (!svc->fwmark && vport != svc->port) {
 494                 if (!svc->port)
 495                         pr_err("Schedule: port zero only supported "
 496                                "in persistent services, "
 497                                "check your ipvs configuration\n");
 498                 return NULL;
 499         }
 500
 501         sched = rcu_dereference(svc->scheduler);
 502         if (sched) {
 503                 /* read svc->sched_data after svc->scheduler */
 504                 smp_rmb();
 505                 dest = sched->schedule(svc, skb, iph);
 506         } else {
 507                 dest = NULL;
 508         }
 509         if (dest == NULL) {
 510                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 511                 return NULL;
 512         }
 513
 514         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 515                  && iph->protocol == IPPROTO_UDP) ?
 516                 IP_VS_CONN_F_ONE_PACKET : 0;
 517
 518         /*
 519          *    Create a connection entry.
 520          */
 521         {
 522                 struct ip_vs_conn_param p;
 523
 524                 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
 525                                       caddr, cport, vaddr, vport, &p);
 526                 cp = ip_vs_conn_new(&p, dest->af, &dest->addr,
 527                                     dest->port ? dest->port : vport,
 528                                     flags, dest, skb->mark);
 529                 if (!cp) {
 530                         *ignored = -1;
 531                         return NULL;
 532                 }
 533         }
 534
 535         IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
 536                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
 537                       ip_vs_fwd_tag(cp),
 538                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 539                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 540                       IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
 541                       cp->flags, refcount_read(&cp->refcnt));
 542
 543         ip_vs_conn_stats(cp, svc);
 544         return cp;
 545 }
 546
 547 static inline int ip_vs_addr_is_unicast(struct net *net, int af,
 548                                         union nf_inet_addr *addr)
 549 {
 550 #ifdef CONFIG_IP_VS_IPV6
 551         if (af == AF_INET6)
 552                 return ipv6_addr_type(&addr->in6) & IPV6_ADDR_UNICAST;
 553 #endif
 554         return (inet_addr_type(net, addr->ip) == RTN_UNICAST);
 555 }
 556
 557 /*
 558  *  Pass or drop the packet.
 559  *  Called by ip_vs_in, when the virtual service is available but
 560  *  no destination is available for a new connection.
 561  */
 562 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 563                 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
 564 {
 565         __be16 _ports[2], *pptr, dport;
 566         struct netns_ipvs *ipvs = svc->ipvs;
 567         struct net *net = ipvs->net;
 568
 569         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports, iph);
 570         if (!pptr)
 571                 return NF_DROP;
 572         dport = likely(!ip_vs_iph_inverse(iph)) ? pptr[1] : pptr[0];
 573
 574         /* if it is fwmark-based service, the cache_bypass sysctl is up
 575            and the destination is a non-local unicast, then create
 576            a cache_bypass connection entry */
 577         if (sysctl_cache_bypass(ipvs) && svc->fwmark &&
 578             !(iph->hdr_flags & (IP_VS_HDR_INVERSE | IP_VS_HDR_ICMP)) &&
 579             ip_vs_addr_is_unicast(net, svc->af, &iph->daddr)) {
 580                 int ret;
 581                 struct ip_vs_conn *cp;
 582                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
 583                                       iph->protocol == IPPROTO_UDP) ?
 584                                       IP_VS_CONN_F_ONE_PACKET : 0;
 585                 union nf_inet_addr daddr =  { .all = { 0, 0, 0, 0 } };
 586
 587                 /* create a new connection entry */
 588                 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
 589                 {
 590                         struct ip_vs_conn_param p;
 591                         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
 592                                               &iph->saddr, pptr[0],
 593                                               &iph->daddr, pptr[1], &p);
 594                         cp = ip_vs_conn_new(&p, svc->af, &daddr, 0,
 595                                             IP_VS_CONN_F_BYPASS | flags,
 596                                             NULL, skb->mark);
 597                         if (!cp)
 598                                 return NF_DROP;
 599                 }
 600
 601                 /* statistics */
 602                 ip_vs_in_stats(cp, skb);
 603
 604                 /* set state */
 605                 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
 606
 607                 /* transmit the first SYN packet */
 608                 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
 609                 /* do not touch skb anymore */
 610
 611                 if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) && cp->control)
 612                         atomic_inc(&cp->control->in_pkts);
 613                 else
 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) != IP_VS_CONN_F_MASQ)
 849                 goto ignore_cp;
 850
 851         /* Ensure the checksum is correct */
 852         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 853                 /* Failed checksum! */
 854                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
 855                               IP_VS_DBG_ADDR(af, snet));
 856                 goto out;
 857         }
 858
 859         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 860             IPPROTO_SCTP == protocol)
 861                 offset += 2 * sizeof(__u16);
 862         if (!skb_make_writable(skb, offset))
 863                 goto out;
 864
 865 #ifdef CONFIG_IP_VS_IPV6
 866         if (af == AF_INET6)
 867                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
 868         else
 869 #endif
 870                 ip_vs_nat_icmp(skb, pp, cp, 1);
 871
 872         if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
 873                 goto out;
 874
 875         /* do the statistics and put it back */
 876         ip_vs_out_stats(cp, skb);
 877
 878         skb->ipvs_property = 1;
 879         if (!(cp->flags & IP_VS_CONN_F_NFCT))
 880                 ip_vs_notrack(skb);
 881         else
 882                 ip_vs_update_conntrack(skb, cp, 0);
 883
 884 ignore_cp:
 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         struct sctp_chunkhdr *sch, schunk;
1037         sch = skb_header_pointer(skb, nh_len + sizeof(struct sctphdr),
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                 struct sctp_chunkhdr *sch, schunk;
1070
1071                 sch = skb_header_pointer(skb, iph->len + sizeof(struct sctphdr),
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 /* Generic function to create new connections for outgoing RS packets
1104  *
1105  * Pre-requisites for successful connection creation:
1106  * 1) Virtual Service is NOT fwmark based:
1107  *    In fwmark-VS actual vaddr and vport are unknown to IPVS
1108  * 2) Real Server and Virtual Service were NOT configured without port:
1109  *    This is to allow match of different VS to the same RS ip-addr
1110  */
1111 struct ip_vs_conn *ip_vs_new_conn_out(struct ip_vs_service *svc,
1112                                       struct ip_vs_dest *dest,
1113                                       struct sk_buff *skb,
1114                                       const struct ip_vs_iphdr *iph,
1115                                       __be16 dport,
1116                                       __be16 cport)
1117 {
1118         struct ip_vs_conn_param param;
1119         struct ip_vs_conn *ct = NULL, *cp = NULL;
1120         const union nf_inet_addr *vaddr, *daddr, *caddr;
1121         union nf_inet_addr snet;
1122         __be16 vport;
1123         unsigned int flags;
1124
1125         EnterFunction(12);
1126         vaddr = &svc->addr;
1127         vport = svc->port;
1128         daddr = &iph->saddr;
1129         caddr = &iph->daddr;
1130
1131         /* check pre-requisites are satisfied */
1132         if (svc->fwmark)
1133                 return NULL;
1134         if (!vport || !dport)
1135                 return NULL;
1136
1137         /* for persistent service first create connection template */
1138         if (svc->flags & IP_VS_SVC_F_PERSISTENT) {
1139                 /* apply netmask the same way ingress-side does */
1140 #ifdef CONFIG_IP_VS_IPV6
1141                 if (svc->af == AF_INET6)
1142                         ipv6_addr_prefix(&snet.in6, &caddr->in6,
1143                                          (__force __u32)svc->netmask);
1144                 else
1145 #endif
1146                         snet.ip = caddr->ip & svc->netmask;
1147                 /* fill params and create template if not existent */
1148                 if (ip_vs_conn_fill_param_persist(svc, skb, iph->protocol,
1149                                                   &snet, 0, vaddr,
1150                                                   vport, &param) < 0)
1151                         return NULL;
1152                 ct = ip_vs_ct_in_get(&param);
1153                 /* check if template exists and points to the same dest */
1154                 if (!ct || !ip_vs_check_template(ct, dest)) {
1155                         ct = ip_vs_conn_new(&param, dest->af, daddr, dport,
1156                                             IP_VS_CONN_F_TEMPLATE, dest, 0);
1157                         if (!ct) {
1158                                 kfree(param.pe_data);
1159                                 return NULL;
1160                         }
1161                         ct->timeout = svc->timeout;
1162                 } else {
1163                         kfree(param.pe_data);
1164                 }
1165         }
1166
1167         /* connection flags */
1168         flags = ((svc->flags & IP_VS_SVC_F_ONEPACKET) &&
1169                  iph->protocol == IPPROTO_UDP) ? IP_VS_CONN_F_ONE_PACKET : 0;
1170         /* create connection */
1171         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
1172                               caddr, cport, vaddr, vport, &param);
1173         cp = ip_vs_conn_new(&param, dest->af, daddr, dport, flags, dest, 0);
1174         if (!cp) {
1175                 if (ct)
1176                         ip_vs_conn_put(ct);
1177                 return NULL;
1178         }
1179         if (ct) {
1180                 ip_vs_control_add(cp, ct);
1181                 ip_vs_conn_put(ct);
1182         }
1183         ip_vs_conn_stats(cp, svc);
1184
1185         /* return connection (will be used to handle outgoing packet) */
1186         IP_VS_DBG_BUF(6, "New connection RS-initiated:%c c:%s:%u v:%s:%u "
1187                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
1188                       ip_vs_fwd_tag(cp),
1189                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
1190                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
1191                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
1192                       cp->flags, refcount_read(&cp->refcnt));
1193         LeaveFunction(12);
1194         return cp;
1195 }
1196
1197 /* Handle outgoing packets which are considered requests initiated by
1198  * real servers, so that subsequent responses from external client can be
1199  * routed to the right real server.
1200  * Used also for outgoing responses in OPS mode.
1201  *
1202  * Connection management is handled by persistent-engine specific callback.
1203  */
1204 static struct ip_vs_conn *__ip_vs_rs_conn_out(unsigned int hooknum,
1205                                               struct netns_ipvs *ipvs,
1206                                               int af, struct sk_buff *skb,
1207                                               const struct ip_vs_iphdr *iph)
1208 {
1209         struct ip_vs_dest *dest;
1210         struct ip_vs_conn *cp = NULL;
1211         __be16 _ports[2], *pptr;
1212
1213         if (hooknum == NF_INET_LOCAL_IN)
1214                 return NULL;
1215
1216         pptr = frag_safe_skb_hp(skb, iph->len,
1217                                 sizeof(_ports), _ports, iph);
1218         if (!pptr)
1219                 return NULL;
1220
1221         dest = ip_vs_find_real_service(ipvs, af, iph->protocol,
1222                                        &iph->saddr, pptr[0]);
1223         if (dest) {
1224                 struct ip_vs_service *svc;
1225                 struct ip_vs_pe *pe;
1226
1227                 svc = rcu_dereference(dest->svc);
1228                 if (svc) {
1229                         pe = rcu_dereference(svc->pe);
1230                         if (pe && pe->conn_out)
1231                                 cp = pe->conn_out(svc, dest, skb, iph,
1232                                                   pptr[0], pptr[1]);
1233                 }
1234         }
1235
1236         return cp;
1237 }
1238
1239 /* Handle response packets: rewrite addresses and send away...
1240  */
1241 static unsigned int
1242 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1243                 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph,
1244                 unsigned int hooknum)
1245 {
1246         struct ip_vs_protocol *pp = pd->pp;
1247
1248         IP_VS_DBG_PKT(11, af, pp, skb, iph->off, "Outgoing packet");
1249
1250         if (!skb_make_writable(skb, iph->len))
1251                 goto drop;
1252
1253         /* mangle the packet */
1254         if (pp->snat_handler && !pp->snat_handler(skb, pp, cp, iph))
1255                 goto drop;
1256
1257 #ifdef CONFIG_IP_VS_IPV6
1258         if (af == AF_INET6)
1259                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1260         else
1261 #endif
1262         {
1263                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1264                 ip_send_check(ip_hdr(skb));
1265         }
1266
1267         /*
1268          * nf_iterate does not expect change in the skb->dst->dev.
1269          * It looks like it is not fatal to enable this code for hooks
1270          * where our handlers are at the end of the chain list and
1271          * when all next handlers use skb->dst->dev and not outdev.
1272          * It will definitely route properly the inout NAT traffic
1273          * when multiple paths are used.
1274          */
1275
1276         /* For policy routing, packets originating from this
1277          * machine itself may be routed differently to packets
1278          * passing through.  We want this packet to be routed as
1279          * if it came from this machine itself.  So re-compute
1280          * the routing information.
1281          */
1282         if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
1283                 goto drop;
1284
1285         IP_VS_DBG_PKT(10, af, pp, skb, iph->off, "After SNAT");
1286
1287         ip_vs_out_stats(cp, skb);
1288         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1289         skb->ipvs_property = 1;
1290         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1291                 ip_vs_notrack(skb);
1292         else
1293                 ip_vs_update_conntrack(skb, cp, 0);
1294         ip_vs_conn_put(cp);
1295
1296         LeaveFunction(11);
1297         return NF_ACCEPT;
1298
1299 drop:
1300         ip_vs_conn_put(cp);
1301         kfree_skb(skb);
1302         LeaveFunction(11);
1303         return NF_STOLEN;
1304 }
1305
1306 /*
1307  *      Check if outgoing packet belongs to the established ip_vs_conn.
1308  */
1309 static unsigned int
1310 ip_vs_out(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1311 {
1312         struct ip_vs_iphdr iph;
1313         struct ip_vs_protocol *pp;
1314         struct ip_vs_proto_data *pd;
1315         struct ip_vs_conn *cp;
1316         struct sock *sk;
1317
1318         EnterFunction(11);
1319
1320         /* Already marked as IPVS request or reply? */
1321         if (skb->ipvs_property)
1322                 return NF_ACCEPT;
1323
1324         sk = skb_to_full_sk(skb);
1325         /* Bad... Do not break raw sockets */
1326         if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
1327                      af == AF_INET)) {
1328
1329                 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
1330                         return NF_ACCEPT;
1331         }
1332
1333         if (unlikely(!skb_dst(skb)))
1334                 return NF_ACCEPT;
1335
1336         if (!ipvs->enable)
1337                 return NF_ACCEPT;
1338
1339         ip_vs_fill_iph_skb(af, skb, false, &iph);
1340 #ifdef CONFIG_IP_VS_IPV6
1341         if (af == AF_INET6) {
1342                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1343                         int related;
1344                         int verdict = ip_vs_out_icmp_v6(ipvs, skb, &related,
1345                                                         hooknum, &iph);
1346
1347                         if (related)
1348                                 return verdict;
1349                 }
1350         } else
1351 #endif
1352                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1353                         int related;
1354                         int verdict = ip_vs_out_icmp(ipvs, skb, &related, hooknum);
1355
1356                         if (related)
1357                                 return verdict;
1358                 }
1359
1360         pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1361         if (unlikely(!pd))
1362                 return NF_ACCEPT;
1363         pp = pd->pp;
1364
1365         /* reassemble IP fragments */
1366 #ifdef CONFIG_IP_VS_IPV6
1367         if (af == AF_INET)
1368 #endif
1369                 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1370                         if (ip_vs_gather_frags(ipvs, skb,
1371                                                ip_vs_defrag_user(hooknum)))
1372                                 return NF_STOLEN;
1373
1374                         ip_vs_fill_iph_skb(AF_INET, skb, false, &iph);
1375                 }
1376
1377         /*
1378          * Check if the packet belongs to an existing entry
1379          */
1380         cp = pp->conn_out_get(ipvs, af, skb, &iph);
1381
1382         if (likely(cp)) {
1383                 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
1384                         goto ignore_cp;
1385                 return handle_response(af, skb, pd, cp, &iph, hooknum);
1386         }
1387
1388         /* Check for real-server-started requests */
1389         if (atomic_read(&ipvs->conn_out_counter)) {
1390                 /* Currently only for UDP:
1391                  * connection oriented protocols typically use
1392                  * ephemeral ports for outgoing connections, so
1393                  * related incoming responses would not match any VS
1394                  */
1395                 if (pp->protocol == IPPROTO_UDP) {
1396                         cp = __ip_vs_rs_conn_out(hooknum, ipvs, af, skb, &iph);
1397                         if (likely(cp))
1398                                 return handle_response(af, skb, pd, cp, &iph,
1399                                                        hooknum);
1400                 }
1401         }
1402
1403         if (sysctl_nat_icmp_send(ipvs) &&
1404             (pp->protocol == IPPROTO_TCP ||
1405              pp->protocol == IPPROTO_UDP ||
1406              pp->protocol == IPPROTO_SCTP)) {
1407                 __be16 _ports[2], *pptr;
1408
1409                 pptr = frag_safe_skb_hp(skb, iph.len,
1410                                          sizeof(_ports), _ports, &iph);
1411                 if (pptr == NULL)
1412                         return NF_ACCEPT;       /* Not for me */
1413                 if (ip_vs_has_real_service(ipvs, af, iph.protocol, &iph.saddr,
1414                                            pptr[0])) {
1415                         /*
1416                          * Notify the real server: there is no
1417                          * existing entry if it is not RST
1418                          * packet or not TCP packet.
1419                          */
1420                         if ((iph.protocol != IPPROTO_TCP &&
1421                              iph.protocol != IPPROTO_SCTP)
1422                              || ((iph.protocol == IPPROTO_TCP
1423                                   && !is_tcp_reset(skb, iph.len))
1424                                  || (iph.protocol == IPPROTO_SCTP
1425                                         && !is_sctp_abort(skb,
1426                                                 iph.len)))) {
1427 #ifdef CONFIG_IP_VS_IPV6
1428                                 if (af == AF_INET6) {
1429                                         if (!skb->dev)
1430                                                 skb->dev = ipvs->net->loopback_dev;
1431                                         icmpv6_send(skb,
1432                                                     ICMPV6_DEST_UNREACH,
1433                                                     ICMPV6_PORT_UNREACH,
1434                                                     0);
1435                                 } else
1436 #endif
1437                                         icmp_send(skb,
1438                                                   ICMP_DEST_UNREACH,
1439                                                   ICMP_PORT_UNREACH, 0);
1440                                 return NF_DROP;
1441                         }
1442                 }
1443         }
1444
1445 out:
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 ignore_cp:
1451         __ip_vs_conn_put(cp);
1452         goto out;
1453 }
1454
1455 /*
1456  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1457  *      used only for VS/NAT.
1458  *      Check if packet is reply for established ip_vs_conn.
1459  */
1460 static unsigned int
1461 ip_vs_reply4(void *priv, struct sk_buff *skb,
1462              const struct nf_hook_state *state)
1463 {
1464         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1465 }
1466
1467 /*
1468  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1469  *      Check if packet is reply for established ip_vs_conn.
1470  */
1471 static unsigned int
1472 ip_vs_local_reply4(void *priv, struct sk_buff *skb,
1473                    const struct nf_hook_state *state)
1474 {
1475         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1476 }
1477
1478 #ifdef CONFIG_IP_VS_IPV6
1479
1480 /*
1481  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1482  *      used only for VS/NAT.
1483  *      Check if packet is reply for established ip_vs_conn.
1484  */
1485 static unsigned int
1486 ip_vs_reply6(void *priv, struct sk_buff *skb,
1487              const struct nf_hook_state *state)
1488 {
1489         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1490 }
1491
1492 /*
1493  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1494  *      Check if packet is reply for established ip_vs_conn.
1495  */
1496 static unsigned int
1497 ip_vs_local_reply6(void *priv, struct sk_buff *skb,
1498                    const struct nf_hook_state *state)
1499 {
1500         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1501 }
1502
1503 #endif
1504
1505 static unsigned int
1506 ip_vs_try_to_schedule(struct netns_ipvs *ipvs, int af, struct sk_buff *skb,
1507                       struct ip_vs_proto_data *pd,
1508                       int *verdict, struct ip_vs_conn **cpp,
1509                       struct ip_vs_iphdr *iph)
1510 {
1511         struct ip_vs_protocol *pp = pd->pp;
1512
1513         if (!iph->fragoffs) {
1514                 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1515                  * replayed fragment zero will already have created the cp
1516                  */
1517
1518                 /* Schedule and create new connection entry into cpp */
1519                 if (!pp->conn_schedule(ipvs, af, skb, pd, verdict, cpp, iph))
1520                         return 0;
1521         }
1522
1523         if (unlikely(!*cpp)) {
1524                 /* sorry, all this trouble for a no-hit :) */
1525                 IP_VS_DBG_PKT(12, af, pp, skb, iph->off,
1526                               "ip_vs_in: packet continues traversal as normal");
1527
1528                 /* Fragment couldn't be mapped to a conn entry */
1529                 if (iph->fragoffs)
1530                         IP_VS_DBG_PKT(7, af, pp, skb, iph->off,
1531                                       "unhandled fragment");
1532
1533                 *verdict = NF_ACCEPT;
1534                 return 0;
1535         }
1536
1537         return 1;
1538 }
1539
1540 /*
1541  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1542  *      Find any that might be relevant, check against existing connections,
1543  *      forward to the right destination host if relevant.
1544  *      Currently handles error types - unreachable, quench, ttl exceeded.
1545  */
1546 static int
1547 ip_vs_in_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb, int *related,
1548               unsigned int hooknum)
1549 {
1550         struct iphdr *iph;
1551         struct icmphdr  _icmph, *ic;
1552         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1553         struct ip_vs_iphdr ciph;
1554         struct ip_vs_conn *cp;
1555         struct ip_vs_protocol *pp;
1556         struct ip_vs_proto_data *pd;
1557         unsigned int offset, offset2, ihl, verdict;
1558         bool ipip, new_cp = false;
1559
1560         *related = 1;
1561
1562         /* reassemble IP fragments */
1563         if (ip_is_fragment(ip_hdr(skb))) {
1564                 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
1565                         return NF_STOLEN;
1566         }
1567
1568         iph = ip_hdr(skb);
1569         offset = ihl = iph->ihl * 4;
1570         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1571         if (ic == NULL)
1572                 return NF_DROP;
1573
1574         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1575                   ic->type, ntohs(icmp_id(ic)),
1576                   &iph->saddr, &iph->daddr);
1577
1578         /*
1579          * Work through seeing if this is for us.
1580          * These checks are supposed to be in an order that means easy
1581          * things are checked first to speed up processing.... however
1582          * this means that some packets will manage to get a long way
1583          * down this stack and then be rejected, but that's life.
1584          */
1585         if ((ic->type != ICMP_DEST_UNREACH) &&
1586             (ic->type != ICMP_SOURCE_QUENCH) &&
1587             (ic->type != ICMP_TIME_EXCEEDED)) {
1588                 *related = 0;
1589                 return NF_ACCEPT;
1590         }
1591
1592         /* Now find the contained IP header */
1593         offset += sizeof(_icmph);
1594         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1595         if (cih == NULL)
1596                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1597
1598         /* Special case for errors for IPIP packets */
1599         ipip = false;
1600         if (cih->protocol == IPPROTO_IPIP) {
1601                 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1602                         return NF_ACCEPT;
1603                 /* Error for our IPIP must arrive at LOCAL_IN */
1604                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1605                         return NF_ACCEPT;
1606                 offset += cih->ihl * 4;
1607                 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1608                 if (cih == NULL)
1609                         return NF_ACCEPT; /* The packet looks wrong, ignore */
1610                 ipip = true;
1611         }
1612
1613         pd = ip_vs_proto_data_get(ipvs, cih->protocol);
1614         if (!pd)
1615                 return NF_ACCEPT;
1616         pp = pd->pp;
1617
1618         /* Is the embedded protocol header present? */
1619         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1620                      pp->dont_defrag))
1621                 return NF_ACCEPT;
1622
1623         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1624                       "Checking incoming ICMP for");
1625
1626         offset2 = offset;
1627         ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, !ipip, &ciph);
1628         offset = ciph.len;
1629
1630         /* The embedded headers contain source and dest in reverse order.
1631          * For IPIP this is error for request, not for reply.
1632          */
1633         cp = pp->conn_in_get(ipvs, AF_INET, skb, &ciph);
1634
1635         if (!cp) {
1636                 int v;
1637
1638                 if (ipip || !sysctl_schedule_icmp(ipvs))
1639                         return NF_ACCEPT;
1640
1641                 if (!ip_vs_try_to_schedule(ipvs, AF_INET, skb, pd, &v, &cp, &ciph))
1642                         return v;
1643                 new_cp = true;
1644         }
1645
1646         verdict = NF_DROP;
1647
1648         /* Ensure the checksum is correct */
1649         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1650                 /* Failed checksum! */
1651                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1652                           &iph->saddr);
1653                 goto out;
1654         }
1655
1656         if (ipip) {
1657                 __be32 info = ic->un.gateway;
1658                 __u8 type = ic->type;
1659                 __u8 code = ic->code;
1660
1661                 /* Update the MTU */
1662                 if (ic->type == ICMP_DEST_UNREACH &&
1663                     ic->code == ICMP_FRAG_NEEDED) {
1664                         struct ip_vs_dest *dest = cp->dest;
1665                         u32 mtu = ntohs(ic->un.frag.mtu);
1666                         __be16 frag_off = cih->frag_off;
1667
1668                         /* Strip outer IP and ICMP, go to IPIP header */
1669                         if (pskb_pull(skb, ihl + sizeof(_icmph)) == NULL)
1670                                 goto ignore_ipip;
1671                         offset2 -= ihl + sizeof(_icmph);
1672                         skb_reset_network_header(skb);
1673                         IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1674                                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1675                         ipv4_update_pmtu(skb, ipvs->net,
1676                                          mtu, 0, 0, 0, 0);
1677                         /* Client uses PMTUD? */
1678                         if (!(frag_off & htons(IP_DF)))
1679                                 goto ignore_ipip;
1680                         /* Prefer the resulting PMTU */
1681                         if (dest) {
1682                                 struct ip_vs_dest_dst *dest_dst;
1683
1684                                 dest_dst = rcu_dereference(dest->dest_dst);
1685                                 if (dest_dst)
1686                                         mtu = dst_mtu(dest_dst->dst_cache);
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         struct sock *sk;
1842
1843         /* Already marked as IPVS request or reply? */
1844         if (skb->ipvs_property)
1845                 return NF_ACCEPT;
1846
1847         /*
1848          *      Big tappo:
1849          *      - remote client: only PACKET_HOST
1850          *      - route: used for struct net when skb->dev is unset
1851          */
1852         if (unlikely((skb->pkt_type != PACKET_HOST &&
1853                       hooknum != NF_INET_LOCAL_OUT) ||
1854                      !skb_dst(skb))) {
1855                 ip_vs_fill_iph_skb(af, skb, false, &iph);
1856                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1857                               " ignored in hook %u\n",
1858                               skb->pkt_type, iph.protocol,
1859                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1860                 return NF_ACCEPT;
1861         }
1862         /* ipvs enabled in this netns ? */
1863         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1864                 return NF_ACCEPT;
1865
1866         ip_vs_fill_iph_skb(af, skb, false, &iph);
1867
1868         /* Bad... Do not break raw sockets */
1869         sk = skb_to_full_sk(skb);
1870         if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
1871                      af == AF_INET)) {
1872
1873                 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
1874                         return NF_ACCEPT;
1875         }
1876
1877 #ifdef CONFIG_IP_VS_IPV6
1878         if (af == AF_INET6) {
1879                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1880                         int related;
1881                         int verdict = ip_vs_in_icmp_v6(ipvs, skb, &related,
1882                                                        hooknum, &iph);
1883
1884                         if (related)
1885                                 return verdict;
1886                 }
1887         } else
1888 #endif
1889                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1890                         int related;
1891                         int verdict = ip_vs_in_icmp(ipvs, skb, &related,
1892                                                     hooknum);
1893
1894                         if (related)
1895                                 return verdict;
1896                 }
1897
1898         /* Protocol supported? */
1899         pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1900         if (unlikely(!pd)) {
1901                 /* The only way we'll see this packet again is if it's
1902                  * encapsulated, so mark it with ipvs_property=1 so we
1903                  * skip it if we're ignoring tunneled packets
1904                  */
1905                 if (sysctl_ignore_tunneled(ipvs))
1906                         skb->ipvs_property = 1;
1907
1908                 return NF_ACCEPT;
1909         }
1910         pp = pd->pp;
1911         /*
1912          * Check if the packet belongs to an existing connection entry
1913          */
1914         cp = pp->conn_in_get(ipvs, af, skb, &iph);
1915
1916         if (!iph.fragoffs && is_new_conn(skb, &iph) && cp) {
1917                 int conn_reuse_mode = sysctl_conn_reuse_mode(ipvs);
1918                 bool old_ct = false, resched = false;
1919
1920                 if (unlikely(sysctl_expire_nodest_conn(ipvs)) && cp->dest &&
1921                     unlikely(!atomic_read(&cp->dest->weight))) {
1922                         resched = true;
1923                         old_ct = ip_vs_conn_uses_old_conntrack(cp, skb);
1924                 } else if (conn_reuse_mode &&
1925                            is_new_conn_expected(cp, conn_reuse_mode)) {
1926                         old_ct = ip_vs_conn_uses_old_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 = !old_ct;
1935                         }
1936                 }
1937
1938                 if (resched) {
1939                         if (!old_ct)
1940                                 cp->flags &= ~IP_VS_CONN_F_NFCT;
1941                         if (!atomic_read(&cp->n_control))
1942                                 ip_vs_conn_expire_now(cp);
1943                         __ip_vs_conn_put(cp);
1944                         if (old_ct)
1945                                 return NF_DROP;
1946                         cp = NULL;
1947                 }
1948         }
1949
1950         if (unlikely(!cp)) {
1951                 int v;
1952
1953                 if (!ip_vs_try_to_schedule(ipvs, af, skb, pd, &v, &cp, &iph))
1954                         return v;
1955         }
1956
1957         IP_VS_DBG_PKT(11, af, pp, skb, iph.off, "Incoming packet");
1958
1959         /* Check the server status */
1960         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1961                 /* the destination server is not available */
1962
1963                 __u32 flags = cp->flags;
1964
1965                 /* when timer already started, silently drop the packet.*/
1966                 if (timer_pending(&cp->timer))
1967                         __ip_vs_conn_put(cp);
1968                 else
1969                         ip_vs_conn_put(cp);
1970
1971                 if (sysctl_expire_nodest_conn(ipvs) &&
1972                     !(flags & IP_VS_CONN_F_ONE_PACKET)) {
1973                         /* try to expire the connection immediately */
1974                         ip_vs_conn_expire_now(cp);
1975                 }
1976
1977                 return NF_DROP;
1978         }
1979
1980         ip_vs_in_stats(cp, skb);
1981         ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1982         if (cp->packet_xmit)
1983                 ret = cp->packet_xmit(skb, cp, pp, &iph);
1984                 /* do not touch skb anymore */
1985         else {
1986                 IP_VS_DBG_RL("warning: packet_xmit is null");
1987                 ret = NF_ACCEPT;
1988         }
1989
1990         /* Increase its packet counter and check if it is needed
1991          * to be synchronized
1992          *
1993          * Sync connection if it is about to close to
1994          * encorage the standby servers to update the connections timeout
1995          *
1996          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1997          */
1998
1999         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
2000                 pkts = sysctl_sync_threshold(ipvs);
2001         else
2002                 pkts = atomic_add_return(1, &cp->in_pkts);
2003
2004         if (ipvs->sync_state & IP_VS_STATE_MASTER)
2005                 ip_vs_sync_conn(ipvs, cp, pkts);
2006         else if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) && cp->control)
2007                 /* increment is done inside ip_vs_sync_conn too */
2008                 atomic_inc(&cp->control->in_pkts);
2009
2010         ip_vs_conn_put(cp);
2011         return ret;
2012 }
2013
2014 /*
2015  *      AF_INET handler in NF_INET_LOCAL_IN chain
2016  *      Schedule and forward packets from remote clients
2017  */
2018 static unsigned int
2019 ip_vs_remote_request4(void *priv, struct sk_buff *skb,
2020                       const struct nf_hook_state *state)
2021 {
2022         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
2023 }
2024
2025 /*
2026  *      AF_INET handler in NF_INET_LOCAL_OUT chain
2027  *      Schedule and forward packets from local clients
2028  */
2029 static unsigned int
2030 ip_vs_local_request4(void *priv, struct sk_buff *skb,
2031                      const struct nf_hook_state *state)
2032 {
2033         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
2034 }
2035
2036 #ifdef CONFIG_IP_VS_IPV6
2037
2038 /*
2039  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
2040  *      Schedule and forward packets from remote clients
2041  */
2042 static unsigned int
2043 ip_vs_remote_request6(void *priv, struct sk_buff *skb,
2044                       const struct nf_hook_state *state)
2045 {
2046         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
2047 }
2048
2049 /*
2050  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
2051  *      Schedule and forward packets from local clients
2052  */
2053 static unsigned int
2054 ip_vs_local_request6(void *priv, struct sk_buff *skb,
2055                      const struct nf_hook_state *state)
2056 {
2057         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
2058 }
2059
2060 #endif
2061
2062
2063 /*
2064  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
2065  *      related packets destined for 0.0.0.0/0.
2066  *      When fwmark-based virtual service is used, such as transparent
2067  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
2068  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
2069  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
2070  *      and send them to ip_vs_in_icmp.
2071  */
2072 static unsigned int
2073 ip_vs_forward_icmp(void *priv, struct sk_buff *skb,
2074                    const struct nf_hook_state *state)
2075 {
2076         int r;
2077         struct netns_ipvs *ipvs = net_ipvs(state->net);
2078
2079         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
2080                 return NF_ACCEPT;
2081
2082         /* ipvs enabled in this netns ? */
2083         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
2084                 return NF_ACCEPT;
2085
2086         return ip_vs_in_icmp(ipvs, skb, &r, state->hook);
2087 }
2088
2089 #ifdef CONFIG_IP_VS_IPV6
2090 static unsigned int
2091 ip_vs_forward_icmp_v6(void *priv, struct sk_buff *skb,
2092                       const struct nf_hook_state *state)
2093 {
2094         int r;
2095         struct netns_ipvs *ipvs = net_ipvs(state->net);
2096         struct ip_vs_iphdr iphdr;
2097
2098         ip_vs_fill_iph_skb(AF_INET6, skb, false, &iphdr);
2099         if (iphdr.protocol != IPPROTO_ICMPV6)
2100                 return NF_ACCEPT;
2101
2102         /* ipvs enabled in this netns ? */
2103         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
2104                 return NF_ACCEPT;
2105
2106         return ip_vs_in_icmp_v6(ipvs, skb, &r, state->hook, &iphdr);
2107 }
2108 #endif
2109
2110
2111 static const struct nf_hook_ops ip_vs_ops[] = {
2112         /* After packet filtering, change source only for VS/NAT */
2113         {
2114                 .hook           = ip_vs_reply4,
2115                 .pf             = NFPROTO_IPV4,
2116                 .hooknum        = NF_INET_LOCAL_IN,
2117                 .priority       = NF_IP_PRI_NAT_SRC - 2,
2118         },
2119         /* After packet filtering, forward packet through VS/DR, VS/TUN,
2120          * or VS/NAT(change destination), so that filtering rules can be
2121          * applied to IPVS. */
2122         {
2123                 .hook           = ip_vs_remote_request4,
2124                 .pf             = NFPROTO_IPV4,
2125                 .hooknum        = NF_INET_LOCAL_IN,
2126                 .priority       = NF_IP_PRI_NAT_SRC - 1,
2127         },
2128         /* Before ip_vs_in, change source only for VS/NAT */
2129         {
2130                 .hook           = ip_vs_local_reply4,
2131                 .pf             = NFPROTO_IPV4,
2132                 .hooknum        = NF_INET_LOCAL_OUT,
2133                 .priority       = NF_IP_PRI_NAT_DST + 1,
2134         },
2135         /* After mangle, schedule and forward local requests */
2136         {
2137                 .hook           = ip_vs_local_request4,
2138                 .pf             = NFPROTO_IPV4,
2139                 .hooknum        = NF_INET_LOCAL_OUT,
2140                 .priority       = NF_IP_PRI_NAT_DST + 2,
2141         },
2142         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2143          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2144         {
2145                 .hook           = ip_vs_forward_icmp,
2146                 .pf             = NFPROTO_IPV4,
2147                 .hooknum        = NF_INET_FORWARD,
2148                 .priority       = 99,
2149         },
2150         /* After packet filtering, change source only for VS/NAT */
2151         {
2152                 .hook           = ip_vs_reply4,
2153                 .pf             = NFPROTO_IPV4,
2154                 .hooknum        = NF_INET_FORWARD,
2155                 .priority       = 100,
2156         },
2157 #ifdef CONFIG_IP_VS_IPV6
2158         /* After packet filtering, change source only for VS/NAT */
2159         {
2160                 .hook           = ip_vs_reply6,
2161                 .pf             = NFPROTO_IPV6,
2162                 .hooknum        = NF_INET_LOCAL_IN,
2163                 .priority       = NF_IP6_PRI_NAT_SRC - 2,
2164         },
2165         /* After packet filtering, forward packet through VS/DR, VS/TUN,
2166          * or VS/NAT(change destination), so that filtering rules can be
2167          * applied to IPVS. */
2168         {
2169                 .hook           = ip_vs_remote_request6,
2170                 .pf             = NFPROTO_IPV6,
2171                 .hooknum        = NF_INET_LOCAL_IN,
2172                 .priority       = NF_IP6_PRI_NAT_SRC - 1,
2173         },
2174         /* Before ip_vs_in, change source only for VS/NAT */
2175         {
2176                 .hook           = ip_vs_local_reply6,
2177                 .pf             = NFPROTO_IPV6,
2178                 .hooknum        = NF_INET_LOCAL_OUT,
2179                 .priority       = NF_IP6_PRI_NAT_DST + 1,
2180         },
2181         /* After mangle, schedule and forward local requests */
2182         {
2183                 .hook           = ip_vs_local_request6,
2184                 .pf             = NFPROTO_IPV6,
2185                 .hooknum        = NF_INET_LOCAL_OUT,
2186                 .priority       = NF_IP6_PRI_NAT_DST + 2,
2187         },
2188         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2189          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2190         {
2191                 .hook           = ip_vs_forward_icmp_v6,
2192                 .pf             = NFPROTO_IPV6,
2193                 .hooknum        = NF_INET_FORWARD,
2194                 .priority       = 99,
2195         },
2196         /* After packet filtering, change source only for VS/NAT */
2197         {
2198                 .hook           = ip_vs_reply6,
2199                 .pf             = NFPROTO_IPV6,
2200                 .hooknum        = NF_INET_FORWARD,
2201                 .priority       = 100,
2202         },
2203 #endif
2204 };
2205 /*
2206  *      Initialize IP Virtual Server netns mem.
2207  */
2208 static int __net_init __ip_vs_init(struct net *net)
2209 {
2210         struct netns_ipvs *ipvs;
2211
2212         ipvs = net_generic(net, ip_vs_net_id);
2213         if (ipvs == NULL)
2214                 return -ENOMEM;
2215
2216         /* Hold the beast until a service is registerd */
2217         ipvs->enable = 0;
2218         ipvs->net = net;
2219         /* Counters used for creating unique names */
2220         ipvs->gen = atomic_read(&ipvs_netns_cnt);
2221         atomic_inc(&ipvs_netns_cnt);
2222         net->ipvs = ipvs;
2223
2224         if (ip_vs_estimator_net_init(ipvs) < 0)
2225                 goto estimator_fail;
2226
2227         if (ip_vs_control_net_init(ipvs) < 0)
2228                 goto control_fail;
2229
2230         if (ip_vs_protocol_net_init(ipvs) < 0)
2231                 goto protocol_fail;
2232
2233         if (ip_vs_app_net_init(ipvs) < 0)
2234                 goto app_fail;
2235
2236         if (ip_vs_conn_net_init(ipvs) < 0)
2237                 goto conn_fail;
2238
2239         if (ip_vs_sync_net_init(ipvs) < 0)
2240                 goto sync_fail;
2241
2242         return 0;
2243 /*
2244  * Error handling
2245  */
2246
2247 sync_fail:
2248         ip_vs_conn_net_cleanup(ipvs);
2249 conn_fail:
2250         ip_vs_app_net_cleanup(ipvs);
2251 app_fail:
2252         ip_vs_protocol_net_cleanup(ipvs);
2253 protocol_fail:
2254         ip_vs_control_net_cleanup(ipvs);
2255 control_fail:
2256         ip_vs_estimator_net_cleanup(ipvs);
2257 estimator_fail:
2258         net->ipvs = NULL;
2259         return -ENOMEM;
2260 }
2261
2262 static void __net_exit __ip_vs_cleanup(struct net *net)
2263 {
2264         struct netns_ipvs *ipvs = net_ipvs(net);
2265
2266         ip_vs_service_net_cleanup(ipvs);        /* ip_vs_flush() with locks */
2267         ip_vs_conn_net_cleanup(ipvs);
2268         ip_vs_app_net_cleanup(ipvs);
2269         ip_vs_protocol_net_cleanup(ipvs);
2270         ip_vs_control_net_cleanup(ipvs);
2271         ip_vs_estimator_net_cleanup(ipvs);
2272         IP_VS_DBG(2, "ipvs netns %d released\n", ipvs->gen);
2273         net->ipvs = NULL;
2274 }
2275
2276 static int __net_init __ip_vs_dev_init(struct net *net)
2277 {
2278         int ret;
2279
2280         ret = nf_register_net_hooks(net, ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2281         if (ret < 0)
2282                 goto hook_fail;
2283         return 0;
2284
2285 hook_fail:
2286         return ret;
2287 }
2288
2289 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
2290 {
2291         struct netns_ipvs *ipvs = net_ipvs(net);
2292         EnterFunction(2);
2293         nf_unregister_net_hooks(net, ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2294         ipvs->enable = 0;       /* Disable packet reception */
2295         smp_wmb();
2296         ip_vs_sync_net_cleanup(ipvs);
2297         LeaveFunction(2);
2298 }
2299
2300 static struct pernet_operations ipvs_core_ops = {
2301         .init = __ip_vs_init,
2302         .exit = __ip_vs_cleanup,
2303         .id   = &ip_vs_net_id,
2304         .size = sizeof(struct netns_ipvs),
2305 };
2306
2307 static struct pernet_operations ipvs_core_dev_ops = {
2308         .init = __ip_vs_dev_init,
2309         .exit = __ip_vs_dev_cleanup,
2310 };
2311
2312 /*
2313  *      Initialize IP Virtual Server
2314  */
2315 static int __init ip_vs_init(void)
2316 {
2317         int ret;
2318
2319         ret = ip_vs_control_init();
2320         if (ret < 0) {
2321                 pr_err("can't setup control.\n");
2322                 goto exit;
2323         }
2324
2325         ip_vs_protocol_init();
2326
2327         ret = ip_vs_conn_init();
2328         if (ret < 0) {
2329                 pr_err("can't setup connection table.\n");
2330                 goto cleanup_protocol;
2331         }
2332
2333         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
2334         if (ret < 0)
2335                 goto cleanup_conn;
2336
2337         ret = register_pernet_device(&ipvs_core_dev_ops);
2338         if (ret < 0)
2339                 goto cleanup_sub;
2340
2341         ret = ip_vs_register_nl_ioctl();
2342         if (ret < 0) {
2343                 pr_err("can't register netlink/ioctl.\n");
2344                 goto cleanup_dev;
2345         }
2346
2347         pr_info("ipvs loaded.\n");
2348
2349         return ret;
2350
2351 cleanup_dev:
2352         unregister_pernet_device(&ipvs_core_dev_ops);
2353 cleanup_sub:
2354         unregister_pernet_subsys(&ipvs_core_ops);
2355 cleanup_conn:
2356         ip_vs_conn_cleanup();
2357 cleanup_protocol:
2358         ip_vs_protocol_cleanup();
2359         ip_vs_control_cleanup();
2360 exit:
2361         return ret;
2362 }
2363
2364 static void __exit ip_vs_cleanup(void)
2365 {
2366         ip_vs_unregister_nl_ioctl();
2367         unregister_pernet_device(&ipvs_core_dev_ops);
2368         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
2369         ip_vs_conn_cleanup();
2370         ip_vs_protocol_cleanup();
2371         ip_vs_control_cleanup();
2372         pr_info("ipvs unloaded.\n");
2373 }
2374
2375 module_init(ip_vs_init);
2376 module_exit(ip_vs_cleanup);
2377 MODULE_LICENSE("GPL");