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                 local_bh_disable();
 123
 124                 s = this_cpu_ptr(dest->stats.cpustats);
 125                 u64_stats_update_begin(&s->syncp);
 126                 s->cnt.inpkts++;
 127                 s->cnt.inbytes += skb->len;
 128                 u64_stats_update_end(&s->syncp);
 129
 130                 svc = rcu_dereference(dest->svc);
 131                 s = this_cpu_ptr(svc->stats.cpustats);
 132                 u64_stats_update_begin(&s->syncp);
 133                 s->cnt.inpkts++;
 134                 s->cnt.inbytes += skb->len;
 135                 u64_stats_update_end(&s->syncp);
 136
 137                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 138                 u64_stats_update_begin(&s->syncp);
 139                 s->cnt.inpkts++;
 140                 s->cnt.inbytes += skb->len;
 141                 u64_stats_update_end(&s->syncp);
 142
 143                 local_bh_enable();
 144         }
 145 }
 146
 147
 148 static inline void
 149 ip_vs_out_stats(struct ip_vs_conn *cp, struct sk_buff *skb)
 150 {
 151         struct ip_vs_dest *dest = cp->dest;
 152         struct netns_ipvs *ipvs = cp->ipvs;
 153
 154         if (dest && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 155                 struct ip_vs_cpu_stats *s;
 156                 struct ip_vs_service *svc;
 157
 158                 local_bh_disable();
 159
 160                 s = this_cpu_ptr(dest->stats.cpustats);
 161                 u64_stats_update_begin(&s->syncp);
 162                 s->cnt.outpkts++;
 163                 s->cnt.outbytes += skb->len;
 164                 u64_stats_update_end(&s->syncp);
 165
 166                 svc = rcu_dereference(dest->svc);
 167                 s = this_cpu_ptr(svc->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                 s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 174                 u64_stats_update_begin(&s->syncp);
 175                 s->cnt.outpkts++;
 176                 s->cnt.outbytes += skb->len;
 177                 u64_stats_update_end(&s->syncp);
 178
 179                 local_bh_enable();
 180         }
 181 }
 182
 183
 184 static inline void
 185 ip_vs_conn_stats(struct ip_vs_conn *cp, struct ip_vs_service *svc)
 186 {
 187         struct netns_ipvs *ipvs = svc->ipvs;
 188         struct ip_vs_cpu_stats *s;
 189
 190         local_bh_disable();
 191
 192         s = this_cpu_ptr(cp->dest->stats.cpustats);
 193         u64_stats_update_begin(&s->syncp);
 194         s->cnt.conns++;
 195         u64_stats_update_end(&s->syncp);
 196
 197         s = this_cpu_ptr(svc->stats.cpustats);
 198         u64_stats_update_begin(&s->syncp);
 199         s->cnt.conns++;
 200         u64_stats_update_end(&s->syncp);
 201
 202         s = this_cpu_ptr(ipvs->tot_stats.cpustats);
 203         u64_stats_update_begin(&s->syncp);
 204         s->cnt.conns++;
 205         u64_stats_update_end(&s->syncp);
 206
 207         local_bh_enable();
 208 }
 209
 210
 211 static inline void
 212 ip_vs_set_state(struct ip_vs_conn *cp, int direction,
 213                 const struct sk_buff *skb,
 214                 struct ip_vs_proto_data *pd)
 215 {
 216         if (likely(pd->pp->state_transition))
 217                 pd->pp->state_transition(cp, direction, skb, pd);
 218 }
 219
 220 static inline int
 221 ip_vs_conn_fill_param_persist(const struct ip_vs_service *svc,
 222                               struct sk_buff *skb, int protocol,
 223                               const union nf_inet_addr *caddr, __be16 cport,
 224                               const union nf_inet_addr *vaddr, __be16 vport,
 225                               struct ip_vs_conn_param *p)
 226 {
 227         ip_vs_conn_fill_param(svc->ipvs, svc->af, protocol, caddr, cport, vaddr,
 228                               vport, p);
 229         p->pe = rcu_dereference(svc->pe);
 230         if (p->pe && p->pe->fill_param)
 231                 return p->pe->fill_param(p, skb);
 232
 233         return 0;
 234 }
 235
 236 /*
 237  *  IPVS persistent scheduling function
 238  *  It creates a connection entry according to its template if exists,
 239  *  or selects a server and creates a connection entry plus a template.
 240  *  Locking: we are svc user (svc->refcnt), so we hold all dests too
 241  *  Protocols supported: TCP, UDP
 242  */
 243 static struct ip_vs_conn *
 244 ip_vs_sched_persist(struct ip_vs_service *svc,
 245                     struct sk_buff *skb, __be16 src_port, __be16 dst_port,
 246                     int *ignored, struct ip_vs_iphdr *iph)
 247 {
 248         struct ip_vs_conn *cp = NULL;
 249         struct ip_vs_dest *dest;
 250         struct ip_vs_conn *ct;
 251         __be16 dport = 0;               /* destination port to forward */
 252         unsigned int flags;
 253         struct ip_vs_conn_param param;
 254         const union nf_inet_addr fwmark = { .ip = htonl(svc->fwmark) };
 255         union nf_inet_addr snet;        /* source network of the client,
 256                                            after masking */
 257         const union nf_inet_addr *src_addr, *dst_addr;
 258
 259         if (likely(!ip_vs_iph_inverse(iph))) {
 260                 src_addr = &iph->saddr;
 261                 dst_addr = &iph->daddr;
 262         } else {
 263                 src_addr = &iph->daddr;
 264                 dst_addr = &iph->saddr;
 265         }
 266
 267
 268         /* Mask saddr with the netmask to adjust template granularity */
 269 #ifdef CONFIG_IP_VS_IPV6
 270         if (svc->af == AF_INET6)
 271                 ipv6_addr_prefix(&snet.in6, &src_addr->in6,
 272                                  (__force __u32) svc->netmask);
 273         else
 274 #endif
 275                 snet.ip = src_addr->ip & svc->netmask;
 276
 277         IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
 278                       "mnet %s\n",
 279                       IP_VS_DBG_ADDR(svc->af, src_addr), ntohs(src_port),
 280                       IP_VS_DBG_ADDR(svc->af, dst_addr), ntohs(dst_port),
 281                       IP_VS_DBG_ADDR(svc->af, &snet));
 282
 283         /*
 284          * As far as we know, FTP is a very complicated network protocol, and
 285          * it uses control connection and data connections. For active FTP,
 286          * FTP server initialize data connection to the client, its source port
 287          * is often 20. For passive FTP, FTP server tells the clients the port
 288          * that it passively listens to,  and the client issues the data
 289          * connection. In the tunneling or direct routing mode, the load
 290          * balancer is on the client-to-server half of connection, the port
 291          * number is unknown to the load balancer. So, a conn template like
 292          * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
 293          * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
 294          * is created for other persistent services.
 295          */
 296         {
 297                 int protocol = iph->protocol;
 298                 const union nf_inet_addr *vaddr = dst_addr;
 299                 __be16 vport = 0;
 300
 301                 if (dst_port == svc->port) {
 302                         /* non-FTP template:
 303                          * <protocol, caddr, 0, vaddr, vport, daddr, dport>
 304                          * FTP template:
 305                          * <protocol, caddr, 0, vaddr, 0, daddr, 0>
 306                          */
 307                         if (svc->port != FTPPORT)
 308                                 vport = dst_port;
 309                 } else {
 310                         /* Note: persistent fwmark-based services and
 311                          * persistent port zero service are handled here.
 312                          * fwmark template:
 313                          * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
 314                          * port zero template:
 315                          * <protocol,caddr,0,vaddr,0,daddr,0>
 316                          */
 317                         if (svc->fwmark) {
 318                                 protocol = IPPROTO_IP;
 319                                 vaddr = &fwmark;
 320                         }
 321                 }
 322                 /* return *ignored = -1 so NF_DROP can be used */
 323                 if (ip_vs_conn_fill_param_persist(svc, skb, protocol, &snet, 0,
 324                                                   vaddr, vport, &param) < 0) {
 325                         *ignored = -1;
 326                         return NULL;
 327                 }
 328         }
 329
 330         /* Check if a template already exists */
 331         ct = ip_vs_ct_in_get(&param);
 332         if (!ct || !ip_vs_check_template(ct, NULL)) {
 333                 struct ip_vs_scheduler *sched;
 334
 335                 /*
 336                  * No template found or the dest of the connection
 337                  * template is not available.
 338                  * return *ignored=0 i.e. ICMP and NF_DROP
 339                  */
 340                 sched = rcu_dereference(svc->scheduler);
 341                 if (sched) {
 342                         /* read svc->sched_data after svc->scheduler */
 343                         smp_rmb();
 344                         dest = sched->schedule(svc, skb, iph);
 345                 } else {
 346                         dest = NULL;
 347                 }
 348                 if (!dest) {
 349                         IP_VS_DBG(1, "p-schedule: no dest found.\n");
 350                         kfree(param.pe_data);
 351                         *ignored = 0;
 352                         return NULL;
 353                 }
 354
 355                 if (dst_port == svc->port && svc->port != FTPPORT)
 356                         dport = dest->port;
 357
 358                 /* Create a template
 359                  * This adds param.pe_data to the template,
 360                  * and thus param.pe_data will be destroyed
 361                  * when the template expires */
 362                 ct = ip_vs_conn_new(&param, dest->af, &dest->addr, dport,
 363                                     IP_VS_CONN_F_TEMPLATE, dest, skb->mark);
 364                 if (ct == NULL) {
 365                         kfree(param.pe_data);
 366                         *ignored = -1;
 367                         return NULL;
 368                 }
 369
 370                 ct->timeout = svc->timeout;
 371         } else {
 372                 /* set destination with the found template */
 373                 dest = ct->dest;
 374                 kfree(param.pe_data);
 375         }
 376
 377         dport = dst_port;
 378         if (dport == svc->port && dest->port)
 379                 dport = dest->port;
 380
 381         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 382                  && iph->protocol == IPPROTO_UDP) ?
 383                 IP_VS_CONN_F_ONE_PACKET : 0;
 384
 385         /*
 386          *    Create a new connection according to the template
 387          */
 388         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol, src_addr,
 389                               src_port, dst_addr, dst_port, &param);
 390
 391         cp = ip_vs_conn_new(&param, dest->af, &dest->addr, dport, flags, dest,
 392                             skb->mark);
 393         if (cp == NULL) {
 394                 ip_vs_conn_put(ct);
 395                 *ignored = -1;
 396                 return NULL;
 397         }
 398
 399         /*
 400          *    Add its control
 401          */
 402         ip_vs_control_add(cp, ct);
 403         ip_vs_conn_put(ct);
 404
 405         ip_vs_conn_stats(cp, svc);
 406         return cp;
 407 }
 408
 409
 410 /*
 411  *  IPVS main scheduling function
 412  *  It selects a server according to the virtual service, and
 413  *  creates a connection entry.
 414  *  Protocols supported: TCP, UDP
 415  *
 416  *  Usage of *ignored
 417  *
 418  * 1 :   protocol tried to schedule (eg. on SYN), found svc but the
 419  *       svc/scheduler decides that this packet should be accepted with
 420  *       NF_ACCEPT because it must not be scheduled.
 421  *
 422  * 0 :   scheduler can not find destination, so try bypass or
 423  *       return ICMP and then NF_DROP (ip_vs_leave).
 424  *
 425  * -1 :  scheduler tried to schedule but fatal error occurred, eg.
 426  *       ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
 427  *       failure such as missing Call-ID, ENOMEM on skb_linearize
 428  *       or pe_data. In this case we should return NF_DROP without
 429  *       any attempts to send ICMP with ip_vs_leave.
 430  */
 431 struct ip_vs_conn *
 432 ip_vs_schedule(struct ip_vs_service *svc, struct sk_buff *skb,
 433                struct ip_vs_proto_data *pd, int *ignored,
 434                struct ip_vs_iphdr *iph)
 435 {
 436         struct ip_vs_protocol *pp = pd->pp;
 437         struct ip_vs_conn *cp = NULL;
 438         struct ip_vs_scheduler *sched;
 439         struct ip_vs_dest *dest;
 440         __be16 _ports[2], *pptr, cport, vport;
 441         const void *caddr, *vaddr;
 442         unsigned int flags;
 443
 444         *ignored = 1;
 445         /*
 446          * IPv6 frags, only the first hit here.
 447          */
 448         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports);
 449         if (pptr == NULL)
 450                 return NULL;
 451
 452         if (likely(!ip_vs_iph_inverse(iph))) {
 453                 cport = pptr[0];
 454                 caddr = &iph->saddr;
 455                 vport = pptr[1];
 456                 vaddr = &iph->daddr;
 457         } else {
 458                 cport = pptr[1];
 459                 caddr = &iph->daddr;
 460                 vport = pptr[0];
 461                 vaddr = &iph->saddr;
 462         }
 463
 464         /*
 465          * FTPDATA needs this check when using local real server.
 466          * Never schedule Active FTPDATA connections from real server.
 467          * For LVS-NAT they must be already created. For other methods
 468          * with persistence the connection is created on SYN+ACK.
 469          */
 470         if (cport == FTPDATA) {
 471                 IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
 472                               "Not scheduling FTPDATA");
 473                 return NULL;
 474         }
 475
 476         /*
 477          *    Do not schedule replies from local real server.
 478          */
 479         if ((!skb->dev || skb->dev->flags & IFF_LOOPBACK)) {
 480                 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
 481                 cp = pp->conn_in_get(svc->ipvs, svc->af, skb, iph);
 482                 iph->hdr_flags ^= IP_VS_HDR_INVERSE;
 483
 484                 if (cp) {
 485                         IP_VS_DBG_PKT(12, svc->af, pp, skb, iph->off,
 486                                       "Not scheduling reply for existing"
 487                                       " connection");
 488                         __ip_vs_conn_put(cp);
 489                         return NULL;
 490                 }
 491         }
 492
 493         /*
 494          *    Persistent service
 495          */
 496         if (svc->flags & IP_VS_SVC_F_PERSISTENT)
 497                 return ip_vs_sched_persist(svc, skb, cport, vport, ignored,
 498                                            iph);
 499
 500         *ignored = 0;
 501
 502         /*
 503          *    Non-persistent service
 504          */
 505         if (!svc->fwmark && vport != svc->port) {
 506                 if (!svc->port)
 507                         pr_err("Schedule: port zero only supported "
 508                                "in persistent services, "
 509                                "check your ipvs configuration\n");
 510                 return NULL;
 511         }
 512
 513         sched = rcu_dereference(svc->scheduler);
 514         if (sched) {
 515                 /* read svc->sched_data after svc->scheduler */
 516                 smp_rmb();
 517                 dest = sched->schedule(svc, skb, iph);
 518         } else {
 519                 dest = NULL;
 520         }
 521         if (dest == NULL) {
 522                 IP_VS_DBG(1, "Schedule: no dest found.\n");
 523                 return NULL;
 524         }
 525
 526         flags = (svc->flags & IP_VS_SVC_F_ONEPACKET
 527                  && iph->protocol == IPPROTO_UDP) ?
 528                 IP_VS_CONN_F_ONE_PACKET : 0;
 529
 530         /*
 531          *    Create a connection entry.
 532          */
 533         {
 534                 struct ip_vs_conn_param p;
 535
 536                 ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
 537                                       caddr, cport, vaddr, vport, &p);
 538                 cp = ip_vs_conn_new(&p, dest->af, &dest->addr,
 539                                     dest->port ? dest->port : vport,
 540                                     flags, dest, skb->mark);
 541                 if (!cp) {
 542                         *ignored = -1;
 543                         return NULL;
 544                 }
 545         }
 546
 547         IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
 548                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
 549                       ip_vs_fwd_tag(cp),
 550                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 551                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 552                       IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
 553                       cp->flags, refcount_read(&cp->refcnt));
 554
 555         ip_vs_conn_stats(cp, svc);
 556         return cp;
 557 }
 558
 559 static inline int ip_vs_addr_is_unicast(struct net *net, int af,
 560                                         union nf_inet_addr *addr)
 561 {
 562 #ifdef CONFIG_IP_VS_IPV6
 563         if (af == AF_INET6)
 564                 return ipv6_addr_type(&addr->in6) & IPV6_ADDR_UNICAST;
 565 #endif
 566         return (inet_addr_type(net, addr->ip) == RTN_UNICAST);
 567 }
 568
 569 /*
 570  *  Pass or drop the packet.
 571  *  Called by ip_vs_in, when the virtual service is available but
 572  *  no destination is available for a new connection.
 573  */
 574 int ip_vs_leave(struct ip_vs_service *svc, struct sk_buff *skb,
 575                 struct ip_vs_proto_data *pd, struct ip_vs_iphdr *iph)
 576 {
 577         __be16 _ports[2], *pptr, dport;
 578         struct netns_ipvs *ipvs = svc->ipvs;
 579         struct net *net = ipvs->net;
 580
 581         pptr = frag_safe_skb_hp(skb, iph->len, sizeof(_ports), _ports);
 582         if (!pptr)
 583                 return NF_DROP;
 584         dport = likely(!ip_vs_iph_inverse(iph)) ? pptr[1] : pptr[0];
 585
 586         /* if it is fwmark-based service, the cache_bypass sysctl is up
 587            and the destination is a non-local unicast, then create
 588            a cache_bypass connection entry */
 589         if (sysctl_cache_bypass(ipvs) && svc->fwmark &&
 590             !(iph->hdr_flags & (IP_VS_HDR_INVERSE | IP_VS_HDR_ICMP)) &&
 591             ip_vs_addr_is_unicast(net, svc->af, &iph->daddr)) {
 592                 int ret;
 593                 struct ip_vs_conn *cp;
 594                 unsigned int flags = (svc->flags & IP_VS_SVC_F_ONEPACKET &&
 595                                       iph->protocol == IPPROTO_UDP) ?
 596                                       IP_VS_CONN_F_ONE_PACKET : 0;
 597                 union nf_inet_addr daddr =  { .all = { 0, 0, 0, 0 } };
 598
 599                 /* create a new connection entry */
 600                 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__);
 601                 {
 602                         struct ip_vs_conn_param p;
 603                         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
 604                                               &iph->saddr, pptr[0],
 605                                               &iph->daddr, pptr[1], &p);
 606                         cp = ip_vs_conn_new(&p, svc->af, &daddr, 0,
 607                                             IP_VS_CONN_F_BYPASS | flags,
 608                                             NULL, skb->mark);
 609                         if (!cp)
 610                                 return NF_DROP;
 611                 }
 612
 613                 /* statistics */
 614                 ip_vs_in_stats(cp, skb);
 615
 616                 /* set state */
 617                 ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
 618
 619                 /* transmit the first SYN packet */
 620                 ret = cp->packet_xmit(skb, cp, pd->pp, iph);
 621                 /* do not touch skb anymore */
 622
 623                 if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) && cp->control)
 624                         atomic_inc(&cp->control->in_pkts);
 625                 else
 626                         atomic_inc(&cp->in_pkts);
 627                 ip_vs_conn_put(cp);
 628                 return ret;
 629         }
 630
 631         /*
 632          * When the virtual ftp service is presented, packets destined
 633          * for other services on the VIP may get here (except services
 634          * listed in the ipvs table), pass the packets, because it is
 635          * not ipvs job to decide to drop the packets.
 636          */
 637         if (svc->port == FTPPORT && dport != FTPPORT)
 638                 return NF_ACCEPT;
 639
 640         if (unlikely(ip_vs_iph_icmp(iph)))
 641                 return NF_DROP;
 642
 643         /*
 644          * Notify the client that the destination is unreachable, and
 645          * release the socket buffer.
 646          * Since it is in IP layer, the TCP socket is not actually
 647          * created, the TCP RST packet cannot be sent, instead that
 648          * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
 649          */
 650 #ifdef CONFIG_IP_VS_IPV6
 651         if (svc->af == AF_INET6) {
 652                 if (!skb->dev)
 653                         skb->dev = net->loopback_dev;
 654                 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 655         } else
 656 #endif
 657                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 658
 659         return NF_DROP;
 660 }
 661
 662 #ifdef CONFIG_SYSCTL
 663
 664 static int sysctl_snat_reroute(struct netns_ipvs *ipvs)
 665 {
 666         return ipvs->sysctl_snat_reroute;
 667 }
 668
 669 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs)
 670 {
 671         return ipvs->sysctl_nat_icmp_send;
 672 }
 673
 674 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs)
 675 {
 676         return ipvs->sysctl_expire_nodest_conn;
 677 }
 678
 679 #else
 680
 681 static int sysctl_snat_reroute(struct netns_ipvs *ipvs) { return 0; }
 682 static int sysctl_nat_icmp_send(struct netns_ipvs *ipvs) { return 0; }
 683 static int sysctl_expire_nodest_conn(struct netns_ipvs *ipvs) { return 0; }
 684
 685 #endif
 686
 687 __sum16 ip_vs_checksum_complete(struct sk_buff *skb, int offset)
 688 {
 689         return csum_fold(skb_checksum(skb, offset, skb->len - offset, 0));
 690 }
 691
 692 static inline enum ip_defrag_users ip_vs_defrag_user(unsigned int hooknum)
 693 {
 694         if (NF_INET_LOCAL_IN == hooknum)
 695                 return IP_DEFRAG_VS_IN;
 696         if (NF_INET_FORWARD == hooknum)
 697                 return IP_DEFRAG_VS_FWD;
 698         return IP_DEFRAG_VS_OUT;
 699 }
 700
 701 static inline int ip_vs_gather_frags(struct netns_ipvs *ipvs,
 702                                      struct sk_buff *skb, u_int32_t user)
 703 {
 704         int err;
 705
 706         local_bh_disable();
 707         err = ip_defrag(ipvs->net, skb, user);
 708         local_bh_enable();
 709         if (!err)
 710                 ip_send_check(ip_hdr(skb));
 711
 712         return err;
 713 }
 714
 715 static int ip_vs_route_me_harder(struct netns_ipvs *ipvs, int af,
 716                                  struct sk_buff *skb, unsigned int hooknum)
 717 {
 718         if (!sysctl_snat_reroute(ipvs))
 719                 return 0;
 720         /* Reroute replies only to remote clients (FORWARD and LOCAL_OUT) */
 721         if (NF_INET_LOCAL_IN == hooknum)
 722                 return 0;
 723 #ifdef CONFIG_IP_VS_IPV6
 724         if (af == AF_INET6) {
 725                 struct dst_entry *dst = skb_dst(skb);
 726
 727                 if (dst->dev && !(dst->dev->flags & IFF_LOOPBACK) &&
 728                     ip6_route_me_harder(ipvs->net, skb->sk, skb) != 0)
 729                         return 1;
 730         } else
 731 #endif
 732                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL) &&
 733                     ip_route_me_harder(ipvs->net, skb->sk, skb, RTN_LOCAL) != 0)
 734                         return 1;
 735
 736         return 0;
 737 }
 738
 739 /*
 740  * Packet has been made sufficiently writable in caller
 741  * - inout: 1=in->out, 0=out->in
 742  */
 743 void ip_vs_nat_icmp(struct sk_buff *skb, struct ip_vs_protocol *pp,
 744                     struct ip_vs_conn *cp, int inout)
 745 {
 746         struct iphdr *iph        = ip_hdr(skb);
 747         unsigned int icmp_offset = iph->ihl*4;
 748         struct icmphdr *icmph    = (struct icmphdr *)(skb_network_header(skb) +
 749                                                       icmp_offset);
 750         struct iphdr *ciph       = (struct iphdr *)(icmph + 1);
 751
 752         if (inout) {
 753                 iph->saddr = cp->vaddr.ip;
 754                 ip_send_check(iph);
 755                 ciph->daddr = cp->vaddr.ip;
 756                 ip_send_check(ciph);
 757         } else {
 758                 iph->daddr = cp->daddr.ip;
 759                 ip_send_check(iph);
 760                 ciph->saddr = cp->daddr.ip;
 761                 ip_send_check(ciph);
 762         }
 763
 764         /* the TCP/UDP/SCTP port */
 765         if (IPPROTO_TCP == ciph->protocol || IPPROTO_UDP == ciph->protocol ||
 766             IPPROTO_SCTP == ciph->protocol) {
 767                 __be16 *ports = (void *)ciph + ciph->ihl*4;
 768
 769                 if (inout)
 770                         ports[1] = cp->vport;
 771                 else
 772                         ports[0] = cp->dport;
 773         }
 774
 775         /* And finally the ICMP checksum */
 776         icmph->checksum = 0;
 777         icmph->checksum = ip_vs_checksum_complete(skb, icmp_offset);
 778         skb->ip_summed = CHECKSUM_UNNECESSARY;
 779
 780         if (inout)
 781                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 782                         "Forwarding altered outgoing ICMP");
 783         else
 784                 IP_VS_DBG_PKT(11, AF_INET, pp, skb, (void *)ciph - (void *)iph,
 785                         "Forwarding altered incoming ICMP");
 786 }
 787
 788 #ifdef CONFIG_IP_VS_IPV6
 789 void ip_vs_nat_icmp_v6(struct sk_buff *skb, struct ip_vs_protocol *pp,
 790                     struct ip_vs_conn *cp, int inout)
 791 {
 792         struct ipv6hdr *iph      = ipv6_hdr(skb);
 793         unsigned int icmp_offset = 0;
 794         unsigned int offs        = 0; /* header offset*/
 795         int protocol;
 796         struct icmp6hdr *icmph;
 797         struct ipv6hdr *ciph;
 798         unsigned short fragoffs;
 799
 800         ipv6_find_hdr(skb, &icmp_offset, IPPROTO_ICMPV6, &fragoffs, NULL);
 801         icmph = (struct icmp6hdr *)(skb_network_header(skb) + icmp_offset);
 802         offs = icmp_offset + sizeof(struct icmp6hdr);
 803         ciph = (struct ipv6hdr *)(skb_network_header(skb) + offs);
 804
 805         protocol = ipv6_find_hdr(skb, &offs, -1, &fragoffs, NULL);
 806
 807         if (inout) {
 808                 iph->saddr = cp->vaddr.in6;
 809                 ciph->daddr = cp->vaddr.in6;
 810         } else {
 811                 iph->daddr = cp->daddr.in6;
 812                 ciph->saddr = cp->daddr.in6;
 813         }
 814
 815         /* the TCP/UDP/SCTP port */
 816         if (!fragoffs && (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 817                           IPPROTO_SCTP == protocol)) {
 818                 __be16 *ports = (void *)(skb_network_header(skb) + offs);
 819
 820                 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__,
 821                               ntohs(inout ? ports[1] : ports[0]),
 822                               ntohs(inout ? cp->vport : cp->dport));
 823                 if (inout)
 824                         ports[1] = cp->vport;
 825                 else
 826                         ports[0] = cp->dport;
 827         }
 828
 829         /* And finally the ICMP checksum */
 830         icmph->icmp6_cksum = ~csum_ipv6_magic(&iph->saddr, &iph->daddr,
 831                                               skb->len - icmp_offset,
 832                                               IPPROTO_ICMPV6, 0);
 833         skb->csum_start = skb_network_header(skb) - skb->head + icmp_offset;
 834         skb->csum_offset = offsetof(struct icmp6hdr, icmp6_cksum);
 835         skb->ip_summed = CHECKSUM_PARTIAL;
 836
 837         if (inout)
 838                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 839                               (void *)ciph - (void *)iph,
 840                               "Forwarding altered outgoing ICMPv6");
 841         else
 842                 IP_VS_DBG_PKT(11, AF_INET6, pp, skb,
 843                               (void *)ciph - (void *)iph,
 844                               "Forwarding altered incoming ICMPv6");
 845 }
 846 #endif
 847
 848 /* Handle relevant response ICMP messages - forward to the right
 849  * destination host.
 850  */
 851 static int handle_response_icmp(int af, struct sk_buff *skb,
 852                                 union nf_inet_addr *snet,
 853                                 __u8 protocol, struct ip_vs_conn *cp,
 854                                 struct ip_vs_protocol *pp,
 855                                 unsigned int offset, unsigned int ihl,
 856                                 unsigned int hooknum)
 857 {
 858         unsigned int verdict = NF_DROP;
 859
 860         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
 861                 goto ignore_cp;
 862
 863         /* Ensure the checksum is correct */
 864         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
 865                 /* Failed checksum! */
 866                 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
 867                               IP_VS_DBG_ADDR(af, snet));
 868                 goto out;
 869         }
 870
 871         if (IPPROTO_TCP == protocol || IPPROTO_UDP == protocol ||
 872             IPPROTO_SCTP == protocol)
 873                 offset += 2 * sizeof(__u16);
 874         if (!skb_make_writable(skb, offset))
 875                 goto out;
 876
 877 #ifdef CONFIG_IP_VS_IPV6
 878         if (af == AF_INET6)
 879                 ip_vs_nat_icmp_v6(skb, pp, cp, 1);
 880         else
 881 #endif
 882                 ip_vs_nat_icmp(skb, pp, cp, 1);
 883
 884         if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
 885                 goto out;
 886
 887         /* do the statistics and put it back */
 888         ip_vs_out_stats(cp, skb);
 889
 890         skb->ipvs_property = 1;
 891         if (!(cp->flags & IP_VS_CONN_F_NFCT))
 892                 ip_vs_notrack(skb);
 893         else
 894                 ip_vs_update_conntrack(skb, cp, 0);
 895
 896 ignore_cp:
 897         verdict = NF_ACCEPT;
 898
 899 out:
 900         __ip_vs_conn_put(cp);
 901
 902         return verdict;
 903 }
 904
 905 /*
 906  *      Handle ICMP messages in the inside-to-outside direction (outgoing).
 907  *      Find any that might be relevant, check against existing connections.
 908  *      Currently handles error types - unreachable, quench, ttl exceeded.
 909  */
 910 static int ip_vs_out_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb,
 911                           int *related, unsigned int hooknum)
 912 {
 913         struct iphdr *iph;
 914         struct icmphdr  _icmph, *ic;
 915         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
 916         struct ip_vs_iphdr ciph;
 917         struct ip_vs_conn *cp;
 918         struct ip_vs_protocol *pp;
 919         unsigned int offset, ihl;
 920         union nf_inet_addr snet;
 921
 922         *related = 1;
 923
 924         /* reassemble IP fragments */
 925         if (ip_is_fragment(ip_hdr(skb))) {
 926                 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
 927                         return NF_STOLEN;
 928         }
 929
 930         iph = ip_hdr(skb);
 931         offset = ihl = iph->ihl * 4;
 932         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
 933         if (ic == NULL)
 934                 return NF_DROP;
 935
 936         IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
 937                   ic->type, ntohs(icmp_id(ic)),
 938                   &iph->saddr, &iph->daddr);
 939
 940         /*
 941          * Work through seeing if this is for us.
 942          * These checks are supposed to be in an order that means easy
 943          * things are checked first to speed up processing.... however
 944          * this means that some packets will manage to get a long way
 945          * down this stack and then be rejected, but that's life.
 946          */
 947         if ((ic->type != ICMP_DEST_UNREACH) &&
 948             (ic->type != ICMP_SOURCE_QUENCH) &&
 949             (ic->type != ICMP_TIME_EXCEEDED)) {
 950                 *related = 0;
 951                 return NF_ACCEPT;
 952         }
 953
 954         /* Now find the contained IP header */
 955         offset += sizeof(_icmph);
 956         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
 957         if (cih == NULL)
 958                 return NF_ACCEPT; /* The packet looks wrong, ignore */
 959
 960         pp = ip_vs_proto_get(cih->protocol);
 961         if (!pp)
 962                 return NF_ACCEPT;
 963
 964         /* Is the embedded protocol header present? */
 965         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
 966                      pp->dont_defrag))
 967                 return NF_ACCEPT;
 968
 969         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
 970                       "Checking outgoing ICMP for");
 971
 972         ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, true, &ciph);
 973
 974         /* The embedded headers contain source and dest in reverse order */
 975         cp = pp->conn_out_get(ipvs, AF_INET, skb, &ciph);
 976         if (!cp)
 977                 return NF_ACCEPT;
 978
 979         snet.ip = iph->saddr;
 980         return handle_response_icmp(AF_INET, skb, &snet, cih->protocol, cp,
 981                                     pp, ciph.len, ihl, hooknum);
 982 }
 983
 984 #ifdef CONFIG_IP_VS_IPV6
 985 static int ip_vs_out_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
 986                              int *related,  unsigned int hooknum,
 987                              struct ip_vs_iphdr *ipvsh)
 988 {
 989         struct icmp6hdr _icmph, *ic;
 990         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
 991         struct ip_vs_conn *cp;
 992         struct ip_vs_protocol *pp;
 993         union nf_inet_addr snet;
 994         unsigned int offset;
 995
 996         *related = 1;
 997         ic = frag_safe_skb_hp(skb, ipvsh->len, sizeof(_icmph), &_icmph);
 998         if (ic == NULL)
 999                 return NF_DROP;
1000
1001         /*
1002          * Work through seeing if this is for us.
1003          * These checks are supposed to be in an order that means easy
1004          * things are checked first to speed up processing.... however
1005          * this means that some packets will manage to get a long way
1006          * down this stack and then be rejected, but that's life.
1007          */
1008         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1009                 *related = 0;
1010                 return NF_ACCEPT;
1011         }
1012         /* Fragment header that is before ICMP header tells us that:
1013          * it's not an error message since they can't be fragmented.
1014          */
1015         if (ipvsh->flags & IP6_FH_F_FRAG)
1016                 return NF_DROP;
1017
1018         IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1019                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1020                   &ipvsh->saddr, &ipvsh->daddr);
1021
1022         if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, ipvsh->len + sizeof(_icmph),
1023                                      true, &ciph))
1024                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1025
1026         pp = ip_vs_proto_get(ciph.protocol);
1027         if (!pp)
1028                 return NF_ACCEPT;
1029
1030         /* The embedded headers contain source and dest in reverse order */
1031         cp = pp->conn_out_get(ipvs, AF_INET6, skb, &ciph);
1032         if (!cp)
1033                 return NF_ACCEPT;
1034
1035         snet.in6 = ciph.saddr.in6;
1036         offset = ciph.len;
1037         return handle_response_icmp(AF_INET6, skb, &snet, ciph.protocol, cp,
1038                                     pp, offset, sizeof(struct ipv6hdr),
1039                                     hooknum);
1040 }
1041 #endif
1042
1043 /*
1044  * Check if sctp chunc is ABORT chunk
1045  */
1046 static inline int is_sctp_abort(const struct sk_buff *skb, int nh_len)
1047 {
1048         struct sctp_chunkhdr *sch, schunk;
1049         sch = skb_header_pointer(skb, nh_len + sizeof(struct sctphdr),
1050                                  sizeof(schunk), &schunk);
1051         if (sch == NULL)
1052                 return 0;
1053         if (sch->type == SCTP_CID_ABORT)
1054                 return 1;
1055         return 0;
1056 }
1057
1058 static inline int is_tcp_reset(const struct sk_buff *skb, int nh_len)
1059 {
1060         struct tcphdr _tcph, *th;
1061
1062         th = skb_header_pointer(skb, nh_len, sizeof(_tcph), &_tcph);
1063         if (th == NULL)
1064                 return 0;
1065         return th->rst;
1066 }
1067
1068 static inline bool is_new_conn(const struct sk_buff *skb,
1069                                struct ip_vs_iphdr *iph)
1070 {
1071         switch (iph->protocol) {
1072         case IPPROTO_TCP: {
1073                 struct tcphdr _tcph, *th;
1074
1075                 th = skb_header_pointer(skb, iph->len, sizeof(_tcph), &_tcph);
1076                 if (th == NULL)
1077                         return false;
1078                 return th->syn;
1079         }
1080         case IPPROTO_SCTP: {
1081                 struct sctp_chunkhdr *sch, schunk;
1082
1083                 sch = skb_header_pointer(skb, iph->len + sizeof(struct sctphdr),
1084                                          sizeof(schunk), &schunk);
1085                 if (sch == NULL)
1086                         return false;
1087                 return sch->type == SCTP_CID_INIT;
1088         }
1089         default:
1090                 return false;
1091         }
1092 }
1093
1094 static inline bool is_new_conn_expected(const struct ip_vs_conn *cp,
1095                                         int conn_reuse_mode)
1096 {
1097         /* Controlled (FTP DATA or persistence)? */
1098         if (cp->control)
1099                 return false;
1100
1101         switch (cp->protocol) {
1102         case IPPROTO_TCP:
1103                 return (cp->state == IP_VS_TCP_S_TIME_WAIT) ||
1104                        (cp->state == IP_VS_TCP_S_CLOSE) ||
1105                         ((conn_reuse_mode & 2) &&
1106                          (cp->state == IP_VS_TCP_S_FIN_WAIT) &&
1107                          (cp->flags & IP_VS_CONN_F_NOOUTPUT));
1108         case IPPROTO_SCTP:
1109                 return cp->state == IP_VS_SCTP_S_CLOSED;
1110         default:
1111                 return false;
1112         }
1113 }
1114
1115 /* Generic function to create new connections for outgoing RS packets
1116  *
1117  * Pre-requisites for successful connection creation:
1118  * 1) Virtual Service is NOT fwmark based:
1119  *    In fwmark-VS actual vaddr and vport are unknown to IPVS
1120  * 2) Real Server and Virtual Service were NOT configured without port:
1121  *    This is to allow match of different VS to the same RS ip-addr
1122  */
1123 struct ip_vs_conn *ip_vs_new_conn_out(struct ip_vs_service *svc,
1124                                       struct ip_vs_dest *dest,
1125                                       struct sk_buff *skb,
1126                                       const struct ip_vs_iphdr *iph,
1127                                       __be16 dport,
1128                                       __be16 cport)
1129 {
1130         struct ip_vs_conn_param param;
1131         struct ip_vs_conn *ct = NULL, *cp = NULL;
1132         const union nf_inet_addr *vaddr, *daddr, *caddr;
1133         union nf_inet_addr snet;
1134         __be16 vport;
1135         unsigned int flags;
1136
1137         EnterFunction(12);
1138         vaddr = &svc->addr;
1139         vport = svc->port;
1140         daddr = &iph->saddr;
1141         caddr = &iph->daddr;
1142
1143         /* check pre-requisites are satisfied */
1144         if (svc->fwmark)
1145                 return NULL;
1146         if (!vport || !dport)
1147                 return NULL;
1148
1149         /* for persistent service first create connection template */
1150         if (svc->flags & IP_VS_SVC_F_PERSISTENT) {
1151                 /* apply netmask the same way ingress-side does */
1152 #ifdef CONFIG_IP_VS_IPV6
1153                 if (svc->af == AF_INET6)
1154                         ipv6_addr_prefix(&snet.in6, &caddr->in6,
1155                                          (__force __u32)svc->netmask);
1156                 else
1157 #endif
1158                         snet.ip = caddr->ip & svc->netmask;
1159                 /* fill params and create template if not existent */
1160                 if (ip_vs_conn_fill_param_persist(svc, skb, iph->protocol,
1161                                                   &snet, 0, vaddr,
1162                                                   vport, &param) < 0)
1163                         return NULL;
1164                 ct = ip_vs_ct_in_get(&param);
1165                 /* check if template exists and points to the same dest */
1166                 if (!ct || !ip_vs_check_template(ct, dest)) {
1167                         ct = ip_vs_conn_new(&param, dest->af, daddr, dport,
1168                                             IP_VS_CONN_F_TEMPLATE, dest, 0);
1169                         if (!ct) {
1170                                 kfree(param.pe_data);
1171                                 return NULL;
1172                         }
1173                         ct->timeout = svc->timeout;
1174                 } else {
1175                         kfree(param.pe_data);
1176                 }
1177         }
1178
1179         /* connection flags */
1180         flags = ((svc->flags & IP_VS_SVC_F_ONEPACKET) &&
1181                  iph->protocol == IPPROTO_UDP) ? IP_VS_CONN_F_ONE_PACKET : 0;
1182         /* create connection */
1183         ip_vs_conn_fill_param(svc->ipvs, svc->af, iph->protocol,
1184                               caddr, cport, vaddr, vport, &param);
1185         cp = ip_vs_conn_new(&param, dest->af, daddr, dport, flags, dest, 0);
1186         if (!cp) {
1187                 if (ct)
1188                         ip_vs_conn_put(ct);
1189                 return NULL;
1190         }
1191         if (ct) {
1192                 ip_vs_control_add(cp, ct);
1193                 ip_vs_conn_put(ct);
1194         }
1195         ip_vs_conn_stats(cp, svc);
1196
1197         /* return connection (will be used to handle outgoing packet) */
1198         IP_VS_DBG_BUF(6, "New connection RS-initiated:%c c:%s:%u v:%s:%u "
1199                       "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
1200                       ip_vs_fwd_tag(cp),
1201                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
1202                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
1203                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
1204                       cp->flags, refcount_read(&cp->refcnt));
1205         LeaveFunction(12);
1206         return cp;
1207 }
1208
1209 /* Handle outgoing packets which are considered requests initiated by
1210  * real servers, so that subsequent responses from external client can be
1211  * routed to the right real server.
1212  * Used also for outgoing responses in OPS mode.
1213  *
1214  * Connection management is handled by persistent-engine specific callback.
1215  */
1216 static struct ip_vs_conn *__ip_vs_rs_conn_out(unsigned int hooknum,
1217                                               struct netns_ipvs *ipvs,
1218                                               int af, struct sk_buff *skb,
1219                                               const struct ip_vs_iphdr *iph)
1220 {
1221         struct ip_vs_dest *dest;
1222         struct ip_vs_conn *cp = NULL;
1223         __be16 _ports[2], *pptr;
1224
1225         if (hooknum == NF_INET_LOCAL_IN)
1226                 return NULL;
1227
1228         pptr = frag_safe_skb_hp(skb, iph->len,
1229                                 sizeof(_ports), _ports);
1230         if (!pptr)
1231                 return NULL;
1232
1233         dest = ip_vs_find_real_service(ipvs, af, iph->protocol,
1234                                        &iph->saddr, pptr[0]);
1235         if (dest) {
1236                 struct ip_vs_service *svc;
1237                 struct ip_vs_pe *pe;
1238
1239                 svc = rcu_dereference(dest->svc);
1240                 if (svc) {
1241                         pe = rcu_dereference(svc->pe);
1242                         if (pe && pe->conn_out)
1243                                 cp = pe->conn_out(svc, dest, skb, iph,
1244                                                   pptr[0], pptr[1]);
1245                 }
1246         }
1247
1248         return cp;
1249 }
1250
1251 /* Handle response packets: rewrite addresses and send away...
1252  */
1253 static unsigned int
1254 handle_response(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
1255                 struct ip_vs_conn *cp, struct ip_vs_iphdr *iph,
1256                 unsigned int hooknum)
1257 {
1258         struct ip_vs_protocol *pp = pd->pp;
1259
1260         IP_VS_DBG_PKT(11, af, pp, skb, iph->off, "Outgoing packet");
1261
1262         if (!skb_make_writable(skb, iph->len))
1263                 goto drop;
1264
1265         /* mangle the packet */
1266         if (pp->snat_handler && !pp->snat_handler(skb, pp, cp, iph))
1267                 goto drop;
1268
1269 #ifdef CONFIG_IP_VS_IPV6
1270         if (af == AF_INET6)
1271                 ipv6_hdr(skb)->saddr = cp->vaddr.in6;
1272         else
1273 #endif
1274         {
1275                 ip_hdr(skb)->saddr = cp->vaddr.ip;
1276                 ip_send_check(ip_hdr(skb));
1277         }
1278
1279         /*
1280          * nf_iterate does not expect change in the skb->dst->dev.
1281          * It looks like it is not fatal to enable this code for hooks
1282          * where our handlers are at the end of the chain list and
1283          * when all next handlers use skb->dst->dev and not outdev.
1284          * It will definitely route properly the inout NAT traffic
1285          * when multiple paths are used.
1286          */
1287
1288         /* For policy routing, packets originating from this
1289          * machine itself may be routed differently to packets
1290          * passing through.  We want this packet to be routed as
1291          * if it came from this machine itself.  So re-compute
1292          * the routing information.
1293          */
1294         if (ip_vs_route_me_harder(cp->ipvs, af, skb, hooknum))
1295                 goto drop;
1296
1297         IP_VS_DBG_PKT(10, af, pp, skb, iph->off, "After SNAT");
1298
1299         ip_vs_out_stats(cp, skb);
1300         ip_vs_set_state(cp, IP_VS_DIR_OUTPUT, skb, pd);
1301         skb->ipvs_property = 1;
1302         if (!(cp->flags & IP_VS_CONN_F_NFCT))
1303                 ip_vs_notrack(skb);
1304         else
1305                 ip_vs_update_conntrack(skb, cp, 0);
1306         ip_vs_conn_put(cp);
1307
1308         LeaveFunction(11);
1309         return NF_ACCEPT;
1310
1311 drop:
1312         ip_vs_conn_put(cp);
1313         kfree_skb(skb);
1314         LeaveFunction(11);
1315         return NF_STOLEN;
1316 }
1317
1318 /*
1319  *      Check if outgoing packet belongs to the established ip_vs_conn.
1320  */
1321 static unsigned int
1322 ip_vs_out(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1323 {
1324         struct ip_vs_iphdr iph;
1325         struct ip_vs_protocol *pp;
1326         struct ip_vs_proto_data *pd;
1327         struct ip_vs_conn *cp;
1328         struct sock *sk;
1329
1330         EnterFunction(11);
1331
1332         /* Already marked as IPVS request or reply? */
1333         if (skb->ipvs_property)
1334                 return NF_ACCEPT;
1335
1336         sk = skb_to_full_sk(skb);
1337         /* Bad... Do not break raw sockets */
1338         if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
1339                      af == AF_INET)) {
1340
1341                 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
1342                         return NF_ACCEPT;
1343         }
1344
1345         if (unlikely(!skb_dst(skb)))
1346                 return NF_ACCEPT;
1347
1348         if (!ipvs->enable)
1349                 return NF_ACCEPT;
1350
1351         ip_vs_fill_iph_skb(af, skb, false, &iph);
1352 #ifdef CONFIG_IP_VS_IPV6
1353         if (af == AF_INET6) {
1354                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1355                         int related;
1356                         int verdict = ip_vs_out_icmp_v6(ipvs, skb, &related,
1357                                                         hooknum, &iph);
1358
1359                         if (related)
1360                                 return verdict;
1361                 }
1362         } else
1363 #endif
1364                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1365                         int related;
1366                         int verdict = ip_vs_out_icmp(ipvs, skb, &related, hooknum);
1367
1368                         if (related)
1369                                 return verdict;
1370                 }
1371
1372         pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1373         if (unlikely(!pd))
1374                 return NF_ACCEPT;
1375         pp = pd->pp;
1376
1377         /* reassemble IP fragments */
1378 #ifdef CONFIG_IP_VS_IPV6
1379         if (af == AF_INET)
1380 #endif
1381                 if (unlikely(ip_is_fragment(ip_hdr(skb)) && !pp->dont_defrag)) {
1382                         if (ip_vs_gather_frags(ipvs, skb,
1383                                                ip_vs_defrag_user(hooknum)))
1384                                 return NF_STOLEN;
1385
1386                         ip_vs_fill_iph_skb(AF_INET, skb, false, &iph);
1387                 }
1388
1389         /*
1390          * Check if the packet belongs to an existing entry
1391          */
1392         cp = pp->conn_out_get(ipvs, af, skb, &iph);
1393
1394         if (likely(cp)) {
1395                 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
1396                         goto ignore_cp;
1397                 return handle_response(af, skb, pd, cp, &iph, hooknum);
1398         }
1399
1400         /* Check for real-server-started requests */
1401         if (atomic_read(&ipvs->conn_out_counter)) {
1402                 /* Currently only for UDP:
1403                  * connection oriented protocols typically use
1404                  * ephemeral ports for outgoing connections, so
1405                  * related incoming responses would not match any VS
1406                  */
1407                 if (pp->protocol == IPPROTO_UDP) {
1408                         cp = __ip_vs_rs_conn_out(hooknum, ipvs, af, skb, &iph);
1409                         if (likely(cp))
1410                                 return handle_response(af, skb, pd, cp, &iph,
1411                                                        hooknum);
1412                 }
1413         }
1414
1415         if (sysctl_nat_icmp_send(ipvs) &&
1416             (pp->protocol == IPPROTO_TCP ||
1417              pp->protocol == IPPROTO_UDP ||
1418              pp->protocol == IPPROTO_SCTP)) {
1419                 __be16 _ports[2], *pptr;
1420
1421                 pptr = frag_safe_skb_hp(skb, iph.len,
1422                                          sizeof(_ports), _ports);
1423                 if (pptr == NULL)
1424                         return NF_ACCEPT;       /* Not for me */
1425                 if (ip_vs_has_real_service(ipvs, af, iph.protocol, &iph.saddr,
1426                                            pptr[0])) {
1427                         /*
1428                          * Notify the real server: there is no
1429                          * existing entry if it is not RST
1430                          * packet or not TCP packet.
1431                          */
1432                         if ((iph.protocol != IPPROTO_TCP &&
1433                              iph.protocol != IPPROTO_SCTP)
1434                              || ((iph.protocol == IPPROTO_TCP
1435                                   && !is_tcp_reset(skb, iph.len))
1436                                  || (iph.protocol == IPPROTO_SCTP
1437                                         && !is_sctp_abort(skb,
1438                                                 iph.len)))) {
1439 #ifdef CONFIG_IP_VS_IPV6
1440                                 if (af == AF_INET6) {
1441                                         if (!skb->dev)
1442                                                 skb->dev = ipvs->net->loopback_dev;
1443                                         icmpv6_send(skb,
1444                                                     ICMPV6_DEST_UNREACH,
1445                                                     ICMPV6_PORT_UNREACH,
1446                                                     0);
1447                                 } else
1448 #endif
1449                                         icmp_send(skb,
1450                                                   ICMP_DEST_UNREACH,
1451                                                   ICMP_PORT_UNREACH, 0);
1452                                 return NF_DROP;
1453                         }
1454                 }
1455         }
1456
1457 out:
1458         IP_VS_DBG_PKT(12, af, pp, skb, iph.off,
1459                       "ip_vs_out: packet continues traversal as normal");
1460         return NF_ACCEPT;
1461
1462 ignore_cp:
1463         __ip_vs_conn_put(cp);
1464         goto out;
1465 }
1466
1467 /*
1468  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1469  *      used only for VS/NAT.
1470  *      Check if packet is reply for established ip_vs_conn.
1471  */
1472 static unsigned int
1473 ip_vs_reply4(void *priv, struct sk_buff *skb,
1474              const struct nf_hook_state *state)
1475 {
1476         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1477 }
1478
1479 /*
1480  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1481  *      Check if packet is reply for established ip_vs_conn.
1482  */
1483 static unsigned int
1484 ip_vs_local_reply4(void *priv, struct sk_buff *skb,
1485                    const struct nf_hook_state *state)
1486 {
1487         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET);
1488 }
1489
1490 #ifdef CONFIG_IP_VS_IPV6
1491
1492 /*
1493  *      It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1494  *      used only for VS/NAT.
1495  *      Check if packet is reply for established ip_vs_conn.
1496  */
1497 static unsigned int
1498 ip_vs_reply6(void *priv, struct sk_buff *skb,
1499              const struct nf_hook_state *state)
1500 {
1501         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1502 }
1503
1504 /*
1505  *      It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1506  *      Check if packet is reply for established ip_vs_conn.
1507  */
1508 static unsigned int
1509 ip_vs_local_reply6(void *priv, struct sk_buff *skb,
1510                    const struct nf_hook_state *state)
1511 {
1512         return ip_vs_out(net_ipvs(state->net), state->hook, skb, AF_INET6);
1513 }
1514
1515 #endif
1516
1517 static unsigned int
1518 ip_vs_try_to_schedule(struct netns_ipvs *ipvs, int af, struct sk_buff *skb,
1519                       struct ip_vs_proto_data *pd,
1520                       int *verdict, struct ip_vs_conn **cpp,
1521                       struct ip_vs_iphdr *iph)
1522 {
1523         struct ip_vs_protocol *pp = pd->pp;
1524
1525         if (!iph->fragoffs) {
1526                 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1527                  * replayed fragment zero will already have created the cp
1528                  */
1529
1530                 /* Schedule and create new connection entry into cpp */
1531                 if (!pp->conn_schedule(ipvs, af, skb, pd, verdict, cpp, iph))
1532                         return 0;
1533         }
1534
1535         if (unlikely(!*cpp)) {
1536                 /* sorry, all this trouble for a no-hit :) */
1537                 IP_VS_DBG_PKT(12, af, pp, skb, iph->off,
1538                               "ip_vs_in: packet continues traversal as normal");
1539
1540                 /* Fragment couldn't be mapped to a conn entry */
1541                 if (iph->fragoffs)
1542                         IP_VS_DBG_PKT(7, af, pp, skb, iph->off,
1543                                       "unhandled fragment");
1544
1545                 *verdict = NF_ACCEPT;
1546                 return 0;
1547         }
1548
1549         return 1;
1550 }
1551
1552 /*
1553  *      Handle ICMP messages in the outside-to-inside direction (incoming).
1554  *      Find any that might be relevant, check against existing connections,
1555  *      forward to the right destination host if relevant.
1556  *      Currently handles error types - unreachable, quench, ttl exceeded.
1557  */
1558 static int
1559 ip_vs_in_icmp(struct netns_ipvs *ipvs, struct sk_buff *skb, int *related,
1560               unsigned int hooknum)
1561 {
1562         struct iphdr *iph;
1563         struct icmphdr  _icmph, *ic;
1564         struct iphdr    _ciph, *cih;    /* The ip header contained within the ICMP */
1565         struct ip_vs_iphdr ciph;
1566         struct ip_vs_conn *cp;
1567         struct ip_vs_protocol *pp;
1568         struct ip_vs_proto_data *pd;
1569         unsigned int offset, offset2, ihl, verdict;
1570         bool ipip, new_cp = false;
1571
1572         *related = 1;
1573
1574         /* reassemble IP fragments */
1575         if (ip_is_fragment(ip_hdr(skb))) {
1576                 if (ip_vs_gather_frags(ipvs, skb, ip_vs_defrag_user(hooknum)))
1577                         return NF_STOLEN;
1578         }
1579
1580         iph = ip_hdr(skb);
1581         offset = ihl = iph->ihl * 4;
1582         ic = skb_header_pointer(skb, offset, sizeof(_icmph), &_icmph);
1583         if (ic == NULL)
1584                 return NF_DROP;
1585
1586         IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1587                   ic->type, ntohs(icmp_id(ic)),
1588                   &iph->saddr, &iph->daddr);
1589
1590         /*
1591          * Work through seeing if this is for us.
1592          * These checks are supposed to be in an order that means easy
1593          * things are checked first to speed up processing.... however
1594          * this means that some packets will manage to get a long way
1595          * down this stack and then be rejected, but that's life.
1596          */
1597         if ((ic->type != ICMP_DEST_UNREACH) &&
1598             (ic->type != ICMP_SOURCE_QUENCH) &&
1599             (ic->type != ICMP_TIME_EXCEEDED)) {
1600                 *related = 0;
1601                 return NF_ACCEPT;
1602         }
1603
1604         /* Now find the contained IP header */
1605         offset += sizeof(_icmph);
1606         cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1607         if (cih == NULL)
1608                 return NF_ACCEPT; /* The packet looks wrong, ignore */
1609
1610         /* Special case for errors for IPIP packets */
1611         ipip = false;
1612         if (cih->protocol == IPPROTO_IPIP) {
1613                 if (unlikely(cih->frag_off & htons(IP_OFFSET)))
1614                         return NF_ACCEPT;
1615                 /* Error for our IPIP must arrive at LOCAL_IN */
1616                 if (!(skb_rtable(skb)->rt_flags & RTCF_LOCAL))
1617                         return NF_ACCEPT;
1618                 offset += cih->ihl * 4;
1619                 cih = skb_header_pointer(skb, offset, sizeof(_ciph), &_ciph);
1620                 if (cih == NULL)
1621                         return NF_ACCEPT; /* The packet looks wrong, ignore */
1622                 ipip = true;
1623         }
1624
1625         pd = ip_vs_proto_data_get(ipvs, cih->protocol);
1626         if (!pd)
1627                 return NF_ACCEPT;
1628         pp = pd->pp;
1629
1630         /* Is the embedded protocol header present? */
1631         if (unlikely(cih->frag_off & htons(IP_OFFSET) &&
1632                      pp->dont_defrag))
1633                 return NF_ACCEPT;
1634
1635         IP_VS_DBG_PKT(11, AF_INET, pp, skb, offset,
1636                       "Checking incoming ICMP for");
1637
1638         offset2 = offset;
1639         ip_vs_fill_iph_skb_icmp(AF_INET, skb, offset, !ipip, &ciph);
1640         offset = ciph.len;
1641
1642         /* The embedded headers contain source and dest in reverse order.
1643          * For IPIP this is error for request, not for reply.
1644          */
1645         cp = pp->conn_in_get(ipvs, AF_INET, skb, &ciph);
1646
1647         if (!cp) {
1648                 int v;
1649
1650                 if (ipip || !sysctl_schedule_icmp(ipvs))
1651                         return NF_ACCEPT;
1652
1653                 if (!ip_vs_try_to_schedule(ipvs, AF_INET, skb, pd, &v, &cp, &ciph))
1654                         return v;
1655                 new_cp = true;
1656         }
1657
1658         verdict = NF_DROP;
1659
1660         /* Ensure the checksum is correct */
1661         if (!skb_csum_unnecessary(skb) && ip_vs_checksum_complete(skb, ihl)) {
1662                 /* Failed checksum! */
1663                 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1664                           &iph->saddr);
1665                 goto out;
1666         }
1667
1668         if (ipip) {
1669                 __be32 info = ic->un.gateway;
1670                 __u8 type = ic->type;
1671                 __u8 code = ic->code;
1672
1673                 /* Update the MTU */
1674                 if (ic->type == ICMP_DEST_UNREACH &&
1675                     ic->code == ICMP_FRAG_NEEDED) {
1676                         struct ip_vs_dest *dest = cp->dest;
1677                         u32 mtu = ntohs(ic->un.frag.mtu);
1678                         __be16 frag_off = cih->frag_off;
1679
1680                         /* Strip outer IP and ICMP, go to IPIP header */
1681                         if (pskb_pull(skb, ihl + sizeof(_icmph)) == NULL)
1682                                 goto ignore_ipip;
1683                         offset2 -= ihl + sizeof(_icmph);
1684                         skb_reset_network_header(skb);
1685                         IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1686                                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, mtu);
1687                         ipv4_update_pmtu(skb, ipvs->net,
1688                                          mtu, 0, 0, 0, 0);
1689                         /* Client uses PMTUD? */
1690                         if (!(frag_off & htons(IP_DF)))
1691                                 goto ignore_ipip;
1692                         /* Prefer the resulting PMTU */
1693                         if (dest) {
1694                                 struct ip_vs_dest_dst *dest_dst;
1695
1696                                 dest_dst = rcu_dereference(dest->dest_dst);
1697                                 if (dest_dst)
1698                                         mtu = dst_mtu(dest_dst->dst_cache);
1699                         }
1700                         if (mtu > 68 + sizeof(struct iphdr))
1701                                 mtu -= sizeof(struct iphdr);
1702                         info = htonl(mtu);
1703                 }
1704                 /* Strip outer IP, ICMP and IPIP, go to IP header of
1705                  * original request.
1706                  */
1707                 if (pskb_pull(skb, offset2) == NULL)
1708                         goto ignore_ipip;
1709                 skb_reset_network_header(skb);
1710                 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1711                         &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1712                         type, code, ntohl(info));
1713                 icmp_send(skb, type, code, info);
1714                 /* ICMP can be shorter but anyways, account it */
1715                 ip_vs_out_stats(cp, skb);
1716
1717 ignore_ipip:
1718                 consume_skb(skb);
1719                 verdict = NF_STOLEN;
1720                 goto out;
1721         }
1722
1723         /* do the statistics and put it back */
1724         ip_vs_in_stats(cp, skb);
1725         if (IPPROTO_TCP == cih->protocol || IPPROTO_UDP == cih->protocol ||
1726             IPPROTO_SCTP == cih->protocol)
1727                 offset += 2 * sizeof(__u16);
1728         verdict = ip_vs_icmp_xmit(skb, cp, pp, offset, hooknum, &ciph);
1729
1730 out:
1731         if (likely(!new_cp))
1732                 __ip_vs_conn_put(cp);
1733         else
1734                 ip_vs_conn_put(cp);
1735
1736         return verdict;
1737 }
1738
1739 #ifdef CONFIG_IP_VS_IPV6
1740 static int ip_vs_in_icmp_v6(struct netns_ipvs *ipvs, struct sk_buff *skb,
1741                             int *related, unsigned int hooknum,
1742                             struct ip_vs_iphdr *iph)
1743 {
1744         struct icmp6hdr _icmph, *ic;
1745         struct ip_vs_iphdr ciph = {.flags = 0, .fragoffs = 0};/*Contained IP */
1746         struct ip_vs_conn *cp;
1747         struct ip_vs_protocol *pp;
1748         struct ip_vs_proto_data *pd;
1749         unsigned int offset, verdict;
1750         bool new_cp = false;
1751
1752         *related = 1;
1753
1754         ic = frag_safe_skb_hp(skb, iph->len, sizeof(_icmph), &_icmph);
1755         if (ic == NULL)
1756                 return NF_DROP;
1757
1758         /*
1759          * Work through seeing if this is for us.
1760          * These checks are supposed to be in an order that means easy
1761          * things are checked first to speed up processing.... however
1762          * this means that some packets will manage to get a long way
1763          * down this stack and then be rejected, but that's life.
1764          */
1765         if (ic->icmp6_type & ICMPV6_INFOMSG_MASK) {
1766                 *related = 0;
1767                 return NF_ACCEPT;
1768         }
1769         /* Fragment header that is before ICMP header tells us that:
1770          * it's not an error message since they can't be fragmented.
1771          */
1772         if (iph->flags & IP6_FH_F_FRAG)
1773                 return NF_DROP;
1774
1775         IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1776                   ic->icmp6_type, ntohs(icmpv6_id(ic)),
1777                   &iph->saddr, &iph->daddr);
1778
1779         offset = iph->len + sizeof(_icmph);
1780         if (!ip_vs_fill_iph_skb_icmp(AF_INET6, skb, offset, true, &ciph))
1781                 return NF_ACCEPT;
1782
1783         pd = ip_vs_proto_data_get(ipvs, ciph.protocol);
1784         if (!pd)
1785                 return NF_ACCEPT;
1786         pp = pd->pp;
1787
1788         /* Cannot handle fragmented embedded protocol */
1789         if (ciph.fragoffs)
1790                 return NF_ACCEPT;
1791
1792         IP_VS_DBG_PKT(11, AF_INET6, pp, skb, offset,
1793                       "Checking incoming ICMPv6 for");
1794
1795         /* The embedded headers contain source and dest in reverse order
1796          * if not from localhost
1797          */
1798         cp = pp->conn_in_get(ipvs, AF_INET6, skb, &ciph);
1799
1800         if (!cp) {
1801                 int v;
1802
1803                 if (!sysctl_schedule_icmp(ipvs))
1804                         return NF_ACCEPT;
1805
1806                 if (!ip_vs_try_to_schedule(ipvs, AF_INET6, skb, pd, &v, &cp, &ciph))
1807                         return v;
1808
1809                 new_cp = true;
1810         }
1811
1812         /* VS/TUN, VS/DR and LOCALNODE just let it go */
1813         if ((hooknum == NF_INET_LOCAL_OUT) &&
1814             (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)) {
1815                 verdict = NF_ACCEPT;
1816                 goto out;
1817         }
1818
1819         /* do the statistics and put it back */
1820         ip_vs_in_stats(cp, skb);
1821
1822         /* Need to mangle contained IPv6 header in ICMPv6 packet */
1823         offset = ciph.len;
1824         if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
1825             IPPROTO_SCTP == ciph.protocol)
1826                 offset += 2 * sizeof(__u16); /* Also mangle ports */
1827
1828         verdict = ip_vs_icmp_xmit_v6(skb, cp, pp, offset, hooknum, &ciph);
1829
1830 out:
1831         if (likely(!new_cp))
1832                 __ip_vs_conn_put(cp);
1833         else
1834                 ip_vs_conn_put(cp);
1835
1836         return verdict;
1837 }
1838 #endif
1839
1840
1841 /*
1842  *      Check if it's for virtual services, look it up,
1843  *      and send it on its way...
1844  */
1845 static unsigned int
1846 ip_vs_in(struct netns_ipvs *ipvs, unsigned int hooknum, struct sk_buff *skb, int af)
1847 {
1848         struct ip_vs_iphdr iph;
1849         struct ip_vs_protocol *pp;
1850         struct ip_vs_proto_data *pd;
1851         struct ip_vs_conn *cp;
1852         int ret, pkts;
1853         struct sock *sk;
1854
1855         /* Already marked as IPVS request or reply? */
1856         if (skb->ipvs_property)
1857                 return NF_ACCEPT;
1858
1859         /*
1860          *      Big tappo:
1861          *      - remote client: only PACKET_HOST
1862          *      - route: used for struct net when skb->dev is unset
1863          */
1864         if (unlikely((skb->pkt_type != PACKET_HOST &&
1865                       hooknum != NF_INET_LOCAL_OUT) ||
1866                      !skb_dst(skb))) {
1867                 ip_vs_fill_iph_skb(af, skb, false, &iph);
1868                 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1869                               " ignored in hook %u\n",
1870                               skb->pkt_type, iph.protocol,
1871                               IP_VS_DBG_ADDR(af, &iph.daddr), hooknum);
1872                 return NF_ACCEPT;
1873         }
1874         /* ipvs enabled in this netns ? */
1875         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
1876                 return NF_ACCEPT;
1877
1878         ip_vs_fill_iph_skb(af, skb, false, &iph);
1879
1880         /* Bad... Do not break raw sockets */
1881         sk = skb_to_full_sk(skb);
1882         if (unlikely(sk && hooknum == NF_INET_LOCAL_OUT &&
1883                      af == AF_INET)) {
1884
1885                 if (sk->sk_family == PF_INET && inet_sk(sk)->nodefrag)
1886                         return NF_ACCEPT;
1887         }
1888
1889 #ifdef CONFIG_IP_VS_IPV6
1890         if (af == AF_INET6) {
1891                 if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
1892                         int related;
1893                         int verdict = ip_vs_in_icmp_v6(ipvs, skb, &related,
1894                                                        hooknum, &iph);
1895
1896                         if (related)
1897                                 return verdict;
1898                 }
1899         } else
1900 #endif
1901                 if (unlikely(iph.protocol == IPPROTO_ICMP)) {
1902                         int related;
1903                         int verdict = ip_vs_in_icmp(ipvs, skb, &related,
1904                                                     hooknum);
1905
1906                         if (related)
1907                                 return verdict;
1908                 }
1909
1910         /* Protocol supported? */
1911         pd = ip_vs_proto_data_get(ipvs, iph.protocol);
1912         if (unlikely(!pd)) {
1913                 /* The only way we'll see this packet again is if it's
1914                  * encapsulated, so mark it with ipvs_property=1 so we
1915                  * skip it if we're ignoring tunneled packets
1916                  */
1917                 if (sysctl_ignore_tunneled(ipvs))
1918                         skb->ipvs_property = 1;
1919
1920                 return NF_ACCEPT;
1921         }
1922         pp = pd->pp;
1923         /*
1924          * Check if the packet belongs to an existing connection entry
1925          */
1926         cp = pp->conn_in_get(ipvs, af, skb, &iph);
1927
1928         if (!iph.fragoffs && is_new_conn(skb, &iph) && cp) {
1929                 int conn_reuse_mode = sysctl_conn_reuse_mode(ipvs);
1930                 bool old_ct = false, resched = false;
1931
1932                 if (unlikely(sysctl_expire_nodest_conn(ipvs)) && cp->dest &&
1933                     unlikely(!atomic_read(&cp->dest->weight))) {
1934                         resched = true;
1935                         old_ct = ip_vs_conn_uses_old_conntrack(cp, skb);
1936                 } else if (conn_reuse_mode &&
1937                            is_new_conn_expected(cp, conn_reuse_mode)) {
1938                         old_ct = ip_vs_conn_uses_old_conntrack(cp, skb);
1939                         if (!atomic_read(&cp->n_control)) {
1940                                 resched = true;
1941                         } else {
1942                                 /* Do not reschedule controlling connection
1943                                  * that uses conntrack while it is still
1944                                  * referenced by controlled connection(s).
1945                                  */
1946                                 resched = !old_ct;
1947                         }
1948                 }
1949
1950                 if (resched) {
1951                         if (!old_ct)
1952                                 cp->flags &= ~IP_VS_CONN_F_NFCT;
1953                         if (!atomic_read(&cp->n_control))
1954                                 ip_vs_conn_expire_now(cp);
1955                         __ip_vs_conn_put(cp);
1956                         if (old_ct)
1957                                 return NF_DROP;
1958                         cp = NULL;
1959                 }
1960         }
1961
1962         if (unlikely(!cp)) {
1963                 int v;
1964
1965                 if (!ip_vs_try_to_schedule(ipvs, af, skb, pd, &v, &cp, &iph))
1966                         return v;
1967         }
1968
1969         IP_VS_DBG_PKT(11, af, pp, skb, iph.off, "Incoming packet");
1970
1971         /* Check the server status */
1972         if (cp->dest && !(cp->dest->flags & IP_VS_DEST_F_AVAILABLE)) {
1973                 /* the destination server is not available */
1974
1975                 __u32 flags = cp->flags;
1976
1977                 /* when timer already started, silently drop the packet.*/
1978                 if (timer_pending(&cp->timer))
1979                         __ip_vs_conn_put(cp);
1980                 else
1981                         ip_vs_conn_put(cp);
1982
1983                 if (sysctl_expire_nodest_conn(ipvs) &&
1984                     !(flags & IP_VS_CONN_F_ONE_PACKET)) {
1985                         /* try to expire the connection immediately */
1986                         ip_vs_conn_expire_now(cp);
1987                 }
1988
1989                 return NF_DROP;
1990         }
1991
1992         ip_vs_in_stats(cp, skb);
1993         ip_vs_set_state(cp, IP_VS_DIR_INPUT, skb, pd);
1994         if (cp->packet_xmit)
1995                 ret = cp->packet_xmit(skb, cp, pp, &iph);
1996                 /* do not touch skb anymore */
1997         else {
1998                 IP_VS_DBG_RL("warning: packet_xmit is null");
1999                 ret = NF_ACCEPT;
2000         }
2001
2002         /* Increase its packet counter and check if it is needed
2003          * to be synchronized
2004          *
2005          * Sync connection if it is about to close to
2006          * encorage the standby servers to update the connections timeout
2007          *
2008          * For ONE_PKT let ip_vs_sync_conn() do the filter work.
2009          */
2010
2011         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
2012                 pkts = sysctl_sync_threshold(ipvs);
2013         else
2014                 pkts = atomic_add_return(1, &cp->in_pkts);
2015
2016         if (ipvs->sync_state & IP_VS_STATE_MASTER)
2017                 ip_vs_sync_conn(ipvs, cp, pkts);
2018         else if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) && cp->control)
2019                 /* increment is done inside ip_vs_sync_conn too */
2020                 atomic_inc(&cp->control->in_pkts);
2021
2022         ip_vs_conn_put(cp);
2023         return ret;
2024 }
2025
2026 /*
2027  *      AF_INET handler in NF_INET_LOCAL_IN chain
2028  *      Schedule and forward packets from remote clients
2029  */
2030 static unsigned int
2031 ip_vs_remote_request4(void *priv, struct sk_buff *skb,
2032                       const struct nf_hook_state *state)
2033 {
2034         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
2035 }
2036
2037 /*
2038  *      AF_INET handler in NF_INET_LOCAL_OUT chain
2039  *      Schedule and forward packets from local clients
2040  */
2041 static unsigned int
2042 ip_vs_local_request4(void *priv, struct sk_buff *skb,
2043                      const struct nf_hook_state *state)
2044 {
2045         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET);
2046 }
2047
2048 #ifdef CONFIG_IP_VS_IPV6
2049
2050 /*
2051  *      AF_INET6 handler in NF_INET_LOCAL_IN chain
2052  *      Schedule and forward packets from remote clients
2053  */
2054 static unsigned int
2055 ip_vs_remote_request6(void *priv, struct sk_buff *skb,
2056                       const struct nf_hook_state *state)
2057 {
2058         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
2059 }
2060
2061 /*
2062  *      AF_INET6 handler in NF_INET_LOCAL_OUT chain
2063  *      Schedule and forward packets from local clients
2064  */
2065 static unsigned int
2066 ip_vs_local_request6(void *priv, struct sk_buff *skb,
2067                      const struct nf_hook_state *state)
2068 {
2069         return ip_vs_in(net_ipvs(state->net), state->hook, skb, AF_INET6);
2070 }
2071
2072 #endif
2073
2074
2075 /*
2076  *      It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
2077  *      related packets destined for 0.0.0.0/0.
2078  *      When fwmark-based virtual service is used, such as transparent
2079  *      cache cluster, TCP packets can be marked and routed to ip_vs_in,
2080  *      but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
2081  *      sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
2082  *      and send them to ip_vs_in_icmp.
2083  */
2084 static unsigned int
2085 ip_vs_forward_icmp(void *priv, struct sk_buff *skb,
2086                    const struct nf_hook_state *state)
2087 {
2088         int r;
2089         struct netns_ipvs *ipvs = net_ipvs(state->net);
2090
2091         if (ip_hdr(skb)->protocol != IPPROTO_ICMP)
2092                 return NF_ACCEPT;
2093
2094         /* ipvs enabled in this netns ? */
2095         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
2096                 return NF_ACCEPT;
2097
2098         return ip_vs_in_icmp(ipvs, skb, &r, state->hook);
2099 }
2100
2101 #ifdef CONFIG_IP_VS_IPV6
2102 static unsigned int
2103 ip_vs_forward_icmp_v6(void *priv, struct sk_buff *skb,
2104                       const struct nf_hook_state *state)
2105 {
2106         int r;
2107         struct netns_ipvs *ipvs = net_ipvs(state->net);
2108         struct ip_vs_iphdr iphdr;
2109
2110         ip_vs_fill_iph_skb(AF_INET6, skb, false, &iphdr);
2111         if (iphdr.protocol != IPPROTO_ICMPV6)
2112                 return NF_ACCEPT;
2113
2114         /* ipvs enabled in this netns ? */
2115         if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
2116                 return NF_ACCEPT;
2117
2118         return ip_vs_in_icmp_v6(ipvs, skb, &r, state->hook, &iphdr);
2119 }
2120 #endif
2121
2122
2123 static const struct nf_hook_ops ip_vs_ops[] = {
2124         /* After packet filtering, change source only for VS/NAT */
2125         {
2126                 .hook           = ip_vs_reply4,
2127                 .pf             = NFPROTO_IPV4,
2128                 .hooknum        = NF_INET_LOCAL_IN,
2129                 .priority       = NF_IP_PRI_NAT_SRC - 2,
2130         },
2131         /* After packet filtering, forward packet through VS/DR, VS/TUN,
2132          * or VS/NAT(change destination), so that filtering rules can be
2133          * applied to IPVS. */
2134         {
2135                 .hook           = ip_vs_remote_request4,
2136                 .pf             = NFPROTO_IPV4,
2137                 .hooknum        = NF_INET_LOCAL_IN,
2138                 .priority       = NF_IP_PRI_NAT_SRC - 1,
2139         },
2140         /* Before ip_vs_in, change source only for VS/NAT */
2141         {
2142                 .hook           = ip_vs_local_reply4,
2143                 .pf             = NFPROTO_IPV4,
2144                 .hooknum        = NF_INET_LOCAL_OUT,
2145                 .priority       = NF_IP_PRI_NAT_DST + 1,
2146         },
2147         /* After mangle, schedule and forward local requests */
2148         {
2149                 .hook           = ip_vs_local_request4,
2150                 .pf             = NFPROTO_IPV4,
2151                 .hooknum        = NF_INET_LOCAL_OUT,
2152                 .priority       = NF_IP_PRI_NAT_DST + 2,
2153         },
2154         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2155          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2156         {
2157                 .hook           = ip_vs_forward_icmp,
2158                 .pf             = NFPROTO_IPV4,
2159                 .hooknum        = NF_INET_FORWARD,
2160                 .priority       = 99,
2161         },
2162         /* After packet filtering, change source only for VS/NAT */
2163         {
2164                 .hook           = ip_vs_reply4,
2165                 .pf             = NFPROTO_IPV4,
2166                 .hooknum        = NF_INET_FORWARD,
2167                 .priority       = 100,
2168         },
2169 #ifdef CONFIG_IP_VS_IPV6
2170         /* After packet filtering, change source only for VS/NAT */
2171         {
2172                 .hook           = ip_vs_reply6,
2173                 .pf             = NFPROTO_IPV6,
2174                 .hooknum        = NF_INET_LOCAL_IN,
2175                 .priority       = NF_IP6_PRI_NAT_SRC - 2,
2176         },
2177         /* After packet filtering, forward packet through VS/DR, VS/TUN,
2178          * or VS/NAT(change destination), so that filtering rules can be
2179          * applied to IPVS. */
2180         {
2181                 .hook           = ip_vs_remote_request6,
2182                 .pf             = NFPROTO_IPV6,
2183                 .hooknum        = NF_INET_LOCAL_IN,
2184                 .priority       = NF_IP6_PRI_NAT_SRC - 1,
2185         },
2186         /* Before ip_vs_in, change source only for VS/NAT */
2187         {
2188                 .hook           = ip_vs_local_reply6,
2189                 .pf             = NFPROTO_IPV6,
2190                 .hooknum        = NF_INET_LOCAL_OUT,
2191                 .priority       = NF_IP6_PRI_NAT_DST + 1,
2192         },
2193         /* After mangle, schedule and forward local requests */
2194         {
2195                 .hook           = ip_vs_local_request6,
2196                 .pf             = NFPROTO_IPV6,
2197                 .hooknum        = NF_INET_LOCAL_OUT,
2198                 .priority       = NF_IP6_PRI_NAT_DST + 2,
2199         },
2200         /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2201          * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2202         {
2203                 .hook           = ip_vs_forward_icmp_v6,
2204                 .pf             = NFPROTO_IPV6,
2205                 .hooknum        = NF_INET_FORWARD,
2206                 .priority       = 99,
2207         },
2208         /* After packet filtering, change source only for VS/NAT */
2209         {
2210                 .hook           = ip_vs_reply6,
2211                 .pf             = NFPROTO_IPV6,
2212                 .hooknum        = NF_INET_FORWARD,
2213                 .priority       = 100,
2214         },
2215 #endif
2216 };
2217 /*
2218  *      Initialize IP Virtual Server netns mem.
2219  */
2220 static int __net_init __ip_vs_init(struct net *net)
2221 {
2222         struct netns_ipvs *ipvs;
2223
2224         ipvs = net_generic(net, ip_vs_net_id);
2225         if (ipvs == NULL)
2226                 return -ENOMEM;
2227
2228         /* Hold the beast until a service is registerd */
2229         ipvs->enable = 0;
2230         ipvs->net = net;
2231         /* Counters used for creating unique names */
2232         ipvs->gen = atomic_read(&ipvs_netns_cnt);
2233         atomic_inc(&ipvs_netns_cnt);
2234         net->ipvs = ipvs;
2235
2236         if (ip_vs_estimator_net_init(ipvs) < 0)
2237                 goto estimator_fail;
2238
2239         if (ip_vs_control_net_init(ipvs) < 0)
2240                 goto control_fail;
2241
2242         if (ip_vs_protocol_net_init(ipvs) < 0)
2243                 goto protocol_fail;
2244
2245         if (ip_vs_app_net_init(ipvs) < 0)
2246                 goto app_fail;
2247
2248         if (ip_vs_conn_net_init(ipvs) < 0)
2249                 goto conn_fail;
2250
2251         if (ip_vs_sync_net_init(ipvs) < 0)
2252                 goto sync_fail;
2253
2254         return 0;
2255 /*
2256  * Error handling
2257  */
2258
2259 sync_fail:
2260         ip_vs_conn_net_cleanup(ipvs);
2261 conn_fail:
2262         ip_vs_app_net_cleanup(ipvs);
2263 app_fail:
2264         ip_vs_protocol_net_cleanup(ipvs);
2265 protocol_fail:
2266         ip_vs_control_net_cleanup(ipvs);
2267 control_fail:
2268         ip_vs_estimator_net_cleanup(ipvs);
2269 estimator_fail:
2270         net->ipvs = NULL;
2271         return -ENOMEM;
2272 }
2273
2274 static void __net_exit __ip_vs_cleanup(struct net *net)
2275 {
2276         struct netns_ipvs *ipvs = net_ipvs(net);
2277
2278         ip_vs_service_net_cleanup(ipvs);        /* ip_vs_flush() with locks */
2279         ip_vs_conn_net_cleanup(ipvs);
2280         ip_vs_app_net_cleanup(ipvs);
2281         ip_vs_protocol_net_cleanup(ipvs);
2282         ip_vs_control_net_cleanup(ipvs);
2283         ip_vs_estimator_net_cleanup(ipvs);
2284         IP_VS_DBG(2, "ipvs netns %d released\n", ipvs->gen);
2285         net->ipvs = NULL;
2286 }
2287
2288 static int __net_init __ip_vs_dev_init(struct net *net)
2289 {
2290         int ret;
2291
2292         ret = nf_register_net_hooks(net, ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2293         if (ret < 0)
2294                 goto hook_fail;
2295         return 0;
2296
2297 hook_fail:
2298         return ret;
2299 }
2300
2301 static void __net_exit __ip_vs_dev_cleanup(struct net *net)
2302 {
2303         struct netns_ipvs *ipvs = net_ipvs(net);
2304         EnterFunction(2);
2305         nf_unregister_net_hooks(net, ip_vs_ops, ARRAY_SIZE(ip_vs_ops));
2306         ipvs->enable = 0;       /* Disable packet reception */
2307         smp_wmb();
2308         ip_vs_sync_net_cleanup(ipvs);
2309         LeaveFunction(2);
2310 }
2311
2312 static struct pernet_operations ipvs_core_ops = {
2313         .init = __ip_vs_init,
2314         .exit = __ip_vs_cleanup,
2315         .id   = &ip_vs_net_id,
2316         .size = sizeof(struct netns_ipvs),
2317 };
2318
2319 static struct pernet_operations ipvs_core_dev_ops = {
2320         .init = __ip_vs_dev_init,
2321         .exit = __ip_vs_dev_cleanup,
2322 };
2323
2324 /*
2325  *      Initialize IP Virtual Server
2326  */
2327 static int __init ip_vs_init(void)
2328 {
2329         int ret;
2330
2331         ret = ip_vs_control_init();
2332         if (ret < 0) {
2333                 pr_err("can't setup control.\n");
2334                 goto exit;
2335         }
2336
2337         ip_vs_protocol_init();
2338
2339         ret = ip_vs_conn_init();
2340         if (ret < 0) {
2341                 pr_err("can't setup connection table.\n");
2342                 goto cleanup_protocol;
2343         }
2344
2345         ret = register_pernet_subsys(&ipvs_core_ops);   /* Alloc ip_vs struct */
2346         if (ret < 0)
2347                 goto cleanup_conn;
2348
2349         ret = register_pernet_device(&ipvs_core_dev_ops);
2350         if (ret < 0)
2351                 goto cleanup_sub;
2352
2353         ret = ip_vs_register_nl_ioctl();
2354         if (ret < 0) {
2355                 pr_err("can't register netlink/ioctl.\n");
2356                 goto cleanup_dev;
2357         }
2358
2359         pr_info("ipvs loaded.\n");
2360
2361         return ret;
2362
2363 cleanup_dev:
2364         unregister_pernet_device(&ipvs_core_dev_ops);
2365 cleanup_sub:
2366         unregister_pernet_subsys(&ipvs_core_ops);
2367 cleanup_conn:
2368         ip_vs_conn_cleanup();
2369 cleanup_protocol:
2370         ip_vs_protocol_cleanup();
2371         ip_vs_control_cleanup();
2372 exit:
2373         return ret;
2374 }
2375
2376 static void __exit ip_vs_cleanup(void)
2377 {
2378         ip_vs_unregister_nl_ioctl();
2379         unregister_pernet_device(&ipvs_core_dev_ops);
2380         unregister_pernet_subsys(&ipvs_core_ops);       /* free ip_vs struct */
2381         ip_vs_conn_cleanup();
2382         ip_vs_protocol_cleanup();
2383         ip_vs_control_cleanup();
2384         pr_info("ipvs unloaded.\n");
2385 }
2386
2387 module_init(ip_vs_init);
2388 module_exit(ip_vs_cleanup);
2389 MODULE_LICENSE("GPL");