net/netfilter/ipvs/ip_vs_xmit.c

   1 /*
   2  * ip_vs_xmit.c: various packet transmitters for IPVS
   3  *
   4  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
   5  *              Julian Anastasov <ja@ssi.bg>
   6  *
   7  *              This program is free software; you can redistribute it and/or
   8  *              modify it under the terms of the GNU General Public License
   9  *              as published by the Free Software Foundation; either version
  10  *              2 of the License, or (at your option) any later version.
  11  *
  12  * Changes:
  13  *
  14  * Description of forwarding methods:
  15  * - all transmitters are called from LOCAL_IN (remote clients) and
  16  * LOCAL_OUT (local clients) but for ICMP can be called from FORWARD
  17  * - not all connections have destination server, for example,
  18  * connections in backup server when fwmark is used
  19  * - bypass connections use daddr from packet
  20  * LOCAL_OUT rules:
  21  * - skb->dev is NULL, skb->protocol is not set (both are set in POST_ROUTING)
  22  * - skb->pkt_type is not set yet
  23  * - the only place where we can see skb->sk != NULL
  24  */
  25
  26 #define KMSG_COMPONENT "IPVS"
  27 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  28
  29 #include <linux/kernel.h>
  30 #include <linux/slab.h>
  31 #include <linux/tcp.h>                  /* for tcphdr */
  32 #include <net/ip.h>
  33 #include <net/tcp.h>                    /* for csum_tcpudp_magic */
  34 #include <net/udp.h>
  35 #include <net/icmp.h>                   /* for icmp_send */
  36 #include <net/route.h>                  /* for ip_route_output */
  37 #include <net/ipv6.h>
  38 #include <net/ip6_route.h>
  39 #include <net/addrconf.h>
  40 #include <linux/icmpv6.h>
  41 #include <linux/netfilter.h>
  42 #include <linux/netfilter_ipv4.h>
  43
  44 #include <net/ip_vs.h>
  45
  46
  47 /*
  48  *      Destination cache to speed up outgoing route lookup
  49  */
  50 static inline void
  51 __ip_vs_dst_set(struct ip_vs_dest *dest, u32 rtos, struct dst_entry *dst,
  52                 u32 dst_cookie)
  53 {
  54         struct dst_entry *old_dst;
  55
  56         old_dst = dest->dst_cache;
  57         dest->dst_cache = dst;
  58         dest->dst_rtos = rtos;
  59         dest->dst_cookie = dst_cookie;
  60         dst_release(old_dst);
  61 }
  62
  63 static inline struct dst_entry *
  64 __ip_vs_dst_check(struct ip_vs_dest *dest, u32 rtos)
  65 {
  66         struct dst_entry *dst = dest->dst_cache;
  67
  68         if (!dst)
  69                 return NULL;
  70         if ((dst->obsolete || rtos != dest->dst_rtos) &&
  71             dst->ops->check(dst, dest->dst_cookie) == NULL) {
  72                 dest->dst_cache = NULL;
  73                 dst_release(dst);
  74                 return NULL;
  75         }
  76         dst_hold(dst);
  77         return dst;
  78 }
  79
  80 /*
  81  * Get route to destination or remote server
  82  * rt_mode: flags, &1=Allow local dest, &2=Allow non-local dest,
  83  *          &4=Allow redirect from remote daddr to local
  84  */
  85 static struct rtable *
  86 __ip_vs_get_out_rt(struct sk_buff *skb, struct ip_vs_dest *dest,
  87                    __be32 daddr, u32 rtos, int rt_mode)
  88 {
  89         struct net *net = dev_net(skb_dst(skb)->dev);
  90         struct rtable *rt;                      /* Route to the other host */
  91         struct rtable *ort;                     /* Original route */
  92         int local;
  93
  94         if (dest) {
  95                 spin_lock(&dest->dst_lock);
  96                 if (!(rt = (struct rtable *)
  97                       __ip_vs_dst_check(dest, rtos))) {
  98                         struct flowi fl = {
  99                                 .oif = 0,
 100                                 .nl_u = {
 101                                         .ip4_u = {
 102                                                 .daddr = dest->addr.ip,
 103                                                 .saddr = 0,
 104                                                 .tos = rtos, } },
 105                         };
 106
 107                         if (ip_route_output_key(net, &rt, &fl)) {
 108                                 spin_unlock(&dest->dst_lock);
 109                                 IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n",
 110                                              &dest->addr.ip);
 111                                 return NULL;
 112                         }
 113                         __ip_vs_dst_set(dest, rtos, dst_clone(&rt->dst), 0);
 114                         IP_VS_DBG(10, "new dst %pI4, refcnt=%d, rtos=%X\n",
 115                                   &dest->addr.ip,
 116                                   atomic_read(&rt->dst.__refcnt), rtos);
 117                 }
 118                 spin_unlock(&dest->dst_lock);
 119         } else {
 120                 struct flowi fl = {
 121                         .oif = 0,
 122                         .nl_u = {
 123                                 .ip4_u = {
 124                                         .daddr = daddr,
 125                                         .saddr = 0,
 126                                         .tos = rtos, } },
 127                 };
 128
 129                 if (ip_route_output_key(net, &rt, &fl)) {
 130                         IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n",
 131                                      &daddr);
 132                         return NULL;
 133                 }
 134         }
 135
 136         local = rt->rt_flags & RTCF_LOCAL;
 137         if (!((local ? 1 : 2) & rt_mode)) {
 138                 IP_VS_DBG_RL("Stopping traffic to %s address, dest: %pI4\n",
 139                              (rt->rt_flags & RTCF_LOCAL) ?
 140                              "local":"non-local", &rt->rt_dst);
 141                 ip_rt_put(rt);
 142                 return NULL;
 143         }
 144         if (local && !(rt_mode & 4) && !((ort = skb_rtable(skb)) &&
 145                                          ort->rt_flags & RTCF_LOCAL)) {
 146                 IP_VS_DBG_RL("Redirect from non-local address %pI4 to local "
 147                              "requires NAT method, dest: %pI4\n",
 148                              &ip_hdr(skb)->daddr, &rt->rt_dst);
 149                 ip_rt_put(rt);
 150                 return NULL;
 151         }
 152         if (unlikely(!local && ipv4_is_loopback(ip_hdr(skb)->saddr))) {
 153                 IP_VS_DBG_RL("Stopping traffic from loopback address %pI4 "
 154                              "to non-local address, dest: %pI4\n",
 155                              &ip_hdr(skb)->saddr, &rt->rt_dst);
 156                 ip_rt_put(rt);
 157                 return NULL;
 158         }
 159
 160         return rt;
 161 }
 162
 163 /* Reroute packet to local IPv4 stack after DNAT */
 164 static int
 165 __ip_vs_reroute_locally(struct sk_buff *skb)
 166 {
 167         struct rtable *rt = skb_rtable(skb);
 168         struct net_device *dev = rt->dst.dev;
 169         struct net *net = dev_net(dev);
 170         struct iphdr *iph = ip_hdr(skb);
 171
 172         if (rt_is_input_route(rt)) {
 173                 unsigned long orefdst = skb->_skb_refdst;
 174
 175                 if (ip_route_input(skb, iph->daddr, iph->saddr,
 176                                    iph->tos, skb->dev))
 177                         return 0;
 178                 refdst_drop(orefdst);
 179         } else {
 180                 struct flowi fl = {
 181                         .oif = 0,
 182                         .nl_u = {
 183                                 .ip4_u = {
 184                                         .daddr = iph->daddr,
 185                                         .saddr = iph->saddr,
 186                                         .tos = RT_TOS(iph->tos),
 187                                 }
 188                         },
 189                         .mark = skb->mark,
 190                 };
 191                 struct rtable *rt;
 192
 193                 if (ip_route_output_key(net, &rt, &fl))
 194                         return 0;
 195                 if (!(rt->rt_flags & RTCF_LOCAL)) {
 196                         ip_rt_put(rt);
 197                         return 0;
 198                 }
 199                 /* Drop old route. */
 200                 skb_dst_drop(skb);
 201                 skb_dst_set(skb, &rt->dst);
 202         }
 203         return 1;
 204 }
 205
 206 #ifdef CONFIG_IP_VS_IPV6
 207
 208 static inline int __ip_vs_is_local_route6(struct rt6_info *rt)
 209 {
 210         return rt->rt6i_dev && rt->rt6i_dev->flags & IFF_LOOPBACK;
 211 }
 212
 213 static struct dst_entry *
 214 __ip_vs_route_output_v6(struct net *net, struct in6_addr *daddr,
 215                         struct in6_addr *ret_saddr, int do_xfrm)
 216 {
 217         struct dst_entry *dst;
 218         struct flowi fl = {
 219                 .oif = 0,
 220                 .nl_u = {
 221                         .ip6_u = {
 222                                 .daddr = *daddr,
 223                         },
 224                 },
 225         };
 226
 227         dst = ip6_route_output(net, NULL, &fl);
 228         if (dst->error)
 229                 goto out_err;
 230         if (!ret_saddr)
 231                 return dst;
 232         if (ipv6_addr_any(&fl.fl6_src) &&
 233             ipv6_dev_get_saddr(net, ip6_dst_idev(dst)->dev,
 234                                &fl.fl6_dst, 0, &fl.fl6_src) < 0)
 235                 goto out_err;
 236         if (do_xfrm && xfrm_lookup(net, &dst, &fl, NULL, 0) < 0)
 237                 goto out_err;
 238         ipv6_addr_copy(ret_saddr, &fl.fl6_src);
 239         return dst;
 240
 241 out_err:
 242         dst_release(dst);
 243         IP_VS_DBG_RL("ip6_route_output error, dest: %pI6\n", daddr);
 244         return NULL;
 245 }
 246
 247 /*
 248  * Get route to destination or remote server
 249  * rt_mode: flags, &1=Allow local dest, &2=Allow non-local dest,
 250  *          &4=Allow redirect from remote daddr to local
 251  */
 252 static struct rt6_info *
 253 __ip_vs_get_out_rt_v6(struct sk_buff *skb, struct ip_vs_dest *dest,
 254                       struct in6_addr *daddr, struct in6_addr *ret_saddr,
 255                       int do_xfrm, int rt_mode)
 256 {
 257         struct net *net = dev_net(skb_dst(skb)->dev);
 258         struct rt6_info *rt;                    /* Route to the other host */
 259         struct rt6_info *ort;                   /* Original route */
 260         struct dst_entry *dst;
 261         int local;
 262
 263         if (dest) {
 264                 spin_lock(&dest->dst_lock);
 265                 rt = (struct rt6_info *)__ip_vs_dst_check(dest, 0);
 266                 if (!rt) {
 267                         u32 cookie;
 268
 269                         dst = __ip_vs_route_output_v6(net, &dest->addr.in6,
 270                                                       &dest->dst_saddr,
 271                                                       do_xfrm);
 272                         if (!dst) {
 273                                 spin_unlock(&dest->dst_lock);
 274                                 return NULL;
 275                         }
 276                         rt = (struct rt6_info *) dst;
 277                         cookie = rt->rt6i_node ? rt->rt6i_node->fn_sernum : 0;
 278                         __ip_vs_dst_set(dest, 0, dst_clone(&rt->dst), cookie);
 279                         IP_VS_DBG(10, "new dst %pI6, src %pI6, refcnt=%d\n",
 280                                   &dest->addr.in6, &dest->dst_saddr,
 281                                   atomic_read(&rt->dst.__refcnt));
 282                 }
 283                 if (ret_saddr)
 284                         ipv6_addr_copy(ret_saddr, &dest->dst_saddr);
 285                 spin_unlock(&dest->dst_lock);
 286         } else {
 287                 dst = __ip_vs_route_output_v6(net, daddr, ret_saddr, do_xfrm);
 288                 if (!dst)
 289                         return NULL;
 290                 rt = (struct rt6_info *) dst;
 291         }
 292
 293         local = __ip_vs_is_local_route6(rt);
 294         if (!((local ? 1 : 2) & rt_mode)) {
 295                 IP_VS_DBG_RL("Stopping traffic to %s address, dest: %pI6\n",
 296                              local ? "local":"non-local", daddr);
 297                 dst_release(&rt->dst);
 298                 return NULL;
 299         }
 300         if (local && !(rt_mode & 4) &&
 301             !((ort = (struct rt6_info *) skb_dst(skb)) &&
 302               __ip_vs_is_local_route6(ort))) {
 303                 IP_VS_DBG_RL("Redirect from non-local address %pI6 to local "
 304                              "requires NAT method, dest: %pI6\n",
 305                              &ipv6_hdr(skb)->daddr, daddr);
 306                 dst_release(&rt->dst);
 307                 return NULL;
 308         }
 309         if (unlikely(!local && (!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
 310                      ipv6_addr_type(&ipv6_hdr(skb)->saddr) &
 311                                     IPV6_ADDR_LOOPBACK)) {
 312                 IP_VS_DBG_RL("Stopping traffic from loopback address %pI6 "
 313                              "to non-local address, dest: %pI6\n",
 314                              &ipv6_hdr(skb)->saddr, daddr);
 315                 dst_release(&rt->dst);
 316                 return NULL;
 317         }
 318
 319         return rt;
 320 }
 321 #endif
 322
 323
 324 /*
 325  *      Release dest->dst_cache before a dest is removed
 326  */
 327 void
 328 ip_vs_dst_reset(struct ip_vs_dest *dest)
 329 {
 330         struct dst_entry *old_dst;
 331
 332         old_dst = dest->dst_cache;
 333         dest->dst_cache = NULL;
 334         dst_release(old_dst);
 335 }
 336
 337 #define IP_VS_XMIT_TUNNEL(skb, cp)                              \
 338 ({                                                              \
 339         int __ret = NF_ACCEPT;                                  \
 340                                                                 \
 341         (skb)->ipvs_property = 1;                               \
 342         if (unlikely((cp)->flags & IP_VS_CONN_F_NFCT))          \
 343                 __ret = ip_vs_confirm_conntrack(skb, cp);       \
 344         if (__ret == NF_ACCEPT) {                               \
 345                 nf_reset(skb);                                  \
 346                 skb_forward_csum(skb);                          \
 347         }                                                       \
 348         __ret;                                                  \
 349 })
 350
 351 #define IP_VS_XMIT_NAT(pf, skb, cp, local)              \
 352 do {                                                    \
 353         (skb)->ipvs_property = 1;                       \
 354         if (likely(!((cp)->flags & IP_VS_CONN_F_NFCT))) \
 355                 ip_vs_notrack(skb);                     \
 356         else                                            \
 357                 ip_vs_update_conntrack(skb, cp, 1);     \
 358         if (local)                                      \
 359                 return NF_ACCEPT;                       \
 360         skb_forward_csum(skb);                          \
 361         NF_HOOK(pf, NF_INET_LOCAL_OUT, (skb), NULL,     \
 362                 skb_dst(skb)->dev, dst_output);         \
 363 } while (0)
 364
 365 #define IP_VS_XMIT(pf, skb, cp, local)                  \
 366 do {                                                    \
 367         (skb)->ipvs_property = 1;                       \
 368         if (likely(!((cp)->flags & IP_VS_CONN_F_NFCT))) \
 369                 ip_vs_notrack(skb);                     \
 370         if (local)                                      \
 371                 return NF_ACCEPT;                       \
 372         skb_forward_csum(skb);                          \
 373         NF_HOOK(pf, NF_INET_LOCAL_OUT, (skb), NULL,     \
 374                 skb_dst(skb)->dev, dst_output);         \
 375 } while (0)
 376
 377
 378 /*
 379  *      NULL transmitter (do nothing except return NF_ACCEPT)
 380  */
 381 int
 382 ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 383                 struct ip_vs_protocol *pp)
 384 {
 385         /* we do not touch skb and do not need pskb ptr */
 386         IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 1);
 387 }
 388
 389
 390 /*
 391  *      Bypass transmitter
 392  *      Let packets bypass the destination when the destination is not
 393  *      available, it may be only used in transparent cache cluster.
 394  */
 395 int
 396 ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 397                   struct ip_vs_protocol *pp)
 398 {
 399         struct rtable *rt;                      /* Route to the other host */
 400         struct iphdr  *iph = ip_hdr(skb);
 401         int    mtu;
 402
 403         EnterFunction(10);
 404
 405         if (!(rt = __ip_vs_get_out_rt(skb, NULL, iph->daddr,
 406                                       RT_TOS(iph->tos), 2)))
 407                 goto tx_error_icmp;
 408
 409         /* MTU checking */
 410         mtu = dst_mtu(&rt->dst);
 411         if ((skb->len > mtu) && (iph->frag_off & htons(IP_DF))) {
 412                 ip_rt_put(rt);
 413                 icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu));
 414                 IP_VS_DBG_RL("%s(): frag needed\n", __func__);
 415                 goto tx_error;
 416         }
 417
 418         /*
 419          * Call ip_send_check because we are not sure it is called
 420          * after ip_defrag. Is copy-on-write needed?
 421          */
 422         if (unlikely((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)) {
 423                 ip_rt_put(rt);
 424                 return NF_STOLEN;
 425         }
 426         ip_send_check(ip_hdr(skb));
 427
 428         /* drop old route */
 429         skb_dst_drop(skb);
 430         skb_dst_set(skb, &rt->dst);
 431
 432         /* Another hack: avoid icmp_send in ip_fragment */
 433         skb->local_df = 1;
 434
 435         IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 0);
 436
 437         LeaveFunction(10);
 438         return NF_STOLEN;
 439
 440  tx_error_icmp:
 441         dst_link_failure(skb);
 442  tx_error:
 443         kfree_skb(skb);
 444         LeaveFunction(10);
 445         return NF_STOLEN;
 446 }
 447
 448 #ifdef CONFIG_IP_VS_IPV6
 449 int
 450 ip_vs_bypass_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
 451                      struct ip_vs_protocol *pp)
 452 {
 453         struct rt6_info *rt;                    /* Route to the other host */
 454         struct ipv6hdr  *iph = ipv6_hdr(skb);
 455         int    mtu;
 456
 457         EnterFunction(10);
 458
 459         if (!(rt = __ip_vs_get_out_rt_v6(skb, NULL, &iph->daddr, NULL, 0, 2)))
 460                 goto tx_error_icmp;
 461
 462         /* MTU checking */
 463         mtu = dst_mtu(&rt->dst);
 464         if (skb->len > mtu) {
 465                 if (!skb->dev) {
 466                         struct net *net = dev_net(skb_dst(skb)->dev);
 467
 468                         skb->dev = net->loopback_dev;
 469                 }
 470                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 471                 dst_release(&rt->dst);
 472                 IP_VS_DBG_RL("%s(): frag needed\n", __func__);
 473                 goto tx_error;
 474         }
 475
 476         /*
 477          * Call ip_send_check because we are not sure it is called
 478          * after ip_defrag. Is copy-on-write needed?
 479          */
 480         skb = skb_share_check(skb, GFP_ATOMIC);
 481         if (unlikely(skb == NULL)) {
 482                 dst_release(&rt->dst);
 483                 return NF_STOLEN;
 484         }
 485
 486         /* drop old route */
 487         skb_dst_drop(skb);
 488         skb_dst_set(skb, &rt->dst);
 489
 490         /* Another hack: avoid icmp_send in ip_fragment */
 491         skb->local_df = 1;
 492
 493         IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 0);
 494
 495         LeaveFunction(10);
 496         return NF_STOLEN;
 497
 498  tx_error_icmp:
 499         dst_link_failure(skb);
 500  tx_error:
 501         kfree_skb(skb);
 502         LeaveFunction(10);
 503         return NF_STOLEN;
 504 }
 505 #endif
 506
 507 /*
 508  *      NAT transmitter (only for outside-to-inside nat forwarding)
 509  *      Not used for related ICMP
 510  */
 511 int
 512 ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 513                struct ip_vs_protocol *pp)
 514 {
 515         struct rtable *rt;              /* Route to the other host */
 516         int mtu;
 517         struct iphdr *iph = ip_hdr(skb);
 518         int local;
 519
 520         EnterFunction(10);
 521
 522         /* check if it is a connection of no-client-port */
 523         if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) {
 524                 __be16 _pt, *p;
 525                 p = skb_header_pointer(skb, iph->ihl*4, sizeof(_pt), &_pt);
 526                 if (p == NULL)
 527                         goto tx_error;
 528                 ip_vs_conn_fill_cport(cp, *p);
 529                 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
 530         }
 531
 532         if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip,
 533                                       RT_TOS(iph->tos), 1|2|4)))
 534                 goto tx_error_icmp;
 535         local = rt->rt_flags & RTCF_LOCAL;
 536         /*
 537          * Avoid duplicate tuple in reply direction for NAT traffic
 538          * to local address when connection is sync-ed
 539          */
 540 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
 541         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
 542                 enum ip_conntrack_info ctinfo;
 543                 struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo);
 544
 545                 if (ct && !nf_ct_is_untracked(ct)) {
 546                         IP_VS_DBG_RL_PKT(10, AF_INET, pp, skb, 0,
 547                                          "ip_vs_nat_xmit(): "
 548                                          "stopping DNAT to local address");
 549                         goto tx_error_put;
 550                 }
 551         }
 552 #endif
 553
 554         /* From world but DNAT to loopback address? */
 555         if (local && ipv4_is_loopback(rt->rt_dst) &&
 556             rt_is_input_route(skb_rtable(skb))) {
 557                 IP_VS_DBG_RL_PKT(1, AF_INET, pp, skb, 0, "ip_vs_nat_xmit(): "
 558                                  "stopping DNAT to loopback address");
 559                 goto tx_error_put;
 560         }
 561
 562         /* MTU checking */
 563         mtu = dst_mtu(&rt->dst);
 564         if ((skb->len > mtu) && (iph->frag_off & htons(IP_DF))) {
 565                 icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu));
 566                 IP_VS_DBG_RL_PKT(0, AF_INET, pp, skb, 0,
 567                                  "ip_vs_nat_xmit(): frag needed for");
 568                 goto tx_error_put;
 569         }
 570
 571         /* copy-on-write the packet before mangling it */
 572         if (!skb_make_writable(skb, sizeof(struct iphdr)))
 573                 goto tx_error_put;
 574
 575         if (skb_cow(skb, rt->dst.dev->hard_header_len))
 576                 goto tx_error_put;
 577
 578         /* mangle the packet */
 579         if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp))
 580                 goto tx_error_put;
 581         ip_hdr(skb)->daddr = cp->daddr.ip;
 582         ip_send_check(ip_hdr(skb));
 583
 584         if (!local) {
 585                 /* drop old route */
 586                 skb_dst_drop(skb);
 587                 skb_dst_set(skb, &rt->dst);
 588         } else {
 589                 ip_rt_put(rt);
 590                 /*
 591                  * Some IPv4 replies get local address from routes,
 592                  * not from iph, so while we DNAT after routing
 593                  * we need this second input/output route.
 594                  */
 595                 if (!__ip_vs_reroute_locally(skb))
 596                         goto tx_error;
 597         }
 598
 599         IP_VS_DBG_PKT(10, AF_INET, pp, skb, 0, "After DNAT");
 600
 601         /* FIXME: when application helper enlarges the packet and the length
 602            is larger than the MTU of outgoing device, there will be still
 603            MTU problem. */
 604
 605         /* Another hack: avoid icmp_send in ip_fragment */
 606         skb->local_df = 1;
 607
 608         IP_VS_XMIT_NAT(NFPROTO_IPV4, skb, cp, local);
 609
 610         LeaveFunction(10);
 611         return NF_STOLEN;
 612
 613   tx_error_icmp:
 614         dst_link_failure(skb);
 615   tx_error:
 616         kfree_skb(skb);
 617         LeaveFunction(10);
 618         return NF_STOLEN;
 619   tx_error_put:
 620         ip_rt_put(rt);
 621         goto tx_error;
 622 }
 623
 624 #ifdef CONFIG_IP_VS_IPV6
 625 int
 626 ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
 627                   struct ip_vs_protocol *pp)
 628 {
 629         struct rt6_info *rt;            /* Route to the other host */
 630         int mtu;
 631         int local;
 632
 633         EnterFunction(10);
 634
 635         /* check if it is a connection of no-client-port */
 636         if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) {
 637                 __be16 _pt, *p;
 638                 p = skb_header_pointer(skb, sizeof(struct ipv6hdr),
 639                                        sizeof(_pt), &_pt);
 640                 if (p == NULL)
 641                         goto tx_error;
 642                 ip_vs_conn_fill_cport(cp, *p);
 643                 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
 644         }
 645
 646         if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, NULL,
 647                                          0, 1|2|4)))
 648                 goto tx_error_icmp;
 649         local = __ip_vs_is_local_route6(rt);
 650         /*
 651          * Avoid duplicate tuple in reply direction for NAT traffic
 652          * to local address when connection is sync-ed
 653          */
 654 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
 655         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
 656                 enum ip_conntrack_info ctinfo;
 657                 struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo);
 658
 659                 if (ct && !nf_ct_is_untracked(ct)) {
 660                         IP_VS_DBG_RL_PKT(10, AF_INET6, pp, skb, 0,
 661                                          "ip_vs_nat_xmit_v6(): "
 662                                          "stopping DNAT to local address");
 663                         goto tx_error_put;
 664                 }
 665         }
 666 #endif
 667
 668         /* From world but DNAT to loopback address? */
 669         if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
 670             ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_LOOPBACK) {
 671                 IP_VS_DBG_RL_PKT(1, AF_INET6, pp, skb, 0,
 672                                  "ip_vs_nat_xmit_v6(): "
 673                                  "stopping DNAT to loopback address");
 674                 goto tx_error_put;
 675         }
 676
 677         /* MTU checking */
 678         mtu = dst_mtu(&rt->dst);
 679         if (skb->len > mtu) {
 680                 if (!skb->dev) {
 681                         struct net *net = dev_net(skb_dst(skb)->dev);
 682
 683                         skb->dev = net->loopback_dev;
 684                 }
 685                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 686                 IP_VS_DBG_RL_PKT(0, AF_INET6, pp, skb, 0,
 687                                  "ip_vs_nat_xmit_v6(): frag needed for");
 688                 goto tx_error_put;
 689         }
 690
 691         /* copy-on-write the packet before mangling it */
 692         if (!skb_make_writable(skb, sizeof(struct ipv6hdr)))
 693                 goto tx_error_put;
 694
 695         if (skb_cow(skb, rt->dst.dev->hard_header_len))
 696                 goto tx_error_put;
 697
 698         /* mangle the packet */
 699         if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp))
 700                 goto tx_error;
 701         ipv6_addr_copy(&ipv6_hdr(skb)->daddr, &cp->daddr.in6);
 702
 703         if (!local || !skb->dev) {
 704                 /* drop the old route when skb is not shared */
 705                 skb_dst_drop(skb);
 706                 skb_dst_set(skb, &rt->dst);
 707         } else {
 708                 /* destined to loopback, do we need to change route? */
 709                 dst_release(&rt->dst);
 710         }
 711
 712         IP_VS_DBG_PKT(10, AF_INET6, pp, skb, 0, "After DNAT");
 713
 714         /* FIXME: when application helper enlarges the packet and the length
 715            is larger than the MTU of outgoing device, there will be still
 716            MTU problem. */
 717
 718         /* Another hack: avoid icmp_send in ip_fragment */
 719         skb->local_df = 1;
 720
 721         IP_VS_XMIT_NAT(NFPROTO_IPV6, skb, cp, local);
 722
 723         LeaveFunction(10);
 724         return NF_STOLEN;
 725
 726 tx_error_icmp:
 727         dst_link_failure(skb);
 728 tx_error:
 729         LeaveFunction(10);
 730         kfree_skb(skb);
 731         return NF_STOLEN;
 732 tx_error_put:
 733         dst_release(&rt->dst);
 734         goto tx_error;
 735 }
 736 #endif
 737
 738
 739 /*
 740  *   IP Tunneling transmitter
 741  *
 742  *   This function encapsulates the packet in a new IP packet, its
 743  *   destination will be set to cp->daddr. Most code of this function
 744  *   is taken from ipip.c.
 745  *
 746  *   It is used in VS/TUN cluster. The load balancer selects a real
 747  *   server from a cluster based on a scheduling algorithm,
 748  *   encapsulates the request packet and forwards it to the selected
 749  *   server. For example, all real servers are configured with
 750  *   "ifconfig tunl0 <Virtual IP Address> up". When the server receives
 751  *   the encapsulated packet, it will decapsulate the packet, processe
 752  *   the request and return the response packets directly to the client
 753  *   without passing the load balancer. This can greatly increase the
 754  *   scalability of virtual server.
 755  *
 756  *   Used for ANY protocol
 757  */
 758 int
 759 ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 760                   struct ip_vs_protocol *pp)
 761 {
 762         struct rtable *rt;                      /* Route to the other host */
 763         struct net_device *tdev;                /* Device to other host */
 764         struct iphdr  *old_iph = ip_hdr(skb);
 765         u8     tos = old_iph->tos;
 766         __be16 df = old_iph->frag_off;
 767         struct iphdr  *iph;                     /* Our new IP header */
 768         unsigned int max_headroom;              /* The extra header space needed */
 769         int    mtu;
 770         int ret;
 771
 772         EnterFunction(10);
 773
 774         if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip,
 775                                       RT_TOS(tos), 1|2)))
 776                 goto tx_error_icmp;
 777         if (rt->rt_flags & RTCF_LOCAL) {
 778                 ip_rt_put(rt);
 779                 IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 1);
 780         }
 781
 782         tdev = rt->dst.dev;
 783
 784         mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr);
 785         if (mtu < 68) {
 786                 IP_VS_DBG_RL("%s(): mtu less than 68\n", __func__);
 787                 goto tx_error_put;
 788         }
 789         if (skb_dst(skb))
 790                 skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu);
 791
 792         df |= (old_iph->frag_off & htons(IP_DF));
 793
 794         if ((old_iph->frag_off & htons(IP_DF))
 795             && mtu < ntohs(old_iph->tot_len)) {
 796                 icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu));
 797                 IP_VS_DBG_RL("%s(): frag needed\n", __func__);
 798                 goto tx_error_put;
 799         }
 800
 801         /*
 802          * Okay, now see if we can stuff it in the buffer as-is.
 803          */
 804         max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct iphdr);
 805
 806         if (skb_headroom(skb) < max_headroom
 807             || skb_cloned(skb) || skb_shared(skb)) {
 808                 struct sk_buff *new_skb =
 809                         skb_realloc_headroom(skb, max_headroom);
 810                 if (!new_skb) {
 811                         ip_rt_put(rt);
 812                         kfree_skb(skb);
 813                         IP_VS_ERR_RL("%s(): no memory\n", __func__);
 814                         return NF_STOLEN;
 815                 }
 816                 kfree_skb(skb);
 817                 skb = new_skb;
 818                 old_iph = ip_hdr(skb);
 819         }
 820
 821         skb->transport_header = skb->network_header;
 822
 823         /* fix old IP header checksum */
 824         ip_send_check(old_iph);
 825
 826         skb_push(skb, sizeof(struct iphdr));
 827         skb_reset_network_header(skb);
 828         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 829
 830         /* drop old route */
 831         skb_dst_drop(skb);
 832         skb_dst_set(skb, &rt->dst);
 833
 834         /*
 835          *      Push down and install the IPIP header.
 836          */
 837         iph                     =       ip_hdr(skb);
 838         iph->version            =       4;
 839         iph->ihl                =       sizeof(struct iphdr)>>2;
 840         iph->frag_off           =       df;
 841         iph->protocol           =       IPPROTO_IPIP;
 842         iph->tos                =       tos;
 843         iph->daddr              =       rt->rt_dst;
 844         iph->saddr              =       rt->rt_src;
 845         iph->ttl                =       old_iph->ttl;
 846         ip_select_ident(iph, &rt->dst, NULL);
 847
 848         /* Another hack: avoid icmp_send in ip_fragment */
 849         skb->local_df = 1;
 850
 851         ret = IP_VS_XMIT_TUNNEL(skb, cp);
 852         if (ret == NF_ACCEPT)
 853                 ip_local_out(skb);
 854         else if (ret == NF_DROP)
 855                 kfree_skb(skb);
 856
 857         LeaveFunction(10);
 858
 859         return NF_STOLEN;
 860
 861   tx_error_icmp:
 862         dst_link_failure(skb);
 863   tx_error:
 864         kfree_skb(skb);
 865         LeaveFunction(10);
 866         return NF_STOLEN;
 867 tx_error_put:
 868         ip_rt_put(rt);
 869         goto tx_error;
 870 }
 871
 872 #ifdef CONFIG_IP_VS_IPV6
 873 int
 874 ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
 875                      struct ip_vs_protocol *pp)
 876 {
 877         struct rt6_info *rt;            /* Route to the other host */
 878         struct in6_addr saddr;          /* Source for tunnel */
 879         struct net_device *tdev;        /* Device to other host */
 880         struct ipv6hdr  *old_iph = ipv6_hdr(skb);
 881         struct ipv6hdr  *iph;           /* Our new IP header */
 882         unsigned int max_headroom;      /* The extra header space needed */
 883         int    mtu;
 884         int ret;
 885
 886         EnterFunction(10);
 887
 888         if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6,
 889                                          &saddr, 1, 1|2)))
 890                 goto tx_error_icmp;
 891         if (__ip_vs_is_local_route6(rt)) {
 892                 dst_release(&rt->dst);
 893                 IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 1);
 894         }
 895
 896         tdev = rt->dst.dev;
 897
 898         mtu = dst_mtu(&rt->dst) - sizeof(struct ipv6hdr);
 899         if (mtu < IPV6_MIN_MTU) {
 900                 IP_VS_DBG_RL("%s(): mtu less than %d\n", __func__,
 901                              IPV6_MIN_MTU);
 902                 goto tx_error_put;
 903         }
 904         if (skb_dst(skb))
 905                 skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu);
 906
 907         if (mtu < ntohs(old_iph->payload_len) + sizeof(struct ipv6hdr)) {
 908                 if (!skb->dev) {
 909                         struct net *net = dev_net(skb_dst(skb)->dev);
 910
 911                         skb->dev = net->loopback_dev;
 912                 }
 913                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 914                 IP_VS_DBG_RL("%s(): frag needed\n", __func__);
 915                 goto tx_error_put;
 916         }
 917
 918         /*
 919          * Okay, now see if we can stuff it in the buffer as-is.
 920          */
 921         max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct ipv6hdr);
 922
 923         if (skb_headroom(skb) < max_headroom
 924             || skb_cloned(skb) || skb_shared(skb)) {
 925                 struct sk_buff *new_skb =
 926                         skb_realloc_headroom(skb, max_headroom);
 927                 if (!new_skb) {
 928                         dst_release(&rt->dst);
 929                         kfree_skb(skb);
 930                         IP_VS_ERR_RL("%s(): no memory\n", __func__);
 931                         return NF_STOLEN;
 932                 }
 933                 kfree_skb(skb);
 934                 skb = new_skb;
 935                 old_iph = ipv6_hdr(skb);
 936         }
 937
 938         skb->transport_header = skb->network_header;
 939
 940         skb_push(skb, sizeof(struct ipv6hdr));
 941         skb_reset_network_header(skb);
 942         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 943
 944         /* drop old route */
 945         skb_dst_drop(skb);
 946         skb_dst_set(skb, &rt->dst);
 947
 948         /*
 949          *      Push down and install the IPIP header.
 950          */
 951         iph                     =       ipv6_hdr(skb);
 952         iph->version            =       6;
 953         iph->nexthdr            =       IPPROTO_IPV6;
 954         iph->payload_len        =       old_iph->payload_len;
 955         be16_add_cpu(&iph->payload_len, sizeof(*old_iph));
 956         iph->priority           =       old_iph->priority;
 957         memset(&iph->flow_lbl, 0, sizeof(iph->flow_lbl));
 958         ipv6_addr_copy(&iph->daddr, &cp->daddr.in6);
 959         ipv6_addr_copy(&iph->saddr, &saddr);
 960         iph->hop_limit          =       old_iph->hop_limit;
 961
 962         /* Another hack: avoid icmp_send in ip_fragment */
 963         skb->local_df = 1;
 964
 965         ret = IP_VS_XMIT_TUNNEL(skb, cp);
 966         if (ret == NF_ACCEPT)
 967                 ip6_local_out(skb);
 968         else if (ret == NF_DROP)
 969                 kfree_skb(skb);
 970
 971         LeaveFunction(10);
 972
 973         return NF_STOLEN;
 974
 975 tx_error_icmp:
 976         dst_link_failure(skb);
 977 tx_error:
 978         kfree_skb(skb);
 979         LeaveFunction(10);
 980         return NF_STOLEN;
 981 tx_error_put:
 982         dst_release(&rt->dst);
 983         goto tx_error;
 984 }
 985 #endif
 986
 987
 988 /*
 989  *      Direct Routing transmitter
 990  *      Used for ANY protocol
 991  */
 992 int
 993 ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 994               struct ip_vs_protocol *pp)
 995 {
 996         struct rtable *rt;                      /* Route to the other host */
 997         struct iphdr  *iph = ip_hdr(skb);
 998         int    mtu;
 999
1000         EnterFunction(10);
1001
1002         if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip,
1003                                       RT_TOS(iph->tos), 1|2)))
1004                 goto tx_error_icmp;
1005         if (rt->rt_flags & RTCF_LOCAL) {
1006                 ip_rt_put(rt);
1007                 IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 1);
1008         }
1009
1010         /* MTU checking */
1011         mtu = dst_mtu(&rt->dst);
1012         if ((iph->frag_off & htons(IP_DF)) && skb->len > mtu) {
1013                 icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu));
1014                 ip_rt_put(rt);
1015                 IP_VS_DBG_RL("%s(): frag needed\n", __func__);
1016                 goto tx_error;
1017         }
1018
1019         /*
1020          * Call ip_send_check because we are not sure it is called
1021          * after ip_defrag. Is copy-on-write needed?
1022          */
1023         if (unlikely((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)) {
1024                 ip_rt_put(rt);
1025                 return NF_STOLEN;
1026         }
1027         ip_send_check(ip_hdr(skb));
1028
1029         /* drop old route */
1030         skb_dst_drop(skb);
1031         skb_dst_set(skb, &rt->dst);
1032
1033         /* Another hack: avoid icmp_send in ip_fragment */
1034         skb->local_df = 1;
1035
1036         IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 0);
1037
1038         LeaveFunction(10);
1039         return NF_STOLEN;
1040
1041   tx_error_icmp:
1042         dst_link_failure(skb);
1043   tx_error:
1044         kfree_skb(skb);
1045         LeaveFunction(10);
1046         return NF_STOLEN;
1047 }
1048
1049 #ifdef CONFIG_IP_VS_IPV6
1050 int
1051 ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1052                  struct ip_vs_protocol *pp)
1053 {
1054         struct rt6_info *rt;                    /* Route to the other host */
1055         int    mtu;
1056
1057         EnterFunction(10);
1058
1059         if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, NULL,
1060                                          0, 1|2)))
1061                 goto tx_error_icmp;
1062         if (__ip_vs_is_local_route6(rt)) {
1063                 dst_release(&rt->dst);
1064                 IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 1);
1065         }
1066
1067         /* MTU checking */
1068         mtu = dst_mtu(&rt->dst);
1069         if (skb->len > mtu) {
1070                 if (!skb->dev) {
1071                         struct net *net = dev_net(skb_dst(skb)->dev);
1072
1073                         skb->dev = net->loopback_dev;
1074                 }
1075                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
1076                 dst_release(&rt->dst);
1077                 IP_VS_DBG_RL("%s(): frag needed\n", __func__);
1078                 goto tx_error;
1079         }
1080
1081         /*
1082          * Call ip_send_check because we are not sure it is called
1083          * after ip_defrag. Is copy-on-write needed?
1084          */
1085         skb = skb_share_check(skb, GFP_ATOMIC);
1086         if (unlikely(skb == NULL)) {
1087                 dst_release(&rt->dst);
1088                 return NF_STOLEN;
1089         }
1090
1091         /* drop old route */
1092         skb_dst_drop(skb);
1093         skb_dst_set(skb, &rt->dst);
1094
1095         /* Another hack: avoid icmp_send in ip_fragment */
1096         skb->local_df = 1;
1097
1098         IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 0);
1099
1100         LeaveFunction(10);
1101         return NF_STOLEN;
1102
1103 tx_error_icmp:
1104         dst_link_failure(skb);
1105 tx_error:
1106         kfree_skb(skb);
1107         LeaveFunction(10);
1108         return NF_STOLEN;
1109 }
1110 #endif
1111
1112
1113 /*
1114  *      ICMP packet transmitter
1115  *      called by the ip_vs_in_icmp
1116  */
1117 int
1118 ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1119                 struct ip_vs_protocol *pp, int offset)
1120 {
1121         struct rtable   *rt;    /* Route to the other host */
1122         int mtu;
1123         int rc;
1124         int local;
1125
1126         EnterFunction(10);
1127
1128         /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1129            forwarded directly here, because there is no need to
1130            translate address/port back */
1131         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1132                 if (cp->packet_xmit)
1133                         rc = cp->packet_xmit(skb, cp, pp);
1134                 else
1135                         rc = NF_ACCEPT;
1136                 /* do not touch skb anymore */
1137                 atomic_inc(&cp->in_pkts);
1138                 goto out;
1139         }
1140
1141         /*
1142          * mangle and send the packet here (only for VS/NAT)
1143          */
1144
1145         if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip,
1146                                       RT_TOS(ip_hdr(skb)->tos), 1|2|4)))
1147                 goto tx_error_icmp;
1148         local = rt->rt_flags & RTCF_LOCAL;
1149
1150         /*
1151          * Avoid duplicate tuple in reply direction for NAT traffic
1152          * to local address when connection is sync-ed
1153          */
1154 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1155         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1156                 enum ip_conntrack_info ctinfo;
1157                 struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo);
1158
1159                 if (ct && !nf_ct_is_untracked(ct)) {
1160                         IP_VS_DBG(10, "%s(): "
1161                                   "stopping DNAT to local address %pI4\n",
1162                                   __func__, &cp->daddr.ip);
1163                         goto tx_error_put;
1164                 }
1165         }
1166 #endif
1167
1168         /* From world but DNAT to loopback address? */
1169         if (local && ipv4_is_loopback(rt->rt_dst) &&
1170             rt_is_input_route(skb_rtable(skb))) {
1171                 IP_VS_DBG(1, "%s(): "
1172                           "stopping DNAT to loopback %pI4\n",
1173                           __func__, &cp->daddr.ip);
1174                 goto tx_error_put;
1175         }
1176
1177         /* MTU checking */
1178         mtu = dst_mtu(&rt->dst);
1179         if ((skb->len > mtu) && (ip_hdr(skb)->frag_off & htons(IP_DF))) {
1180                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
1181                 IP_VS_DBG_RL("%s(): frag needed\n", __func__);
1182                 goto tx_error_put;
1183         }
1184
1185         /* copy-on-write the packet before mangling it */
1186         if (!skb_make_writable(skb, offset))
1187                 goto tx_error_put;
1188
1189         if (skb_cow(skb, rt->dst.dev->hard_header_len))
1190                 goto tx_error_put;
1191
1192         ip_vs_nat_icmp(skb, pp, cp, 0);
1193
1194         if (!local) {
1195                 /* drop the old route when skb is not shared */
1196                 skb_dst_drop(skb);
1197                 skb_dst_set(skb, &rt->dst);
1198         } else {
1199                 ip_rt_put(rt);
1200                 /*
1201                  * Some IPv4 replies get local address from routes,
1202                  * not from iph, so while we DNAT after routing
1203                  * we need this second input/output route.
1204                  */
1205                 if (!__ip_vs_reroute_locally(skb))
1206                         goto tx_error;
1207         }
1208
1209         /* Another hack: avoid icmp_send in ip_fragment */
1210         skb->local_df = 1;
1211
1212         IP_VS_XMIT_NAT(NFPROTO_IPV4, skb, cp, local);
1213
1214         rc = NF_STOLEN;
1215         goto out;
1216
1217   tx_error_icmp:
1218         dst_link_failure(skb);
1219   tx_error:
1220         dev_kfree_skb(skb);
1221         rc = NF_STOLEN;
1222   out:
1223         LeaveFunction(10);
1224         return rc;
1225   tx_error_put:
1226         ip_rt_put(rt);
1227         goto tx_error;
1228 }
1229
1230 #ifdef CONFIG_IP_VS_IPV6
1231 int
1232 ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1233                 struct ip_vs_protocol *pp, int offset)
1234 {
1235         struct rt6_info *rt;    /* Route to the other host */
1236         int mtu;
1237         int rc;
1238         int local;
1239
1240         EnterFunction(10);
1241
1242         /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1243            forwarded directly here, because there is no need to
1244            translate address/port back */
1245         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1246                 if (cp->packet_xmit)
1247                         rc = cp->packet_xmit(skb, cp, pp);
1248                 else
1249                         rc = NF_ACCEPT;
1250                 /* do not touch skb anymore */
1251                 atomic_inc(&cp->in_pkts);
1252                 goto out;
1253         }
1254
1255         /*
1256          * mangle and send the packet here (only for VS/NAT)
1257          */
1258
1259         if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, NULL,
1260                                          0, 1|2|4)))
1261                 goto tx_error_icmp;
1262
1263         local = __ip_vs_is_local_route6(rt);
1264         /*
1265          * Avoid duplicate tuple in reply direction for NAT traffic
1266          * to local address when connection is sync-ed
1267          */
1268 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1269         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1270                 enum ip_conntrack_info ctinfo;
1271                 struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo);
1272
1273                 if (ct && !nf_ct_is_untracked(ct)) {
1274                         IP_VS_DBG(10, "%s(): "
1275                                   "stopping DNAT to local address %pI6\n",
1276                                   __func__, &cp->daddr.in6);
1277                         goto tx_error_put;
1278                 }
1279         }
1280 #endif
1281
1282         /* From world but DNAT to loopback address? */
1283         if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
1284             ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_LOOPBACK) {
1285                 IP_VS_DBG(1, "%s(): "
1286                           "stopping DNAT to loopback %pI6\n",
1287                           __func__, &cp->daddr.in6);
1288                 goto tx_error_put;
1289         }
1290
1291         /* MTU checking */
1292         mtu = dst_mtu(&rt->dst);
1293         if (skb->len > mtu) {
1294                 if (!skb->dev) {
1295                         struct net *net = dev_net(skb_dst(skb)->dev);
1296
1297                         skb->dev = net->loopback_dev;
1298                 }
1299                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
1300                 IP_VS_DBG_RL("%s(): frag needed\n", __func__);
1301                 goto tx_error_put;
1302         }
1303
1304         /* copy-on-write the packet before mangling it */
1305         if (!skb_make_writable(skb, offset))
1306                 goto tx_error_put;
1307
1308         if (skb_cow(skb, rt->dst.dev->hard_header_len))
1309                 goto tx_error_put;
1310
1311         ip_vs_nat_icmp_v6(skb, pp, cp, 0);
1312
1313         if (!local || !skb->dev) {
1314                 /* drop the old route when skb is not shared */
1315                 skb_dst_drop(skb);
1316                 skb_dst_set(skb, &rt->dst);
1317         } else {
1318                 /* destined to loopback, do we need to change route? */
1319                 dst_release(&rt->dst);
1320         }
1321
1322         /* Another hack: avoid icmp_send in ip_fragment */
1323         skb->local_df = 1;
1324
1325         IP_VS_XMIT_NAT(NFPROTO_IPV6, skb, cp, local);
1326
1327         rc = NF_STOLEN;
1328         goto out;
1329
1330 tx_error_icmp:
1331         dst_link_failure(skb);
1332 tx_error:
1333         dev_kfree_skb(skb);
1334         rc = NF_STOLEN;
1335 out:
1336         LeaveFunction(10);
1337         return rc;
1338 tx_error_put:
1339         dst_release(&rt->dst);
1340         goto tx_error;
1341 }
1342 #endif