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
 192                 if (ip_route_output_key(net, &rt, &fl))
 193                         return 0;
 194                 if (!(rt->rt_flags & RTCF_LOCAL)) {
 195                         ip_rt_put(rt);
 196                         return 0;
 197                 }
 198                 /* Drop old route. */
 199                 skb_dst_drop(skb);
 200                 skb_dst_set(skb, &rt->dst);
 201         }
 202         return 1;
 203 }
 204
 205 #ifdef CONFIG_IP_VS_IPV6
 206
 207 static inline int __ip_vs_is_local_route6(struct rt6_info *rt)
 208 {
 209         return rt->rt6i_dev && rt->rt6i_dev->flags & IFF_LOOPBACK;
 210 }
 211
 212 static struct dst_entry *
 213 __ip_vs_route_output_v6(struct net *net, struct in6_addr *daddr,
 214                         struct in6_addr *ret_saddr, int do_xfrm)
 215 {
 216         struct dst_entry *dst;
 217         struct flowi fl = {
 218                 .oif = 0,
 219                 .nl_u = {
 220                         .ip6_u = {
 221                                 .daddr = *daddr,
 222                         },
 223                 },
 224         };
 225
 226         dst = ip6_route_output(net, NULL, &fl);
 227         if (dst->error)
 228                 goto out_err;
 229         if (!ret_saddr)
 230                 return dst;
 231         if (ipv6_addr_any(&fl.fl6_src) &&
 232             ipv6_dev_get_saddr(net, ip6_dst_idev(dst)->dev,
 233                                &fl.fl6_dst, 0, &fl.fl6_src) < 0)
 234                 goto out_err;
 235         if (do_xfrm && xfrm_lookup(net, &dst, &fl, NULL, 0) < 0)
 236                 goto out_err;
 237         ipv6_addr_copy(ret_saddr, &fl.fl6_src);
 238         return dst;
 239
 240 out_err:
 241         dst_release(dst);
 242         IP_VS_DBG_RL("ip6_route_output error, dest: %pI6\n", daddr);
 243         return NULL;
 244 }
 245
 246 /*
 247  * Get route to destination or remote server
 248  * rt_mode: flags, &1=Allow local dest, &2=Allow non-local dest,
 249  *          &4=Allow redirect from remote daddr to local
 250  */
 251 static struct rt6_info *
 252 __ip_vs_get_out_rt_v6(struct sk_buff *skb, struct ip_vs_dest *dest,
 253                       struct in6_addr *daddr, struct in6_addr *ret_saddr,
 254                       int do_xfrm, int rt_mode)
 255 {
 256         struct net *net = dev_net(skb_dst(skb)->dev);
 257         struct rt6_info *rt;                    /* Route to the other host */
 258         struct rt6_info *ort;                   /* Original route */
 259         struct dst_entry *dst;
 260         int local;
 261
 262         if (dest) {
 263                 spin_lock(&dest->dst_lock);
 264                 rt = (struct rt6_info *)__ip_vs_dst_check(dest, 0);
 265                 if (!rt) {
 266                         u32 cookie;
 267
 268                         dst = __ip_vs_route_output_v6(net, &dest->addr.in6,
 269                                                       &dest->dst_saddr,
 270                                                       do_xfrm);
 271                         if (!dst) {
 272                                 spin_unlock(&dest->dst_lock);
 273                                 return NULL;
 274                         }
 275                         rt = (struct rt6_info *) dst;
 276                         cookie = rt->rt6i_node ? rt->rt6i_node->fn_sernum : 0;
 277                         __ip_vs_dst_set(dest, 0, dst_clone(&rt->dst), cookie);
 278                         IP_VS_DBG(10, "new dst %pI6, src %pI6, refcnt=%d\n",
 279                                   &dest->addr.in6, &dest->dst_saddr,
 280                                   atomic_read(&rt->dst.__refcnt));
 281                 }
 282                 if (ret_saddr)
 283                         ipv6_addr_copy(ret_saddr, &dest->dst_saddr);
 284                 spin_unlock(&dest->dst_lock);
 285         } else {
 286                 dst = __ip_vs_route_output_v6(net, daddr, ret_saddr, do_xfrm);
 287                 if (!dst)
 288                         return NULL;
 289                 rt = (struct rt6_info *) dst;
 290         }
 291
 292         local = __ip_vs_is_local_route6(rt);
 293         if (!((local ? 1 : 2) & rt_mode)) {
 294                 IP_VS_DBG_RL("Stopping traffic to %s address, dest: %pI6\n",
 295                              local ? "local":"non-local", daddr);
 296                 dst_release(&rt->dst);
 297                 return NULL;
 298         }
 299         if (local && !(rt_mode & 4) &&
 300             !((ort = (struct rt6_info *) skb_dst(skb)) &&
 301               __ip_vs_is_local_route6(ort))) {
 302                 IP_VS_DBG_RL("Redirect from non-local address %pI6 to local "
 303                              "requires NAT method, dest: %pI6\n",
 304                              &ipv6_hdr(skb)->daddr, daddr);
 305                 dst_release(&rt->dst);
 306                 return NULL;
 307         }
 308         if (unlikely(!local && (!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
 309                      ipv6_addr_type(&ipv6_hdr(skb)->saddr) &
 310                                     IPV6_ADDR_LOOPBACK)) {
 311                 IP_VS_DBG_RL("Stopping traffic from loopback address %pI6 "
 312                              "to non-local address, dest: %pI6\n",
 313                              &ipv6_hdr(skb)->saddr, daddr);
 314                 dst_release(&rt->dst);
 315                 return NULL;
 316         }
 317
 318         return rt;
 319 }
 320 #endif
 321
 322
 323 /*
 324  *      Release dest->dst_cache before a dest is removed
 325  */
 326 void
 327 ip_vs_dst_reset(struct ip_vs_dest *dest)
 328 {
 329         struct dst_entry *old_dst;
 330
 331         old_dst = dest->dst_cache;
 332         dest->dst_cache = NULL;
 333         dst_release(old_dst);
 334 }
 335
 336 #define IP_VS_XMIT_TUNNEL(skb, cp)                              \
 337 ({                                                              \
 338         int __ret = NF_ACCEPT;                                  \
 339                                                                 \
 340         (skb)->ipvs_property = 1;                               \
 341         if (unlikely((cp)->flags & IP_VS_CONN_F_NFCT))          \
 342                 __ret = ip_vs_confirm_conntrack(skb, cp);       \
 343         if (__ret == NF_ACCEPT) {                               \
 344                 nf_reset(skb);                                  \
 345                 skb_forward_csum(skb);                          \
 346         }                                                       \
 347         __ret;                                                  \
 348 })
 349
 350 #define IP_VS_XMIT_NAT(pf, skb, cp, local)              \
 351 do {                                                    \
 352         (skb)->ipvs_property = 1;                       \
 353         if (likely(!((cp)->flags & IP_VS_CONN_F_NFCT))) \
 354                 ip_vs_notrack(skb);                     \
 355         else                                            \
 356                 ip_vs_update_conntrack(skb, cp, 1);     \
 357         if (local)                                      \
 358                 return NF_ACCEPT;                       \
 359         skb_forward_csum(skb);                          \
 360         NF_HOOK(pf, NF_INET_LOCAL_OUT, (skb), NULL,     \
 361                 skb_dst(skb)->dev, dst_output);         \
 362 } while (0)
 363
 364 #define IP_VS_XMIT(pf, skb, cp, local)                  \
 365 do {                                                    \
 366         (skb)->ipvs_property = 1;                       \
 367         if (likely(!((cp)->flags & IP_VS_CONN_F_NFCT))) \
 368                 ip_vs_notrack(skb);                     \
 369         if (local)                                      \
 370                 return NF_ACCEPT;                       \
 371         skb_forward_csum(skb);                          \
 372         NF_HOOK(pf, NF_INET_LOCAL_OUT, (skb), NULL,     \
 373                 skb_dst(skb)->dev, dst_output);         \
 374 } while (0)
 375
 376
 377 /*
 378  *      NULL transmitter (do nothing except return NF_ACCEPT)
 379  */
 380 int
 381 ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 382                 struct ip_vs_protocol *pp)
 383 {
 384         /* we do not touch skb and do not need pskb ptr */
 385         IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 1);
 386 }
 387
 388
 389 /*
 390  *      Bypass transmitter
 391  *      Let packets bypass the destination when the destination is not
 392  *      available, it may be only used in transparent cache cluster.
 393  */
 394 int
 395 ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 396                   struct ip_vs_protocol *pp)
 397 {
 398         struct rtable *rt;                      /* Route to the other host */
 399         struct iphdr  *iph = ip_hdr(skb);
 400         int    mtu;
 401
 402         EnterFunction(10);
 403
 404         if (!(rt = __ip_vs_get_out_rt(skb, NULL, iph->daddr,
 405                                       RT_TOS(iph->tos), 2)))
 406                 goto tx_error_icmp;
 407
 408         /* MTU checking */
 409         mtu = dst_mtu(&rt->dst);
 410         if ((skb->len > mtu) && (iph->frag_off & htons(IP_DF)) &&
 411             !skb_is_gso(skb)) {
 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 && !skb_is_gso(skb)) {
 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             !skb_is_gso(skb)) {
 566                 icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu));
 567                 IP_VS_DBG_RL_PKT(0, AF_INET, pp, skb, 0,
 568                                  "ip_vs_nat_xmit(): frag needed for");
 569                 goto tx_error_put;
 570         }
 571
 572         /* copy-on-write the packet before mangling it */
 573         if (!skb_make_writable(skb, sizeof(struct iphdr)))
 574                 goto tx_error_put;
 575
 576         if (skb_cow(skb, rt->dst.dev->hard_header_len))
 577                 goto tx_error_put;
 578
 579         /* mangle the packet */
 580         if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp))
 581                 goto tx_error_put;
 582         ip_hdr(skb)->daddr = cp->daddr.ip;
 583         ip_send_check(ip_hdr(skb));
 584
 585         if (!local) {
 586                 /* drop old route */
 587                 skb_dst_drop(skb);
 588                 skb_dst_set(skb, &rt->dst);
 589         } else {
 590                 ip_rt_put(rt);
 591                 /*
 592                  * Some IPv4 replies get local address from routes,
 593                  * not from iph, so while we DNAT after routing
 594                  * we need this second input/output route.
 595                  */
 596                 if (!__ip_vs_reroute_locally(skb))
 597                         goto tx_error;
 598         }
 599
 600         IP_VS_DBG_PKT(10, AF_INET, pp, skb, 0, "After DNAT");
 601
 602         /* FIXME: when application helper enlarges the packet and the length
 603            is larger than the MTU of outgoing device, there will be still
 604            MTU problem. */
 605
 606         /* Another hack: avoid icmp_send in ip_fragment */
 607         skb->local_df = 1;
 608
 609         IP_VS_XMIT_NAT(NFPROTO_IPV4, skb, cp, local);
 610
 611         LeaveFunction(10);
 612         return NF_STOLEN;
 613
 614   tx_error_icmp:
 615         dst_link_failure(skb);
 616   tx_error:
 617         kfree_skb(skb);
 618         LeaveFunction(10);
 619         return NF_STOLEN;
 620   tx_error_put:
 621         ip_rt_put(rt);
 622         goto tx_error;
 623 }
 624
 625 #ifdef CONFIG_IP_VS_IPV6
 626 int
 627 ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
 628                   struct ip_vs_protocol *pp)
 629 {
 630         struct rt6_info *rt;            /* Route to the other host */
 631         int mtu;
 632         int local;
 633
 634         EnterFunction(10);
 635
 636         /* check if it is a connection of no-client-port */
 637         if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) {
 638                 __be16 _pt, *p;
 639                 p = skb_header_pointer(skb, sizeof(struct ipv6hdr),
 640                                        sizeof(_pt), &_pt);
 641                 if (p == NULL)
 642                         goto tx_error;
 643                 ip_vs_conn_fill_cport(cp, *p);
 644                 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
 645         }
 646
 647         if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, NULL,
 648                                          0, 1|2|4)))
 649                 goto tx_error_icmp;
 650         local = __ip_vs_is_local_route6(rt);
 651         /*
 652          * Avoid duplicate tuple in reply direction for NAT traffic
 653          * to local address when connection is sync-ed
 654          */
 655 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
 656         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
 657                 enum ip_conntrack_info ctinfo;
 658                 struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo);
 659
 660                 if (ct && !nf_ct_is_untracked(ct)) {
 661                         IP_VS_DBG_RL_PKT(10, AF_INET6, pp, skb, 0,
 662                                          "ip_vs_nat_xmit_v6(): "
 663                                          "stopping DNAT to local address");
 664                         goto tx_error_put;
 665                 }
 666         }
 667 #endif
 668
 669         /* From world but DNAT to loopback address? */
 670         if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
 671             ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_LOOPBACK) {
 672                 IP_VS_DBG_RL_PKT(1, AF_INET6, pp, skb, 0,
 673                                  "ip_vs_nat_xmit_v6(): "
 674                                  "stopping DNAT to loopback address");
 675                 goto tx_error_put;
 676         }
 677
 678         /* MTU checking */
 679         mtu = dst_mtu(&rt->dst);
 680         if (skb->len > mtu && !skb_is_gso(skb)) {
 681                 if (!skb->dev) {
 682                         struct net *net = dev_net(skb_dst(skb)->dev);
 683
 684                         skb->dev = net->loopback_dev;
 685                 }
 686                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 687                 IP_VS_DBG_RL_PKT(0, AF_INET6, pp, skb, 0,
 688                                  "ip_vs_nat_xmit_v6(): frag needed for");
 689                 goto tx_error_put;
 690         }
 691
 692         /* copy-on-write the packet before mangling it */
 693         if (!skb_make_writable(skb, sizeof(struct ipv6hdr)))
 694                 goto tx_error_put;
 695
 696         if (skb_cow(skb, rt->dst.dev->hard_header_len))
 697                 goto tx_error_put;
 698
 699         /* mangle the packet */
 700         if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp))
 701                 goto tx_error;
 702         ipv6_addr_copy(&ipv6_hdr(skb)->daddr, &cp->daddr.in6);
 703
 704         if (!local || !skb->dev) {
 705                 /* drop the old route when skb is not shared */
 706                 skb_dst_drop(skb);
 707                 skb_dst_set(skb, &rt->dst);
 708         } else {
 709                 /* destined to loopback, do we need to change route? */
 710                 dst_release(&rt->dst);
 711         }
 712
 713         IP_VS_DBG_PKT(10, AF_INET6, pp, skb, 0, "After DNAT");
 714
 715         /* FIXME: when application helper enlarges the packet and the length
 716            is larger than the MTU of outgoing device, there will be still
 717            MTU problem. */
 718
 719         /* Another hack: avoid icmp_send in ip_fragment */
 720         skb->local_df = 1;
 721
 722         IP_VS_XMIT_NAT(NFPROTO_IPV6, skb, cp, local);
 723
 724         LeaveFunction(10);
 725         return NF_STOLEN;
 726
 727 tx_error_icmp:
 728         dst_link_failure(skb);
 729 tx_error:
 730         LeaveFunction(10);
 731         kfree_skb(skb);
 732         return NF_STOLEN;
 733 tx_error_put:
 734         dst_release(&rt->dst);
 735         goto tx_error;
 736 }
 737 #endif
 738
 739
 740 /*
 741  *   IP Tunneling transmitter
 742  *
 743  *   This function encapsulates the packet in a new IP packet, its
 744  *   destination will be set to cp->daddr. Most code of this function
 745  *   is taken from ipip.c.
 746  *
 747  *   It is used in VS/TUN cluster. The load balancer selects a real
 748  *   server from a cluster based on a scheduling algorithm,
 749  *   encapsulates the request packet and forwards it to the selected
 750  *   server. For example, all real servers are configured with
 751  *   "ifconfig tunl0 <Virtual IP Address> up". When the server receives
 752  *   the encapsulated packet, it will decapsulate the packet, processe
 753  *   the request and return the response packets directly to the client
 754  *   without passing the load balancer. This can greatly increase the
 755  *   scalability of virtual server.
 756  *
 757  *   Used for ANY protocol
 758  */
 759 int
 760 ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 761                   struct ip_vs_protocol *pp)
 762 {
 763         struct rtable *rt;                      /* Route to the other host */
 764         struct net_device *tdev;                /* Device to other host */
 765         struct iphdr  *old_iph = ip_hdr(skb);
 766         u8     tos = old_iph->tos;
 767         __be16 df = old_iph->frag_off;
 768         struct iphdr  *iph;                     /* Our new IP header */
 769         unsigned int max_headroom;              /* The extra header space needed */
 770         int    mtu;
 771         int ret;
 772
 773         EnterFunction(10);
 774
 775         if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip,
 776                                       RT_TOS(tos), 1|2)))
 777                 goto tx_error_icmp;
 778         if (rt->rt_flags & RTCF_LOCAL) {
 779                 ip_rt_put(rt);
 780                 IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 1);
 781         }
 782
 783         tdev = rt->dst.dev;
 784
 785         mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr);
 786         if (mtu < 68) {
 787                 IP_VS_DBG_RL("%s(): mtu less than 68\n", __func__);
 788                 goto tx_error_put;
 789         }
 790         if (skb_dst(skb))
 791                 skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu);
 792
 793         df |= (old_iph->frag_off & htons(IP_DF));
 794
 795         if ((old_iph->frag_off & htons(IP_DF) &&
 796             mtu < ntohs(old_iph->tot_len) && !skb_is_gso(skb))) {
 797                 icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu));
 798                 IP_VS_DBG_RL("%s(): frag needed\n", __func__);
 799                 goto tx_error_put;
 800         }
 801
 802         /*
 803          * Okay, now see if we can stuff it in the buffer as-is.
 804          */
 805         max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct iphdr);
 806
 807         if (skb_headroom(skb) < max_headroom
 808             || skb_cloned(skb) || skb_shared(skb)) {
 809                 struct sk_buff *new_skb =
 810                         skb_realloc_headroom(skb, max_headroom);
 811                 if (!new_skb) {
 812                         ip_rt_put(rt);
 813                         kfree_skb(skb);
 814                         IP_VS_ERR_RL("%s(): no memory\n", __func__);
 815                         return NF_STOLEN;
 816                 }
 817                 kfree_skb(skb);
 818                 skb = new_skb;
 819                 old_iph = ip_hdr(skb);
 820         }
 821
 822         skb->transport_header = skb->network_header;
 823
 824         /* fix old IP header checksum */
 825         ip_send_check(old_iph);
 826
 827         skb_push(skb, sizeof(struct iphdr));
 828         skb_reset_network_header(skb);
 829         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 830
 831         /* drop old route */
 832         skb_dst_drop(skb);
 833         skb_dst_set(skb, &rt->dst);
 834
 835         /*
 836          *      Push down and install the IPIP header.
 837          */
 838         iph                     =       ip_hdr(skb);
 839         iph->version            =       4;
 840         iph->ihl                =       sizeof(struct iphdr)>>2;
 841         iph->frag_off           =       df;
 842         iph->protocol           =       IPPROTO_IPIP;
 843         iph->tos                =       tos;
 844         iph->daddr              =       rt->rt_dst;
 845         iph->saddr              =       rt->rt_src;
 846         iph->ttl                =       old_iph->ttl;
 847         ip_select_ident(iph, &rt->dst, NULL);
 848
 849         /* Another hack: avoid icmp_send in ip_fragment */
 850         skb->local_df = 1;
 851
 852         ret = IP_VS_XMIT_TUNNEL(skb, cp);
 853         if (ret == NF_ACCEPT)
 854                 ip_local_out(skb);
 855         else if (ret == NF_DROP)
 856                 kfree_skb(skb);
 857
 858         LeaveFunction(10);
 859
 860         return NF_STOLEN;
 861
 862   tx_error_icmp:
 863         dst_link_failure(skb);
 864   tx_error:
 865         kfree_skb(skb);
 866         LeaveFunction(10);
 867         return NF_STOLEN;
 868 tx_error_put:
 869         ip_rt_put(rt);
 870         goto tx_error;
 871 }
 872
 873 #ifdef CONFIG_IP_VS_IPV6
 874 int
 875 ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
 876                      struct ip_vs_protocol *pp)
 877 {
 878         struct rt6_info *rt;            /* Route to the other host */
 879         struct in6_addr saddr;          /* Source for tunnel */
 880         struct net_device *tdev;        /* Device to other host */
 881         struct ipv6hdr  *old_iph = ipv6_hdr(skb);
 882         struct ipv6hdr  *iph;           /* Our new IP header */
 883         unsigned int max_headroom;      /* The extra header space needed */
 884         int    mtu;
 885         int ret;
 886
 887         EnterFunction(10);
 888
 889         if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6,
 890                                          &saddr, 1, 1|2)))
 891                 goto tx_error_icmp;
 892         if (__ip_vs_is_local_route6(rt)) {
 893                 dst_release(&rt->dst);
 894                 IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 1);
 895         }
 896
 897         tdev = rt->dst.dev;
 898
 899         mtu = dst_mtu(&rt->dst) - sizeof(struct ipv6hdr);
 900         if (mtu < IPV6_MIN_MTU) {
 901                 IP_VS_DBG_RL("%s(): mtu less than %d\n", __func__,
 902                              IPV6_MIN_MTU);
 903                 goto tx_error_put;
 904         }
 905         if (skb_dst(skb))
 906                 skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu);
 907
 908         if (mtu < ntohs(old_iph->payload_len) + sizeof(struct ipv6hdr) &&
 909             !skb_is_gso(skb)) {
 910                 if (!skb->dev) {
 911                         struct net *net = dev_net(skb_dst(skb)->dev);
 912
 913                         skb->dev = net->loopback_dev;
 914                 }
 915                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 916                 IP_VS_DBG_RL("%s(): frag needed\n", __func__);
 917                 goto tx_error_put;
 918         }
 919
 920         /*
 921          * Okay, now see if we can stuff it in the buffer as-is.
 922          */
 923         max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct ipv6hdr);
 924
 925         if (skb_headroom(skb) < max_headroom
 926             || skb_cloned(skb) || skb_shared(skb)) {
 927                 struct sk_buff *new_skb =
 928                         skb_realloc_headroom(skb, max_headroom);
 929                 if (!new_skb) {
 930                         dst_release(&rt->dst);
 931                         kfree_skb(skb);
 932                         IP_VS_ERR_RL("%s(): no memory\n", __func__);
 933                         return NF_STOLEN;
 934                 }
 935                 kfree_skb(skb);
 936                 skb = new_skb;
 937                 old_iph = ipv6_hdr(skb);
 938         }
 939
 940         skb->transport_header = skb->network_header;
 941
 942         skb_push(skb, sizeof(struct ipv6hdr));
 943         skb_reset_network_header(skb);
 944         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 945
 946         /* drop old route */
 947         skb_dst_drop(skb);
 948         skb_dst_set(skb, &rt->dst);
 949
 950         /*
 951          *      Push down and install the IPIP header.
 952          */
 953         iph                     =       ipv6_hdr(skb);
 954         iph->version            =       6;
 955         iph->nexthdr            =       IPPROTO_IPV6;
 956         iph->payload_len        =       old_iph->payload_len;
 957         be16_add_cpu(&iph->payload_len, sizeof(*old_iph));
 958         iph->priority           =       old_iph->priority;
 959         memset(&iph->flow_lbl, 0, sizeof(iph->flow_lbl));
 960         ipv6_addr_copy(&iph->daddr, &cp->daddr.in6);
 961         ipv6_addr_copy(&iph->saddr, &saddr);
 962         iph->hop_limit          =       old_iph->hop_limit;
 963
 964         /* Another hack: avoid icmp_send in ip_fragment */
 965         skb->local_df = 1;
 966
 967         ret = IP_VS_XMIT_TUNNEL(skb, cp);
 968         if (ret == NF_ACCEPT)
 969                 ip6_local_out(skb);
 970         else if (ret == NF_DROP)
 971                 kfree_skb(skb);
 972
 973         LeaveFunction(10);
 974
 975         return NF_STOLEN;
 976
 977 tx_error_icmp:
 978         dst_link_failure(skb);
 979 tx_error:
 980         kfree_skb(skb);
 981         LeaveFunction(10);
 982         return NF_STOLEN;
 983 tx_error_put:
 984         dst_release(&rt->dst);
 985         goto tx_error;
 986 }
 987 #endif
 988
 989
 990 /*
 991  *      Direct Routing transmitter
 992  *      Used for ANY protocol
 993  */
 994 int
 995 ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 996               struct ip_vs_protocol *pp)
 997 {
 998         struct rtable *rt;                      /* Route to the other host */
 999         struct iphdr  *iph = ip_hdr(skb);
1000         int    mtu;
1001
1002         EnterFunction(10);
1003
1004         if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip,
1005                                       RT_TOS(iph->tos), 1|2)))
1006                 goto tx_error_icmp;
1007         if (rt->rt_flags & RTCF_LOCAL) {
1008                 ip_rt_put(rt);
1009                 IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 1);
1010         }
1011
1012         /* MTU checking */
1013         mtu = dst_mtu(&rt->dst);
1014         if ((iph->frag_off & htons(IP_DF)) && skb->len > mtu &&
1015             !skb_is_gso(skb)) {
1016                 icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu));
1017                 ip_rt_put(rt);
1018                 IP_VS_DBG_RL("%s(): frag needed\n", __func__);
1019                 goto tx_error;
1020         }
1021
1022         /*
1023          * Call ip_send_check because we are not sure it is called
1024          * after ip_defrag. Is copy-on-write needed?
1025          */
1026         if (unlikely((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)) {
1027                 ip_rt_put(rt);
1028                 return NF_STOLEN;
1029         }
1030         ip_send_check(ip_hdr(skb));
1031
1032         /* drop old route */
1033         skb_dst_drop(skb);
1034         skb_dst_set(skb, &rt->dst);
1035
1036         /* Another hack: avoid icmp_send in ip_fragment */
1037         skb->local_df = 1;
1038
1039         IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 0);
1040
1041         LeaveFunction(10);
1042         return NF_STOLEN;
1043
1044   tx_error_icmp:
1045         dst_link_failure(skb);
1046   tx_error:
1047         kfree_skb(skb);
1048         LeaveFunction(10);
1049         return NF_STOLEN;
1050 }
1051
1052 #ifdef CONFIG_IP_VS_IPV6
1053 int
1054 ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1055                  struct ip_vs_protocol *pp)
1056 {
1057         struct rt6_info *rt;                    /* Route to the other host */
1058         int    mtu;
1059
1060         EnterFunction(10);
1061
1062         if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, NULL,
1063                                          0, 1|2)))
1064                 goto tx_error_icmp;
1065         if (__ip_vs_is_local_route6(rt)) {
1066                 dst_release(&rt->dst);
1067                 IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 1);
1068         }
1069
1070         /* MTU checking */
1071         mtu = dst_mtu(&rt->dst);
1072         if (skb->len > mtu) {
1073                 if (!skb->dev) {
1074                         struct net *net = dev_net(skb_dst(skb)->dev);
1075
1076                         skb->dev = net->loopback_dev;
1077                 }
1078                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
1079                 dst_release(&rt->dst);
1080                 IP_VS_DBG_RL("%s(): frag needed\n", __func__);
1081                 goto tx_error;
1082         }
1083
1084         /*
1085          * Call ip_send_check because we are not sure it is called
1086          * after ip_defrag. Is copy-on-write needed?
1087          */
1088         skb = skb_share_check(skb, GFP_ATOMIC);
1089         if (unlikely(skb == NULL)) {
1090                 dst_release(&rt->dst);
1091                 return NF_STOLEN;
1092         }
1093
1094         /* drop old route */
1095         skb_dst_drop(skb);
1096         skb_dst_set(skb, &rt->dst);
1097
1098         /* Another hack: avoid icmp_send in ip_fragment */
1099         skb->local_df = 1;
1100
1101         IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 0);
1102
1103         LeaveFunction(10);
1104         return NF_STOLEN;
1105
1106 tx_error_icmp:
1107         dst_link_failure(skb);
1108 tx_error:
1109         kfree_skb(skb);
1110         LeaveFunction(10);
1111         return NF_STOLEN;
1112 }
1113 #endif
1114
1115
1116 /*
1117  *      ICMP packet transmitter
1118  *      called by the ip_vs_in_icmp
1119  */
1120 int
1121 ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1122                 struct ip_vs_protocol *pp, int offset)
1123 {
1124         struct rtable   *rt;    /* Route to the other host */
1125         int mtu;
1126         int rc;
1127         int local;
1128
1129         EnterFunction(10);
1130
1131         /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1132            forwarded directly here, because there is no need to
1133            translate address/port back */
1134         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1135                 if (cp->packet_xmit)
1136                         rc = cp->packet_xmit(skb, cp, pp);
1137                 else
1138                         rc = NF_ACCEPT;
1139                 /* do not touch skb anymore */
1140                 atomic_inc(&cp->in_pkts);
1141                 goto out;
1142         }
1143
1144         /*
1145          * mangle and send the packet here (only for VS/NAT)
1146          */
1147
1148         if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip,
1149                                       RT_TOS(ip_hdr(skb)->tos), 1|2|4)))
1150                 goto tx_error_icmp;
1151         local = rt->rt_flags & RTCF_LOCAL;
1152
1153         /*
1154          * Avoid duplicate tuple in reply direction for NAT traffic
1155          * to local address when connection is sync-ed
1156          */
1157 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1158         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1159                 enum ip_conntrack_info ctinfo;
1160                 struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo);
1161
1162                 if (ct && !nf_ct_is_untracked(ct)) {
1163                         IP_VS_DBG(10, "%s(): "
1164                                   "stopping DNAT to local address %pI4\n",
1165                                   __func__, &cp->daddr.ip);
1166                         goto tx_error_put;
1167                 }
1168         }
1169 #endif
1170
1171         /* From world but DNAT to loopback address? */
1172         if (local && ipv4_is_loopback(rt->rt_dst) &&
1173             rt_is_input_route(skb_rtable(skb))) {
1174                 IP_VS_DBG(1, "%s(): "
1175                           "stopping DNAT to loopback %pI4\n",
1176                           __func__, &cp->daddr.ip);
1177                 goto tx_error_put;
1178         }
1179
1180         /* MTU checking */
1181         mtu = dst_mtu(&rt->dst);
1182         if ((skb->len > mtu) && (ip_hdr(skb)->frag_off & htons(IP_DF)) &&
1183             !skb_is_gso(skb)) {
1184                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
1185                 IP_VS_DBG_RL("%s(): frag needed\n", __func__);
1186                 goto tx_error_put;
1187         }
1188
1189         /* copy-on-write the packet before mangling it */
1190         if (!skb_make_writable(skb, offset))
1191                 goto tx_error_put;
1192
1193         if (skb_cow(skb, rt->dst.dev->hard_header_len))
1194                 goto tx_error_put;
1195
1196         ip_vs_nat_icmp(skb, pp, cp, 0);
1197
1198         if (!local) {
1199                 /* drop the old route when skb is not shared */
1200                 skb_dst_drop(skb);
1201                 skb_dst_set(skb, &rt->dst);
1202         } else {
1203                 ip_rt_put(rt);
1204                 /*
1205                  * Some IPv4 replies get local address from routes,
1206                  * not from iph, so while we DNAT after routing
1207                  * we need this second input/output route.
1208                  */
1209                 if (!__ip_vs_reroute_locally(skb))
1210                         goto tx_error;
1211         }
1212
1213         /* Another hack: avoid icmp_send in ip_fragment */
1214         skb->local_df = 1;
1215
1216         IP_VS_XMIT_NAT(NFPROTO_IPV4, skb, cp, local);
1217
1218         rc = NF_STOLEN;
1219         goto out;
1220
1221   tx_error_icmp:
1222         dst_link_failure(skb);
1223   tx_error:
1224         dev_kfree_skb(skb);
1225         rc = NF_STOLEN;
1226   out:
1227         LeaveFunction(10);
1228         return rc;
1229   tx_error_put:
1230         ip_rt_put(rt);
1231         goto tx_error;
1232 }
1233
1234 #ifdef CONFIG_IP_VS_IPV6
1235 int
1236 ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1237                 struct ip_vs_protocol *pp, int offset)
1238 {
1239         struct rt6_info *rt;    /* Route to the other host */
1240         int mtu;
1241         int rc;
1242         int local;
1243
1244         EnterFunction(10);
1245
1246         /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1247            forwarded directly here, because there is no need to
1248            translate address/port back */
1249         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1250                 if (cp->packet_xmit)
1251                         rc = cp->packet_xmit(skb, cp, pp);
1252                 else
1253                         rc = NF_ACCEPT;
1254                 /* do not touch skb anymore */
1255                 atomic_inc(&cp->in_pkts);
1256                 goto out;
1257         }
1258
1259         /*
1260          * mangle and send the packet here (only for VS/NAT)
1261          */
1262
1263         if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, NULL,
1264                                          0, 1|2|4)))
1265                 goto tx_error_icmp;
1266
1267         local = __ip_vs_is_local_route6(rt);
1268         /*
1269          * Avoid duplicate tuple in reply direction for NAT traffic
1270          * to local address when connection is sync-ed
1271          */
1272 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1273         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1274                 enum ip_conntrack_info ctinfo;
1275                 struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo);
1276
1277                 if (ct && !nf_ct_is_untracked(ct)) {
1278                         IP_VS_DBG(10, "%s(): "
1279                                   "stopping DNAT to local address %pI6\n",
1280                                   __func__, &cp->daddr.in6);
1281                         goto tx_error_put;
1282                 }
1283         }
1284 #endif
1285
1286         /* From world but DNAT to loopback address? */
1287         if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
1288             ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_LOOPBACK) {
1289                 IP_VS_DBG(1, "%s(): "
1290                           "stopping DNAT to loopback %pI6\n",
1291                           __func__, &cp->daddr.in6);
1292                 goto tx_error_put;
1293         }
1294
1295         /* MTU checking */
1296         mtu = dst_mtu(&rt->dst);
1297         if (skb->len > mtu && !skb_is_gso(skb)) {
1298                 if (!skb->dev) {
1299                         struct net *net = dev_net(skb_dst(skb)->dev);
1300
1301                         skb->dev = net->loopback_dev;
1302                 }
1303                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
1304                 IP_VS_DBG_RL("%s(): frag needed\n", __func__);
1305                 goto tx_error_put;
1306         }
1307
1308         /* copy-on-write the packet before mangling it */
1309         if (!skb_make_writable(skb, offset))
1310                 goto tx_error_put;
1311
1312         if (skb_cow(skb, rt->dst.dev->hard_header_len))
1313                 goto tx_error_put;
1314
1315         ip_vs_nat_icmp_v6(skb, pp, cp, 0);
1316
1317         if (!local || !skb->dev) {
1318                 /* drop the old route when skb is not shared */
1319                 skb_dst_drop(skb);
1320                 skb_dst_set(skb, &rt->dst);
1321         } else {
1322                 /* destined to loopback, do we need to change route? */
1323                 dst_release(&rt->dst);
1324         }
1325
1326         /* Another hack: avoid icmp_send in ip_fragment */
1327         skb->local_df = 1;
1328
1329         IP_VS_XMIT_NAT(NFPROTO_IPV6, skb, cp, local);
1330
1331         rc = NF_STOLEN;
1332         goto out;
1333
1334 tx_error_icmp:
1335         dst_link_failure(skb);
1336 tx_error:
1337         dev_kfree_skb(skb);
1338         rc = NF_STOLEN;
1339 out:
1340         LeaveFunction(10);
1341         return rc;
1342 tx_error_put:
1343         dst_release(&rt->dst);
1344         goto tx_error;
1345 }
1346 #endif