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->fl.iif) {
 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, pp, skb, 0, "ip_vs_nat_xmit(): "
 547                                          "stopping DNAT to local address");
 548                         goto tx_error_put;
 549                 }
 550         }
 551 #endif
 552
 553         /* From world but DNAT to loopback address? */
 554         if (local && ipv4_is_loopback(rt->rt_dst) && skb_rtable(skb)->fl.iif) {
 555                 IP_VS_DBG_RL_PKT(1, pp, skb, 0, "ip_vs_nat_xmit(): "
 556                                  "stopping DNAT to loopback address");
 557                 goto tx_error_put;
 558         }
 559
 560         /* MTU checking */
 561         mtu = dst_mtu(&rt->dst);
 562         if ((skb->len > mtu) && (iph->frag_off & htons(IP_DF))) {
 563                 icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu));
 564                 IP_VS_DBG_RL_PKT(0, pp, skb, 0, "ip_vs_nat_xmit(): frag needed for");
 565                 goto tx_error_put;
 566         }
 567
 568         /* copy-on-write the packet before mangling it */
 569         if (!skb_make_writable(skb, sizeof(struct iphdr)))
 570                 goto tx_error_put;
 571
 572         if (skb_cow(skb, rt->dst.dev->hard_header_len))
 573                 goto tx_error_put;
 574
 575         /* mangle the packet */
 576         if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp))
 577                 goto tx_error_put;
 578         ip_hdr(skb)->daddr = cp->daddr.ip;
 579         ip_send_check(ip_hdr(skb));
 580
 581         if (!local) {
 582                 /* drop old route */
 583                 skb_dst_drop(skb);
 584                 skb_dst_set(skb, &rt->dst);
 585         } else {
 586                 ip_rt_put(rt);
 587                 /*
 588                  * Some IPv4 replies get local address from routes,
 589                  * not from iph, so while we DNAT after routing
 590                  * we need this second input/output route.
 591                  */
 592                 if (!__ip_vs_reroute_locally(skb))
 593                         goto tx_error;
 594         }
 595
 596         IP_VS_DBG_PKT(10, pp, skb, 0, "After DNAT");
 597
 598         /* FIXME: when application helper enlarges the packet and the length
 599            is larger than the MTU of outgoing device, there will be still
 600            MTU problem. */
 601
 602         /* Another hack: avoid icmp_send in ip_fragment */
 603         skb->local_df = 1;
 604
 605         IP_VS_XMIT_NAT(NFPROTO_IPV4, skb, cp, local);
 606
 607         LeaveFunction(10);
 608         return NF_STOLEN;
 609
 610   tx_error_icmp:
 611         dst_link_failure(skb);
 612   tx_error:
 613         kfree_skb(skb);
 614         LeaveFunction(10);
 615         return NF_STOLEN;
 616   tx_error_put:
 617         ip_rt_put(rt);
 618         goto tx_error;
 619 }
 620
 621 #ifdef CONFIG_IP_VS_IPV6
 622 int
 623 ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
 624                   struct ip_vs_protocol *pp)
 625 {
 626         struct rt6_info *rt;            /* Route to the other host */
 627         int mtu;
 628         int local;
 629
 630         EnterFunction(10);
 631
 632         /* check if it is a connection of no-client-port */
 633         if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) {
 634                 __be16 _pt, *p;
 635                 p = skb_header_pointer(skb, sizeof(struct ipv6hdr),
 636                                        sizeof(_pt), &_pt);
 637                 if (p == NULL)
 638                         goto tx_error;
 639                 ip_vs_conn_fill_cport(cp, *p);
 640                 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
 641         }
 642
 643         if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, NULL,
 644                                          0, 1|2|4)))
 645                 goto tx_error_icmp;
 646         local = __ip_vs_is_local_route6(rt);
 647         /*
 648          * Avoid duplicate tuple in reply direction for NAT traffic
 649          * to local address when connection is sync-ed
 650          */
 651 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
 652         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
 653                 enum ip_conntrack_info ctinfo;
 654                 struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo);
 655
 656                 if (ct && !nf_ct_is_untracked(ct)) {
 657                         IP_VS_DBG_RL_PKT(10, pp, skb, 0,
 658                                          "ip_vs_nat_xmit_v6(): "
 659                                          "stopping DNAT to local address");
 660                         goto tx_error_put;
 661                 }
 662         }
 663 #endif
 664
 665         /* From world but DNAT to loopback address? */
 666         if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
 667             ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_LOOPBACK) {
 668                 IP_VS_DBG_RL_PKT(1, pp, skb, 0,
 669                                  "ip_vs_nat_xmit_v6(): "
 670                                  "stopping DNAT to loopback address");
 671                 goto tx_error_put;
 672         }
 673
 674         /* MTU checking */
 675         mtu = dst_mtu(&rt->dst);
 676         if (skb->len > mtu) {
 677                 if (!skb->dev) {
 678                         struct net *net = dev_net(skb_dst(skb)->dev);
 679
 680                         skb->dev = net->loopback_dev;
 681                 }
 682                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 683                 IP_VS_DBG_RL_PKT(0, pp, skb, 0,
 684                                  "ip_vs_nat_xmit_v6(): frag needed for");
 685                 goto tx_error_put;
 686         }
 687
 688         /* copy-on-write the packet before mangling it */
 689         if (!skb_make_writable(skb, sizeof(struct ipv6hdr)))
 690                 goto tx_error_put;
 691
 692         if (skb_cow(skb, rt->dst.dev->hard_header_len))
 693                 goto tx_error_put;
 694
 695         /* mangle the packet */
 696         if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp))
 697                 goto tx_error;
 698         ipv6_addr_copy(&ipv6_hdr(skb)->daddr, &cp->daddr.in6);
 699
 700         if (!local || !skb->dev) {
 701                 /* drop the old route when skb is not shared */
 702                 skb_dst_drop(skb);
 703                 skb_dst_set(skb, &rt->dst);
 704         } else {
 705                 /* destined to loopback, do we need to change route? */
 706                 dst_release(&rt->dst);
 707         }
 708
 709         IP_VS_DBG_PKT(10, pp, skb, 0, "After DNAT");
 710
 711         /* FIXME: when application helper enlarges the packet and the length
 712            is larger than the MTU of outgoing device, there will be still
 713            MTU problem. */
 714
 715         /* Another hack: avoid icmp_send in ip_fragment */
 716         skb->local_df = 1;
 717
 718         IP_VS_XMIT_NAT(NFPROTO_IPV6, skb, cp, local);
 719
 720         LeaveFunction(10);
 721         return NF_STOLEN;
 722
 723 tx_error_icmp:
 724         dst_link_failure(skb);
 725 tx_error:
 726         LeaveFunction(10);
 727         kfree_skb(skb);
 728         return NF_STOLEN;
 729 tx_error_put:
 730         dst_release(&rt->dst);
 731         goto tx_error;
 732 }
 733 #endif
 734
 735
 736 /*
 737  *   IP Tunneling transmitter
 738  *
 739  *   This function encapsulates the packet in a new IP packet, its
 740  *   destination will be set to cp->daddr. Most code of this function
 741  *   is taken from ipip.c.
 742  *
 743  *   It is used in VS/TUN cluster. The load balancer selects a real
 744  *   server from a cluster based on a scheduling algorithm,
 745  *   encapsulates the request packet and forwards it to the selected
 746  *   server. For example, all real servers are configured with
 747  *   "ifconfig tunl0 <Virtual IP Address> up". When the server receives
 748  *   the encapsulated packet, it will decapsulate the packet, processe
 749  *   the request and return the response packets directly to the client
 750  *   without passing the load balancer. This can greatly increase the
 751  *   scalability of virtual server.
 752  *
 753  *   Used for ANY protocol
 754  */
 755 int
 756 ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 757                   struct ip_vs_protocol *pp)
 758 {
 759         struct rtable *rt;                      /* Route to the other host */
 760         struct net_device *tdev;                /* Device to other host */
 761         struct iphdr  *old_iph = ip_hdr(skb);
 762         u8     tos = old_iph->tos;
 763         __be16 df = old_iph->frag_off;
 764         struct iphdr  *iph;                     /* Our new IP header */
 765         unsigned int max_headroom;              /* The extra header space needed */
 766         int    mtu;
 767         int ret;
 768
 769         EnterFunction(10);
 770
 771         if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip,
 772                                       RT_TOS(tos), 1|2)))
 773                 goto tx_error_icmp;
 774         if (rt->rt_flags & RTCF_LOCAL) {
 775                 ip_rt_put(rt);
 776                 IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 1);
 777         }
 778
 779         tdev = rt->dst.dev;
 780
 781         mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr);
 782         if (mtu < 68) {
 783                 IP_VS_DBG_RL("%s(): mtu less than 68\n", __func__);
 784                 goto tx_error_put;
 785         }
 786         if (skb_dst(skb))
 787                 skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu);
 788
 789         df |= (old_iph->frag_off & htons(IP_DF));
 790
 791         if ((old_iph->frag_off & htons(IP_DF))
 792             && mtu < ntohs(old_iph->tot_len)) {
 793                 icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu));
 794                 IP_VS_DBG_RL("%s(): frag needed\n", __func__);
 795                 goto tx_error_put;
 796         }
 797
 798         /*
 799          * Okay, now see if we can stuff it in the buffer as-is.
 800          */
 801         max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct iphdr);
 802
 803         if (skb_headroom(skb) < max_headroom
 804             || skb_cloned(skb) || skb_shared(skb)) {
 805                 struct sk_buff *new_skb =
 806                         skb_realloc_headroom(skb, max_headroom);
 807                 if (!new_skb) {
 808                         ip_rt_put(rt);
 809                         kfree_skb(skb);
 810                         IP_VS_ERR_RL("%s(): no memory\n", __func__);
 811                         return NF_STOLEN;
 812                 }
 813                 kfree_skb(skb);
 814                 skb = new_skb;
 815                 old_iph = ip_hdr(skb);
 816         }
 817
 818         skb->transport_header = skb->network_header;
 819
 820         /* fix old IP header checksum */
 821         ip_send_check(old_iph);
 822
 823         skb_push(skb, sizeof(struct iphdr));
 824         skb_reset_network_header(skb);
 825         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 826
 827         /* drop old route */
 828         skb_dst_drop(skb);
 829         skb_dst_set(skb, &rt->dst);
 830
 831         /*
 832          *      Push down and install the IPIP header.
 833          */
 834         iph                     =       ip_hdr(skb);
 835         iph->version            =       4;
 836         iph->ihl                =       sizeof(struct iphdr)>>2;
 837         iph->frag_off           =       df;
 838         iph->protocol           =       IPPROTO_IPIP;
 839         iph->tos                =       tos;
 840         iph->daddr              =       rt->rt_dst;
 841         iph->saddr              =       rt->rt_src;
 842         iph->ttl                =       old_iph->ttl;
 843         ip_select_ident(iph, &rt->dst, NULL);
 844
 845         /* Another hack: avoid icmp_send in ip_fragment */
 846         skb->local_df = 1;
 847
 848         ret = IP_VS_XMIT_TUNNEL(skb, cp);
 849         if (ret == NF_ACCEPT)
 850                 ip_local_out(skb);
 851         else if (ret == NF_DROP)
 852                 kfree_skb(skb);
 853
 854         LeaveFunction(10);
 855
 856         return NF_STOLEN;
 857
 858   tx_error_icmp:
 859         dst_link_failure(skb);
 860   tx_error:
 861         kfree_skb(skb);
 862         LeaveFunction(10);
 863         return NF_STOLEN;
 864 tx_error_put:
 865         ip_rt_put(rt);
 866         goto tx_error;
 867 }
 868
 869 #ifdef CONFIG_IP_VS_IPV6
 870 int
 871 ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
 872                      struct ip_vs_protocol *pp)
 873 {
 874         struct rt6_info *rt;            /* Route to the other host */
 875         struct in6_addr saddr;          /* Source for tunnel */
 876         struct net_device *tdev;        /* Device to other host */
 877         struct ipv6hdr  *old_iph = ipv6_hdr(skb);
 878         struct ipv6hdr  *iph;           /* Our new IP header */
 879         unsigned int max_headroom;      /* The extra header space needed */
 880         int    mtu;
 881         int ret;
 882
 883         EnterFunction(10);
 884
 885         if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6,
 886                                          &saddr, 1, 1|2)))
 887                 goto tx_error_icmp;
 888         if (__ip_vs_is_local_route6(rt)) {
 889                 dst_release(&rt->dst);
 890                 IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 1);
 891         }
 892
 893         tdev = rt->dst.dev;
 894
 895         mtu = dst_mtu(&rt->dst) - sizeof(struct ipv6hdr);
 896         if (mtu < IPV6_MIN_MTU) {
 897                 IP_VS_DBG_RL("%s(): mtu less than %d\n", __func__,
 898                              IPV6_MIN_MTU);
 899                 goto tx_error_put;
 900         }
 901         if (skb_dst(skb))
 902                 skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu);
 903
 904         if (mtu < ntohs(old_iph->payload_len) + sizeof(struct ipv6hdr)) {
 905                 if (!skb->dev) {
 906                         struct net *net = dev_net(skb_dst(skb)->dev);
 907
 908                         skb->dev = net->loopback_dev;
 909                 }
 910                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 911                 IP_VS_DBG_RL("%s(): frag needed\n", __func__);
 912                 goto tx_error_put;
 913         }
 914
 915         /*
 916          * Okay, now see if we can stuff it in the buffer as-is.
 917          */
 918         max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct ipv6hdr);
 919
 920         if (skb_headroom(skb) < max_headroom
 921             || skb_cloned(skb) || skb_shared(skb)) {
 922                 struct sk_buff *new_skb =
 923                         skb_realloc_headroom(skb, max_headroom);
 924                 if (!new_skb) {
 925                         dst_release(&rt->dst);
 926                         kfree_skb(skb);
 927                         IP_VS_ERR_RL("%s(): no memory\n", __func__);
 928                         return NF_STOLEN;
 929                 }
 930                 kfree_skb(skb);
 931                 skb = new_skb;
 932                 old_iph = ipv6_hdr(skb);
 933         }
 934
 935         skb->transport_header = skb->network_header;
 936
 937         skb_push(skb, sizeof(struct ipv6hdr));
 938         skb_reset_network_header(skb);
 939         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 940
 941         /* drop old route */
 942         skb_dst_drop(skb);
 943         skb_dst_set(skb, &rt->dst);
 944
 945         /*
 946          *      Push down and install the IPIP header.
 947          */
 948         iph                     =       ipv6_hdr(skb);
 949         iph->version            =       6;
 950         iph->nexthdr            =       IPPROTO_IPV6;
 951         iph->payload_len        =       old_iph->payload_len;
 952         be16_add_cpu(&iph->payload_len, sizeof(*old_iph));
 953         iph->priority           =       old_iph->priority;
 954         memset(&iph->flow_lbl, 0, sizeof(iph->flow_lbl));
 955         ipv6_addr_copy(&iph->daddr, &cp->daddr.in6);
 956         ipv6_addr_copy(&iph->saddr, &saddr);
 957         iph->hop_limit          =       old_iph->hop_limit;
 958
 959         /* Another hack: avoid icmp_send in ip_fragment */
 960         skb->local_df = 1;
 961
 962         ret = IP_VS_XMIT_TUNNEL(skb, cp);
 963         if (ret == NF_ACCEPT)
 964                 ip6_local_out(skb);
 965         else if (ret == NF_DROP)
 966                 kfree_skb(skb);
 967
 968         LeaveFunction(10);
 969
 970         return NF_STOLEN;
 971
 972 tx_error_icmp:
 973         dst_link_failure(skb);
 974 tx_error:
 975         kfree_skb(skb);
 976         LeaveFunction(10);
 977         return NF_STOLEN;
 978 tx_error_put:
 979         dst_release(&rt->dst);
 980         goto tx_error;
 981 }
 982 #endif
 983
 984
 985 /*
 986  *      Direct Routing transmitter
 987  *      Used for ANY protocol
 988  */
 989 int
 990 ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 991               struct ip_vs_protocol *pp)
 992 {
 993         struct rtable *rt;                      /* Route to the other host */
 994         struct iphdr  *iph = ip_hdr(skb);
 995         int    mtu;
 996
 997         EnterFunction(10);
 998
 999         if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip,
1000                                       RT_TOS(iph->tos), 1|2)))
1001                 goto tx_error_icmp;
1002         if (rt->rt_flags & RTCF_LOCAL) {
1003                 ip_rt_put(rt);
1004                 IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 1);
1005         }
1006
1007         /* MTU checking */
1008         mtu = dst_mtu(&rt->dst);
1009         if ((iph->frag_off & htons(IP_DF)) && skb->len > mtu) {
1010                 icmp_send(skb, ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED, htonl(mtu));
1011                 ip_rt_put(rt);
1012                 IP_VS_DBG_RL("%s(): frag needed\n", __func__);
1013                 goto tx_error;
1014         }
1015
1016         /*
1017          * Call ip_send_check because we are not sure it is called
1018          * after ip_defrag. Is copy-on-write needed?
1019          */
1020         if (unlikely((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)) {
1021                 ip_rt_put(rt);
1022                 return NF_STOLEN;
1023         }
1024         ip_send_check(ip_hdr(skb));
1025
1026         /* drop old route */
1027         skb_dst_drop(skb);
1028         skb_dst_set(skb, &rt->dst);
1029
1030         /* Another hack: avoid icmp_send in ip_fragment */
1031         skb->local_df = 1;
1032
1033         IP_VS_XMIT(NFPROTO_IPV4, skb, cp, 0);
1034
1035         LeaveFunction(10);
1036         return NF_STOLEN;
1037
1038   tx_error_icmp:
1039         dst_link_failure(skb);
1040   tx_error:
1041         kfree_skb(skb);
1042         LeaveFunction(10);
1043         return NF_STOLEN;
1044 }
1045
1046 #ifdef CONFIG_IP_VS_IPV6
1047 int
1048 ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1049                  struct ip_vs_protocol *pp)
1050 {
1051         struct rt6_info *rt;                    /* Route to the other host */
1052         int    mtu;
1053
1054         EnterFunction(10);
1055
1056         if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, NULL,
1057                                          0, 1|2)))
1058                 goto tx_error_icmp;
1059         if (__ip_vs_is_local_route6(rt)) {
1060                 dst_release(&rt->dst);
1061                 IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 1);
1062         }
1063
1064         /* MTU checking */
1065         mtu = dst_mtu(&rt->dst);
1066         if (skb->len > mtu) {
1067                 if (!skb->dev) {
1068                         struct net *net = dev_net(skb_dst(skb)->dev);
1069
1070                         skb->dev = net->loopback_dev;
1071                 }
1072                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
1073                 dst_release(&rt->dst);
1074                 IP_VS_DBG_RL("%s(): frag needed\n", __func__);
1075                 goto tx_error;
1076         }
1077
1078         /*
1079          * Call ip_send_check because we are not sure it is called
1080          * after ip_defrag. Is copy-on-write needed?
1081          */
1082         skb = skb_share_check(skb, GFP_ATOMIC);
1083         if (unlikely(skb == NULL)) {
1084                 dst_release(&rt->dst);
1085                 return NF_STOLEN;
1086         }
1087
1088         /* drop old route */
1089         skb_dst_drop(skb);
1090         skb_dst_set(skb, &rt->dst);
1091
1092         /* Another hack: avoid icmp_send in ip_fragment */
1093         skb->local_df = 1;
1094
1095         IP_VS_XMIT(NFPROTO_IPV6, skb, cp, 0);
1096
1097         LeaveFunction(10);
1098         return NF_STOLEN;
1099
1100 tx_error_icmp:
1101         dst_link_failure(skb);
1102 tx_error:
1103         kfree_skb(skb);
1104         LeaveFunction(10);
1105         return NF_STOLEN;
1106 }
1107 #endif
1108
1109
1110 /*
1111  *      ICMP packet transmitter
1112  *      called by the ip_vs_in_icmp
1113  */
1114 int
1115 ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1116                 struct ip_vs_protocol *pp, int offset)
1117 {
1118         struct rtable   *rt;    /* Route to the other host */
1119         int mtu;
1120         int rc;
1121         int local;
1122
1123         EnterFunction(10);
1124
1125         /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1126            forwarded directly here, because there is no need to
1127            translate address/port back */
1128         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1129                 if (cp->packet_xmit)
1130                         rc = cp->packet_xmit(skb, cp, pp);
1131                 else
1132                         rc = NF_ACCEPT;
1133                 /* do not touch skb anymore */
1134                 atomic_inc(&cp->in_pkts);
1135                 goto out;
1136         }
1137
1138         /*
1139          * mangle and send the packet here (only for VS/NAT)
1140          */
1141
1142         if (!(rt = __ip_vs_get_out_rt(skb, cp->dest, cp->daddr.ip,
1143                                       RT_TOS(ip_hdr(skb)->tos), 1|2|4)))
1144                 goto tx_error_icmp;
1145         local = rt->rt_flags & RTCF_LOCAL;
1146
1147         /*
1148          * Avoid duplicate tuple in reply direction for NAT traffic
1149          * to local address when connection is sync-ed
1150          */
1151 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1152         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1153                 enum ip_conntrack_info ctinfo;
1154                 struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo);
1155
1156                 if (ct && !nf_ct_is_untracked(ct)) {
1157                         IP_VS_DBG(10, "%s(): "
1158                                   "stopping DNAT to local address %pI4\n",
1159                                   __func__, &cp->daddr.ip);
1160                         goto tx_error_put;
1161                 }
1162         }
1163 #endif
1164
1165         /* From world but DNAT to loopback address? */
1166         if (local && ipv4_is_loopback(rt->rt_dst) && skb_rtable(skb)->fl.iif) {
1167                 IP_VS_DBG(1, "%s(): "
1168                           "stopping DNAT to loopback %pI4\n",
1169                           __func__, &cp->daddr.ip);
1170                 goto tx_error_put;
1171         }
1172
1173         /* MTU checking */
1174         mtu = dst_mtu(&rt->dst);
1175         if ((skb->len > mtu) && (ip_hdr(skb)->frag_off & htons(IP_DF))) {
1176                 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
1177                 IP_VS_DBG_RL("%s(): frag needed\n", __func__);
1178                 goto tx_error_put;
1179         }
1180
1181         /* copy-on-write the packet before mangling it */
1182         if (!skb_make_writable(skb, offset))
1183                 goto tx_error_put;
1184
1185         if (skb_cow(skb, rt->dst.dev->hard_header_len))
1186                 goto tx_error_put;
1187
1188         ip_vs_nat_icmp(skb, pp, cp, 0);
1189
1190         if (!local) {
1191                 /* drop the old route when skb is not shared */
1192                 skb_dst_drop(skb);
1193                 skb_dst_set(skb, &rt->dst);
1194         } else {
1195                 ip_rt_put(rt);
1196                 /*
1197                  * Some IPv4 replies get local address from routes,
1198                  * not from iph, so while we DNAT after routing
1199                  * we need this second input/output route.
1200                  */
1201                 if (!__ip_vs_reroute_locally(skb))
1202                         goto tx_error;
1203         }
1204
1205         /* Another hack: avoid icmp_send in ip_fragment */
1206         skb->local_df = 1;
1207
1208         IP_VS_XMIT_NAT(NFPROTO_IPV4, skb, cp, local);
1209
1210         rc = NF_STOLEN;
1211         goto out;
1212
1213   tx_error_icmp:
1214         dst_link_failure(skb);
1215   tx_error:
1216         dev_kfree_skb(skb);
1217         rc = NF_STOLEN;
1218   out:
1219         LeaveFunction(10);
1220         return rc;
1221   tx_error_put:
1222         ip_rt_put(rt);
1223         goto tx_error;
1224 }
1225
1226 #ifdef CONFIG_IP_VS_IPV6
1227 int
1228 ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1229                 struct ip_vs_protocol *pp, int offset)
1230 {
1231         struct rt6_info *rt;    /* Route to the other host */
1232         int mtu;
1233         int rc;
1234         int local;
1235
1236         EnterFunction(10);
1237
1238         /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1239            forwarded directly here, because there is no need to
1240            translate address/port back */
1241         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1242                 if (cp->packet_xmit)
1243                         rc = cp->packet_xmit(skb, cp, pp);
1244                 else
1245                         rc = NF_ACCEPT;
1246                 /* do not touch skb anymore */
1247                 atomic_inc(&cp->in_pkts);
1248                 goto out;
1249         }
1250
1251         /*
1252          * mangle and send the packet here (only for VS/NAT)
1253          */
1254
1255         if (!(rt = __ip_vs_get_out_rt_v6(skb, cp->dest, &cp->daddr.in6, NULL,
1256                                          0, 1|2|4)))
1257                 goto tx_error_icmp;
1258
1259         local = __ip_vs_is_local_route6(rt);
1260         /*
1261          * Avoid duplicate tuple in reply direction for NAT traffic
1262          * to local address when connection is sync-ed
1263          */
1264 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1265         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1266                 enum ip_conntrack_info ctinfo;
1267                 struct nf_conn *ct = ct = nf_ct_get(skb, &ctinfo);
1268
1269                 if (ct && !nf_ct_is_untracked(ct)) {
1270                         IP_VS_DBG(10, "%s(): "
1271                                   "stopping DNAT to local address %pI6\n",
1272                                   __func__, &cp->daddr.in6);
1273                         goto tx_error_put;
1274                 }
1275         }
1276 #endif
1277
1278         /* From world but DNAT to loopback address? */
1279         if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
1280             ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_LOOPBACK) {
1281                 IP_VS_DBG(1, "%s(): "
1282                           "stopping DNAT to loopback %pI6\n",
1283                           __func__, &cp->daddr.in6);
1284                 goto tx_error_put;
1285         }
1286
1287         /* MTU checking */
1288         mtu = dst_mtu(&rt->dst);
1289         if (skb->len > mtu) {
1290                 if (!skb->dev) {
1291                         struct net *net = dev_net(skb_dst(skb)->dev);
1292
1293                         skb->dev = net->loopback_dev;
1294                 }
1295                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
1296                 IP_VS_DBG_RL("%s(): frag needed\n", __func__);
1297                 goto tx_error_put;
1298         }
1299
1300         /* copy-on-write the packet before mangling it */
1301         if (!skb_make_writable(skb, offset))
1302                 goto tx_error_put;
1303
1304         if (skb_cow(skb, rt->dst.dev->hard_header_len))
1305                 goto tx_error_put;
1306
1307         ip_vs_nat_icmp_v6(skb, pp, cp, 0);
1308
1309         if (!local || !skb->dev) {
1310                 /* drop the old route when skb is not shared */
1311                 skb_dst_drop(skb);
1312                 skb_dst_set(skb, &rt->dst);
1313         } else {
1314                 /* destined to loopback, do we need to change route? */
1315                 dst_release(&rt->dst);
1316         }
1317
1318         /* Another hack: avoid icmp_send in ip_fragment */
1319         skb->local_df = 1;
1320
1321         IP_VS_XMIT_NAT(NFPROTO_IPV6, skb, cp, local);
1322
1323         rc = NF_STOLEN;
1324         goto out;
1325
1326 tx_error_icmp:
1327         dst_link_failure(skb);
1328 tx_error:
1329         dev_kfree_skb(skb);
1330         rc = NF_STOLEN;
1331 out:
1332         LeaveFunction(10);
1333         return rc;
1334 tx_error_put:
1335         dst_release(&rt->dst);
1336         goto tx_error;
1337 }
1338 #endif