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  * - we can use dst without ref while sending in RCU section, we use
  21  * ref when returning NF_ACCEPT for NAT-ed packet via loopback
  22  * LOCAL_OUT rules:
  23  * - skb->dev is NULL, skb->protocol is not set (both are set in POST_ROUTING)
  24  * - skb->pkt_type is not set yet
  25  * - the only place where we can see skb->sk != NULL
  26  */
  27
  28 #define KMSG_COMPONENT "IPVS"
  29 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  30
  31 #include <linux/kernel.h>
  32 #include <linux/slab.h>
  33 #include <linux/tcp.h>                  /* for tcphdr */
  34 #include <net/ip.h>
  35 #include <net/tcp.h>                    /* for csum_tcpudp_magic */
  36 #include <net/udp.h>
  37 #include <net/icmp.h>                   /* for icmp_send */
  38 #include <net/route.h>                  /* for ip_route_output */
  39 #include <net/ipv6.h>
  40 #include <net/ip6_route.h>
  41 #include <net/ip_tunnels.h>
  42 #include <net/addrconf.h>
  43 #include <linux/icmpv6.h>
  44 #include <linux/netfilter.h>
  45 #include <linux/netfilter_ipv4.h>
  46
  47 #include <net/ip_vs.h>
  48
  49 enum {
  50         IP_VS_RT_MODE_LOCAL     = 1, /* Allow local dest */
  51         IP_VS_RT_MODE_NON_LOCAL = 2, /* Allow non-local dest */
  52         IP_VS_RT_MODE_RDR       = 4, /* Allow redirect from remote daddr to
  53                                       * local
  54                                       */
  55         IP_VS_RT_MODE_CONNECT   = 8, /* Always bind route to saddr */
  56         IP_VS_RT_MODE_KNOWN_NH  = 16,/* Route via remote addr */
  57         IP_VS_RT_MODE_TUNNEL    = 32,/* Tunnel mode */
  58 };
  59
  60 static inline struct ip_vs_dest_dst *ip_vs_dest_dst_alloc(void)
  61 {
  62         return kmalloc(sizeof(struct ip_vs_dest_dst), GFP_ATOMIC);
  63 }
  64
  65 static inline void ip_vs_dest_dst_free(struct ip_vs_dest_dst *dest_dst)
  66 {
  67         kfree(dest_dst);
  68 }
  69
  70 /*
  71  *      Destination cache to speed up outgoing route lookup
  72  */
  73 static inline void
  74 __ip_vs_dst_set(struct ip_vs_dest *dest, struct ip_vs_dest_dst *dest_dst,
  75                 struct dst_entry *dst, u32 dst_cookie)
  76 {
  77         struct ip_vs_dest_dst *old;
  78
  79         old = rcu_dereference_protected(dest->dest_dst,
  80                                         lockdep_is_held(&dest->dst_lock));
  81
  82         if (dest_dst) {
  83                 dest_dst->dst_cache = dst;
  84                 dest_dst->dst_cookie = dst_cookie;
  85         }
  86         rcu_assign_pointer(dest->dest_dst, dest_dst);
  87
  88         if (old)
  89                 call_rcu(&old->rcu_head, ip_vs_dest_dst_rcu_free);
  90 }
  91
  92 static inline struct ip_vs_dest_dst *
  93 __ip_vs_dst_check(struct ip_vs_dest *dest)
  94 {
  95         struct ip_vs_dest_dst *dest_dst = rcu_dereference(dest->dest_dst);
  96         struct dst_entry *dst;
  97
  98         if (!dest_dst)
  99                 return NULL;
 100         dst = dest_dst->dst_cache;
 101         if (dst->obsolete &&
 102             dst->ops->check(dst, dest_dst->dst_cookie) == NULL)
 103                 return NULL;
 104         return dest_dst;
 105 }
 106
 107 static inline bool
 108 __mtu_check_toobig_v6(const struct sk_buff *skb, u32 mtu)
 109 {
 110         if (IP6CB(skb)->frag_max_size) {
 111                 /* frag_max_size tell us that, this packet have been
 112                  * defragmented by netfilter IPv6 conntrack module.
 113                  */
 114                 if (IP6CB(skb)->frag_max_size > mtu)
 115                         return true; /* largest fragment violate MTU */
 116         }
 117         else if (skb->len > mtu && !skb_is_gso(skb)) {
 118                 return true; /* Packet size violate MTU size */
 119         }
 120         return false;
 121 }
 122
 123 /* Get route to daddr, update *saddr, optionally bind route to saddr */
 124 static struct rtable *do_output_route4(struct net *net, __be32 daddr,
 125                                        int rt_mode, __be32 *saddr)
 126 {
 127         struct flowi4 fl4;
 128         struct rtable *rt;
 129         int loop = 0;
 130
 131         memset(&fl4, 0, sizeof(fl4));
 132         fl4.daddr = daddr;
 133         fl4.saddr = (rt_mode & IP_VS_RT_MODE_CONNECT) ? *saddr : 0;
 134         fl4.flowi4_flags = (rt_mode & IP_VS_RT_MODE_KNOWN_NH) ?
 135                            FLOWI_FLAG_KNOWN_NH : 0;
 136
 137 retry:
 138         rt = ip_route_output_key(net, &fl4);
 139         if (IS_ERR(rt)) {
 140                 /* Invalid saddr ? */
 141                 if (PTR_ERR(rt) == -EINVAL && *saddr &&
 142                     rt_mode & IP_VS_RT_MODE_CONNECT && !loop) {
 143                         *saddr = 0;
 144                         flowi4_update_output(&fl4, 0, 0, daddr, 0);
 145                         goto retry;
 146                 }
 147                 IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n", &daddr);
 148                 return NULL;
 149         } else if (!*saddr && rt_mode & IP_VS_RT_MODE_CONNECT && fl4.saddr) {
 150                 ip_rt_put(rt);
 151                 *saddr = fl4.saddr;
 152                 flowi4_update_output(&fl4, 0, 0, daddr, fl4.saddr);
 153                 loop++;
 154                 goto retry;
 155         }
 156         *saddr = fl4.saddr;
 157         return rt;
 158 }
 159
 160 #ifdef CONFIG_IP_VS_IPV6
 161 static inline int __ip_vs_is_local_route6(struct rt6_info *rt)
 162 {
 163         return rt->dst.dev && rt->dst.dev->flags & IFF_LOOPBACK;
 164 }
 165 #endif
 166
 167 static inline bool crosses_local_route_boundary(int skb_af, struct sk_buff *skb,
 168                                                 int rt_mode,
 169                                                 bool new_rt_is_local)
 170 {
 171         bool rt_mode_allow_local = !!(rt_mode & IP_VS_RT_MODE_LOCAL);
 172         bool rt_mode_allow_non_local = !!(rt_mode & IP_VS_RT_MODE_LOCAL);
 173         bool rt_mode_allow_redirect = !!(rt_mode & IP_VS_RT_MODE_RDR);
 174         bool source_is_loopback;
 175         bool old_rt_is_local;
 176
 177 #ifdef CONFIG_IP_VS_IPV6
 178         if (skb_af == AF_INET6) {
 179                 int addr_type = ipv6_addr_type(&ipv6_hdr(skb)->saddr);
 180
 181                 source_is_loopback =
 182                         (!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
 183                         (addr_type & IPV6_ADDR_LOOPBACK);
 184                 old_rt_is_local = __ip_vs_is_local_route6(
 185                         (struct rt6_info *)skb_dst(skb));
 186         } else
 187 #endif
 188         {
 189                 source_is_loopback = ipv4_is_loopback(ip_hdr(skb)->saddr);
 190                 old_rt_is_local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
 191         }
 192
 193         if (unlikely(new_rt_is_local)) {
 194                 if (!rt_mode_allow_local)
 195                         return true;
 196                 if (!rt_mode_allow_redirect && !old_rt_is_local)
 197                         return true;
 198         } else {
 199                 if (!rt_mode_allow_non_local)
 200                         return true;
 201                 if (source_is_loopback)
 202                         return true;
 203         }
 204         return false;
 205 }
 206
 207 static inline void maybe_update_pmtu(int skb_af, struct sk_buff *skb, int mtu)
 208 {
 209         struct sock *sk = skb->sk;
 210         struct rtable *ort = skb_rtable(skb);
 211
 212         if (!skb->dev && sk && sk->sk_state != TCP_TIME_WAIT)
 213                 ort->dst.ops->update_pmtu(&ort->dst, sk, NULL, mtu);
 214 }
 215
 216 static inline bool ensure_mtu_is_adequate(int skb_af, int rt_mode,
 217                                           struct ip_vs_iphdr *ipvsh,
 218                                           struct sk_buff *skb, int mtu)
 219 {
 220 #ifdef CONFIG_IP_VS_IPV6
 221         if (skb_af == AF_INET6) {
 222                 struct net *net = dev_net(skb_dst(skb)->dev);
 223
 224                 if (unlikely(__mtu_check_toobig_v6(skb, mtu))) {
 225                         if (!skb->dev)
 226                                 skb->dev = net->loopback_dev;
 227                         /* only send ICMP too big on first fragment */
 228                         if (!ipvsh->fragoffs)
 229                                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 230                         IP_VS_DBG(1, "frag needed for %pI6c\n",
 231                                   &ipv6_hdr(skb)->saddr);
 232                         return false;
 233                 }
 234         } else
 235 #endif
 236         {
 237                 struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 238
 239                 /* If we're going to tunnel the packet and pmtu discovery
 240                  * is disabled, we'll just fragment it anyway
 241                  */
 242                 if ((rt_mode & IP_VS_RT_MODE_TUNNEL) && !sysctl_pmtu_disc(ipvs))
 243                         return true;
 244
 245                 if (unlikely(ip_hdr(skb)->frag_off & htons(IP_DF) &&
 246                              skb->len > mtu && !skb_is_gso(skb))) {
 247                         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
 248                                   htonl(mtu));
 249                         IP_VS_DBG(1, "frag needed for %pI4\n",
 250                                   &ip_hdr(skb)->saddr);
 251                         return false;
 252                 }
 253         }
 254
 255         return true;
 256 }
 257
 258 /* Get route to destination or remote server */
 259 static int
 260 __ip_vs_get_out_rt(int skb_af, struct sk_buff *skb, struct ip_vs_dest *dest,
 261                    __be32 daddr, int rt_mode, __be32 *ret_saddr,
 262                    struct ip_vs_iphdr *ipvsh)
 263 {
 264         struct net *net = dev_net(skb_dst(skb)->dev);
 265         struct ip_vs_dest_dst *dest_dst;
 266         struct rtable *rt;                      /* Route to the other host */
 267         int mtu;
 268         int local, noref = 1;
 269
 270         if (dest) {
 271                 dest_dst = __ip_vs_dst_check(dest);
 272                 if (likely(dest_dst))
 273                         rt = (struct rtable *) dest_dst->dst_cache;
 274                 else {
 275                         dest_dst = ip_vs_dest_dst_alloc();
 276                         spin_lock_bh(&dest->dst_lock);
 277                         if (!dest_dst) {
 278                                 __ip_vs_dst_set(dest, NULL, NULL, 0);
 279                                 spin_unlock_bh(&dest->dst_lock);
 280                                 goto err_unreach;
 281                         }
 282                         rt = do_output_route4(net, dest->addr.ip, rt_mode,
 283                                               &dest_dst->dst_saddr.ip);
 284                         if (!rt) {
 285                                 __ip_vs_dst_set(dest, NULL, NULL, 0);
 286                                 spin_unlock_bh(&dest->dst_lock);
 287                                 ip_vs_dest_dst_free(dest_dst);
 288                                 goto err_unreach;
 289                         }
 290                         __ip_vs_dst_set(dest, dest_dst, &rt->dst, 0);
 291                         spin_unlock_bh(&dest->dst_lock);
 292                         IP_VS_DBG(10, "new dst %pI4, src %pI4, refcnt=%d\n",
 293                                   &dest->addr.ip, &dest_dst->dst_saddr.ip,
 294                                   atomic_read(&rt->dst.__refcnt));
 295                 }
 296                 if (ret_saddr)
 297                         *ret_saddr = dest_dst->dst_saddr.ip;
 298         } else {
 299                 __be32 saddr = htonl(INADDR_ANY);
 300
 301                 noref = 0;
 302
 303                 /* For such unconfigured boxes avoid many route lookups
 304                  * for performance reasons because we do not remember saddr
 305                  */
 306                 rt_mode &= ~IP_VS_RT_MODE_CONNECT;
 307                 rt = do_output_route4(net, daddr, rt_mode, &saddr);
 308                 if (!rt)
 309                         goto err_unreach;
 310                 if (ret_saddr)
 311                         *ret_saddr = saddr;
 312         }
 313
 314         local = (rt->rt_flags & RTCF_LOCAL) ? 1 : 0;
 315         if (unlikely(crosses_local_route_boundary(skb_af, skb, rt_mode,
 316                                                   local))) {
 317                 IP_VS_DBG_RL("We are crossing local and non-local addresses"
 318                              " daddr=%pI4\n", &daddr);
 319                 goto err_put;
 320         }
 321
 322         if (unlikely(local)) {
 323                 /* skb to local stack, preserve old route */
 324                 if (!noref)
 325                         ip_rt_put(rt);
 326                 return local;
 327         }
 328
 329         if (likely(!(rt_mode & IP_VS_RT_MODE_TUNNEL))) {
 330                 mtu = dst_mtu(&rt->dst);
 331         } else {
 332                 mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr);
 333                 if (mtu < 68) {
 334                         IP_VS_DBG_RL("%s(): mtu less than 68\n", __func__);
 335                         goto err_put;
 336                 }
 337                 maybe_update_pmtu(skb_af, skb, mtu);
 338         }
 339
 340         if (!ensure_mtu_is_adequate(skb_af, rt_mode, ipvsh, skb, mtu))
 341                 goto err_put;
 342
 343         skb_dst_drop(skb);
 344         if (noref) {
 345                 if (!local)
 346                         skb_dst_set_noref_force(skb, &rt->dst);
 347                 else
 348                         skb_dst_set(skb, dst_clone(&rt->dst));
 349         } else
 350                 skb_dst_set(skb, &rt->dst);
 351
 352         return local;
 353
 354 err_put:
 355         if (!noref)
 356                 ip_rt_put(rt);
 357         return -1;
 358
 359 err_unreach:
 360         dst_link_failure(skb);
 361         return -1;
 362 }
 363
 364 #ifdef CONFIG_IP_VS_IPV6
 365 static struct dst_entry *
 366 __ip_vs_route_output_v6(struct net *net, struct in6_addr *daddr,
 367                         struct in6_addr *ret_saddr, int do_xfrm)
 368 {
 369         struct dst_entry *dst;
 370         struct flowi6 fl6 = {
 371                 .daddr = *daddr,
 372         };
 373
 374         dst = ip6_route_output(net, NULL, &fl6);
 375         if (dst->error)
 376                 goto out_err;
 377         if (!ret_saddr)
 378                 return dst;
 379         if (ipv6_addr_any(&fl6.saddr) &&
 380             ipv6_dev_get_saddr(net, ip6_dst_idev(dst)->dev,
 381                                &fl6.daddr, 0, &fl6.saddr) < 0)
 382                 goto out_err;
 383         if (do_xfrm) {
 384                 dst = xfrm_lookup(net, dst, flowi6_to_flowi(&fl6), NULL, 0);
 385                 if (IS_ERR(dst)) {
 386                         dst = NULL;
 387                         goto out_err;
 388                 }
 389         }
 390         *ret_saddr = fl6.saddr;
 391         return dst;
 392
 393 out_err:
 394         dst_release(dst);
 395         IP_VS_DBG_RL("ip6_route_output error, dest: %pI6\n", daddr);
 396         return NULL;
 397 }
 398
 399 /*
 400  * Get route to destination or remote server
 401  */
 402 static int
 403 __ip_vs_get_out_rt_v6(int skb_af, struct sk_buff *skb, struct ip_vs_dest *dest,
 404                       struct in6_addr *daddr, struct in6_addr *ret_saddr,
 405                       struct ip_vs_iphdr *ipvsh, int do_xfrm, int rt_mode)
 406 {
 407         struct net *net = dev_net(skb_dst(skb)->dev);
 408         struct ip_vs_dest_dst *dest_dst;
 409         struct rt6_info *rt;                    /* Route to the other host */
 410         struct dst_entry *dst;
 411         int mtu;
 412         int local, noref = 1;
 413
 414         if (dest) {
 415                 dest_dst = __ip_vs_dst_check(dest);
 416                 if (likely(dest_dst))
 417                         rt = (struct rt6_info *) dest_dst->dst_cache;
 418                 else {
 419                         u32 cookie;
 420
 421                         dest_dst = ip_vs_dest_dst_alloc();
 422                         spin_lock_bh(&dest->dst_lock);
 423                         if (!dest_dst) {
 424                                 __ip_vs_dst_set(dest, NULL, NULL, 0);
 425                                 spin_unlock_bh(&dest->dst_lock);
 426                                 goto err_unreach;
 427                         }
 428                         dst = __ip_vs_route_output_v6(net, &dest->addr.in6,
 429                                                       &dest_dst->dst_saddr.in6,
 430                                                       do_xfrm);
 431                         if (!dst) {
 432                                 __ip_vs_dst_set(dest, NULL, NULL, 0);
 433                                 spin_unlock_bh(&dest->dst_lock);
 434                                 ip_vs_dest_dst_free(dest_dst);
 435                                 goto err_unreach;
 436                         }
 437                         rt = (struct rt6_info *) dst;
 438                         cookie = rt->rt6i_node ? rt->rt6i_node->fn_sernum : 0;
 439                         __ip_vs_dst_set(dest, dest_dst, &rt->dst, cookie);
 440                         spin_unlock_bh(&dest->dst_lock);
 441                         IP_VS_DBG(10, "new dst %pI6, src %pI6, refcnt=%d\n",
 442                                   &dest->addr.in6, &dest_dst->dst_saddr.in6,
 443                                   atomic_read(&rt->dst.__refcnt));
 444                 }
 445                 if (ret_saddr)
 446                         *ret_saddr = dest_dst->dst_saddr.in6;
 447         } else {
 448                 noref = 0;
 449                 dst = __ip_vs_route_output_v6(net, daddr, ret_saddr, do_xfrm);
 450                 if (!dst)
 451                         goto err_unreach;
 452                 rt = (struct rt6_info *) dst;
 453         }
 454
 455         local = __ip_vs_is_local_route6(rt);
 456
 457         if (unlikely(crosses_local_route_boundary(skb_af, skb, rt_mode,
 458                                                   local))) {
 459                 IP_VS_DBG_RL("We are crossing local and non-local addresses"
 460                              " daddr=%pI6\n", daddr);
 461                 goto err_put;
 462         }
 463
 464         if (unlikely(local)) {
 465                 /* skb to local stack, preserve old route */
 466                 if (!noref)
 467                         dst_release(&rt->dst);
 468                 return local;
 469         }
 470
 471         /* MTU checking */
 472         if (likely(!(rt_mode & IP_VS_RT_MODE_TUNNEL)))
 473                 mtu = dst_mtu(&rt->dst);
 474         else {
 475                 mtu = dst_mtu(&rt->dst) - sizeof(struct ipv6hdr);
 476                 if (mtu < IPV6_MIN_MTU) {
 477                         IP_VS_DBG_RL("%s(): mtu less than %d\n", __func__,
 478                                      IPV6_MIN_MTU);
 479                         goto err_put;
 480                 }
 481                 maybe_update_pmtu(skb_af, skb, mtu);
 482         }
 483
 484         if (!ensure_mtu_is_adequate(skb_af, rt_mode, ipvsh, skb, mtu))
 485                 goto err_put;
 486
 487         skb_dst_drop(skb);
 488         if (noref) {
 489                 if (!local)
 490                         skb_dst_set_noref_force(skb, &rt->dst);
 491                 else
 492                         skb_dst_set(skb, dst_clone(&rt->dst));
 493         } else
 494                 skb_dst_set(skb, &rt->dst);
 495
 496         return local;
 497
 498 err_put:
 499         if (!noref)
 500                 dst_release(&rt->dst);
 501         return -1;
 502
 503 err_unreach:
 504         dst_link_failure(skb);
 505         return -1;
 506 }
 507 #endif
 508
 509
 510 /* return NF_ACCEPT to allow forwarding or other NF_xxx on error */
 511 static inline int ip_vs_tunnel_xmit_prepare(struct sk_buff *skb,
 512                                             struct ip_vs_conn *cp)
 513 {
 514         int ret = NF_ACCEPT;
 515
 516         skb->ipvs_property = 1;
 517         if (unlikely(cp->flags & IP_VS_CONN_F_NFCT))
 518                 ret = ip_vs_confirm_conntrack(skb);
 519         if (ret == NF_ACCEPT) {
 520                 nf_reset(skb);
 521                 skb_forward_csum(skb);
 522         }
 523         return ret;
 524 }
 525
 526 /* return NF_STOLEN (sent) or NF_ACCEPT if local=1 (not sent) */
 527 static inline int ip_vs_nat_send_or_cont(int pf, struct sk_buff *skb,
 528                                          struct ip_vs_conn *cp, int local)
 529 {
 530         int ret = NF_STOLEN;
 531
 532         skb->ipvs_property = 1;
 533         if (likely(!(cp->flags & IP_VS_CONN_F_NFCT)))
 534                 ip_vs_notrack(skb);
 535         else
 536                 ip_vs_update_conntrack(skb, cp, 1);
 537         if (!local) {
 538                 skb_forward_csum(skb);
 539                 NF_HOOK(pf, NF_INET_LOCAL_OUT, skb, NULL, skb_dst(skb)->dev,
 540                         dst_output);
 541         } else
 542                 ret = NF_ACCEPT;
 543         return ret;
 544 }
 545
 546 /* return NF_STOLEN (sent) or NF_ACCEPT if local=1 (not sent) */
 547 static inline int ip_vs_send_or_cont(int pf, struct sk_buff *skb,
 548                                      struct ip_vs_conn *cp, int local)
 549 {
 550         int ret = NF_STOLEN;
 551
 552         skb->ipvs_property = 1;
 553         if (likely(!(cp->flags & IP_VS_CONN_F_NFCT)))
 554                 ip_vs_notrack(skb);
 555         if (!local) {
 556                 skb_forward_csum(skb);
 557                 NF_HOOK(pf, NF_INET_LOCAL_OUT, skb, NULL, skb_dst(skb)->dev,
 558                         dst_output);
 559         } else
 560                 ret = NF_ACCEPT;
 561         return ret;
 562 }
 563
 564
 565 /*
 566  *      NULL transmitter (do nothing except return NF_ACCEPT)
 567  */
 568 int
 569 ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 570                 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
 571 {
 572         /* we do not touch skb and do not need pskb ptr */
 573         return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
 574 }
 575
 576
 577 /*
 578  *      Bypass transmitter
 579  *      Let packets bypass the destination when the destination is not
 580  *      available, it may be only used in transparent cache cluster.
 581  */
 582 int
 583 ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 584                   struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
 585 {
 586         struct iphdr  *iph = ip_hdr(skb);
 587
 588         EnterFunction(10);
 589
 590         rcu_read_lock();
 591         if (__ip_vs_get_out_rt(cp->af, skb, NULL, iph->daddr,
 592                                IP_VS_RT_MODE_NON_LOCAL, NULL, ipvsh) < 0)
 593                 goto tx_error;
 594
 595         ip_send_check(iph);
 596
 597         /* Another hack: avoid icmp_send in ip_fragment */
 598         skb->ignore_df = 1;
 599
 600         ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 0);
 601         rcu_read_unlock();
 602
 603         LeaveFunction(10);
 604         return NF_STOLEN;
 605
 606  tx_error:
 607         kfree_skb(skb);
 608         rcu_read_unlock();
 609         LeaveFunction(10);
 610         return NF_STOLEN;
 611 }
 612
 613 #ifdef CONFIG_IP_VS_IPV6
 614 int
 615 ip_vs_bypass_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
 616                      struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
 617 {
 618         EnterFunction(10);
 619
 620         rcu_read_lock();
 621         if (__ip_vs_get_out_rt_v6(cp->af, skb, NULL, &ipvsh->daddr.in6, NULL,
 622                                   ipvsh, 0, IP_VS_RT_MODE_NON_LOCAL) < 0)
 623                 goto tx_error;
 624
 625         /* Another hack: avoid icmp_send in ip_fragment */
 626         skb->ignore_df = 1;
 627
 628         ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 0);
 629         rcu_read_unlock();
 630
 631         LeaveFunction(10);
 632         return NF_STOLEN;
 633
 634  tx_error:
 635         kfree_skb(skb);
 636         rcu_read_unlock();
 637         LeaveFunction(10);
 638         return NF_STOLEN;
 639 }
 640 #endif
 641
 642 /*
 643  *      NAT transmitter (only for outside-to-inside nat forwarding)
 644  *      Not used for related ICMP
 645  */
 646 int
 647 ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 648                struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
 649 {
 650         struct rtable *rt;              /* Route to the other host */
 651         int local, rc, was_input;
 652
 653         EnterFunction(10);
 654
 655         rcu_read_lock();
 656         /* check if it is a connection of no-client-port */
 657         if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) {
 658                 __be16 _pt, *p;
 659
 660                 p = skb_header_pointer(skb, ipvsh->len, sizeof(_pt), &_pt);
 661                 if (p == NULL)
 662                         goto tx_error;
 663                 ip_vs_conn_fill_cport(cp, *p);
 664                 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
 665         }
 666
 667         was_input = rt_is_input_route(skb_rtable(skb));
 668         local = __ip_vs_get_out_rt(cp->af, skb, cp->dest, cp->daddr.ip,
 669                                    IP_VS_RT_MODE_LOCAL |
 670                                    IP_VS_RT_MODE_NON_LOCAL |
 671                                    IP_VS_RT_MODE_RDR, NULL, ipvsh);
 672         if (local < 0)
 673                 goto tx_error;
 674         rt = skb_rtable(skb);
 675         /*
 676          * Avoid duplicate tuple in reply direction for NAT traffic
 677          * to local address when connection is sync-ed
 678          */
 679 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 680         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
 681                 enum ip_conntrack_info ctinfo;
 682                 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
 683
 684                 if (ct && !nf_ct_is_untracked(ct)) {
 685                         IP_VS_DBG_RL_PKT(10, AF_INET, pp, skb, 0,
 686                                          "ip_vs_nat_xmit(): "
 687                                          "stopping DNAT to local address");
 688                         goto tx_error;
 689                 }
 690         }
 691 #endif
 692
 693         /* From world but DNAT to loopback address? */
 694         if (local && ipv4_is_loopback(cp->daddr.ip) && was_input) {
 695                 IP_VS_DBG_RL_PKT(1, AF_INET, pp, skb, 0, "ip_vs_nat_xmit(): "
 696                                  "stopping DNAT to loopback address");
 697                 goto tx_error;
 698         }
 699
 700         /* copy-on-write the packet before mangling it */
 701         if (!skb_make_writable(skb, sizeof(struct iphdr)))
 702                 goto tx_error;
 703
 704         if (skb_cow(skb, rt->dst.dev->hard_header_len))
 705                 goto tx_error;
 706
 707         /* mangle the packet */
 708         if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp, ipvsh))
 709                 goto tx_error;
 710         ip_hdr(skb)->daddr = cp->daddr.ip;
 711         ip_send_check(ip_hdr(skb));
 712
 713         IP_VS_DBG_PKT(10, AF_INET, 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->ignore_df = 1;
 721
 722         rc = ip_vs_nat_send_or_cont(NFPROTO_IPV4, skb, cp, local);
 723         rcu_read_unlock();
 724
 725         LeaveFunction(10);
 726         return rc;
 727
 728   tx_error:
 729         kfree_skb(skb);
 730         rcu_read_unlock();
 731         LeaveFunction(10);
 732         return NF_STOLEN;
 733 }
 734
 735 #ifdef CONFIG_IP_VS_IPV6
 736 int
 737 ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
 738                   struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
 739 {
 740         struct rt6_info *rt;            /* Route to the other host */
 741         int local, rc;
 742
 743         EnterFunction(10);
 744
 745         rcu_read_lock();
 746         /* check if it is a connection of no-client-port */
 747         if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT && !ipvsh->fragoffs)) {
 748                 __be16 _pt, *p;
 749                 p = skb_header_pointer(skb, ipvsh->len, sizeof(_pt), &_pt);
 750                 if (p == NULL)
 751                         goto tx_error;
 752                 ip_vs_conn_fill_cport(cp, *p);
 753                 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
 754         }
 755
 756         local = __ip_vs_get_out_rt_v6(cp->af, skb, cp->dest, &cp->daddr.in6,
 757                                       NULL, ipvsh, 0,
 758                                       IP_VS_RT_MODE_LOCAL |
 759                                       IP_VS_RT_MODE_NON_LOCAL |
 760                                       IP_VS_RT_MODE_RDR);
 761         if (local < 0)
 762                 goto tx_error;
 763         rt = (struct rt6_info *) skb_dst(skb);
 764         /*
 765          * Avoid duplicate tuple in reply direction for NAT traffic
 766          * to local address when connection is sync-ed
 767          */
 768 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 769         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
 770                 enum ip_conntrack_info ctinfo;
 771                 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
 772
 773                 if (ct && !nf_ct_is_untracked(ct)) {
 774                         IP_VS_DBG_RL_PKT(10, AF_INET6, pp, skb, 0,
 775                                          "ip_vs_nat_xmit_v6(): "
 776                                          "stopping DNAT to local address");
 777                         goto tx_error;
 778                 }
 779         }
 780 #endif
 781
 782         /* From world but DNAT to loopback address? */
 783         if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
 784             ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_LOOPBACK) {
 785                 IP_VS_DBG_RL_PKT(1, AF_INET6, pp, skb, 0,
 786                                  "ip_vs_nat_xmit_v6(): "
 787                                  "stopping DNAT to loopback address");
 788                 goto tx_error;
 789         }
 790
 791         /* copy-on-write the packet before mangling it */
 792         if (!skb_make_writable(skb, sizeof(struct ipv6hdr)))
 793                 goto tx_error;
 794
 795         if (skb_cow(skb, rt->dst.dev->hard_header_len))
 796                 goto tx_error;
 797
 798         /* mangle the packet */
 799         if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp, ipvsh))
 800                 goto tx_error;
 801         ipv6_hdr(skb)->daddr = cp->daddr.in6;
 802
 803         IP_VS_DBG_PKT(10, AF_INET6, pp, skb, 0, "After DNAT");
 804
 805         /* FIXME: when application helper enlarges the packet and the length
 806            is larger than the MTU of outgoing device, there will be still
 807            MTU problem. */
 808
 809         /* Another hack: avoid icmp_send in ip_fragment */
 810         skb->ignore_df = 1;
 811
 812         rc = ip_vs_nat_send_or_cont(NFPROTO_IPV6, skb, cp, local);
 813         rcu_read_unlock();
 814
 815         LeaveFunction(10);
 816         return rc;
 817
 818 tx_error:
 819         LeaveFunction(10);
 820         kfree_skb(skb);
 821         rcu_read_unlock();
 822         return NF_STOLEN;
 823 }
 824 #endif
 825
 826 /* When forwarding a packet, we must ensure that we've got enough headroom
 827  * for the encapsulation packet in the skb.  This also gives us an
 828  * opportunity to figure out what the payload_len, dsfield, ttl, and df
 829  * values should be, so that we won't need to look at the old ip header
 830  * again
 831  */
 832 static struct sk_buff *
 833 ip_vs_prepare_tunneled_skb(struct sk_buff *skb, int skb_af,
 834                            unsigned int max_headroom, __u8 *next_protocol,
 835                            __u32 *payload_len, __u8 *dsfield, __u8 *ttl,
 836                            __be16 *df)
 837 {
 838         struct sk_buff *new_skb = NULL;
 839         struct iphdr *old_iph = NULL;
 840 #ifdef CONFIG_IP_VS_IPV6
 841         struct ipv6hdr *old_ipv6h = NULL;
 842 #endif
 843
 844         if (skb_headroom(skb) < max_headroom || skb_cloned(skb)) {
 845                 new_skb = skb_realloc_headroom(skb, max_headroom);
 846                 if (!new_skb)
 847                         goto error;
 848                 if (skb->sk)
 849                         skb_set_owner_w(new_skb, skb->sk);
 850                 consume_skb(skb);
 851                 skb = new_skb;
 852         }
 853
 854 #ifdef CONFIG_IP_VS_IPV6
 855         if (skb_af == AF_INET6) {
 856                 old_ipv6h = ipv6_hdr(skb);
 857                 *next_protocol = IPPROTO_IPV6;
 858                 if (payload_len)
 859                         *payload_len =
 860                                 ntohs(old_ipv6h->payload_len) +
 861                                 sizeof(*old_ipv6h);
 862                 *dsfield = ipv6_get_dsfield(old_ipv6h);
 863                 *ttl = old_ipv6h->hop_limit;
 864                 if (df)
 865                         *df = 0;
 866         } else
 867 #endif
 868         {
 869                 old_iph = ip_hdr(skb);
 870                 /* Copy DF, reset fragment offset and MF */
 871                 if (df)
 872                         *df = (old_iph->frag_off & htons(IP_DF));
 873                 *next_protocol = IPPROTO_IPIP;
 874
 875                 /* fix old IP header checksum */
 876                 ip_send_check(old_iph);
 877                 *dsfield = ipv4_get_dsfield(old_iph);
 878                 *ttl = old_iph->ttl;
 879                 if (payload_len)
 880                         *payload_len = ntohs(old_iph->tot_len);
 881         }
 882
 883         return skb;
 884 error:
 885         kfree_skb(skb);
 886         return ERR_PTR(-ENOMEM);
 887 }
 888
 889 static inline int __tun_gso_type_mask(int encaps_af, int orig_af)
 890 {
 891         if (encaps_af == AF_INET) {
 892                 if (orig_af == AF_INET)
 893                         return SKB_GSO_IPIP;
 894
 895                 return SKB_GSO_SIT;
 896         }
 897
 898         /* GSO: we need to provide proper SKB_GSO_ value for IPv6:
 899          * SKB_GSO_SIT/IPV6
 900          */
 901         return 0;
 902 }
 903
 904 /*
 905  *   IP Tunneling transmitter
 906  *
 907  *   This function encapsulates the packet in a new IP packet, its
 908  *   destination will be set to cp->daddr. Most code of this function
 909  *   is taken from ipip.c.
 910  *
 911  *   It is used in VS/TUN cluster. The load balancer selects a real
 912  *   server from a cluster based on a scheduling algorithm,
 913  *   encapsulates the request packet and forwards it to the selected
 914  *   server. For example, all real servers are configured with
 915  *   "ifconfig tunl0 <Virtual IP Address> up". When the server receives
 916  *   the encapsulated packet, it will decapsulate the packet, processe
 917  *   the request and return the response packets directly to the client
 918  *   without passing the load balancer. This can greatly increase the
 919  *   scalability of virtual server.
 920  *
 921  *   Used for ANY protocol
 922  */
 923 int
 924 ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 925                   struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
 926 {
 927         struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
 928         struct rtable *rt;                      /* Route to the other host */
 929         __be32 saddr;                           /* Source for tunnel */
 930         struct net_device *tdev;                /* Device to other host */
 931         __u8 next_protocol = 0;
 932         __u8 dsfield = 0;
 933         __u8 ttl = 0;
 934         __be16 df = 0;
 935         __be16 *dfp = NULL;
 936         struct iphdr  *iph;                     /* Our new IP header */
 937         unsigned int max_headroom;              /* The extra header space needed */
 938         int ret, local;
 939
 940         EnterFunction(10);
 941
 942         rcu_read_lock();
 943         local = __ip_vs_get_out_rt(cp->af, skb, cp->dest, cp->daddr.ip,
 944                                    IP_VS_RT_MODE_LOCAL |
 945                                    IP_VS_RT_MODE_NON_LOCAL |
 946                                    IP_VS_RT_MODE_CONNECT |
 947                                    IP_VS_RT_MODE_TUNNEL, &saddr, ipvsh);
 948         if (local < 0)
 949                 goto tx_error;
 950         if (local) {
 951                 rcu_read_unlock();
 952                 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
 953         }
 954
 955         rt = skb_rtable(skb);
 956         tdev = rt->dst.dev;
 957
 958         /*
 959          * Okay, now see if we can stuff it in the buffer as-is.
 960          */
 961         max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct iphdr);
 962
 963         /* We only care about the df field if sysctl_pmtu_disc(ipvs) is set */
 964         dfp = sysctl_pmtu_disc(ipvs) ? &df : NULL;
 965         skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
 966                                          &next_protocol, NULL, &dsfield,
 967                                          &ttl, dfp);
 968         if (IS_ERR(skb))
 969                 goto tx_error;
 970
 971         skb = iptunnel_handle_offloads(
 972                 skb, false, __tun_gso_type_mask(AF_INET, cp->af));
 973         if (IS_ERR(skb))
 974                 goto tx_error;
 975
 976         skb->transport_header = skb->network_header;
 977
 978         skb_push(skb, sizeof(struct iphdr));
 979         skb_reset_network_header(skb);
 980         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 981
 982         /*
 983          *      Push down and install the IPIP header.
 984          */
 985         iph                     =       ip_hdr(skb);
 986         iph->version            =       4;
 987         iph->ihl                =       sizeof(struct iphdr)>>2;
 988         iph->frag_off           =       df;
 989         iph->protocol           =       next_protocol;
 990         iph->tos                =       dsfield;
 991         iph->daddr              =       cp->daddr.ip;
 992         iph->saddr              =       saddr;
 993         iph->ttl                =       ttl;
 994         ip_select_ident(skb, NULL);
 995
 996         /* Another hack: avoid icmp_send in ip_fragment */
 997         skb->ignore_df = 1;
 998
 999         ret = ip_vs_tunnel_xmit_prepare(skb, cp);
1000         if (ret == NF_ACCEPT)
1001                 ip_local_out(skb);
1002         else if (ret == NF_DROP)
1003                 kfree_skb(skb);
1004         rcu_read_unlock();
1005
1006         LeaveFunction(10);
1007
1008         return NF_STOLEN;
1009
1010   tx_error:
1011         if (!IS_ERR(skb))
1012                 kfree_skb(skb);
1013         rcu_read_unlock();
1014         LeaveFunction(10);
1015         return NF_STOLEN;
1016 }
1017
1018 #ifdef CONFIG_IP_VS_IPV6
1019 int
1020 ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1021                      struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1022 {
1023         struct rt6_info *rt;            /* Route to the other host */
1024         struct in6_addr saddr;          /* Source for tunnel */
1025         struct net_device *tdev;        /* Device to other host */
1026         __u8 next_protocol = 0;
1027         __u32 payload_len = 0;
1028         __u8 dsfield = 0;
1029         __u8 ttl = 0;
1030         struct ipv6hdr  *iph;           /* Our new IP header */
1031         unsigned int max_headroom;      /* The extra header space needed */
1032         int ret, local;
1033
1034         EnterFunction(10);
1035
1036         rcu_read_lock();
1037         local = __ip_vs_get_out_rt_v6(cp->af, skb, cp->dest, &cp->daddr.in6,
1038                                       &saddr, ipvsh, 1,
1039                                       IP_VS_RT_MODE_LOCAL |
1040                                       IP_VS_RT_MODE_NON_LOCAL |
1041                                       IP_VS_RT_MODE_TUNNEL);
1042         if (local < 0)
1043                 goto tx_error;
1044         if (local) {
1045                 rcu_read_unlock();
1046                 return ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 1);
1047         }
1048
1049         rt = (struct rt6_info *) skb_dst(skb);
1050         tdev = rt->dst.dev;
1051
1052         /*
1053          * Okay, now see if we can stuff it in the buffer as-is.
1054          */
1055         max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct ipv6hdr);
1056
1057         skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
1058                                          &next_protocol, &payload_len,
1059                                          &dsfield, &ttl, NULL);
1060         if (IS_ERR(skb))
1061                 goto tx_error;
1062
1063         skb = iptunnel_handle_offloads(
1064                 skb, false, __tun_gso_type_mask(AF_INET6, cp->af));
1065         if (IS_ERR(skb))
1066                 goto tx_error;
1067
1068         skb->transport_header = skb->network_header;
1069
1070         skb_push(skb, sizeof(struct ipv6hdr));
1071         skb_reset_network_header(skb);
1072         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1073
1074         /*
1075          *      Push down and install the IPIP header.
1076          */
1077         iph                     =       ipv6_hdr(skb);
1078         iph->version            =       6;
1079         iph->nexthdr            =       next_protocol;
1080         iph->payload_len        =       htons(payload_len);
1081         memset(&iph->flow_lbl, 0, sizeof(iph->flow_lbl));
1082         ipv6_change_dsfield(iph, 0, dsfield);
1083         iph->daddr = cp->daddr.in6;
1084         iph->saddr = saddr;
1085         iph->hop_limit          =       ttl;
1086
1087         /* Another hack: avoid icmp_send in ip_fragment */
1088         skb->ignore_df = 1;
1089
1090         ret = ip_vs_tunnel_xmit_prepare(skb, cp);
1091         if (ret == NF_ACCEPT)
1092                 ip6_local_out(skb);
1093         else if (ret == NF_DROP)
1094                 kfree_skb(skb);
1095         rcu_read_unlock();
1096
1097         LeaveFunction(10);
1098
1099         return NF_STOLEN;
1100
1101 tx_error:
1102         if (!IS_ERR(skb))
1103                 kfree_skb(skb);
1104         rcu_read_unlock();
1105         LeaveFunction(10);
1106         return NF_STOLEN;
1107 }
1108 #endif
1109
1110
1111 /*
1112  *      Direct Routing transmitter
1113  *      Used for ANY protocol
1114  */
1115 int
1116 ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1117               struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1118 {
1119         int local;
1120
1121         EnterFunction(10);
1122
1123         rcu_read_lock();
1124         local = __ip_vs_get_out_rt(cp->af, skb, cp->dest, cp->daddr.ip,
1125                                    IP_VS_RT_MODE_LOCAL |
1126                                    IP_VS_RT_MODE_NON_LOCAL |
1127                                    IP_VS_RT_MODE_KNOWN_NH, NULL, ipvsh);
1128         if (local < 0)
1129                 goto tx_error;
1130         if (local) {
1131                 rcu_read_unlock();
1132                 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
1133         }
1134
1135         ip_send_check(ip_hdr(skb));
1136
1137         /* Another hack: avoid icmp_send in ip_fragment */
1138         skb->ignore_df = 1;
1139
1140         ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 0);
1141         rcu_read_unlock();
1142
1143         LeaveFunction(10);
1144         return NF_STOLEN;
1145
1146   tx_error:
1147         kfree_skb(skb);
1148         rcu_read_unlock();
1149         LeaveFunction(10);
1150         return NF_STOLEN;
1151 }
1152
1153 #ifdef CONFIG_IP_VS_IPV6
1154 int
1155 ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1156                  struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1157 {
1158         int local;
1159
1160         EnterFunction(10);
1161
1162         rcu_read_lock();
1163         local = __ip_vs_get_out_rt_v6(cp->af, skb, cp->dest, &cp->daddr.in6,
1164                                       NULL, ipvsh, 0,
1165                                       IP_VS_RT_MODE_LOCAL |
1166                                       IP_VS_RT_MODE_NON_LOCAL);
1167         if (local < 0)
1168                 goto tx_error;
1169         if (local) {
1170                 rcu_read_unlock();
1171                 return ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 1);
1172         }
1173
1174         /* Another hack: avoid icmp_send in ip_fragment */
1175         skb->ignore_df = 1;
1176
1177         ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 0);
1178         rcu_read_unlock();
1179
1180         LeaveFunction(10);
1181         return NF_STOLEN;
1182
1183 tx_error:
1184         kfree_skb(skb);
1185         rcu_read_unlock();
1186         LeaveFunction(10);
1187         return NF_STOLEN;
1188 }
1189 #endif
1190
1191
1192 /*
1193  *      ICMP packet transmitter
1194  *      called by the ip_vs_in_icmp
1195  */
1196 int
1197 ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1198                 struct ip_vs_protocol *pp, int offset, unsigned int hooknum,
1199                 struct ip_vs_iphdr *iph)
1200 {
1201         struct rtable   *rt;    /* Route to the other host */
1202         int rc;
1203         int local;
1204         int rt_mode, was_input;
1205
1206         EnterFunction(10);
1207
1208         /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1209            forwarded directly here, because there is no need to
1210            translate address/port back */
1211         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1212                 if (cp->packet_xmit)
1213                         rc = cp->packet_xmit(skb, cp, pp, iph);
1214                 else
1215                         rc = NF_ACCEPT;
1216                 /* do not touch skb anymore */
1217                 atomic_inc(&cp->in_pkts);
1218                 goto out;
1219         }
1220
1221         /*
1222          * mangle and send the packet here (only for VS/NAT)
1223          */
1224         was_input = rt_is_input_route(skb_rtable(skb));
1225
1226         /* LOCALNODE from FORWARD hook is not supported */
1227         rt_mode = (hooknum != NF_INET_FORWARD) ?
1228                   IP_VS_RT_MODE_LOCAL | IP_VS_RT_MODE_NON_LOCAL |
1229                   IP_VS_RT_MODE_RDR : IP_VS_RT_MODE_NON_LOCAL;
1230         rcu_read_lock();
1231         local = __ip_vs_get_out_rt(cp->af, skb, cp->dest, cp->daddr.ip, rt_mode,
1232                                    NULL, iph);
1233         if (local < 0)
1234                 goto tx_error;
1235         rt = skb_rtable(skb);
1236
1237         /*
1238          * Avoid duplicate tuple in reply direction for NAT traffic
1239          * to local address when connection is sync-ed
1240          */
1241 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1242         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1243                 enum ip_conntrack_info ctinfo;
1244                 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
1245
1246                 if (ct && !nf_ct_is_untracked(ct)) {
1247                         IP_VS_DBG(10, "%s(): "
1248                                   "stopping DNAT to local address %pI4\n",
1249                                   __func__, &cp->daddr.ip);
1250                         goto tx_error;
1251                 }
1252         }
1253 #endif
1254
1255         /* From world but DNAT to loopback address? */
1256         if (local && ipv4_is_loopback(cp->daddr.ip) && was_input) {
1257                 IP_VS_DBG(1, "%s(): "
1258                           "stopping DNAT to loopback %pI4\n",
1259                           __func__, &cp->daddr.ip);
1260                 goto tx_error;
1261         }
1262
1263         /* copy-on-write the packet before mangling it */
1264         if (!skb_make_writable(skb, offset))
1265                 goto tx_error;
1266
1267         if (skb_cow(skb, rt->dst.dev->hard_header_len))
1268                 goto tx_error;
1269
1270         ip_vs_nat_icmp(skb, pp, cp, 0);
1271
1272         /* Another hack: avoid icmp_send in ip_fragment */
1273         skb->ignore_df = 1;
1274
1275         rc = ip_vs_nat_send_or_cont(NFPROTO_IPV4, skb, cp, local);
1276         rcu_read_unlock();
1277         goto out;
1278
1279   tx_error:
1280         kfree_skb(skb);
1281         rcu_read_unlock();
1282         rc = NF_STOLEN;
1283   out:
1284         LeaveFunction(10);
1285         return rc;
1286 }
1287
1288 #ifdef CONFIG_IP_VS_IPV6
1289 int
1290 ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1291                 struct ip_vs_protocol *pp, int offset, unsigned int hooknum,
1292                 struct ip_vs_iphdr *ipvsh)
1293 {
1294         struct rt6_info *rt;    /* Route to the other host */
1295         int rc;
1296         int local;
1297         int rt_mode;
1298
1299         EnterFunction(10);
1300
1301         /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1302            forwarded directly here, because there is no need to
1303            translate address/port back */
1304         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1305                 if (cp->packet_xmit)
1306                         rc = cp->packet_xmit(skb, cp, pp, ipvsh);
1307                 else
1308                         rc = NF_ACCEPT;
1309                 /* do not touch skb anymore */
1310                 atomic_inc(&cp->in_pkts);
1311                 goto out;
1312         }
1313
1314         /*
1315          * mangle and send the packet here (only for VS/NAT)
1316          */
1317
1318         /* LOCALNODE from FORWARD hook is not supported */
1319         rt_mode = (hooknum != NF_INET_FORWARD) ?
1320                   IP_VS_RT_MODE_LOCAL | IP_VS_RT_MODE_NON_LOCAL |
1321                   IP_VS_RT_MODE_RDR : IP_VS_RT_MODE_NON_LOCAL;
1322         rcu_read_lock();
1323         local = __ip_vs_get_out_rt_v6(cp->af, skb, cp->dest, &cp->daddr.in6,
1324                                       NULL, ipvsh, 0, rt_mode);
1325         if (local < 0)
1326                 goto tx_error;
1327         rt = (struct rt6_info *) skb_dst(skb);
1328         /*
1329          * Avoid duplicate tuple in reply direction for NAT traffic
1330          * to local address when connection is sync-ed
1331          */
1332 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1333         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1334                 enum ip_conntrack_info ctinfo;
1335                 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
1336
1337                 if (ct && !nf_ct_is_untracked(ct)) {
1338                         IP_VS_DBG(10, "%s(): "
1339                                   "stopping DNAT to local address %pI6\n",
1340                                   __func__, &cp->daddr.in6);
1341                         goto tx_error;
1342                 }
1343         }
1344 #endif
1345
1346         /* From world but DNAT to loopback address? */
1347         if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
1348             ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_LOOPBACK) {
1349                 IP_VS_DBG(1, "%s(): "
1350                           "stopping DNAT to loopback %pI6\n",
1351                           __func__, &cp->daddr.in6);
1352                 goto tx_error;
1353         }
1354
1355         /* copy-on-write the packet before mangling it */
1356         if (!skb_make_writable(skb, offset))
1357                 goto tx_error;
1358
1359         if (skb_cow(skb, rt->dst.dev->hard_header_len))
1360                 goto tx_error;
1361
1362         ip_vs_nat_icmp_v6(skb, pp, cp, 0);
1363
1364         /* Another hack: avoid icmp_send in ip_fragment */
1365         skb->ignore_df = 1;
1366
1367         rc = ip_vs_nat_send_or_cont(NFPROTO_IPV6, skb, cp, local);
1368         rcu_read_unlock();
1369         goto out;
1370
1371 tx_error:
1372         kfree_skb(skb);
1373         rcu_read_unlock();
1374         rc = NF_STOLEN;
1375 out:
1376         LeaveFunction(10);
1377         return rc;
1378 }
1379 #endif