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.flowi4_flags = (rt_mode & IP_VS_RT_MODE_KNOWN_NH) ?
 134                            FLOWI_FLAG_KNOWN_NH : 0;
 135
 136 retry:
 137         rt = ip_route_output_key(net, &fl4);
 138         if (IS_ERR(rt)) {
 139                 /* Invalid saddr ? */
 140                 if (PTR_ERR(rt) == -EINVAL && *saddr &&
 141                     rt_mode & IP_VS_RT_MODE_CONNECT && !loop) {
 142                         *saddr = 0;
 143                         flowi4_update_output(&fl4, 0, 0, daddr, 0);
 144                         goto retry;
 145                 }
 146                 IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n", &daddr);
 147                 return NULL;
 148         } else if (!*saddr && rt_mode & IP_VS_RT_MODE_CONNECT && fl4.saddr) {
 149                 ip_rt_put(rt);
 150                 *saddr = fl4.saddr;
 151                 flowi4_update_output(&fl4, 0, 0, daddr, fl4.saddr);
 152                 loop++;
 153                 goto retry;
 154         }
 155         *saddr = fl4.saddr;
 156         return rt;
 157 }
 158
 159 #ifdef CONFIG_IP_VS_IPV6
 160 static inline int __ip_vs_is_local_route6(struct rt6_info *rt)
 161 {
 162         return rt->dst.dev && rt->dst.dev->flags & IFF_LOOPBACK;
 163 }
 164 #endif
 165
 166 static inline bool crosses_local_route_boundary(int skb_af, struct sk_buff *skb,
 167                                                 int rt_mode,
 168                                                 bool new_rt_is_local)
 169 {
 170         bool rt_mode_allow_local = !!(rt_mode & IP_VS_RT_MODE_LOCAL);
 171         bool rt_mode_allow_non_local = !!(rt_mode & IP_VS_RT_MODE_LOCAL);
 172         bool rt_mode_allow_redirect = !!(rt_mode & IP_VS_RT_MODE_RDR);
 173         bool source_is_loopback;
 174         bool old_rt_is_local;
 175
 176 #ifdef CONFIG_IP_VS_IPV6
 177         if (skb_af == AF_INET6) {
 178                 int addr_type = ipv6_addr_type(&ipv6_hdr(skb)->saddr);
 179
 180                 source_is_loopback =
 181                         (!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
 182                         (addr_type & IPV6_ADDR_LOOPBACK);
 183                 old_rt_is_local = __ip_vs_is_local_route6(
 184                         (struct rt6_info *)skb_dst(skb));
 185         } else
 186 #endif
 187         {
 188                 source_is_loopback = ipv4_is_loopback(ip_hdr(skb)->saddr);
 189                 old_rt_is_local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
 190         }
 191
 192         if (unlikely(new_rt_is_local)) {
 193                 if (!rt_mode_allow_local)
 194                         return true;
 195                 if (!rt_mode_allow_redirect && !old_rt_is_local)
 196                         return true;
 197         } else {
 198                 if (!rt_mode_allow_non_local)
 199                         return true;
 200                 if (source_is_loopback)
 201                         return true;
 202         }
 203         return false;
 204 }
 205
 206 static inline void maybe_update_pmtu(int skb_af, struct sk_buff *skb, int mtu)
 207 {
 208         struct sock *sk = skb->sk;
 209         struct rtable *ort = skb_rtable(skb);
 210
 211         if (!skb->dev && sk && sk_fullsock(sk))
 212                 ort->dst.ops->update_pmtu(&ort->dst, sk, NULL, mtu);
 213 }
 214
 215 static inline bool ensure_mtu_is_adequate(struct netns_ipvs *ipvs, int skb_af,
 216                                           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 = ipvs->net;
 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 && !ip_vs_iph_icmp(ipvsh))
 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                 /* If we're going to tunnel the packet and pmtu discovery
 238                  * is disabled, we'll just fragment it anyway
 239                  */
 240                 if ((rt_mode & IP_VS_RT_MODE_TUNNEL) && !sysctl_pmtu_disc(ipvs))
 241                         return true;
 242
 243                 if (unlikely(ip_hdr(skb)->frag_off & htons(IP_DF) &&
 244                              skb->len > mtu && !skb_is_gso(skb) &&
 245                              !ip_vs_iph_icmp(ipvsh))) {
 246                         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
 247                                   htonl(mtu));
 248                         IP_VS_DBG(1, "frag needed for %pI4\n",
 249                                   &ip_hdr(skb)->saddr);
 250                         return false;
 251                 }
 252         }
 253
 254         return true;
 255 }
 256
 257 /* Get route to destination or remote server */
 258 static int
 259 __ip_vs_get_out_rt(struct netns_ipvs *ipvs, int skb_af, struct sk_buff *skb,
 260                    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 = ipvs->net;
 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(ipvs, 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(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, int rt_mode)
 368 {
 369         struct dst_entry *dst;
 370         struct flowi6 fl6 = {
 371                 .daddr = *daddr,
 372         };
 373
 374         if (rt_mode & IP_VS_RT_MODE_KNOWN_NH)
 375                 fl6.flowi6_flags = FLOWI_FLAG_KNOWN_NH;
 376
 377         dst = ip6_route_output(net, NULL, &fl6);
 378         if (dst->error)
 379                 goto out_err;
 380         if (!ret_saddr)
 381                 return dst;
 382         if (ipv6_addr_any(&fl6.saddr) &&
 383             ipv6_dev_get_saddr(net, ip6_dst_idev(dst)->dev,
 384                                &fl6.daddr, 0, &fl6.saddr) < 0)
 385                 goto out_err;
 386         if (do_xfrm) {
 387                 dst = xfrm_lookup(net, dst, flowi6_to_flowi(&fl6), NULL, 0);
 388                 if (IS_ERR(dst)) {
 389                         dst = NULL;
 390                         goto out_err;
 391                 }
 392         }
 393         *ret_saddr = fl6.saddr;
 394         return dst;
 395
 396 out_err:
 397         dst_release(dst);
 398         IP_VS_DBG_RL("ip6_route_output error, dest: %pI6\n", daddr);
 399         return NULL;
 400 }
 401
 402 /*
 403  * Get route to destination or remote server
 404  */
 405 static int
 406 __ip_vs_get_out_rt_v6(struct netns_ipvs *ipvs, int skb_af, struct sk_buff *skb,
 407                       struct ip_vs_dest *dest,
 408                       struct in6_addr *daddr, struct in6_addr *ret_saddr,
 409                       struct ip_vs_iphdr *ipvsh, int do_xfrm, int rt_mode)
 410 {
 411         struct net *net = ipvs->net;
 412         struct ip_vs_dest_dst *dest_dst;
 413         struct rt6_info *rt;                    /* Route to the other host */
 414         struct dst_entry *dst;
 415         int mtu;
 416         int local, noref = 1;
 417
 418         if (dest) {
 419                 dest_dst = __ip_vs_dst_check(dest);
 420                 if (likely(dest_dst))
 421                         rt = (struct rt6_info *) dest_dst->dst_cache;
 422                 else {
 423                         u32 cookie;
 424
 425                         dest_dst = ip_vs_dest_dst_alloc();
 426                         spin_lock_bh(&dest->dst_lock);
 427                         if (!dest_dst) {
 428                                 __ip_vs_dst_set(dest, NULL, NULL, 0);
 429                                 spin_unlock_bh(&dest->dst_lock);
 430                                 goto err_unreach;
 431                         }
 432                         dst = __ip_vs_route_output_v6(net, &dest->addr.in6,
 433                                                       &dest_dst->dst_saddr.in6,
 434                                                       do_xfrm, rt_mode);
 435                         if (!dst) {
 436                                 __ip_vs_dst_set(dest, NULL, NULL, 0);
 437                                 spin_unlock_bh(&dest->dst_lock);
 438                                 ip_vs_dest_dst_free(dest_dst);
 439                                 goto err_unreach;
 440                         }
 441                         rt = (struct rt6_info *) dst;
 442                         cookie = rt6_get_cookie(rt);
 443                         __ip_vs_dst_set(dest, dest_dst, &rt->dst, cookie);
 444                         spin_unlock_bh(&dest->dst_lock);
 445                         IP_VS_DBG(10, "new dst %pI6, src %pI6, refcnt=%d\n",
 446                                   &dest->addr.in6, &dest_dst->dst_saddr.in6,
 447                                   atomic_read(&rt->dst.__refcnt));
 448                 }
 449                 if (ret_saddr)
 450                         *ret_saddr = dest_dst->dst_saddr.in6;
 451         } else {
 452                 noref = 0;
 453                 dst = __ip_vs_route_output_v6(net, daddr, ret_saddr, do_xfrm,
 454                                               rt_mode);
 455                 if (!dst)
 456                         goto err_unreach;
 457                 rt = (struct rt6_info *) dst;
 458         }
 459
 460         local = __ip_vs_is_local_route6(rt);
 461
 462         if (unlikely(crosses_local_route_boundary(skb_af, skb, rt_mode,
 463                                                   local))) {
 464                 IP_VS_DBG_RL("We are crossing local and non-local addresses"
 465                              " daddr=%pI6\n", daddr);
 466                 goto err_put;
 467         }
 468
 469         if (unlikely(local)) {
 470                 /* skb to local stack, preserve old route */
 471                 if (!noref)
 472                         dst_release(&rt->dst);
 473                 return local;
 474         }
 475
 476         /* MTU checking */
 477         if (likely(!(rt_mode & IP_VS_RT_MODE_TUNNEL)))
 478                 mtu = dst_mtu(&rt->dst);
 479         else {
 480                 mtu = dst_mtu(&rt->dst) - sizeof(struct ipv6hdr);
 481                 if (mtu < IPV6_MIN_MTU) {
 482                         IP_VS_DBG_RL("%s(): mtu less than %d\n", __func__,
 483                                      IPV6_MIN_MTU);
 484                         goto err_put;
 485                 }
 486                 maybe_update_pmtu(skb_af, skb, mtu);
 487         }
 488
 489         if (!ensure_mtu_is_adequate(ipvs, skb_af, rt_mode, ipvsh, skb, mtu))
 490                 goto err_put;
 491
 492         skb_dst_drop(skb);
 493         if (noref) {
 494                 if (!local)
 495                         skb_dst_set_noref(skb, &rt->dst);
 496                 else
 497                         skb_dst_set(skb, dst_clone(&rt->dst));
 498         } else
 499                 skb_dst_set(skb, &rt->dst);
 500
 501         return local;
 502
 503 err_put:
 504         if (!noref)
 505                 dst_release(&rt->dst);
 506         return -1;
 507
 508 err_unreach:
 509         /* The ip6_link_failure function requires the dev field to be set
 510          * in order to get the net (further for the sake of fwmark
 511          * reflection).
 512          */
 513         if (!skb->dev)
 514                 skb->dev = skb_dst(skb)->dev;
 515
 516         dst_link_failure(skb);
 517         return -1;
 518 }
 519 #endif
 520
 521
 522 /* return NF_ACCEPT to allow forwarding or other NF_xxx on error */
 523 static inline int ip_vs_tunnel_xmit_prepare(struct sk_buff *skb,
 524                                             struct ip_vs_conn *cp)
 525 {
 526         int ret = NF_ACCEPT;
 527
 528         skb->ipvs_property = 1;
 529         if (unlikely(cp->flags & IP_VS_CONN_F_NFCT))
 530                 ret = ip_vs_confirm_conntrack(skb);
 531         if (ret == NF_ACCEPT) {
 532                 nf_reset(skb);
 533                 skb_forward_csum(skb);
 534         }
 535         return ret;
 536 }
 537
 538 /* In the event of a remote destination, it's possible that we would have
 539  * matches against an old socket (particularly a TIME-WAIT socket). This
 540  * causes havoc down the line (ip_local_out et. al. expect regular sockets
 541  * and invalid memory accesses will happen) so simply drop the association
 542  * in this case.
 543 */
 544 static inline void ip_vs_drop_early_demux_sk(struct sk_buff *skb)
 545 {
 546         /* If dev is set, the packet came from the LOCAL_IN callback and
 547          * not from a local TCP socket.
 548          */
 549         if (skb->dev)
 550                 skb_orphan(skb);
 551 }
 552
 553 /* return NF_STOLEN (sent) or NF_ACCEPT if local=1 (not sent) */
 554 static inline int ip_vs_nat_send_or_cont(int pf, struct sk_buff *skb,
 555                                          struct ip_vs_conn *cp, int local)
 556 {
 557         int ret = NF_STOLEN;
 558
 559         skb->ipvs_property = 1;
 560         if (likely(!(cp->flags & IP_VS_CONN_F_NFCT)))
 561                 ip_vs_notrack(skb);
 562         else
 563                 ip_vs_update_conntrack(skb, cp, 1);
 564
 565         /* Remove the early_demux association unless it's bound for the
 566          * exact same port and address on this host after translation.
 567          */
 568         if (!local || cp->vport != cp->dport ||
 569             !ip_vs_addr_equal(cp->af, &cp->vaddr, &cp->daddr))
 570                 ip_vs_drop_early_demux_sk(skb);
 571
 572         if (!local) {
 573                 skb_forward_csum(skb);
 574                 NF_HOOK(pf, NF_INET_LOCAL_OUT, cp->ipvs->net, NULL, skb,
 575                         NULL, skb_dst(skb)->dev, dst_output);
 576         } else
 577                 ret = NF_ACCEPT;
 578
 579         return ret;
 580 }
 581
 582 /* return NF_STOLEN (sent) or NF_ACCEPT if local=1 (not sent) */
 583 static inline int ip_vs_send_or_cont(int pf, struct sk_buff *skb,
 584                                      struct ip_vs_conn *cp, int local)
 585 {
 586         int ret = NF_STOLEN;
 587
 588         skb->ipvs_property = 1;
 589         if (likely(!(cp->flags & IP_VS_CONN_F_NFCT)))
 590                 ip_vs_notrack(skb);
 591         if (!local) {
 592                 ip_vs_drop_early_demux_sk(skb);
 593                 skb_forward_csum(skb);
 594                 NF_HOOK(pf, NF_INET_LOCAL_OUT, cp->ipvs->net, NULL, skb,
 595                         NULL, skb_dst(skb)->dev, dst_output);
 596         } else
 597                 ret = NF_ACCEPT;
 598         return ret;
 599 }
 600
 601
 602 /*
 603  *      NULL transmitter (do nothing except return NF_ACCEPT)
 604  */
 605 int
 606 ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 607                 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
 608 {
 609         /* we do not touch skb and do not need pskb ptr */
 610         return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
 611 }
 612
 613
 614 /*
 615  *      Bypass transmitter
 616  *      Let packets bypass the destination when the destination is not
 617  *      available, it may be only used in transparent cache cluster.
 618  */
 619 int
 620 ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 621                   struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
 622 {
 623         struct iphdr  *iph = ip_hdr(skb);
 624
 625         EnterFunction(10);
 626
 627         rcu_read_lock();
 628         if (__ip_vs_get_out_rt(cp->ipvs, cp->af, skb, NULL, iph->daddr,
 629                                IP_VS_RT_MODE_NON_LOCAL, NULL, ipvsh) < 0)
 630                 goto tx_error;
 631
 632         ip_send_check(iph);
 633
 634         /* Another hack: avoid icmp_send in ip_fragment */
 635         skb->ignore_df = 1;
 636
 637         ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 0);
 638         rcu_read_unlock();
 639
 640         LeaveFunction(10);
 641         return NF_STOLEN;
 642
 643  tx_error:
 644         kfree_skb(skb);
 645         rcu_read_unlock();
 646         LeaveFunction(10);
 647         return NF_STOLEN;
 648 }
 649
 650 #ifdef CONFIG_IP_VS_IPV6
 651 int
 652 ip_vs_bypass_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
 653                      struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
 654 {
 655         struct ipv6hdr *iph = ipv6_hdr(skb);
 656
 657         EnterFunction(10);
 658
 659         rcu_read_lock();
 660         if (__ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, NULL,
 661                                   &iph->daddr, NULL,
 662                                   ipvsh, 0, IP_VS_RT_MODE_NON_LOCAL) < 0)
 663                 goto tx_error;
 664
 665         /* Another hack: avoid icmp_send in ip_fragment */
 666         skb->ignore_df = 1;
 667
 668         ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 0);
 669         rcu_read_unlock();
 670
 671         LeaveFunction(10);
 672         return NF_STOLEN;
 673
 674  tx_error:
 675         kfree_skb(skb);
 676         rcu_read_unlock();
 677         LeaveFunction(10);
 678         return NF_STOLEN;
 679 }
 680 #endif
 681
 682 /*
 683  *      NAT transmitter (only for outside-to-inside nat forwarding)
 684  *      Not used for related ICMP
 685  */
 686 int
 687 ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 688                struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
 689 {
 690         struct rtable *rt;              /* Route to the other host */
 691         int local, rc, was_input;
 692
 693         EnterFunction(10);
 694
 695         rcu_read_lock();
 696         /* check if it is a connection of no-client-port */
 697         if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) {
 698                 __be16 _pt, *p;
 699
 700                 p = skb_header_pointer(skb, ipvsh->len, sizeof(_pt), &_pt);
 701                 if (p == NULL)
 702                         goto tx_error;
 703                 ip_vs_conn_fill_cport(cp, *p);
 704                 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
 705         }
 706
 707         was_input = rt_is_input_route(skb_rtable(skb));
 708         local = __ip_vs_get_out_rt(cp->ipvs, cp->af, skb, cp->dest, cp->daddr.ip,
 709                                    IP_VS_RT_MODE_LOCAL |
 710                                    IP_VS_RT_MODE_NON_LOCAL |
 711                                    IP_VS_RT_MODE_RDR, NULL, ipvsh);
 712         if (local < 0)
 713                 goto tx_error;
 714         rt = skb_rtable(skb);
 715         /*
 716          * Avoid duplicate tuple in reply direction for NAT traffic
 717          * to local address when connection is sync-ed
 718          */
 719 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 720         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
 721                 enum ip_conntrack_info ctinfo;
 722                 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
 723
 724                 if (ct && !nf_ct_is_untracked(ct)) {
 725                         IP_VS_DBG_RL_PKT(10, AF_INET, pp, skb, ipvsh->off,
 726                                          "ip_vs_nat_xmit(): "
 727                                          "stopping DNAT to local address");
 728                         goto tx_error;
 729                 }
 730         }
 731 #endif
 732
 733         /* From world but DNAT to loopback address? */
 734         if (local && ipv4_is_loopback(cp->daddr.ip) && was_input) {
 735                 IP_VS_DBG_RL_PKT(1, AF_INET, pp, skb, ipvsh->off,
 736                                  "ip_vs_nat_xmit(): stopping DNAT to loopback "
 737                                  "address");
 738                 goto tx_error;
 739         }
 740
 741         /* copy-on-write the packet before mangling it */
 742         if (!skb_make_writable(skb, sizeof(struct iphdr)))
 743                 goto tx_error;
 744
 745         if (skb_cow(skb, rt->dst.dev->hard_header_len))
 746                 goto tx_error;
 747
 748         /* mangle the packet */
 749         if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp, ipvsh))
 750                 goto tx_error;
 751         ip_hdr(skb)->daddr = cp->daddr.ip;
 752         ip_send_check(ip_hdr(skb));
 753
 754         IP_VS_DBG_PKT(10, AF_INET, pp, skb, ipvsh->off, "After DNAT");
 755
 756         /* FIXME: when application helper enlarges the packet and the length
 757            is larger than the MTU of outgoing device, there will be still
 758            MTU problem. */
 759
 760         /* Another hack: avoid icmp_send in ip_fragment */
 761         skb->ignore_df = 1;
 762
 763         rc = ip_vs_nat_send_or_cont(NFPROTO_IPV4, skb, cp, local);
 764         rcu_read_unlock();
 765
 766         LeaveFunction(10);
 767         return rc;
 768
 769   tx_error:
 770         kfree_skb(skb);
 771         rcu_read_unlock();
 772         LeaveFunction(10);
 773         return NF_STOLEN;
 774 }
 775
 776 #ifdef CONFIG_IP_VS_IPV6
 777 int
 778 ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
 779                   struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
 780 {
 781         struct rt6_info *rt;            /* Route to the other host */
 782         int local, rc;
 783
 784         EnterFunction(10);
 785
 786         rcu_read_lock();
 787         /* check if it is a connection of no-client-port */
 788         if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT && !ipvsh->fragoffs)) {
 789                 __be16 _pt, *p;
 790                 p = skb_header_pointer(skb, ipvsh->len, sizeof(_pt), &_pt);
 791                 if (p == NULL)
 792                         goto tx_error;
 793                 ip_vs_conn_fill_cport(cp, *p);
 794                 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
 795         }
 796
 797         local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
 798                                       &cp->daddr.in6,
 799                                       NULL, ipvsh, 0,
 800                                       IP_VS_RT_MODE_LOCAL |
 801                                       IP_VS_RT_MODE_NON_LOCAL |
 802                                       IP_VS_RT_MODE_RDR);
 803         if (local < 0)
 804                 goto tx_error;
 805         rt = (struct rt6_info *) skb_dst(skb);
 806         /*
 807          * Avoid duplicate tuple in reply direction for NAT traffic
 808          * to local address when connection is sync-ed
 809          */
 810 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 811         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
 812                 enum ip_conntrack_info ctinfo;
 813                 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
 814
 815                 if (ct && !nf_ct_is_untracked(ct)) {
 816                         IP_VS_DBG_RL_PKT(10, AF_INET6, pp, skb, ipvsh->off,
 817                                          "ip_vs_nat_xmit_v6(): "
 818                                          "stopping DNAT to local address");
 819                         goto tx_error;
 820                 }
 821         }
 822 #endif
 823
 824         /* From world but DNAT to loopback address? */
 825         if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
 826             ipv6_addr_type(&cp->daddr.in6) & IPV6_ADDR_LOOPBACK) {
 827                 IP_VS_DBG_RL_PKT(1, AF_INET6, pp, skb, ipvsh->off,
 828                                  "ip_vs_nat_xmit_v6(): "
 829                                  "stopping DNAT to loopback address");
 830                 goto tx_error;
 831         }
 832
 833         /* copy-on-write the packet before mangling it */
 834         if (!skb_make_writable(skb, sizeof(struct ipv6hdr)))
 835                 goto tx_error;
 836
 837         if (skb_cow(skb, rt->dst.dev->hard_header_len))
 838                 goto tx_error;
 839
 840         /* mangle the packet */
 841         if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp, ipvsh))
 842                 goto tx_error;
 843         ipv6_hdr(skb)->daddr = cp->daddr.in6;
 844
 845         IP_VS_DBG_PKT(10, AF_INET6, pp, skb, ipvsh->off, "After DNAT");
 846
 847         /* FIXME: when application helper enlarges the packet and the length
 848            is larger than the MTU of outgoing device, there will be still
 849            MTU problem. */
 850
 851         /* Another hack: avoid icmp_send in ip_fragment */
 852         skb->ignore_df = 1;
 853
 854         rc = ip_vs_nat_send_or_cont(NFPROTO_IPV6, skb, cp, local);
 855         rcu_read_unlock();
 856
 857         LeaveFunction(10);
 858         return rc;
 859
 860 tx_error:
 861         LeaveFunction(10);
 862         kfree_skb(skb);
 863         rcu_read_unlock();
 864         return NF_STOLEN;
 865 }
 866 #endif
 867
 868 /* When forwarding a packet, we must ensure that we've got enough headroom
 869  * for the encapsulation packet in the skb.  This also gives us an
 870  * opportunity to figure out what the payload_len, dsfield, ttl, and df
 871  * values should be, so that we won't need to look at the old ip header
 872  * again
 873  */
 874 static struct sk_buff *
 875 ip_vs_prepare_tunneled_skb(struct sk_buff *skb, int skb_af,
 876                            unsigned int max_headroom, __u8 *next_protocol,
 877                            __u32 *payload_len, __u8 *dsfield, __u8 *ttl,
 878                            __be16 *df)
 879 {
 880         struct sk_buff *new_skb = NULL;
 881         struct iphdr *old_iph = NULL;
 882 #ifdef CONFIG_IP_VS_IPV6
 883         struct ipv6hdr *old_ipv6h = NULL;
 884 #endif
 885
 886         ip_vs_drop_early_demux_sk(skb);
 887
 888         if (skb_headroom(skb) < max_headroom || skb_cloned(skb)) {
 889                 new_skb = skb_realloc_headroom(skb, max_headroom);
 890                 if (!new_skb)
 891                         goto error;
 892                 if (skb->sk)
 893                         skb_set_owner_w(new_skb, skb->sk);
 894                 consume_skb(skb);
 895                 skb = new_skb;
 896         }
 897
 898 #ifdef CONFIG_IP_VS_IPV6
 899         if (skb_af == AF_INET6) {
 900                 old_ipv6h = ipv6_hdr(skb);
 901                 *next_protocol = IPPROTO_IPV6;
 902                 if (payload_len)
 903                         *payload_len =
 904                                 ntohs(old_ipv6h->payload_len) +
 905                                 sizeof(*old_ipv6h);
 906                 *dsfield = ipv6_get_dsfield(old_ipv6h);
 907                 *ttl = old_ipv6h->hop_limit;
 908                 if (df)
 909                         *df = 0;
 910         } else
 911 #endif
 912         {
 913                 old_iph = ip_hdr(skb);
 914                 /* Copy DF, reset fragment offset and MF */
 915                 if (df)
 916                         *df = (old_iph->frag_off & htons(IP_DF));
 917                 *next_protocol = IPPROTO_IPIP;
 918
 919                 /* fix old IP header checksum */
 920                 ip_send_check(old_iph);
 921                 *dsfield = ipv4_get_dsfield(old_iph);
 922                 *ttl = old_iph->ttl;
 923                 if (payload_len)
 924                         *payload_len = ntohs(old_iph->tot_len);
 925         }
 926
 927         return skb;
 928 error:
 929         kfree_skb(skb);
 930         return ERR_PTR(-ENOMEM);
 931 }
 932
 933 static inline int __tun_gso_type_mask(int encaps_af, int orig_af)
 934 {
 935         if (encaps_af == AF_INET) {
 936                 if (orig_af == AF_INET)
 937                         return SKB_GSO_IPIP;
 938
 939                 return SKB_GSO_SIT;
 940         }
 941
 942         /* GSO: we need to provide proper SKB_GSO_ value for IPv6:
 943          * SKB_GSO_SIT/IPV6
 944          */
 945         return 0;
 946 }
 947
 948 /*
 949  *   IP Tunneling transmitter
 950  *
 951  *   This function encapsulates the packet in a new IP packet, its
 952  *   destination will be set to cp->daddr. Most code of this function
 953  *   is taken from ipip.c.
 954  *
 955  *   It is used in VS/TUN cluster. The load balancer selects a real
 956  *   server from a cluster based on a scheduling algorithm,
 957  *   encapsulates the request packet and forwards it to the selected
 958  *   server. For example, all real servers are configured with
 959  *   "ifconfig tunl0 <Virtual IP Address> up". When the server receives
 960  *   the encapsulated packet, it will decapsulate the packet, processe
 961  *   the request and return the response packets directly to the client
 962  *   without passing the load balancer. This can greatly increase the
 963  *   scalability of virtual server.
 964  *
 965  *   Used for ANY protocol
 966  */
 967 int
 968 ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
 969                   struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
 970 {
 971         struct netns_ipvs *ipvs = cp->ipvs;
 972         struct net *net = ipvs->net;
 973         struct rtable *rt;                      /* Route to the other host */
 974         __be32 saddr;                           /* Source for tunnel */
 975         struct net_device *tdev;                /* Device to other host */
 976         __u8 next_protocol = 0;
 977         __u8 dsfield = 0;
 978         __u8 ttl = 0;
 979         __be16 df = 0;
 980         __be16 *dfp = NULL;
 981         struct iphdr  *iph;                     /* Our new IP header */
 982         unsigned int max_headroom;              /* The extra header space needed */
 983         int ret, local;
 984
 985         EnterFunction(10);
 986
 987         rcu_read_lock();
 988         local = __ip_vs_get_out_rt(ipvs, cp->af, skb, cp->dest, cp->daddr.ip,
 989                                    IP_VS_RT_MODE_LOCAL |
 990                                    IP_VS_RT_MODE_NON_LOCAL |
 991                                    IP_VS_RT_MODE_CONNECT |
 992                                    IP_VS_RT_MODE_TUNNEL, &saddr, ipvsh);
 993         if (local < 0)
 994                 goto tx_error;
 995         if (local) {
 996                 rcu_read_unlock();
 997                 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
 998         }
 999
1000         rt = skb_rtable(skb);
1001         tdev = rt->dst.dev;
1002
1003         /*
1004          * Okay, now see if we can stuff it in the buffer as-is.
1005          */
1006         max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct iphdr);
1007
1008         /* We only care about the df field if sysctl_pmtu_disc(ipvs) is set */
1009         dfp = sysctl_pmtu_disc(ipvs) ? &df : NULL;
1010         skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
1011                                          &next_protocol, NULL, &dsfield,
1012                                          &ttl, dfp);
1013         if (IS_ERR(skb))
1014                 goto tx_error;
1015
1016         skb = iptunnel_handle_offloads(skb, __tun_gso_type_mask(AF_INET, cp->af));
1017         if (IS_ERR(skb))
1018                 goto tx_error;
1019
1020         skb->transport_header = skb->network_header;
1021
1022         skb_push(skb, sizeof(struct iphdr));
1023         skb_reset_network_header(skb);
1024         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1025
1026         /*
1027          *      Push down and install the IPIP header.
1028          */
1029         iph                     =       ip_hdr(skb);
1030         iph->version            =       4;
1031         iph->ihl                =       sizeof(struct iphdr)>>2;
1032         iph->frag_off           =       df;
1033         iph->protocol           =       next_protocol;
1034         iph->tos                =       dsfield;
1035         iph->daddr              =       cp->daddr.ip;
1036         iph->saddr              =       saddr;
1037         iph->ttl                =       ttl;
1038         ip_select_ident(net, skb, NULL);
1039
1040         /* Another hack: avoid icmp_send in ip_fragment */
1041         skb->ignore_df = 1;
1042
1043         ret = ip_vs_tunnel_xmit_prepare(skb, cp);
1044         if (ret == NF_ACCEPT)
1045                 ip_local_out(net, skb->sk, skb);
1046         else if (ret == NF_DROP)
1047                 kfree_skb(skb);
1048         rcu_read_unlock();
1049
1050         LeaveFunction(10);
1051
1052         return NF_STOLEN;
1053
1054   tx_error:
1055         if (!IS_ERR(skb))
1056                 kfree_skb(skb);
1057         rcu_read_unlock();
1058         LeaveFunction(10);
1059         return NF_STOLEN;
1060 }
1061
1062 #ifdef CONFIG_IP_VS_IPV6
1063 int
1064 ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1065                      struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1066 {
1067         struct rt6_info *rt;            /* Route to the other host */
1068         struct in6_addr saddr;          /* Source for tunnel */
1069         struct net_device *tdev;        /* Device to other host */
1070         __u8 next_protocol = 0;
1071         __u32 payload_len = 0;
1072         __u8 dsfield = 0;
1073         __u8 ttl = 0;
1074         struct ipv6hdr  *iph;           /* Our new IP header */
1075         unsigned int max_headroom;      /* The extra header space needed */
1076         int ret, local;
1077
1078         EnterFunction(10);
1079
1080         rcu_read_lock();
1081         local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
1082                                       &cp->daddr.in6,
1083                                       &saddr, ipvsh, 1,
1084                                       IP_VS_RT_MODE_LOCAL |
1085                                       IP_VS_RT_MODE_NON_LOCAL |
1086                                       IP_VS_RT_MODE_TUNNEL);
1087         if (local < 0)
1088                 goto tx_error;
1089         if (local) {
1090                 rcu_read_unlock();
1091                 return ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 1);
1092         }
1093
1094         rt = (struct rt6_info *) skb_dst(skb);
1095         tdev = rt->dst.dev;
1096
1097         /*
1098          * Okay, now see if we can stuff it in the buffer as-is.
1099          */
1100         max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct ipv6hdr);
1101
1102         skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
1103                                          &next_protocol, &payload_len,
1104                                          &dsfield, &ttl, NULL);
1105         if (IS_ERR(skb))
1106                 goto tx_error;
1107
1108         skb = iptunnel_handle_offloads(skb, __tun_gso_type_mask(AF_INET6, cp->af));
1109         if (IS_ERR(skb))
1110                 goto tx_error;
1111
1112         skb->transport_header = skb->network_header;
1113
1114         skb_push(skb, sizeof(struct ipv6hdr));
1115         skb_reset_network_header(skb);
1116         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1117
1118         /*
1119          *      Push down and install the IPIP header.
1120          */
1121         iph                     =       ipv6_hdr(skb);
1122         iph->version            =       6;
1123         iph->nexthdr            =       next_protocol;
1124         iph->payload_len        =       htons(payload_len);
1125         memset(&iph->flow_lbl, 0, sizeof(iph->flow_lbl));
1126         ipv6_change_dsfield(iph, 0, dsfield);
1127         iph->daddr = cp->daddr.in6;
1128         iph->saddr = saddr;
1129         iph->hop_limit          =       ttl;
1130
1131         /* Another hack: avoid icmp_send in ip_fragment */
1132         skb->ignore_df = 1;
1133
1134         ret = ip_vs_tunnel_xmit_prepare(skb, cp);
1135         if (ret == NF_ACCEPT)
1136                 ip6_local_out(cp->ipvs->net, skb->sk, skb);
1137         else if (ret == NF_DROP)
1138                 kfree_skb(skb);
1139         rcu_read_unlock();
1140
1141         LeaveFunction(10);
1142
1143         return NF_STOLEN;
1144
1145 tx_error:
1146         if (!IS_ERR(skb))
1147                 kfree_skb(skb);
1148         rcu_read_unlock();
1149         LeaveFunction(10);
1150         return NF_STOLEN;
1151 }
1152 #endif
1153
1154
1155 /*
1156  *      Direct Routing transmitter
1157  *      Used for ANY protocol
1158  */
1159 int
1160 ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1161               struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1162 {
1163         int local;
1164
1165         EnterFunction(10);
1166
1167         rcu_read_lock();
1168         local = __ip_vs_get_out_rt(cp->ipvs, cp->af, skb, cp->dest, cp->daddr.ip,
1169                                    IP_VS_RT_MODE_LOCAL |
1170                                    IP_VS_RT_MODE_NON_LOCAL |
1171                                    IP_VS_RT_MODE_KNOWN_NH, NULL, ipvsh);
1172         if (local < 0)
1173                 goto tx_error;
1174         if (local) {
1175                 rcu_read_unlock();
1176                 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
1177         }
1178
1179         ip_send_check(ip_hdr(skb));
1180
1181         /* Another hack: avoid icmp_send in ip_fragment */
1182         skb->ignore_df = 1;
1183
1184         ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 0);
1185         rcu_read_unlock();
1186
1187         LeaveFunction(10);
1188         return NF_STOLEN;
1189
1190   tx_error:
1191         kfree_skb(skb);
1192         rcu_read_unlock();
1193         LeaveFunction(10);
1194         return NF_STOLEN;
1195 }
1196
1197 #ifdef CONFIG_IP_VS_IPV6
1198 int
1199 ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1200                  struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1201 {
1202         int local;
1203
1204         EnterFunction(10);
1205
1206         rcu_read_lock();
1207         local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
1208                                       &cp->daddr.in6,
1209                                       NULL, ipvsh, 0,
1210                                       IP_VS_RT_MODE_LOCAL |
1211                                       IP_VS_RT_MODE_NON_LOCAL |
1212                                       IP_VS_RT_MODE_KNOWN_NH);
1213         if (local < 0)
1214                 goto tx_error;
1215         if (local) {
1216                 rcu_read_unlock();
1217                 return ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 1);
1218         }
1219
1220         /* Another hack: avoid icmp_send in ip_fragment */
1221         skb->ignore_df = 1;
1222
1223         ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 0);
1224         rcu_read_unlock();
1225
1226         LeaveFunction(10);
1227         return NF_STOLEN;
1228
1229 tx_error:
1230         kfree_skb(skb);
1231         rcu_read_unlock();
1232         LeaveFunction(10);
1233         return NF_STOLEN;
1234 }
1235 #endif
1236
1237
1238 /*
1239  *      ICMP packet transmitter
1240  *      called by the ip_vs_in_icmp
1241  */
1242 int
1243 ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1244                 struct ip_vs_protocol *pp, int offset, unsigned int hooknum,
1245                 struct ip_vs_iphdr *iph)
1246 {
1247         struct rtable   *rt;    /* Route to the other host */
1248         int rc;
1249         int local;
1250         int rt_mode, was_input;
1251
1252         EnterFunction(10);
1253
1254         /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1255            forwarded directly here, because there is no need to
1256            translate address/port back */
1257         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1258                 if (cp->packet_xmit)
1259                         rc = cp->packet_xmit(skb, cp, pp, iph);
1260                 else
1261                         rc = NF_ACCEPT;
1262                 /* do not touch skb anymore */
1263                 atomic_inc(&cp->in_pkts);
1264                 goto out;
1265         }
1266
1267         /*
1268          * mangle and send the packet here (only for VS/NAT)
1269          */
1270         was_input = rt_is_input_route(skb_rtable(skb));
1271
1272         /* LOCALNODE from FORWARD hook is not supported */
1273         rt_mode = (hooknum != NF_INET_FORWARD) ?
1274                   IP_VS_RT_MODE_LOCAL | IP_VS_RT_MODE_NON_LOCAL |
1275                   IP_VS_RT_MODE_RDR : IP_VS_RT_MODE_NON_LOCAL;
1276         rcu_read_lock();
1277         local = __ip_vs_get_out_rt(cp->ipvs, cp->af, skb, cp->dest, cp->daddr.ip, rt_mode,
1278                                    NULL, iph);
1279         if (local < 0)
1280                 goto tx_error;
1281         rt = skb_rtable(skb);
1282
1283         /*
1284          * Avoid duplicate tuple in reply direction for NAT traffic
1285          * to local address when connection is sync-ed
1286          */
1287 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1288         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1289                 enum ip_conntrack_info ctinfo;
1290                 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
1291
1292                 if (ct && !nf_ct_is_untracked(ct)) {
1293                         IP_VS_DBG(10, "%s(): "
1294                                   "stopping DNAT to local address %pI4\n",
1295                                   __func__, &cp->daddr.ip);
1296                         goto tx_error;
1297                 }
1298         }
1299 #endif
1300
1301         /* From world but DNAT to loopback address? */
1302         if (local && ipv4_is_loopback(cp->daddr.ip) && was_input) {
1303                 IP_VS_DBG(1, "%s(): "
1304                           "stopping DNAT to loopback %pI4\n",
1305                           __func__, &cp->daddr.ip);
1306                 goto tx_error;
1307         }
1308
1309         /* copy-on-write the packet before mangling it */
1310         if (!skb_make_writable(skb, offset))
1311                 goto tx_error;
1312
1313         if (skb_cow(skb, rt->dst.dev->hard_header_len))
1314                 goto tx_error;
1315
1316         ip_vs_nat_icmp(skb, pp, cp, 0);
1317
1318         /* Another hack: avoid icmp_send in ip_fragment */
1319         skb->ignore_df = 1;
1320
1321         rc = ip_vs_nat_send_or_cont(NFPROTO_IPV4, skb, cp, local);
1322         rcu_read_unlock();
1323         goto out;
1324
1325   tx_error:
1326         kfree_skb(skb);
1327         rcu_read_unlock();
1328         rc = NF_STOLEN;
1329   out:
1330         LeaveFunction(10);
1331         return rc;
1332 }
1333
1334 #ifdef CONFIG_IP_VS_IPV6
1335 int
1336 ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1337                 struct ip_vs_protocol *pp, int offset, unsigned int hooknum,
1338                 struct ip_vs_iphdr *ipvsh)
1339 {
1340         struct rt6_info *rt;    /* Route to the other host */
1341         int rc;
1342         int local;
1343         int rt_mode;
1344
1345         EnterFunction(10);
1346
1347         /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1348            forwarded directly here, because there is no need to
1349            translate address/port back */
1350         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1351                 if (cp->packet_xmit)
1352                         rc = cp->packet_xmit(skb, cp, pp, ipvsh);
1353                 else
1354                         rc = NF_ACCEPT;
1355                 /* do not touch skb anymore */
1356                 atomic_inc(&cp->in_pkts);
1357                 goto out;
1358         }
1359
1360         /*
1361          * mangle and send the packet here (only for VS/NAT)
1362          */
1363
1364         /* LOCALNODE from FORWARD hook is not supported */
1365         rt_mode = (hooknum != NF_INET_FORWARD) ?
1366                   IP_VS_RT_MODE_LOCAL | IP_VS_RT_MODE_NON_LOCAL |
1367                   IP_VS_RT_MODE_RDR : IP_VS_RT_MODE_NON_LOCAL;
1368         rcu_read_lock();
1369         local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
1370                                       &cp->daddr.in6, NULL, ipvsh, 0, rt_mode);
1371         if (local < 0)
1372                 goto tx_error;
1373         rt = (struct rt6_info *) skb_dst(skb);
1374         /*
1375          * Avoid duplicate tuple in reply direction for NAT traffic
1376          * to local address when connection is sync-ed
1377          */
1378 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1379         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1380                 enum ip_conntrack_info ctinfo;
1381                 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
1382
1383                 if (ct && !nf_ct_is_untracked(ct)) {
1384                         IP_VS_DBG(10, "%s(): "
1385                                   "stopping DNAT to local address %pI6\n",
1386                                   __func__, &cp->daddr.in6);
1387                         goto tx_error;
1388                 }
1389         }
1390 #endif
1391
1392         /* From world but DNAT to loopback address? */
1393         if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
1394             ipv6_addr_type(&cp->daddr.in6) & IPV6_ADDR_LOOPBACK) {
1395                 IP_VS_DBG(1, "%s(): "
1396                           "stopping DNAT to loopback %pI6\n",
1397                           __func__, &cp->daddr.in6);
1398                 goto tx_error;
1399         }
1400
1401         /* copy-on-write the packet before mangling it */
1402         if (!skb_make_writable(skb, offset))
1403                 goto tx_error;
1404
1405         if (skb_cow(skb, rt->dst.dev->hard_header_len))
1406                 goto tx_error;
1407
1408         ip_vs_nat_icmp_v6(skb, pp, cp, 0);
1409
1410         /* Another hack: avoid icmp_send in ip_fragment */
1411         skb->ignore_df = 1;
1412
1413         rc = ip_vs_nat_send_or_cont(NFPROTO_IPV6, skb, cp, local);
1414         rcu_read_unlock();
1415         goto out;
1416
1417 tx_error:
1418         kfree_skb(skb);
1419         rcu_read_unlock();
1420         rc = NF_STOLEN;
1421 out:
1422         LeaveFunction(10);
1423         return rc;
1424 }
1425 #endif