net/netfilter/nf_flow_table_ip.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 #include <linux/kernel.h>
   3 #include <linux/init.h>
   4 #include <linux/module.h>
   5 #include <linux/netfilter.h>
   6 #include <linux/rhashtable.h>
   7 #include <linux/ip.h>
   8 #include <linux/ipv6.h>
   9 #include <linux/netdevice.h>
  10 #include <linux/if_ether.h>
  11 #include <net/ip.h>
  12 #include <net/ipv6.h>
  13 #include <net/ip6_route.h>
  14 #include <net/neighbour.h>
  15 #include <net/netfilter/nf_flow_table.h>
  16 #include <net/netfilter/nf_conntrack_acct.h>
  17 /* For layer 4 checksum field offset. */
  18 #include <linux/tcp.h>
  19 #include <linux/udp.h>
  20
  21 static int nf_flow_state_check(struct flow_offload *flow, int proto,
  22                                struct sk_buff *skb, unsigned int thoff)
  23 {
  24         struct tcphdr *tcph;
  25
  26         if (proto != IPPROTO_TCP)
  27                 return 0;
  28
  29         tcph = (void *)(skb_network_header(skb) + thoff);
  30         if (unlikely(tcph->fin || tcph->rst)) {
  31                 flow_offload_teardown(flow);
  32                 return -1;
  33         }
  34
  35         return 0;
  36 }
  37
  38 static void nf_flow_nat_ip_tcp(struct sk_buff *skb, unsigned int thoff,
  39                                __be32 addr, __be32 new_addr)
  40 {
  41         struct tcphdr *tcph;
  42
  43         tcph = (void *)(skb_network_header(skb) + thoff);
  44         inet_proto_csum_replace4(&tcph->check, skb, addr, new_addr, true);
  45 }
  46
  47 static void nf_flow_nat_ip_udp(struct sk_buff *skb, unsigned int thoff,
  48                                __be32 addr, __be32 new_addr)
  49 {
  50         struct udphdr *udph;
  51
  52         udph = (void *)(skb_network_header(skb) + thoff);
  53         if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
  54                 inet_proto_csum_replace4(&udph->check, skb, addr,
  55                                          new_addr, true);
  56                 if (!udph->check)
  57                         udph->check = CSUM_MANGLED_0;
  58         }
  59 }
  60
  61 static void nf_flow_nat_ip_l4proto(struct sk_buff *skb, struct iphdr *iph,
  62                                    unsigned int thoff, __be32 addr,
  63                                    __be32 new_addr)
  64 {
  65         switch (iph->protocol) {
  66         case IPPROTO_TCP:
  67                 nf_flow_nat_ip_tcp(skb, thoff, addr, new_addr);
  68                 break;
  69         case IPPROTO_UDP:
  70                 nf_flow_nat_ip_udp(skb, thoff, addr, new_addr);
  71                 break;
  72         }
  73 }
  74
  75 static void nf_flow_snat_ip(const struct flow_offload *flow,
  76                             struct sk_buff *skb, struct iphdr *iph,
  77                             unsigned int thoff, enum flow_offload_tuple_dir dir)
  78 {
  79         __be32 addr, new_addr;
  80
  81         switch (dir) {
  82         case FLOW_OFFLOAD_DIR_ORIGINAL:
  83                 addr = iph->saddr;
  84                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_v4.s_addr;
  85                 iph->saddr = new_addr;
  86                 break;
  87         case FLOW_OFFLOAD_DIR_REPLY:
  88                 addr = iph->daddr;
  89                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_v4.s_addr;
  90                 iph->daddr = new_addr;
  91                 break;
  92         }
  93         csum_replace4(&iph->check, addr, new_addr);
  94
  95         nf_flow_nat_ip_l4proto(skb, iph, thoff, addr, new_addr);
  96 }
  97
  98 static void nf_flow_dnat_ip(const struct flow_offload *flow,
  99                             struct sk_buff *skb, struct iphdr *iph,
 100                             unsigned int thoff, enum flow_offload_tuple_dir dir)
 101 {
 102         __be32 addr, new_addr;
 103
 104         switch (dir) {
 105         case FLOW_OFFLOAD_DIR_ORIGINAL:
 106                 addr = iph->daddr;
 107                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_v4.s_addr;
 108                 iph->daddr = new_addr;
 109                 break;
 110         case FLOW_OFFLOAD_DIR_REPLY:
 111                 addr = iph->saddr;
 112                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_v4.s_addr;
 113                 iph->saddr = new_addr;
 114                 break;
 115         }
 116         csum_replace4(&iph->check, addr, new_addr);
 117
 118         nf_flow_nat_ip_l4proto(skb, iph, thoff, addr, new_addr);
 119 }
 120
 121 static void nf_flow_nat_ip(const struct flow_offload *flow, struct sk_buff *skb,
 122                           unsigned int thoff, enum flow_offload_tuple_dir dir,
 123                           struct iphdr *iph)
 124 {
 125         if (test_bit(NF_FLOW_SNAT, &flow->flags)) {
 126                 nf_flow_snat_port(flow, skb, thoff, iph->protocol, dir);
 127                 nf_flow_snat_ip(flow, skb, iph, thoff, dir);
 128         }
 129         if (test_bit(NF_FLOW_DNAT, &flow->flags)) {
 130                 nf_flow_dnat_port(flow, skb, thoff, iph->protocol, dir);
 131                 nf_flow_dnat_ip(flow, skb, iph, thoff, dir);
 132         }
 133 }
 134
 135 static bool ip_has_options(unsigned int thoff)
 136 {
 137         return thoff != sizeof(struct iphdr);
 138 }
 139
 140 static void nf_flow_tuple_encap(struct sk_buff *skb,
 141                                 struct flow_offload_tuple *tuple)
 142 {
 143         struct vlan_ethhdr *veth;
 144         struct pppoe_hdr *phdr;
 145         int i = 0;
 146
 147         if (skb_vlan_tag_present(skb)) {
 148                 tuple->encap[i].id = skb_vlan_tag_get(skb);
 149                 tuple->encap[i].proto = skb->vlan_proto;
 150                 i++;
 151         }
 152         switch (skb->protocol) {
 153         case htons(ETH_P_8021Q):
 154                 veth = (struct vlan_ethhdr *)skb_mac_header(skb);
 155                 tuple->encap[i].id = ntohs(veth->h_vlan_TCI);
 156                 tuple->encap[i].proto = skb->protocol;
 157                 break;
 158         case htons(ETH_P_PPP_SES):
 159                 phdr = (struct pppoe_hdr *)skb_mac_header(skb);
 160                 tuple->encap[i].id = ntohs(phdr->sid);
 161                 tuple->encap[i].proto = skb->protocol;
 162                 break;
 163         }
 164 }
 165
 166 static int nf_flow_tuple_ip(struct sk_buff *skb, const struct net_device *dev,
 167                             struct flow_offload_tuple *tuple, u32 *hdrsize,
 168                             u32 offset)
 169 {
 170         struct flow_ports *ports;
 171         unsigned int thoff;
 172         struct iphdr *iph;
 173         u8 ipproto;
 174
 175         if (!pskb_may_pull(skb, sizeof(*iph) + offset))
 176                 return -1;
 177
 178         iph = (struct iphdr *)(skb_network_header(skb) + offset);
 179         thoff = (iph->ihl * 4);
 180
 181         if (ip_is_fragment(iph) ||
 182             unlikely(ip_has_options(thoff)))
 183                 return -1;
 184
 185         thoff += offset;
 186
 187         ipproto = iph->protocol;
 188         switch (ipproto) {
 189         case IPPROTO_TCP:
 190                 *hdrsize = sizeof(struct tcphdr);
 191                 break;
 192         case IPPROTO_UDP:
 193                 *hdrsize = sizeof(struct udphdr);
 194                 break;
 195 #ifdef CONFIG_NF_CT_PROTO_GRE
 196         case IPPROTO_GRE:
 197                 *hdrsize = sizeof(struct gre_base_hdr);
 198                 break;
 199 #endif
 200         default:
 201                 return -1;
 202         }
 203
 204         if (iph->ttl <= 1)
 205                 return -1;
 206
 207         if (!pskb_may_pull(skb, thoff + *hdrsize))
 208                 return -1;
 209
 210         switch (ipproto) {
 211         case IPPROTO_TCP:
 212         case IPPROTO_UDP:
 213                 ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
 214                 tuple->src_port         = ports->source;
 215                 tuple->dst_port         = ports->dest;
 216                 break;
 217         case IPPROTO_GRE: {
 218                 struct gre_base_hdr *greh;
 219
 220                 greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff);
 221                 if ((greh->flags & GRE_VERSION) != GRE_VERSION_0)
 222                         return -1;
 223                 break;
 224         }
 225         }
 226
 227         iph = (struct iphdr *)(skb_network_header(skb) + offset);
 228
 229         tuple->src_v4.s_addr    = iph->saddr;
 230         tuple->dst_v4.s_addr    = iph->daddr;
 231         tuple->l3proto          = AF_INET;
 232         tuple->l4proto          = ipproto;
 233         tuple->iifidx           = dev->ifindex;
 234         nf_flow_tuple_encap(skb, tuple);
 235
 236         return 0;
 237 }
 238
 239 /* Based on ip_exceeds_mtu(). */
 240 static bool nf_flow_exceeds_mtu(const struct sk_buff *skb, unsigned int mtu)
 241 {
 242         if (skb->len <= mtu)
 243                 return false;
 244
 245         if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
 246                 return false;
 247
 248         return true;
 249 }
 250
 251 static unsigned int nf_flow_xmit_xfrm(struct sk_buff *skb,
 252                                       const struct nf_hook_state *state,
 253                                       struct dst_entry *dst)
 254 {
 255         skb_orphan(skb);
 256         skb_dst_set_noref(skb, dst);
 257         dst_output(state->net, state->sk, skb);
 258         return NF_STOLEN;
 259 }
 260
 261 static bool nf_flow_skb_encap_protocol(const struct sk_buff *skb, __be16 proto,
 262                                        u32 *offset)
 263 {
 264         struct vlan_ethhdr *veth;
 265
 266         switch (skb->protocol) {
 267         case htons(ETH_P_8021Q):
 268                 veth = (struct vlan_ethhdr *)skb_mac_header(skb);
 269                 if (veth->h_vlan_encapsulated_proto == proto) {
 270                         *offset += VLAN_HLEN;
 271                         return true;
 272                 }
 273                 break;
 274         case htons(ETH_P_PPP_SES):
 275                 if (nf_flow_pppoe_proto(skb) == proto) {
 276                         *offset += PPPOE_SES_HLEN;
 277                         return true;
 278                 }
 279                 break;
 280         }
 281
 282         return false;
 283 }
 284
 285 static void nf_flow_encap_pop(struct sk_buff *skb,
 286                               struct flow_offload_tuple_rhash *tuplehash)
 287 {
 288         struct vlan_hdr *vlan_hdr;
 289         int i;
 290
 291         for (i = 0; i < tuplehash->tuple.encap_num; i++) {
 292                 if (skb_vlan_tag_present(skb)) {
 293                         __vlan_hwaccel_clear_tag(skb);
 294                         continue;
 295                 }
 296                 switch (skb->protocol) {
 297                 case htons(ETH_P_8021Q):
 298                         vlan_hdr = (struct vlan_hdr *)skb->data;
 299                         __skb_pull(skb, VLAN_HLEN);
 300                         vlan_set_encap_proto(skb, vlan_hdr);
 301                         skb_reset_network_header(skb);
 302                         break;
 303                 case htons(ETH_P_PPP_SES):
 304                         skb->protocol = nf_flow_pppoe_proto(skb);
 305                         skb_pull(skb, PPPOE_SES_HLEN);
 306                         skb_reset_network_header(skb);
 307                         break;
 308                 }
 309         }
 310 }
 311
 312 static unsigned int nf_flow_queue_xmit(struct net *net, struct sk_buff *skb,
 313                                        const struct flow_offload_tuple_rhash *tuplehash,
 314                                        unsigned short type)
 315 {
 316         struct net_device *outdev;
 317
 318         outdev = dev_get_by_index_rcu(net, tuplehash->tuple.out.ifidx);
 319         if (!outdev)
 320                 return NF_DROP;
 321
 322         skb->dev = outdev;
 323         dev_hard_header(skb, skb->dev, type, tuplehash->tuple.out.h_dest,
 324                         tuplehash->tuple.out.h_source, skb->len);
 325         dev_queue_xmit(skb);
 326
 327         return NF_STOLEN;
 328 }
 329
 330 unsigned int
 331 nf_flow_offload_ip_hook(void *priv, struct sk_buff *skb,
 332                         const struct nf_hook_state *state)
 333 {
 334         struct flow_offload_tuple_rhash *tuplehash;
 335         struct nf_flowtable *flow_table = priv;
 336         struct flow_offload_tuple tuple = {};
 337         enum flow_offload_tuple_dir dir;
 338         struct flow_offload *flow;
 339         struct net_device *outdev;
 340         u32 hdrsize, offset = 0;
 341         unsigned int thoff, mtu;
 342         struct rtable *rt;
 343         struct iphdr *iph;
 344         __be32 nexthop;
 345         int ret;
 346
 347         if (skb->protocol != htons(ETH_P_IP) &&
 348             !nf_flow_skb_encap_protocol(skb, htons(ETH_P_IP), &offset))
 349                 return NF_ACCEPT;
 350
 351         if (nf_flow_tuple_ip(skb, state->in, &tuple, &hdrsize, offset) < 0)
 352                 return NF_ACCEPT;
 353
 354         tuplehash = flow_offload_lookup(flow_table, &tuple);
 355         if (tuplehash == NULL)
 356                 return NF_ACCEPT;
 357
 358         dir = tuplehash->tuple.dir;
 359         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 360
 361         mtu = flow->tuplehash[dir].tuple.mtu + offset;
 362         if (unlikely(nf_flow_exceeds_mtu(skb, mtu)))
 363                 return NF_ACCEPT;
 364
 365         iph = (struct iphdr *)(skb_network_header(skb) + offset);
 366         thoff = (iph->ihl * 4) + offset;
 367         if (nf_flow_state_check(flow, iph->protocol, skb, thoff))
 368                 return NF_ACCEPT;
 369
 370         if (skb_try_make_writable(skb, thoff + hdrsize))
 371                 return NF_DROP;
 372
 373         flow_offload_refresh(flow_table, flow);
 374
 375         nf_flow_encap_pop(skb, tuplehash);
 376         thoff -= offset;
 377
 378         iph = ip_hdr(skb);
 379         nf_flow_nat_ip(flow, skb, thoff, dir, iph);
 380
 381         ip_decrease_ttl(iph);
 382         skb_clear_tstamp(skb);
 383
 384         if (flow_table->flags & NF_FLOWTABLE_COUNTER)
 385                 nf_ct_acct_update(flow->ct, tuplehash->tuple.dir, skb->len);
 386
 387         if (unlikely(tuplehash->tuple.xmit_type == FLOW_OFFLOAD_XMIT_XFRM)) {
 388                 rt = (struct rtable *)tuplehash->tuple.dst_cache;
 389                 memset(skb->cb, 0, sizeof(struct inet_skb_parm));
 390                 IPCB(skb)->iif = skb->dev->ifindex;
 391                 IPCB(skb)->flags = IPSKB_FORWARDED;
 392                 return nf_flow_xmit_xfrm(skb, state, &rt->dst);
 393         }
 394
 395         switch (tuplehash->tuple.xmit_type) {
 396         case FLOW_OFFLOAD_XMIT_NEIGH:
 397                 rt = (struct rtable *)tuplehash->tuple.dst_cache;
 398                 outdev = rt->dst.dev;
 399                 skb->dev = outdev;
 400                 nexthop = rt_nexthop(rt, flow->tuplehash[!dir].tuple.src_v4.s_addr);
 401                 skb_dst_set_noref(skb, &rt->dst);
 402                 neigh_xmit(NEIGH_ARP_TABLE, outdev, &nexthop, skb);
 403                 ret = NF_STOLEN;
 404                 break;
 405         case FLOW_OFFLOAD_XMIT_DIRECT:
 406                 ret = nf_flow_queue_xmit(state->net, skb, tuplehash, ETH_P_IP);
 407                 if (ret == NF_DROP)
 408                         flow_offload_teardown(flow);
 409                 break;
 410         }
 411
 412         return ret;
 413 }
 414 EXPORT_SYMBOL_GPL(nf_flow_offload_ip_hook);
 415
 416 static void nf_flow_nat_ipv6_tcp(struct sk_buff *skb, unsigned int thoff,
 417                                  struct in6_addr *addr,
 418                                  struct in6_addr *new_addr,
 419                                  struct ipv6hdr *ip6h)
 420 {
 421         struct tcphdr *tcph;
 422
 423         tcph = (void *)(skb_network_header(skb) + thoff);
 424         inet_proto_csum_replace16(&tcph->check, skb, addr->s6_addr32,
 425                                   new_addr->s6_addr32, true);
 426 }
 427
 428 static void nf_flow_nat_ipv6_udp(struct sk_buff *skb, unsigned int thoff,
 429                                  struct in6_addr *addr,
 430                                  struct in6_addr *new_addr)
 431 {
 432         struct udphdr *udph;
 433
 434         udph = (void *)(skb_network_header(skb) + thoff);
 435         if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 436                 inet_proto_csum_replace16(&udph->check, skb, addr->s6_addr32,
 437                                           new_addr->s6_addr32, true);
 438                 if (!udph->check)
 439                         udph->check = CSUM_MANGLED_0;
 440         }
 441 }
 442
 443 static void nf_flow_nat_ipv6_l4proto(struct sk_buff *skb, struct ipv6hdr *ip6h,
 444                                      unsigned int thoff, struct in6_addr *addr,
 445                                      struct in6_addr *new_addr)
 446 {
 447         switch (ip6h->nexthdr) {
 448         case IPPROTO_TCP:
 449                 nf_flow_nat_ipv6_tcp(skb, thoff, addr, new_addr, ip6h);
 450                 break;
 451         case IPPROTO_UDP:
 452                 nf_flow_nat_ipv6_udp(skb, thoff, addr, new_addr);
 453                 break;
 454         }
 455 }
 456
 457 static void nf_flow_snat_ipv6(const struct flow_offload *flow,
 458                               struct sk_buff *skb, struct ipv6hdr *ip6h,
 459                               unsigned int thoff,
 460                               enum flow_offload_tuple_dir dir)
 461 {
 462         struct in6_addr addr, new_addr;
 463
 464         switch (dir) {
 465         case FLOW_OFFLOAD_DIR_ORIGINAL:
 466                 addr = ip6h->saddr;
 467                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_v6;
 468                 ip6h->saddr = new_addr;
 469                 break;
 470         case FLOW_OFFLOAD_DIR_REPLY:
 471                 addr = ip6h->daddr;
 472                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_v6;
 473                 ip6h->daddr = new_addr;
 474                 break;
 475         }
 476
 477         nf_flow_nat_ipv6_l4proto(skb, ip6h, thoff, &addr, &new_addr);
 478 }
 479
 480 static void nf_flow_dnat_ipv6(const struct flow_offload *flow,
 481                               struct sk_buff *skb, struct ipv6hdr *ip6h,
 482                               unsigned int thoff,
 483                               enum flow_offload_tuple_dir dir)
 484 {
 485         struct in6_addr addr, new_addr;
 486
 487         switch (dir) {
 488         case FLOW_OFFLOAD_DIR_ORIGINAL:
 489                 addr = ip6h->daddr;
 490                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_v6;
 491                 ip6h->daddr = new_addr;
 492                 break;
 493         case FLOW_OFFLOAD_DIR_REPLY:
 494                 addr = ip6h->saddr;
 495                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_v6;
 496                 ip6h->saddr = new_addr;
 497                 break;
 498         }
 499
 500         nf_flow_nat_ipv6_l4proto(skb, ip6h, thoff, &addr, &new_addr);
 501 }
 502
 503 static void nf_flow_nat_ipv6(const struct flow_offload *flow,
 504                              struct sk_buff *skb,
 505                              enum flow_offload_tuple_dir dir,
 506                              struct ipv6hdr *ip6h)
 507 {
 508         unsigned int thoff = sizeof(*ip6h);
 509
 510         if (test_bit(NF_FLOW_SNAT, &flow->flags)) {
 511                 nf_flow_snat_port(flow, skb, thoff, ip6h->nexthdr, dir);
 512                 nf_flow_snat_ipv6(flow, skb, ip6h, thoff, dir);
 513         }
 514         if (test_bit(NF_FLOW_DNAT, &flow->flags)) {
 515                 nf_flow_dnat_port(flow, skb, thoff, ip6h->nexthdr, dir);
 516                 nf_flow_dnat_ipv6(flow, skb, ip6h, thoff, dir);
 517         }
 518 }
 519
 520 static int nf_flow_tuple_ipv6(struct sk_buff *skb, const struct net_device *dev,
 521                               struct flow_offload_tuple *tuple, u32 *hdrsize,
 522                               u32 offset)
 523 {
 524         struct flow_ports *ports;
 525         struct ipv6hdr *ip6h;
 526         unsigned int thoff;
 527         u8 nexthdr;
 528
 529         thoff = sizeof(*ip6h) + offset;
 530         if (!pskb_may_pull(skb, thoff))
 531                 return -1;
 532
 533         ip6h = (struct ipv6hdr *)(skb_network_header(skb) + offset);
 534
 535         nexthdr = ip6h->nexthdr;
 536         switch (nexthdr) {
 537         case IPPROTO_TCP:
 538                 *hdrsize = sizeof(struct tcphdr);
 539                 break;
 540         case IPPROTO_UDP:
 541                 *hdrsize = sizeof(struct udphdr);
 542                 break;
 543 #ifdef CONFIG_NF_CT_PROTO_GRE
 544         case IPPROTO_GRE:
 545                 *hdrsize = sizeof(struct gre_base_hdr);
 546                 break;
 547 #endif
 548         default:
 549                 return -1;
 550         }
 551
 552         if (ip6h->hop_limit <= 1)
 553                 return -1;
 554
 555         if (!pskb_may_pull(skb, thoff + *hdrsize))
 556                 return -1;
 557
 558         switch (nexthdr) {
 559         case IPPROTO_TCP:
 560         case IPPROTO_UDP:
 561                 ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
 562                 tuple->src_port         = ports->source;
 563                 tuple->dst_port         = ports->dest;
 564                 break;
 565         case IPPROTO_GRE: {
 566                 struct gre_base_hdr *greh;
 567
 568                 greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff);
 569                 if ((greh->flags & GRE_VERSION) != GRE_VERSION_0)
 570                         return -1;
 571                 break;
 572         }
 573         }
 574
 575         ip6h = (struct ipv6hdr *)(skb_network_header(skb) + offset);
 576
 577         tuple->src_v6           = ip6h->saddr;
 578         tuple->dst_v6           = ip6h->daddr;
 579         tuple->l3proto          = AF_INET6;
 580         tuple->l4proto          = nexthdr;
 581         tuple->iifidx           = dev->ifindex;
 582         nf_flow_tuple_encap(skb, tuple);
 583
 584         return 0;
 585 }
 586
 587 unsigned int
 588 nf_flow_offload_ipv6_hook(void *priv, struct sk_buff *skb,
 589                           const struct nf_hook_state *state)
 590 {
 591         struct flow_offload_tuple_rhash *tuplehash;
 592         struct nf_flowtable *flow_table = priv;
 593         struct flow_offload_tuple tuple = {};
 594         enum flow_offload_tuple_dir dir;
 595         const struct in6_addr *nexthop;
 596         struct flow_offload *flow;
 597         struct net_device *outdev;
 598         unsigned int thoff, mtu;
 599         u32 hdrsize, offset = 0;
 600         struct ipv6hdr *ip6h;
 601         struct rt6_info *rt;
 602         int ret;
 603
 604         if (skb->protocol != htons(ETH_P_IPV6) &&
 605             !nf_flow_skb_encap_protocol(skb, htons(ETH_P_IPV6), &offset))
 606                 return NF_ACCEPT;
 607
 608         if (nf_flow_tuple_ipv6(skb, state->in, &tuple, &hdrsize, offset) < 0)
 609                 return NF_ACCEPT;
 610
 611         tuplehash = flow_offload_lookup(flow_table, &tuple);
 612         if (tuplehash == NULL)
 613                 return NF_ACCEPT;
 614
 615         dir = tuplehash->tuple.dir;
 616         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 617
 618         mtu = flow->tuplehash[dir].tuple.mtu + offset;
 619         if (unlikely(nf_flow_exceeds_mtu(skb, mtu)))
 620                 return NF_ACCEPT;
 621
 622         ip6h = (struct ipv6hdr *)(skb_network_header(skb) + offset);
 623         thoff = sizeof(*ip6h) + offset;
 624         if (nf_flow_state_check(flow, ip6h->nexthdr, skb, thoff))
 625                 return NF_ACCEPT;
 626
 627         if (skb_try_make_writable(skb, thoff + hdrsize))
 628                 return NF_DROP;
 629
 630         flow_offload_refresh(flow_table, flow);
 631
 632         nf_flow_encap_pop(skb, tuplehash);
 633
 634         ip6h = ipv6_hdr(skb);
 635         nf_flow_nat_ipv6(flow, skb, dir, ip6h);
 636
 637         ip6h->hop_limit--;
 638         skb_clear_tstamp(skb);
 639
 640         if (flow_table->flags & NF_FLOWTABLE_COUNTER)
 641                 nf_ct_acct_update(flow->ct, tuplehash->tuple.dir, skb->len);
 642
 643         if (unlikely(tuplehash->tuple.xmit_type == FLOW_OFFLOAD_XMIT_XFRM)) {
 644                 rt = (struct rt6_info *)tuplehash->tuple.dst_cache;
 645                 memset(skb->cb, 0, sizeof(struct inet6_skb_parm));
 646                 IP6CB(skb)->iif = skb->dev->ifindex;
 647                 IP6CB(skb)->flags = IP6SKB_FORWARDED;
 648                 return nf_flow_xmit_xfrm(skb, state, &rt->dst);
 649         }
 650
 651         switch (tuplehash->tuple.xmit_type) {
 652         case FLOW_OFFLOAD_XMIT_NEIGH:
 653                 rt = (struct rt6_info *)tuplehash->tuple.dst_cache;
 654                 outdev = rt->dst.dev;
 655                 skb->dev = outdev;
 656                 nexthop = rt6_nexthop(rt, &flow->tuplehash[!dir].tuple.src_v6);
 657                 skb_dst_set_noref(skb, &rt->dst);
 658                 neigh_xmit(NEIGH_ND_TABLE, outdev, nexthop, skb);
 659                 ret = NF_STOLEN;
 660                 break;
 661         case FLOW_OFFLOAD_XMIT_DIRECT:
 662                 ret = nf_flow_queue_xmit(state->net, skb, tuplehash, ETH_P_IPV6);
 663                 if (ret == NF_DROP)
 664                         flow_offload_teardown(flow);
 665                 break;
 666         }
 667
 668         return ret;
 669 }
 670 EXPORT_SYMBOL_GPL(nf_flow_offload_ipv6_hook);