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 struct nf_flowtable_ctx {
 167         const struct net_device *in;
 168         u32                     offset;
 169         u32                     hdrsize;
 170 };
 171
 172 static int nf_flow_tuple_ip(struct nf_flowtable_ctx *ctx, struct sk_buff *skb,
 173                             struct flow_offload_tuple *tuple)
 174 {
 175         struct flow_ports *ports;
 176         unsigned int thoff;
 177         struct iphdr *iph;
 178         u8 ipproto;
 179
 180         if (!pskb_may_pull(skb, sizeof(*iph) + ctx->offset))
 181                 return -1;
 182
 183         iph = (struct iphdr *)(skb_network_header(skb) + ctx->offset);
 184         thoff = (iph->ihl * 4);
 185
 186         if (ip_is_fragment(iph) ||
 187             unlikely(ip_has_options(thoff)))
 188                 return -1;
 189
 190         thoff += ctx->offset;
 191
 192         ipproto = iph->protocol;
 193         switch (ipproto) {
 194         case IPPROTO_TCP:
 195                 ctx->hdrsize = sizeof(struct tcphdr);
 196                 break;
 197         case IPPROTO_UDP:
 198                 ctx->hdrsize = sizeof(struct udphdr);
 199                 break;
 200 #ifdef CONFIG_NF_CT_PROTO_GRE
 201         case IPPROTO_GRE:
 202                 ctx->hdrsize = sizeof(struct gre_base_hdr);
 203                 break;
 204 #endif
 205         default:
 206                 return -1;
 207         }
 208
 209         if (iph->ttl <= 1)
 210                 return -1;
 211
 212         if (!pskb_may_pull(skb, thoff + ctx->hdrsize))
 213                 return -1;
 214
 215         switch (ipproto) {
 216         case IPPROTO_TCP:
 217         case IPPROTO_UDP:
 218                 ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
 219                 tuple->src_port         = ports->source;
 220                 tuple->dst_port         = ports->dest;
 221                 break;
 222         case IPPROTO_GRE: {
 223                 struct gre_base_hdr *greh;
 224
 225                 greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff);
 226                 if ((greh->flags & GRE_VERSION) != GRE_VERSION_0)
 227                         return -1;
 228                 break;
 229         }
 230         }
 231
 232         iph = (struct iphdr *)(skb_network_header(skb) + ctx->offset);
 233
 234         tuple->src_v4.s_addr    = iph->saddr;
 235         tuple->dst_v4.s_addr    = iph->daddr;
 236         tuple->l3proto          = AF_INET;
 237         tuple->l4proto          = ipproto;
 238         tuple->iifidx           = ctx->in->ifindex;
 239         nf_flow_tuple_encap(skb, tuple);
 240
 241         return 0;
 242 }
 243
 244 /* Based on ip_exceeds_mtu(). */
 245 static bool nf_flow_exceeds_mtu(const struct sk_buff *skb, unsigned int mtu)
 246 {
 247         if (skb->len <= mtu)
 248                 return false;
 249
 250         if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
 251                 return false;
 252
 253         return true;
 254 }
 255
 256 static inline bool nf_flow_dst_check(struct flow_offload_tuple *tuple)
 257 {
 258         if (tuple->xmit_type != FLOW_OFFLOAD_XMIT_NEIGH &&
 259             tuple->xmit_type != FLOW_OFFLOAD_XMIT_XFRM)
 260                 return true;
 261
 262         return dst_check(tuple->dst_cache, tuple->dst_cookie);
 263 }
 264
 265 static unsigned int nf_flow_xmit_xfrm(struct sk_buff *skb,
 266                                       const struct nf_hook_state *state,
 267                                       struct dst_entry *dst)
 268 {
 269         skb_orphan(skb);
 270         skb_dst_set_noref(skb, dst);
 271         dst_output(state->net, state->sk, skb);
 272         return NF_STOLEN;
 273 }
 274
 275 static bool nf_flow_skb_encap_protocol(const struct sk_buff *skb, __be16 proto,
 276                                        u32 *offset)
 277 {
 278         struct vlan_ethhdr *veth;
 279
 280         switch (skb->protocol) {
 281         case htons(ETH_P_8021Q):
 282                 veth = (struct vlan_ethhdr *)skb_mac_header(skb);
 283                 if (veth->h_vlan_encapsulated_proto == proto) {
 284                         *offset += VLAN_HLEN;
 285                         return true;
 286                 }
 287                 break;
 288         case htons(ETH_P_PPP_SES):
 289                 if (nf_flow_pppoe_proto(skb) == proto) {
 290                         *offset += PPPOE_SES_HLEN;
 291                         return true;
 292                 }
 293                 break;
 294         }
 295
 296         return false;
 297 }
 298
 299 static void nf_flow_encap_pop(struct sk_buff *skb,
 300                               struct flow_offload_tuple_rhash *tuplehash)
 301 {
 302         struct vlan_hdr *vlan_hdr;
 303         int i;
 304
 305         for (i = 0; i < tuplehash->tuple.encap_num; i++) {
 306                 if (skb_vlan_tag_present(skb)) {
 307                         __vlan_hwaccel_clear_tag(skb);
 308                         continue;
 309                 }
 310                 switch (skb->protocol) {
 311                 case htons(ETH_P_8021Q):
 312                         vlan_hdr = (struct vlan_hdr *)skb->data;
 313                         __skb_pull(skb, VLAN_HLEN);
 314                         vlan_set_encap_proto(skb, vlan_hdr);
 315                         skb_reset_network_header(skb);
 316                         break;
 317                 case htons(ETH_P_PPP_SES):
 318                         skb->protocol = nf_flow_pppoe_proto(skb);
 319                         skb_pull(skb, PPPOE_SES_HLEN);
 320                         skb_reset_network_header(skb);
 321                         break;
 322                 }
 323         }
 324 }
 325
 326 static unsigned int nf_flow_queue_xmit(struct net *net, struct sk_buff *skb,
 327                                        const struct flow_offload_tuple_rhash *tuplehash,
 328                                        unsigned short type)
 329 {
 330         struct net_device *outdev;
 331
 332         outdev = dev_get_by_index_rcu(net, tuplehash->tuple.out.ifidx);
 333         if (!outdev)
 334                 return NF_DROP;
 335
 336         skb->dev = outdev;
 337         dev_hard_header(skb, skb->dev, type, tuplehash->tuple.out.h_dest,
 338                         tuplehash->tuple.out.h_source, skb->len);
 339         dev_queue_xmit(skb);
 340
 341         return NF_STOLEN;
 342 }
 343
 344 static struct flow_offload_tuple_rhash *
 345 nf_flow_offload_lookup(struct nf_flowtable_ctx *ctx,
 346                        struct nf_flowtable *flow_table, struct sk_buff *skb)
 347 {
 348         struct flow_offload_tuple tuple = {};
 349
 350         if (skb->protocol != htons(ETH_P_IP) &&
 351             !nf_flow_skb_encap_protocol(skb, htons(ETH_P_IP), &ctx->offset))
 352                 return NULL;
 353
 354         if (nf_flow_tuple_ip(ctx, skb, &tuple) < 0)
 355                 return NULL;
 356
 357         return flow_offload_lookup(flow_table, &tuple);
 358 }
 359
 360 static int nf_flow_offload_forward(struct nf_flowtable_ctx *ctx,
 361                                    struct nf_flowtable *flow_table,
 362                                    struct flow_offload_tuple_rhash *tuplehash,
 363                                    struct sk_buff *skb)
 364 {
 365         enum flow_offload_tuple_dir dir;
 366         struct flow_offload *flow;
 367         unsigned int thoff, mtu;
 368         struct iphdr *iph;
 369
 370         dir = tuplehash->tuple.dir;
 371         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 372
 373         mtu = flow->tuplehash[dir].tuple.mtu + ctx->offset;
 374         if (unlikely(nf_flow_exceeds_mtu(skb, mtu)))
 375                 return 0;
 376
 377         iph = (struct iphdr *)(skb_network_header(skb) + ctx->offset);
 378         thoff = (iph->ihl * 4) + ctx->offset;
 379         if (nf_flow_state_check(flow, iph->protocol, skb, thoff))
 380                 return 0;
 381
 382         if (!nf_flow_dst_check(&tuplehash->tuple)) {
 383                 flow_offload_teardown(flow);
 384                 return 0;
 385         }
 386
 387         if (skb_try_make_writable(skb, thoff + ctx->hdrsize))
 388                 return -1;
 389
 390         flow_offload_refresh(flow_table, flow);
 391
 392         nf_flow_encap_pop(skb, tuplehash);
 393         thoff -= ctx->offset;
 394
 395         iph = ip_hdr(skb);
 396         nf_flow_nat_ip(flow, skb, thoff, dir, iph);
 397
 398         ip_decrease_ttl(iph);
 399         skb_clear_tstamp(skb);
 400
 401         if (flow_table->flags & NF_FLOWTABLE_COUNTER)
 402                 nf_ct_acct_update(flow->ct, tuplehash->tuple.dir, skb->len);
 403
 404         return 1;
 405 }
 406
 407 unsigned int
 408 nf_flow_offload_ip_hook(void *priv, struct sk_buff *skb,
 409                         const struct nf_hook_state *state)
 410 {
 411         struct flow_offload_tuple_rhash *tuplehash;
 412         struct nf_flowtable *flow_table = priv;
 413         enum flow_offload_tuple_dir dir;
 414         struct nf_flowtable_ctx ctx = {
 415                 .in     = state->in,
 416         };
 417         struct flow_offload *flow;
 418         struct net_device *outdev;
 419         struct rtable *rt;
 420         __be32 nexthop;
 421         int ret;
 422
 423         tuplehash = nf_flow_offload_lookup(&ctx, flow_table, skb);
 424         if (!tuplehash)
 425                 return NF_ACCEPT;
 426
 427         ret = nf_flow_offload_forward(&ctx, flow_table, tuplehash, skb);
 428         if (ret < 0)
 429                 return NF_DROP;
 430         else if (ret == 0)
 431                 return NF_ACCEPT;
 432
 433         if (unlikely(tuplehash->tuple.xmit_type == FLOW_OFFLOAD_XMIT_XFRM)) {
 434                 rt = (struct rtable *)tuplehash->tuple.dst_cache;
 435                 memset(skb->cb, 0, sizeof(struct inet_skb_parm));
 436                 IPCB(skb)->iif = skb->dev->ifindex;
 437                 IPCB(skb)->flags = IPSKB_FORWARDED;
 438                 return nf_flow_xmit_xfrm(skb, state, &rt->dst);
 439         }
 440
 441         dir = tuplehash->tuple.dir;
 442         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 443
 444         switch (tuplehash->tuple.xmit_type) {
 445         case FLOW_OFFLOAD_XMIT_NEIGH:
 446                 rt = (struct rtable *)tuplehash->tuple.dst_cache;
 447                 outdev = rt->dst.dev;
 448                 skb->dev = outdev;
 449                 nexthop = rt_nexthop(rt, flow->tuplehash[!dir].tuple.src_v4.s_addr);
 450                 skb_dst_set_noref(skb, &rt->dst);
 451                 neigh_xmit(NEIGH_ARP_TABLE, outdev, &nexthop, skb);
 452                 ret = NF_STOLEN;
 453                 break;
 454         case FLOW_OFFLOAD_XMIT_DIRECT:
 455                 ret = nf_flow_queue_xmit(state->net, skb, tuplehash, ETH_P_IP);
 456                 if (ret == NF_DROP)
 457                         flow_offload_teardown(flow);
 458                 break;
 459         default:
 460                 WARN_ON_ONCE(1);
 461                 ret = NF_DROP;
 462                 break;
 463         }
 464
 465         return ret;
 466 }
 467 EXPORT_SYMBOL_GPL(nf_flow_offload_ip_hook);
 468
 469 static void nf_flow_nat_ipv6_tcp(struct sk_buff *skb, unsigned int thoff,
 470                                  struct in6_addr *addr,
 471                                  struct in6_addr *new_addr,
 472                                  struct ipv6hdr *ip6h)
 473 {
 474         struct tcphdr *tcph;
 475
 476         tcph = (void *)(skb_network_header(skb) + thoff);
 477         inet_proto_csum_replace16(&tcph->check, skb, addr->s6_addr32,
 478                                   new_addr->s6_addr32, true);
 479 }
 480
 481 static void nf_flow_nat_ipv6_udp(struct sk_buff *skb, unsigned int thoff,
 482                                  struct in6_addr *addr,
 483                                  struct in6_addr *new_addr)
 484 {
 485         struct udphdr *udph;
 486
 487         udph = (void *)(skb_network_header(skb) + thoff);
 488         if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 489                 inet_proto_csum_replace16(&udph->check, skb, addr->s6_addr32,
 490                                           new_addr->s6_addr32, true);
 491                 if (!udph->check)
 492                         udph->check = CSUM_MANGLED_0;
 493         }
 494 }
 495
 496 static void nf_flow_nat_ipv6_l4proto(struct sk_buff *skb, struct ipv6hdr *ip6h,
 497                                      unsigned int thoff, struct in6_addr *addr,
 498                                      struct in6_addr *new_addr)
 499 {
 500         switch (ip6h->nexthdr) {
 501         case IPPROTO_TCP:
 502                 nf_flow_nat_ipv6_tcp(skb, thoff, addr, new_addr, ip6h);
 503                 break;
 504         case IPPROTO_UDP:
 505                 nf_flow_nat_ipv6_udp(skb, thoff, addr, new_addr);
 506                 break;
 507         }
 508 }
 509
 510 static void nf_flow_snat_ipv6(const struct flow_offload *flow,
 511                               struct sk_buff *skb, struct ipv6hdr *ip6h,
 512                               unsigned int thoff,
 513                               enum flow_offload_tuple_dir dir)
 514 {
 515         struct in6_addr addr, new_addr;
 516
 517         switch (dir) {
 518         case FLOW_OFFLOAD_DIR_ORIGINAL:
 519                 addr = ip6h->saddr;
 520                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_v6;
 521                 ip6h->saddr = new_addr;
 522                 break;
 523         case FLOW_OFFLOAD_DIR_REPLY:
 524                 addr = ip6h->daddr;
 525                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_v6;
 526                 ip6h->daddr = new_addr;
 527                 break;
 528         }
 529
 530         nf_flow_nat_ipv6_l4proto(skb, ip6h, thoff, &addr, &new_addr);
 531 }
 532
 533 static void nf_flow_dnat_ipv6(const struct flow_offload *flow,
 534                               struct sk_buff *skb, struct ipv6hdr *ip6h,
 535                               unsigned int thoff,
 536                               enum flow_offload_tuple_dir dir)
 537 {
 538         struct in6_addr addr, new_addr;
 539
 540         switch (dir) {
 541         case FLOW_OFFLOAD_DIR_ORIGINAL:
 542                 addr = ip6h->daddr;
 543                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_v6;
 544                 ip6h->daddr = new_addr;
 545                 break;
 546         case FLOW_OFFLOAD_DIR_REPLY:
 547                 addr = ip6h->saddr;
 548                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_v6;
 549                 ip6h->saddr = new_addr;
 550                 break;
 551         }
 552
 553         nf_flow_nat_ipv6_l4proto(skb, ip6h, thoff, &addr, &new_addr);
 554 }
 555
 556 static void nf_flow_nat_ipv6(const struct flow_offload *flow,
 557                              struct sk_buff *skb,
 558                              enum flow_offload_tuple_dir dir,
 559                              struct ipv6hdr *ip6h)
 560 {
 561         unsigned int thoff = sizeof(*ip6h);
 562
 563         if (test_bit(NF_FLOW_SNAT, &flow->flags)) {
 564                 nf_flow_snat_port(flow, skb, thoff, ip6h->nexthdr, dir);
 565                 nf_flow_snat_ipv6(flow, skb, ip6h, thoff, dir);
 566         }
 567         if (test_bit(NF_FLOW_DNAT, &flow->flags)) {
 568                 nf_flow_dnat_port(flow, skb, thoff, ip6h->nexthdr, dir);
 569                 nf_flow_dnat_ipv6(flow, skb, ip6h, thoff, dir);
 570         }
 571 }
 572
 573 static int nf_flow_tuple_ipv6(struct nf_flowtable_ctx *ctx, struct sk_buff *skb,
 574                               struct flow_offload_tuple *tuple)
 575 {
 576         struct flow_ports *ports;
 577         struct ipv6hdr *ip6h;
 578         unsigned int thoff;
 579         u8 nexthdr;
 580
 581         thoff = sizeof(*ip6h) + ctx->offset;
 582         if (!pskb_may_pull(skb, thoff))
 583                 return -1;
 584
 585         ip6h = (struct ipv6hdr *)(skb_network_header(skb) + ctx->offset);
 586
 587         nexthdr = ip6h->nexthdr;
 588         switch (nexthdr) {
 589         case IPPROTO_TCP:
 590                 ctx->hdrsize = sizeof(struct tcphdr);
 591                 break;
 592         case IPPROTO_UDP:
 593                 ctx->hdrsize = sizeof(struct udphdr);
 594                 break;
 595 #ifdef CONFIG_NF_CT_PROTO_GRE
 596         case IPPROTO_GRE:
 597                 ctx->hdrsize = sizeof(struct gre_base_hdr);
 598                 break;
 599 #endif
 600         default:
 601                 return -1;
 602         }
 603
 604         if (ip6h->hop_limit <= 1)
 605                 return -1;
 606
 607         if (!pskb_may_pull(skb, thoff + ctx->hdrsize))
 608                 return -1;
 609
 610         switch (nexthdr) {
 611         case IPPROTO_TCP:
 612         case IPPROTO_UDP:
 613                 ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
 614                 tuple->src_port         = ports->source;
 615                 tuple->dst_port         = ports->dest;
 616                 break;
 617         case IPPROTO_GRE: {
 618                 struct gre_base_hdr *greh;
 619
 620                 greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff);
 621                 if ((greh->flags & GRE_VERSION) != GRE_VERSION_0)
 622                         return -1;
 623                 break;
 624         }
 625         }
 626
 627         ip6h = (struct ipv6hdr *)(skb_network_header(skb) + ctx->offset);
 628
 629         tuple->src_v6           = ip6h->saddr;
 630         tuple->dst_v6           = ip6h->daddr;
 631         tuple->l3proto          = AF_INET6;
 632         tuple->l4proto          = nexthdr;
 633         tuple->iifidx           = ctx->in->ifindex;
 634         nf_flow_tuple_encap(skb, tuple);
 635
 636         return 0;
 637 }
 638
 639 static int nf_flow_offload_ipv6_forward(struct nf_flowtable_ctx *ctx,
 640                                         struct nf_flowtable *flow_table,
 641                                         struct flow_offload_tuple_rhash *tuplehash,
 642                                         struct sk_buff *skb)
 643 {
 644         enum flow_offload_tuple_dir dir;
 645         struct flow_offload *flow;
 646         unsigned int thoff, mtu;
 647         struct ipv6hdr *ip6h;
 648
 649         dir = tuplehash->tuple.dir;
 650         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 651
 652         mtu = flow->tuplehash[dir].tuple.mtu + ctx->offset;
 653         if (unlikely(nf_flow_exceeds_mtu(skb, mtu)))
 654                 return 0;
 655
 656         ip6h = (struct ipv6hdr *)(skb_network_header(skb) + ctx->offset);
 657         thoff = sizeof(*ip6h) + ctx->offset;
 658         if (nf_flow_state_check(flow, ip6h->nexthdr, skb, thoff))
 659                 return 0;
 660
 661         if (!nf_flow_dst_check(&tuplehash->tuple)) {
 662                 flow_offload_teardown(flow);
 663                 return 0;
 664         }
 665
 666         if (skb_try_make_writable(skb, thoff + ctx->hdrsize))
 667                 return -1;
 668
 669         flow_offload_refresh(flow_table, flow);
 670
 671         nf_flow_encap_pop(skb, tuplehash);
 672
 673         ip6h = ipv6_hdr(skb);
 674         nf_flow_nat_ipv6(flow, skb, dir, ip6h);
 675
 676         ip6h->hop_limit--;
 677         skb_clear_tstamp(skb);
 678
 679         if (flow_table->flags & NF_FLOWTABLE_COUNTER)
 680                 nf_ct_acct_update(flow->ct, tuplehash->tuple.dir, skb->len);
 681
 682         return 1;
 683 }
 684
 685 static struct flow_offload_tuple_rhash *
 686 nf_flow_offload_ipv6_lookup(struct nf_flowtable_ctx *ctx,
 687                             struct nf_flowtable *flow_table,
 688                             struct sk_buff *skb)
 689 {
 690         struct flow_offload_tuple tuple = {};
 691
 692         if (skb->protocol != htons(ETH_P_IPV6) &&
 693             !nf_flow_skb_encap_protocol(skb, htons(ETH_P_IPV6), &ctx->offset))
 694                 return NULL;
 695
 696         if (nf_flow_tuple_ipv6(ctx, skb, &tuple) < 0)
 697                 return NULL;
 698
 699         return flow_offload_lookup(flow_table, &tuple);
 700 }
 701
 702 unsigned int
 703 nf_flow_offload_ipv6_hook(void *priv, struct sk_buff *skb,
 704                           const struct nf_hook_state *state)
 705 {
 706         struct flow_offload_tuple_rhash *tuplehash;
 707         struct nf_flowtable *flow_table = priv;
 708         enum flow_offload_tuple_dir dir;
 709         struct nf_flowtable_ctx ctx = {
 710                 .in     = state->in,
 711         };
 712         const struct in6_addr *nexthop;
 713         struct flow_offload *flow;
 714         struct net_device *outdev;
 715         struct rt6_info *rt;
 716         int ret;
 717
 718         tuplehash = nf_flow_offload_ipv6_lookup(&ctx, flow_table, skb);
 719         if (tuplehash == NULL)
 720                 return NF_ACCEPT;
 721
 722         ret = nf_flow_offload_ipv6_forward(&ctx, flow_table, tuplehash, skb);
 723         if (ret < 0)
 724                 return NF_DROP;
 725         else if (ret == 0)
 726                 return NF_ACCEPT;
 727
 728         if (unlikely(tuplehash->tuple.xmit_type == FLOW_OFFLOAD_XMIT_XFRM)) {
 729                 rt = (struct rt6_info *)tuplehash->tuple.dst_cache;
 730                 memset(skb->cb, 0, sizeof(struct inet6_skb_parm));
 731                 IP6CB(skb)->iif = skb->dev->ifindex;
 732                 IP6CB(skb)->flags = IP6SKB_FORWARDED;
 733                 return nf_flow_xmit_xfrm(skb, state, &rt->dst);
 734         }
 735
 736         dir = tuplehash->tuple.dir;
 737         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 738
 739         switch (tuplehash->tuple.xmit_type) {
 740         case FLOW_OFFLOAD_XMIT_NEIGH:
 741                 rt = (struct rt6_info *)tuplehash->tuple.dst_cache;
 742                 outdev = rt->dst.dev;
 743                 skb->dev = outdev;
 744                 nexthop = rt6_nexthop(rt, &flow->tuplehash[!dir].tuple.src_v6);
 745                 skb_dst_set_noref(skb, &rt->dst);
 746                 neigh_xmit(NEIGH_ND_TABLE, outdev, nexthop, skb);
 747                 ret = NF_STOLEN;
 748                 break;
 749         case FLOW_OFFLOAD_XMIT_DIRECT:
 750                 ret = nf_flow_queue_xmit(state->net, skb, tuplehash, ETH_P_IPV6);
 751                 if (ret == NF_DROP)
 752                         flow_offload_teardown(flow);
 753                 break;
 754         default:
 755                 WARN_ON_ONCE(1);
 756                 ret = NF_DROP;
 757                 break;
 758         }
 759
 760         return ret;
 761 }
 762 EXPORT_SYMBOL_GPL(nf_flow_offload_ipv6_hook);