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