net/netfilter/nf_flow_table_ip.c

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