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         thoff = iph->ihl * 4;
 185         if (!pskb_may_pull(skb, thoff + sizeof(*ports)))
 186                 return -1;
 187
 188         ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
 189
 190         tuple->src_v4.s_addr    = iph->saddr;
 191         tuple->dst_v4.s_addr    = iph->daddr;
 192         tuple->src_port         = ports->source;
 193         tuple->dst_port         = ports->dest;
 194         tuple->l3proto          = AF_INET;
 195         tuple->l4proto          = iph->protocol;
 196         tuple->iifidx           = dev->ifindex;
 197
 198         return 0;
 199 }
 200
 201 /* Based on ip_exceeds_mtu(). */
 202 static bool nf_flow_exceeds_mtu(const struct sk_buff *skb, unsigned int mtu)
 203 {
 204         if (skb->len <= mtu)
 205                 return false;
 206
 207         if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
 208                 return false;
 209
 210         return true;
 211 }
 212
 213 unsigned int
 214 nf_flow_offload_ip_hook(void *priv, struct sk_buff *skb,
 215                         const struct nf_hook_state *state)
 216 {
 217         struct flow_offload_tuple_rhash *tuplehash;
 218         struct nf_flowtable *flow_table = priv;
 219         struct flow_offload_tuple tuple = {};
 220         enum flow_offload_tuple_dir dir;
 221         struct flow_offload *flow;
 222         struct net_device *outdev;
 223         struct rtable *rt;
 224         unsigned int thoff;
 225         struct iphdr *iph;
 226         __be32 nexthop;
 227
 228         if (skb->protocol != htons(ETH_P_IP))
 229                 return NF_ACCEPT;
 230
 231         if (nf_flow_tuple_ip(skb, state->in, &tuple) < 0)
 232                 return NF_ACCEPT;
 233
 234         tuplehash = flow_offload_lookup(flow_table, &tuple);
 235         if (tuplehash == NULL)
 236                 return NF_ACCEPT;
 237
 238         dir = tuplehash->tuple.dir;
 239         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 240         rt = (struct rtable *)flow->tuplehash[dir].tuple.dst_cache;
 241         outdev = rt->dst.dev;
 242
 243         if (unlikely(nf_flow_exceeds_mtu(skb, flow->tuplehash[dir].tuple.mtu)) &&
 244             (ip_hdr(skb)->frag_off & htons(IP_DF)) != 0)
 245                 return NF_ACCEPT;
 246
 247         if (skb_try_make_writable(skb, sizeof(*iph)))
 248                 return NF_DROP;
 249
 250         thoff = ip_hdr(skb)->ihl * 4;
 251         if (nf_flow_state_check(flow, ip_hdr(skb)->protocol, skb, thoff))
 252                 return NF_ACCEPT;
 253
 254         if (nf_flow_nat_ip(flow, skb, thoff, dir) < 0)
 255                 return NF_DROP;
 256
 257         flow->timeout = (u32)jiffies + NF_FLOW_TIMEOUT;
 258         iph = ip_hdr(skb);
 259         ip_decrease_ttl(iph);
 260
 261         skb->dev = outdev;
 262         nexthop = rt_nexthop(rt, flow->tuplehash[!dir].tuple.src_v4.s_addr);
 263         skb_dst_set_noref(skb, &rt->dst);
 264         neigh_xmit(NEIGH_ARP_TABLE, outdev, &nexthop, skb);
 265
 266         return NF_STOLEN;
 267 }
 268 EXPORT_SYMBOL_GPL(nf_flow_offload_ip_hook);
 269
 270 static int nf_flow_nat_ipv6_tcp(struct sk_buff *skb, unsigned int thoff,
 271                                 struct in6_addr *addr,
 272                                 struct in6_addr *new_addr)
 273 {
 274         struct tcphdr *tcph;
 275
 276         if (!pskb_may_pull(skb, thoff + sizeof(*tcph)) ||
 277             skb_try_make_writable(skb, thoff + sizeof(*tcph)))
 278                 return -1;
 279
 280         tcph = (void *)(skb_network_header(skb) + thoff);
 281         inet_proto_csum_replace16(&tcph->check, skb, addr->s6_addr32,
 282                                   new_addr->s6_addr32, true);
 283
 284         return 0;
 285 }
 286
 287 static int nf_flow_nat_ipv6_udp(struct sk_buff *skb, unsigned int thoff,
 288                                 struct in6_addr *addr,
 289                                 struct in6_addr *new_addr)
 290 {
 291         struct udphdr *udph;
 292
 293         if (!pskb_may_pull(skb, thoff + sizeof(*udph)) ||
 294             skb_try_make_writable(skb, thoff + sizeof(*udph)))
 295                 return -1;
 296
 297         udph = (void *)(skb_network_header(skb) + thoff);
 298         if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 299                 inet_proto_csum_replace16(&udph->check, skb, addr->s6_addr32,
 300                                           new_addr->s6_addr32, true);
 301                 if (!udph->check)
 302                         udph->check = CSUM_MANGLED_0;
 303         }
 304
 305         return 0;
 306 }
 307
 308 static int nf_flow_nat_ipv6_l4proto(struct sk_buff *skb, struct ipv6hdr *ip6h,
 309                                     unsigned int thoff, struct in6_addr *addr,
 310                                     struct in6_addr *new_addr)
 311 {
 312         switch (ip6h->nexthdr) {
 313         case IPPROTO_TCP:
 314                 if (nf_flow_nat_ipv6_tcp(skb, thoff, addr, new_addr) < 0)
 315                         return NF_DROP;
 316                 break;
 317         case IPPROTO_UDP:
 318                 if (nf_flow_nat_ipv6_udp(skb, thoff, addr, new_addr) < 0)
 319                         return NF_DROP;
 320                 break;
 321         }
 322
 323         return 0;
 324 }
 325
 326 static int nf_flow_snat_ipv6(const struct flow_offload *flow,
 327                              struct sk_buff *skb, struct ipv6hdr *ip6h,
 328                              unsigned int thoff,
 329                              enum flow_offload_tuple_dir dir)
 330 {
 331         struct in6_addr addr, new_addr;
 332
 333         switch (dir) {
 334         case FLOW_OFFLOAD_DIR_ORIGINAL:
 335                 addr = ip6h->saddr;
 336                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_v6;
 337                 ip6h->saddr = new_addr;
 338                 break;
 339         case FLOW_OFFLOAD_DIR_REPLY:
 340                 addr = ip6h->daddr;
 341                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_v6;
 342                 ip6h->daddr = new_addr;
 343                 break;
 344         default:
 345                 return -1;
 346         }
 347
 348         return nf_flow_nat_ipv6_l4proto(skb, ip6h, thoff, &addr, &new_addr);
 349 }
 350
 351 static int nf_flow_dnat_ipv6(const struct flow_offload *flow,
 352                              struct sk_buff *skb, struct ipv6hdr *ip6h,
 353                              unsigned int thoff,
 354                              enum flow_offload_tuple_dir dir)
 355 {
 356         struct in6_addr addr, new_addr;
 357
 358         switch (dir) {
 359         case FLOW_OFFLOAD_DIR_ORIGINAL:
 360                 addr = ip6h->daddr;
 361                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_v6;
 362                 ip6h->daddr = new_addr;
 363                 break;
 364         case FLOW_OFFLOAD_DIR_REPLY:
 365                 addr = ip6h->saddr;
 366                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_v6;
 367                 ip6h->saddr = new_addr;
 368                 break;
 369         default:
 370                 return -1;
 371         }
 372
 373         return nf_flow_nat_ipv6_l4proto(skb, ip6h, thoff, &addr, &new_addr);
 374 }
 375
 376 static int nf_flow_nat_ipv6(const struct flow_offload *flow,
 377                             struct sk_buff *skb,
 378                             enum flow_offload_tuple_dir dir)
 379 {
 380         struct ipv6hdr *ip6h = ipv6_hdr(skb);
 381         unsigned int thoff = sizeof(*ip6h);
 382
 383         if (flow->flags & FLOW_OFFLOAD_SNAT &&
 384             (nf_flow_snat_port(flow, skb, thoff, ip6h->nexthdr, dir) < 0 ||
 385              nf_flow_snat_ipv6(flow, skb, ip6h, thoff, dir) < 0))
 386                 return -1;
 387         if (flow->flags & FLOW_OFFLOAD_DNAT &&
 388             (nf_flow_dnat_port(flow, skb, thoff, ip6h->nexthdr, dir) < 0 ||
 389              nf_flow_dnat_ipv6(flow, skb, ip6h, thoff, dir) < 0))
 390                 return -1;
 391
 392         return 0;
 393 }
 394
 395 static int nf_flow_tuple_ipv6(struct sk_buff *skb, const struct net_device *dev,
 396                               struct flow_offload_tuple *tuple)
 397 {
 398         struct flow_ports *ports;
 399         struct ipv6hdr *ip6h;
 400         unsigned int thoff;
 401
 402         if (!pskb_may_pull(skb, sizeof(*ip6h)))
 403                 return -1;
 404
 405         ip6h = ipv6_hdr(skb);
 406
 407         if (ip6h->nexthdr != IPPROTO_TCP &&
 408             ip6h->nexthdr != IPPROTO_UDP)
 409                 return -1;
 410
 411         thoff = sizeof(*ip6h);
 412         if (!pskb_may_pull(skb, thoff + sizeof(*ports)))
 413                 return -1;
 414
 415         ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
 416
 417         tuple->src_v6           = ip6h->saddr;
 418         tuple->dst_v6           = ip6h->daddr;
 419         tuple->src_port         = ports->source;
 420         tuple->dst_port         = ports->dest;
 421         tuple->l3proto          = AF_INET6;
 422         tuple->l4proto          = ip6h->nexthdr;
 423         tuple->iifidx           = dev->ifindex;
 424
 425         return 0;
 426 }
 427
 428 unsigned int
 429 nf_flow_offload_ipv6_hook(void *priv, struct sk_buff *skb,
 430                           const struct nf_hook_state *state)
 431 {
 432         struct flow_offload_tuple_rhash *tuplehash;
 433         struct nf_flowtable *flow_table = priv;
 434         struct flow_offload_tuple tuple = {};
 435         enum flow_offload_tuple_dir dir;
 436         struct flow_offload *flow;
 437         struct net_device *outdev;
 438         struct in6_addr *nexthop;
 439         struct ipv6hdr *ip6h;
 440         struct rt6_info *rt;
 441
 442         if (skb->protocol != htons(ETH_P_IPV6))
 443                 return NF_ACCEPT;
 444
 445         if (nf_flow_tuple_ipv6(skb, state->in, &tuple) < 0)
 446                 return NF_ACCEPT;
 447
 448         tuplehash = flow_offload_lookup(flow_table, &tuple);
 449         if (tuplehash == NULL)
 450                 return NF_ACCEPT;
 451
 452         dir = tuplehash->tuple.dir;
 453         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 454         rt = (struct rt6_info *)flow->tuplehash[dir].tuple.dst_cache;
 455         outdev = rt->dst.dev;
 456
 457         if (unlikely(nf_flow_exceeds_mtu(skb, flow->tuplehash[dir].tuple.mtu)))
 458                 return NF_ACCEPT;
 459
 460         if (nf_flow_state_check(flow, ipv6_hdr(skb)->nexthdr, skb,
 461                                 sizeof(*ip6h)))
 462                 return NF_ACCEPT;
 463
 464         if (skb_try_make_writable(skb, sizeof(*ip6h)))
 465                 return NF_DROP;
 466
 467         if (nf_flow_nat_ipv6(flow, skb, dir) < 0)
 468                 return NF_DROP;
 469
 470         flow->timeout = (u32)jiffies + NF_FLOW_TIMEOUT;
 471         ip6h = ipv6_hdr(skb);
 472         ip6h->hop_limit--;
 473
 474         skb->dev = outdev;
 475         nexthop = rt6_nexthop(rt, &flow->tuplehash[!dir].tuple.src_v6);
 476         skb_dst_set_noref(skb, &rt->dst);
 477         neigh_xmit(NEIGH_ND_TABLE, outdev, nexthop, skb);
 478
 479         return NF_STOLEN;
 480 }
 481 EXPORT_SYMBOL_GPL(nf_flow_offload_ipv6_hook);