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 unsigned int
 211 nf_flow_offload_ip_hook(void *priv, struct sk_buff *skb,
 212                         const struct nf_hook_state *state)
 213 {
 214         struct flow_offload_tuple_rhash *tuplehash;
 215         struct nf_flowtable *flow_table = priv;
 216         struct flow_offload_tuple tuple = {};
 217         enum flow_offload_tuple_dir dir;
 218         struct flow_offload *flow;
 219         struct net_device *outdev;
 220         struct rtable *rt;
 221         unsigned int thoff;
 222         struct iphdr *iph;
 223         __be32 nexthop;
 224         u32 hdrsize;
 225
 226         if (skb->protocol != htons(ETH_P_IP))
 227                 return NF_ACCEPT;
 228
 229         if (nf_flow_tuple_ip(skb, state->in, &tuple, &hdrsize) < 0)
 230                 return NF_ACCEPT;
 231
 232         tuplehash = flow_offload_lookup(flow_table, &tuple);
 233         if (tuplehash == NULL)
 234                 return NF_ACCEPT;
 235
 236         dir = tuplehash->tuple.dir;
 237         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 238         rt = (struct rtable *)flow->tuplehash[dir].tuple.dst_cache;
 239         outdev = rt->dst.dev;
 240
 241         if (unlikely(nf_flow_exceeds_mtu(skb, flow->tuplehash[dir].tuple.mtu)))
 242                 return NF_ACCEPT;
 243
 244         iph = ip_hdr(skb);
 245         thoff = iph->ihl * 4;
 246         if (nf_flow_state_check(flow, iph->protocol, skb, thoff))
 247                 return NF_ACCEPT;
 248
 249         if (!dst_check(&rt->dst, 0)) {
 250                 flow_offload_teardown(flow);
 251                 return NF_ACCEPT;
 252         }
 253
 254         if (skb_try_make_writable(skb, thoff + hdrsize))
 255                 return NF_DROP;
 256
 257         flow_offload_refresh(flow_table, flow);
 258
 259         iph = ip_hdr(skb);
 260         nf_flow_nat_ip(flow, skb, thoff, dir, iph);
 261
 262         ip_decrease_ttl(iph);
 263         skb->tstamp = 0;
 264
 265         if (flow_table->flags & NF_FLOWTABLE_COUNTER)
 266                 nf_ct_acct_update(flow->ct, tuplehash->tuple.dir, skb->len);
 267
 268         if (unlikely(dst_xfrm(&rt->dst))) {
 269                 memset(skb->cb, 0, sizeof(struct inet_skb_parm));
 270                 IPCB(skb)->iif = skb->dev->ifindex;
 271                 IPCB(skb)->flags = IPSKB_FORWARDED;
 272                 return nf_flow_xmit_xfrm(skb, state, &rt->dst);
 273         }
 274
 275         skb->dev = outdev;
 276         nexthop = rt_nexthop(rt, flow->tuplehash[!dir].tuple.src_v4.s_addr);
 277         skb_dst_set_noref(skb, &rt->dst);
 278         neigh_xmit(NEIGH_ARP_TABLE, outdev, &nexthop, skb);
 279
 280         return NF_STOLEN;
 281 }
 282 EXPORT_SYMBOL_GPL(nf_flow_offload_ip_hook);
 283
 284 static void nf_flow_nat_ipv6_tcp(struct sk_buff *skb, unsigned int thoff,
 285                                  struct in6_addr *addr,
 286                                  struct in6_addr *new_addr,
 287                                  struct ipv6hdr *ip6h)
 288 {
 289         struct tcphdr *tcph;
 290
 291         tcph = (void *)(skb_network_header(skb) + thoff);
 292         inet_proto_csum_replace16(&tcph->check, skb, addr->s6_addr32,
 293                                   new_addr->s6_addr32, true);
 294 }
 295
 296 static void nf_flow_nat_ipv6_udp(struct sk_buff *skb, unsigned int thoff,
 297                                  struct in6_addr *addr,
 298                                  struct in6_addr *new_addr)
 299 {
 300         struct udphdr *udph;
 301
 302         udph = (void *)(skb_network_header(skb) + thoff);
 303         if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 304                 inet_proto_csum_replace16(&udph->check, skb, addr->s6_addr32,
 305                                           new_addr->s6_addr32, true);
 306                 if (!udph->check)
 307                         udph->check = CSUM_MANGLED_0;
 308         }
 309 }
 310
 311 static void nf_flow_nat_ipv6_l4proto(struct sk_buff *skb, struct ipv6hdr *ip6h,
 312                                      unsigned int thoff, struct in6_addr *addr,
 313                                      struct in6_addr *new_addr)
 314 {
 315         switch (ip6h->nexthdr) {
 316         case IPPROTO_TCP:
 317                 nf_flow_nat_ipv6_tcp(skb, thoff, addr, new_addr, ip6h);
 318                 break;
 319         case IPPROTO_UDP:
 320                 nf_flow_nat_ipv6_udp(skb, thoff, addr, new_addr);
 321                 break;
 322         }
 323 }
 324
 325 static void nf_flow_snat_ipv6(const struct flow_offload *flow,
 326                               struct sk_buff *skb, struct ipv6hdr *ip6h,
 327                               unsigned int thoff,
 328                               enum flow_offload_tuple_dir dir)
 329 {
 330         struct in6_addr addr, new_addr;
 331
 332         switch (dir) {
 333         case FLOW_OFFLOAD_DIR_ORIGINAL:
 334                 addr = ip6h->saddr;
 335                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_v6;
 336                 ip6h->saddr = new_addr;
 337                 break;
 338         case FLOW_OFFLOAD_DIR_REPLY:
 339                 addr = ip6h->daddr;
 340                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_v6;
 341                 ip6h->daddr = new_addr;
 342                 break;
 343         }
 344
 345         nf_flow_nat_ipv6_l4proto(skb, ip6h, thoff, &addr, &new_addr);
 346 }
 347
 348 static void nf_flow_dnat_ipv6(const struct flow_offload *flow,
 349                               struct sk_buff *skb, struct ipv6hdr *ip6h,
 350                               unsigned int thoff,
 351                               enum flow_offload_tuple_dir dir)
 352 {
 353         struct in6_addr addr, new_addr;
 354
 355         switch (dir) {
 356         case FLOW_OFFLOAD_DIR_ORIGINAL:
 357                 addr = ip6h->daddr;
 358                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_v6;
 359                 ip6h->daddr = new_addr;
 360                 break;
 361         case FLOW_OFFLOAD_DIR_REPLY:
 362                 addr = ip6h->saddr;
 363                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_v6;
 364                 ip6h->saddr = new_addr;
 365                 break;
 366         }
 367
 368         nf_flow_nat_ipv6_l4proto(skb, ip6h, thoff, &addr, &new_addr);
 369 }
 370
 371 static void nf_flow_nat_ipv6(const struct flow_offload *flow,
 372                              struct sk_buff *skb,
 373                              enum flow_offload_tuple_dir dir,
 374                              struct ipv6hdr *ip6h)
 375 {
 376         unsigned int thoff = sizeof(*ip6h);
 377
 378         if (test_bit(NF_FLOW_SNAT, &flow->flags)) {
 379                 nf_flow_snat_port(flow, skb, thoff, ip6h->nexthdr, dir);
 380                 nf_flow_snat_ipv6(flow, skb, ip6h, thoff, dir);
 381         }
 382         if (test_bit(NF_FLOW_DNAT, &flow->flags)) {
 383                 nf_flow_dnat_port(flow, skb, thoff, ip6h->nexthdr, dir);
 384                 nf_flow_dnat_ipv6(flow, skb, ip6h, thoff, dir);
 385         }
 386 }
 387
 388 static int nf_flow_tuple_ipv6(struct sk_buff *skb, const struct net_device *dev,
 389                               struct flow_offload_tuple *tuple, u32 *hdrsize)
 390 {
 391         struct flow_ports *ports;
 392         struct ipv6hdr *ip6h;
 393         unsigned int thoff;
 394
 395         if (!pskb_may_pull(skb, sizeof(*ip6h)))
 396                 return -1;
 397
 398         ip6h = ipv6_hdr(skb);
 399
 400         switch (ip6h->nexthdr) {
 401         case IPPROTO_TCP:
 402                 *hdrsize = sizeof(struct tcphdr);
 403                 break;
 404         case IPPROTO_UDP:
 405                 *hdrsize = sizeof(struct udphdr);
 406                 break;
 407         default:
 408                 return -1;
 409         }
 410
 411         if (ip6h->hop_limit <= 1)
 412                 return -1;
 413
 414         thoff = sizeof(*ip6h);
 415         if (!pskb_may_pull(skb, thoff + *hdrsize))
 416                 return -1;
 417
 418         ip6h = ipv6_hdr(skb);
 419         ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
 420
 421         tuple->src_v6           = ip6h->saddr;
 422         tuple->dst_v6           = ip6h->daddr;
 423         tuple->src_port         = ports->source;
 424         tuple->dst_port         = ports->dest;
 425         tuple->l3proto          = AF_INET6;
 426         tuple->l4proto          = ip6h->nexthdr;
 427         tuple->iifidx           = dev->ifindex;
 428
 429         return 0;
 430 }
 431
 432 unsigned int
 433 nf_flow_offload_ipv6_hook(void *priv, struct sk_buff *skb,
 434                           const struct nf_hook_state *state)
 435 {
 436         struct flow_offload_tuple_rhash *tuplehash;
 437         struct nf_flowtable *flow_table = priv;
 438         struct flow_offload_tuple tuple = {};
 439         enum flow_offload_tuple_dir dir;
 440         const struct in6_addr *nexthop;
 441         struct flow_offload *flow;
 442         struct net_device *outdev;
 443         struct ipv6hdr *ip6h;
 444         struct rt6_info *rt;
 445         u32 hdrsize;
 446
 447         if (skb->protocol != htons(ETH_P_IPV6))
 448                 return NF_ACCEPT;
 449
 450         if (nf_flow_tuple_ipv6(skb, state->in, &tuple, &hdrsize) < 0)
 451                 return NF_ACCEPT;
 452
 453         tuplehash = flow_offload_lookup(flow_table, &tuple);
 454         if (tuplehash == NULL)
 455                 return NF_ACCEPT;
 456
 457         dir = tuplehash->tuple.dir;
 458         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 459         rt = (struct rt6_info *)flow->tuplehash[dir].tuple.dst_cache;
 460         outdev = rt->dst.dev;
 461
 462         if (unlikely(nf_flow_exceeds_mtu(skb, flow->tuplehash[dir].tuple.mtu)))
 463                 return NF_ACCEPT;
 464
 465         if (nf_flow_state_check(flow, ipv6_hdr(skb)->nexthdr, skb,
 466                                 sizeof(*ip6h)))
 467                 return NF_ACCEPT;
 468
 469         if (!dst_check(&rt->dst, 0)) {
 470                 flow_offload_teardown(flow);
 471                 return NF_ACCEPT;
 472         }
 473
 474         if (skb_try_make_writable(skb, sizeof(*ip6h) + hdrsize))
 475                 return NF_DROP;
 476
 477         flow_offload_refresh(flow_table, flow);
 478
 479         ip6h = ipv6_hdr(skb);
 480         nf_flow_nat_ipv6(flow, skb, dir, ip6h);
 481
 482         ip6h->hop_limit--;
 483         skb->tstamp = 0;
 484
 485         if (flow_table->flags & NF_FLOWTABLE_COUNTER)
 486                 nf_ct_acct_update(flow->ct, tuplehash->tuple.dir, skb->len);
 487
 488         if (unlikely(dst_xfrm(&rt->dst))) {
 489                 memset(skb->cb, 0, sizeof(struct inet6_skb_parm));
 490                 IP6CB(skb)->iif = skb->dev->ifindex;
 491                 IP6CB(skb)->flags = IP6SKB_FORWARDED;
 492                 return nf_flow_xmit_xfrm(skb, state, &rt->dst);
 493         }
 494
 495         skb->dev = outdev;
 496         nexthop = rt6_nexthop(rt, &flow->tuplehash[!dir].tuple.src_v6);
 497         skb_dst_set_noref(skb, &rt->dst);
 498         neigh_xmit(NEIGH_ND_TABLE, outdev, nexthop, skb);
 499
 500         return NF_STOLEN;
 501 }
 502 EXPORT_SYMBOL_GPL(nf_flow_offload_ipv6_hook);