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 int 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         if (skb_try_make_writable(skb, thoff + sizeof(*tcph)))
  43                 return -1;
  44
  45         tcph = (void *)(skb_network_header(skb) + thoff);
  46         inet_proto_csum_replace4(&tcph->check, skb, addr, new_addr, true);
  47
  48         return 0;
  49 }
  50
  51 static int nf_flow_nat_ip_udp(struct sk_buff *skb, unsigned int thoff,
  52                               __be32 addr, __be32 new_addr)
  53 {
  54         struct udphdr *udph;
  55
  56         if (skb_try_make_writable(skb, thoff + sizeof(*udph)))
  57                 return -1;
  58
  59         udph = (void *)(skb_network_header(skb) + thoff);
  60         if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
  61                 inet_proto_csum_replace4(&udph->check, skb, addr,
  62                                          new_addr, true);
  63                 if (!udph->check)
  64                         udph->check = CSUM_MANGLED_0;
  65         }
  66
  67         return 0;
  68 }
  69
  70 static int nf_flow_nat_ip_l4proto(struct sk_buff *skb, struct iphdr *iph,
  71                                   unsigned int thoff, __be32 addr,
  72                                   __be32 new_addr)
  73 {
  74         switch (iph->protocol) {
  75         case IPPROTO_TCP:
  76                 if (nf_flow_nat_ip_tcp(skb, thoff, addr, new_addr) < 0)
  77                         return NF_DROP;
  78                 break;
  79         case IPPROTO_UDP:
  80                 if (nf_flow_nat_ip_udp(skb, thoff, addr, new_addr) < 0)
  81                         return NF_DROP;
  82                 break;
  83         }
  84
  85         return 0;
  86 }
  87
  88 static int nf_flow_snat_ip(const struct flow_offload *flow, struct sk_buff *skb,
  89                            struct iphdr *iph, unsigned int thoff,
  90                            enum flow_offload_tuple_dir dir)
  91 {
  92         __be32 addr, new_addr;
  93
  94         switch (dir) {
  95         case FLOW_OFFLOAD_DIR_ORIGINAL:
  96                 addr = iph->saddr;
  97                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_v4.s_addr;
  98                 iph->saddr = new_addr;
  99                 break;
 100         case FLOW_OFFLOAD_DIR_REPLY:
 101                 addr = iph->daddr;
 102                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_v4.s_addr;
 103                 iph->daddr = new_addr;
 104                 break;
 105         default:
 106                 return -1;
 107         }
 108         csum_replace4(&iph->check, addr, new_addr);
 109
 110         return nf_flow_nat_ip_l4proto(skb, iph, thoff, addr, new_addr);
 111 }
 112
 113 static int nf_flow_dnat_ip(const struct flow_offload *flow, struct sk_buff *skb,
 114                            struct iphdr *iph, unsigned int thoff,
 115                            enum flow_offload_tuple_dir dir)
 116 {
 117         __be32 addr, new_addr;
 118
 119         switch (dir) {
 120         case FLOW_OFFLOAD_DIR_ORIGINAL:
 121                 addr = iph->daddr;
 122                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_v4.s_addr;
 123                 iph->daddr = new_addr;
 124                 break;
 125         case FLOW_OFFLOAD_DIR_REPLY:
 126                 addr = iph->saddr;
 127                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_v4.s_addr;
 128                 iph->saddr = new_addr;
 129                 break;
 130         default:
 131                 return -1;
 132         }
 133         csum_replace4(&iph->check, addr, new_addr);
 134
 135         return nf_flow_nat_ip_l4proto(skb, iph, thoff, addr, new_addr);
 136 }
 137
 138 static int nf_flow_nat_ip(const struct flow_offload *flow, struct sk_buff *skb,
 139                           unsigned int thoff, enum flow_offload_tuple_dir dir)
 140 {
 141         struct iphdr *iph = ip_hdr(skb);
 142
 143         if (test_bit(NF_FLOW_SNAT, &flow->flags) &&
 144             (nf_flow_snat_port(flow, skb, thoff, iph->protocol, dir) < 0 ||
 145              nf_flow_snat_ip(flow, skb, ip_hdr(skb), thoff, dir) < 0))
 146                 return -1;
 147
 148         iph = ip_hdr(skb);
 149         if (test_bit(NF_FLOW_DNAT, &flow->flags) &&
 150             (nf_flow_dnat_port(flow, skb, thoff, iph->protocol, dir) < 0 ||
 151              nf_flow_dnat_ip(flow, skb, ip_hdr(skb), thoff, dir) < 0))
 152                 return -1;
 153
 154         return 0;
 155 }
 156
 157 static bool ip_has_options(unsigned int thoff)
 158 {
 159         return thoff != sizeof(struct iphdr);
 160 }
 161
 162 static int nf_flow_tuple_ip(struct sk_buff *skb, const struct net_device *dev,
 163                             struct flow_offload_tuple *tuple)
 164 {
 165         unsigned int thoff, hdrsize;
 166         struct flow_ports *ports;
 167         struct iphdr *iph;
 168
 169         if (!pskb_may_pull(skb, sizeof(*iph)))
 170                 return -1;
 171
 172         iph = ip_hdr(skb);
 173         thoff = iph->ihl * 4;
 174
 175         if (ip_is_fragment(iph) ||
 176             unlikely(ip_has_options(thoff)))
 177                 return -1;
 178
 179         switch (iph->protocol) {
 180         case IPPROTO_TCP:
 181                 hdrsize = sizeof(struct tcphdr);
 182                 break;
 183         case IPPROTO_UDP:
 184                 hdrsize = sizeof(struct udphdr);
 185                 break;
 186         default:
 187                 return -1;
 188         }
 189
 190         if (iph->ttl <= 1)
 191                 return -1;
 192
 193         thoff = iph->ihl * 4;
 194         if (!pskb_may_pull(skb, thoff + hdrsize))
 195                 return -1;
 196
 197         iph = ip_hdr(skb);
 198         ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
 199
 200         tuple->src_v4.s_addr    = iph->saddr;
 201         tuple->dst_v4.s_addr    = iph->daddr;
 202         tuple->src_port         = ports->source;
 203         tuple->dst_port         = ports->dest;
 204         tuple->l3proto          = AF_INET;
 205         tuple->l4proto          = iph->protocol;
 206         tuple->iifidx           = dev->ifindex;
 207
 208         return 0;
 209 }
 210
 211 /* Based on ip_exceeds_mtu(). */
 212 static bool nf_flow_exceeds_mtu(const struct sk_buff *skb, unsigned int mtu)
 213 {
 214         if (skb->len <= mtu)
 215                 return false;
 216
 217         if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
 218                 return false;
 219
 220         return true;
 221 }
 222
 223 static int nf_flow_offload_dst_check(struct dst_entry *dst)
 224 {
 225         if (unlikely(dst_xfrm(dst)))
 226                 return dst_check(dst, 0) ? 0 : -1;
 227
 228         return 0;
 229 }
 230
 231 static unsigned int nf_flow_xmit_xfrm(struct sk_buff *skb,
 232                                       const struct nf_hook_state *state,
 233                                       struct dst_entry *dst)
 234 {
 235         skb_orphan(skb);
 236         skb_dst_set_noref(skb, dst);
 237         dst_output(state->net, state->sk, skb);
 238         return NF_STOLEN;
 239 }
 240
 241 unsigned int
 242 nf_flow_offload_ip_hook(void *priv, struct sk_buff *skb,
 243                         const struct nf_hook_state *state)
 244 {
 245         struct flow_offload_tuple_rhash *tuplehash;
 246         struct nf_flowtable *flow_table = priv;
 247         struct flow_offload_tuple tuple = {};
 248         enum flow_offload_tuple_dir dir;
 249         struct flow_offload *flow;
 250         struct net_device *outdev;
 251         struct rtable *rt;
 252         unsigned int thoff;
 253         struct iphdr *iph;
 254         __be32 nexthop;
 255
 256         if (skb->protocol != htons(ETH_P_IP))
 257                 return NF_ACCEPT;
 258
 259         if (nf_flow_tuple_ip(skb, state->in, &tuple) < 0)
 260                 return NF_ACCEPT;
 261
 262         tuplehash = flow_offload_lookup(flow_table, &tuple);
 263         if (tuplehash == NULL)
 264                 return NF_ACCEPT;
 265
 266         dir = tuplehash->tuple.dir;
 267         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 268         rt = (struct rtable *)flow->tuplehash[dir].tuple.dst_cache;
 269         outdev = rt->dst.dev;
 270
 271         if (unlikely(nf_flow_exceeds_mtu(skb, flow->tuplehash[dir].tuple.mtu)))
 272                 return NF_ACCEPT;
 273
 274         if (skb_try_make_writable(skb, sizeof(*iph)))
 275                 return NF_DROP;
 276
 277         thoff = ip_hdr(skb)->ihl * 4;
 278         if (nf_flow_state_check(flow, ip_hdr(skb)->protocol, skb, thoff))
 279                 return NF_ACCEPT;
 280
 281         flow_offload_refresh(flow_table, flow);
 282
 283         if (nf_flow_offload_dst_check(&rt->dst)) {
 284                 flow_offload_teardown(flow);
 285                 return NF_ACCEPT;
 286         }
 287
 288         if (nf_flow_nat_ip(flow, skb, thoff, dir) < 0)
 289                 return NF_DROP;
 290
 291         iph = ip_hdr(skb);
 292         ip_decrease_ttl(iph);
 293         skb->tstamp = 0;
 294
 295         if (flow_table->flags & NF_FLOWTABLE_COUNTER)
 296                 nf_ct_acct_update(flow->ct, tuplehash->tuple.dir, skb->len);
 297
 298         if (unlikely(dst_xfrm(&rt->dst))) {
 299                 memset(skb->cb, 0, sizeof(struct inet_skb_parm));
 300                 IPCB(skb)->iif = skb->dev->ifindex;
 301                 IPCB(skb)->flags = IPSKB_FORWARDED;
 302                 return nf_flow_xmit_xfrm(skb, state, &rt->dst);
 303         }
 304
 305         skb->dev = outdev;
 306         nexthop = rt_nexthop(rt, flow->tuplehash[!dir].tuple.src_v4.s_addr);
 307         skb_dst_set_noref(skb, &rt->dst);
 308         neigh_xmit(NEIGH_ARP_TABLE, outdev, &nexthop, skb);
 309
 310         return NF_STOLEN;
 311 }
 312 EXPORT_SYMBOL_GPL(nf_flow_offload_ip_hook);
 313
 314 static int nf_flow_nat_ipv6_tcp(struct sk_buff *skb, unsigned int thoff,
 315                                 struct in6_addr *addr,
 316                                 struct in6_addr *new_addr)
 317 {
 318         struct tcphdr *tcph;
 319
 320         if (skb_try_make_writable(skb, thoff + sizeof(*tcph)))
 321                 return -1;
 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         return 0;
 328 }
 329
 330 static int nf_flow_nat_ipv6_udp(struct sk_buff *skb, unsigned int thoff,
 331                                 struct in6_addr *addr,
 332                                 struct in6_addr *new_addr)
 333 {
 334         struct udphdr *udph;
 335
 336         if (skb_try_make_writable(skb, thoff + sizeof(*udph)))
 337                 return -1;
 338
 339         udph = (void *)(skb_network_header(skb) + thoff);
 340         if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 341                 inet_proto_csum_replace16(&udph->check, skb, addr->s6_addr32,
 342                                           new_addr->s6_addr32, true);
 343                 if (!udph->check)
 344                         udph->check = CSUM_MANGLED_0;
 345         }
 346
 347         return 0;
 348 }
 349
 350 static int nf_flow_nat_ipv6_l4proto(struct sk_buff *skb, struct ipv6hdr *ip6h,
 351                                     unsigned int thoff, struct in6_addr *addr,
 352                                     struct in6_addr *new_addr)
 353 {
 354         switch (ip6h->nexthdr) {
 355         case IPPROTO_TCP:
 356                 if (nf_flow_nat_ipv6_tcp(skb, thoff, addr, new_addr) < 0)
 357                         return NF_DROP;
 358                 break;
 359         case IPPROTO_UDP:
 360                 if (nf_flow_nat_ipv6_udp(skb, thoff, addr, new_addr) < 0)
 361                         return NF_DROP;
 362                 break;
 363         }
 364
 365         return 0;
 366 }
 367
 368 static int nf_flow_snat_ipv6(const struct flow_offload *flow,
 369                              struct sk_buff *skb, struct ipv6hdr *ip6h,
 370                              unsigned int thoff,
 371                              enum flow_offload_tuple_dir dir)
 372 {
 373         struct in6_addr addr, new_addr;
 374
 375         switch (dir) {
 376         case FLOW_OFFLOAD_DIR_ORIGINAL:
 377                 addr = ip6h->saddr;
 378                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_v6;
 379                 ip6h->saddr = new_addr;
 380                 break;
 381         case FLOW_OFFLOAD_DIR_REPLY:
 382                 addr = ip6h->daddr;
 383                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_v6;
 384                 ip6h->daddr = new_addr;
 385                 break;
 386         default:
 387                 return -1;
 388         }
 389
 390         return nf_flow_nat_ipv6_l4proto(skb, ip6h, thoff, &addr, &new_addr);
 391 }
 392
 393 static int nf_flow_dnat_ipv6(const struct flow_offload *flow,
 394                              struct sk_buff *skb, struct ipv6hdr *ip6h,
 395                              unsigned int thoff,
 396                              enum flow_offload_tuple_dir dir)
 397 {
 398         struct in6_addr addr, new_addr;
 399
 400         switch (dir) {
 401         case FLOW_OFFLOAD_DIR_ORIGINAL:
 402                 addr = ip6h->daddr;
 403                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_v6;
 404                 ip6h->daddr = new_addr;
 405                 break;
 406         case FLOW_OFFLOAD_DIR_REPLY:
 407                 addr = ip6h->saddr;
 408                 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_v6;
 409                 ip6h->saddr = new_addr;
 410                 break;
 411         default:
 412                 return -1;
 413         }
 414
 415         return nf_flow_nat_ipv6_l4proto(skb, ip6h, thoff, &addr, &new_addr);
 416 }
 417
 418 static int nf_flow_nat_ipv6(const struct flow_offload *flow,
 419                             struct sk_buff *skb,
 420                             enum flow_offload_tuple_dir dir)
 421 {
 422         struct ipv6hdr *ip6h = ipv6_hdr(skb);
 423         unsigned int thoff = sizeof(*ip6h);
 424
 425         if (test_bit(NF_FLOW_SNAT, &flow->flags) &&
 426             (nf_flow_snat_port(flow, skb, thoff, ip6h->nexthdr, dir) < 0 ||
 427              nf_flow_snat_ipv6(flow, skb, ipv6_hdr(skb), thoff, dir) < 0))
 428                 return -1;
 429
 430         ip6h = ipv6_hdr(skb);
 431         if (test_bit(NF_FLOW_DNAT, &flow->flags) &&
 432             (nf_flow_dnat_port(flow, skb, thoff, ip6h->nexthdr, dir) < 0 ||
 433              nf_flow_dnat_ipv6(flow, skb, ipv6_hdr(skb), thoff, dir) < 0))
 434                 return -1;
 435
 436         return 0;
 437 }
 438
 439 static int nf_flow_tuple_ipv6(struct sk_buff *skb, const struct net_device *dev,
 440                               struct flow_offload_tuple *tuple)
 441 {
 442         unsigned int thoff, hdrsize;
 443         struct flow_ports *ports;
 444         struct ipv6hdr *ip6h;
 445
 446         if (!pskb_may_pull(skb, sizeof(*ip6h)))
 447                 return -1;
 448
 449         ip6h = ipv6_hdr(skb);
 450
 451         switch (ip6h->nexthdr) {
 452         case IPPROTO_TCP:
 453                 hdrsize = sizeof(struct tcphdr);
 454                 break;
 455         case IPPROTO_UDP:
 456                 hdrsize = sizeof(struct udphdr);
 457                 break;
 458         default:
 459                 return -1;
 460         }
 461
 462         if (ip6h->hop_limit <= 1)
 463                 return -1;
 464
 465         thoff = sizeof(*ip6h);
 466         if (!pskb_may_pull(skb, thoff + hdrsize))
 467                 return -1;
 468
 469         ip6h = ipv6_hdr(skb);
 470         ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
 471
 472         tuple->src_v6           = ip6h->saddr;
 473         tuple->dst_v6           = ip6h->daddr;
 474         tuple->src_port         = ports->source;
 475         tuple->dst_port         = ports->dest;
 476         tuple->l3proto          = AF_INET6;
 477         tuple->l4proto          = ip6h->nexthdr;
 478         tuple->iifidx           = dev->ifindex;
 479
 480         return 0;
 481 }
 482
 483 unsigned int
 484 nf_flow_offload_ipv6_hook(void *priv, struct sk_buff *skb,
 485                           const struct nf_hook_state *state)
 486 {
 487         struct flow_offload_tuple_rhash *tuplehash;
 488         struct nf_flowtable *flow_table = priv;
 489         struct flow_offload_tuple tuple = {};
 490         enum flow_offload_tuple_dir dir;
 491         const struct in6_addr *nexthop;
 492         struct flow_offload *flow;
 493         struct net_device *outdev;
 494         struct ipv6hdr *ip6h;
 495         struct rt6_info *rt;
 496
 497         if (skb->protocol != htons(ETH_P_IPV6))
 498                 return NF_ACCEPT;
 499
 500         if (nf_flow_tuple_ipv6(skb, state->in, &tuple) < 0)
 501                 return NF_ACCEPT;
 502
 503         tuplehash = flow_offload_lookup(flow_table, &tuple);
 504         if (tuplehash == NULL)
 505                 return NF_ACCEPT;
 506
 507         dir = tuplehash->tuple.dir;
 508         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 509         rt = (struct rt6_info *)flow->tuplehash[dir].tuple.dst_cache;
 510         outdev = rt->dst.dev;
 511
 512         if (unlikely(nf_flow_exceeds_mtu(skb, flow->tuplehash[dir].tuple.mtu)))
 513                 return NF_ACCEPT;
 514
 515         if (nf_flow_state_check(flow, ipv6_hdr(skb)->nexthdr, skb,
 516                                 sizeof(*ip6h)))
 517                 return NF_ACCEPT;
 518
 519         flow_offload_refresh(flow_table, flow);
 520
 521         if (nf_flow_offload_dst_check(&rt->dst)) {
 522                 flow_offload_teardown(flow);
 523                 return NF_ACCEPT;
 524         }
 525
 526         if (skb_try_make_writable(skb, sizeof(*ip6h)))
 527                 return NF_DROP;
 528
 529         if (nf_flow_nat_ipv6(flow, skb, dir) < 0)
 530                 return NF_DROP;
 531
 532         ip6h = ipv6_hdr(skb);
 533         ip6h->hop_limit--;
 534         skb->tstamp = 0;
 535
 536         if (flow_table->flags & NF_FLOWTABLE_COUNTER)
 537                 nf_ct_acct_update(flow->ct, tuplehash->tuple.dir, skb->len);
 538
 539         if (unlikely(dst_xfrm(&rt->dst))) {
 540                 memset(skb->cb, 0, sizeof(struct inet6_skb_parm));
 541                 IP6CB(skb)->iif = skb->dev->ifindex;
 542                 IP6CB(skb)->flags = IP6SKB_FORWARDED;
 543                 return nf_flow_xmit_xfrm(skb, state, &rt->dst);
 544         }
 545
 546         skb->dev = outdev;
 547         nexthop = rt6_nexthop(rt, &flow->tuplehash[!dir].tuple.src_v6);
 548         skb_dst_set_noref(skb, &rt->dst);
 549         neigh_xmit(NEIGH_ND_TABLE, outdev, nexthop, skb);
 550
 551         return NF_STOLEN;
 552 }
 553 EXPORT_SYMBOL_GPL(nf_flow_offload_ipv6_hook);