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