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