net/netfilter/nf_nat_proto.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* (C) 1999-2001 Paul `Rusty' Russell
   3  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
   4  */
   5
   6 #include <linux/types.h>
   7 #include <linux/export.h>
   8 #include <linux/init.h>
   9 #include <linux/udp.h>
  10 #include <linux/tcp.h>
  11 #include <linux/icmp.h>
  12 #include <linux/icmpv6.h>
  13
  14 #include <linux/dccp.h>
  15 #include <linux/sctp.h>
  16 #include <net/sctp/checksum.h>
  17
  18 #include <linux/netfilter.h>
  19 #include <net/netfilter/nf_nat.h>
  20
  21 #include <linux/ipv6.h>
  22 #include <linux/netfilter_ipv6.h>
  23 #include <net/checksum.h>
  24 #include <net/ip6_checksum.h>
  25 #include <net/ip6_route.h>
  26 #include <net/xfrm.h>
  27 #include <net/ipv6.h>
  28
  29 #include <net/netfilter/nf_conntrack_core.h>
  30 #include <net/netfilter/nf_conntrack.h>
  31 #include <linux/netfilter/nfnetlink_conntrack.h>
  32
  33 static void nf_csum_update(struct sk_buff *skb,
  34                            unsigned int iphdroff, __sum16 *check,
  35                            const struct nf_conntrack_tuple *t,
  36                            enum nf_nat_manip_type maniptype);
  37
  38 static void
  39 __udp_manip_pkt(struct sk_buff *skb,
  40                 unsigned int iphdroff, struct udphdr *hdr,
  41                 const struct nf_conntrack_tuple *tuple,
  42                 enum nf_nat_manip_type maniptype, bool do_csum)
  43 {
  44         __be16 *portptr, newport;
  45
  46         if (maniptype == NF_NAT_MANIP_SRC) {
  47                 /* Get rid of src port */
  48                 newport = tuple->src.u.udp.port;
  49                 portptr = &hdr->source;
  50         } else {
  51                 /* Get rid of dst port */
  52                 newport = tuple->dst.u.udp.port;
  53                 portptr = &hdr->dest;
  54         }
  55         if (do_csum) {
  56                 nf_csum_update(skb, iphdroff, &hdr->check, tuple, maniptype);
  57                 inet_proto_csum_replace2(&hdr->check, skb, *portptr, newport,
  58                                          false);
  59                 if (!hdr->check)
  60                         hdr->check = CSUM_MANGLED_0;
  61         }
  62         *portptr = newport;
  63 }
  64
  65 static bool udp_manip_pkt(struct sk_buff *skb,
  66                           unsigned int iphdroff, unsigned int hdroff,
  67                           const struct nf_conntrack_tuple *tuple,
  68                           enum nf_nat_manip_type maniptype)
  69 {
  70         struct udphdr *hdr;
  71
  72         if (skb_ensure_writable(skb, hdroff + sizeof(*hdr)))
  73                 return false;
  74
  75         hdr = (struct udphdr *)(skb->data + hdroff);
  76         __udp_manip_pkt(skb, iphdroff, hdr, tuple, maniptype, !!hdr->check);
  77
  78         return true;
  79 }
  80
  81 static bool udplite_manip_pkt(struct sk_buff *skb,
  82                               unsigned int iphdroff, unsigned int hdroff,
  83                               const struct nf_conntrack_tuple *tuple,
  84                               enum nf_nat_manip_type maniptype)
  85 {
  86 #ifdef CONFIG_NF_CT_PROTO_UDPLITE
  87         struct udphdr *hdr;
  88
  89         if (skb_ensure_writable(skb, hdroff + sizeof(*hdr)))
  90                 return false;
  91
  92         hdr = (struct udphdr *)(skb->data + hdroff);
  93         __udp_manip_pkt(skb, iphdroff, hdr, tuple, maniptype, true);
  94 #endif
  95         return true;
  96 }
  97
  98 static bool
  99 sctp_manip_pkt(struct sk_buff *skb,
 100                unsigned int iphdroff, unsigned int hdroff,
 101                const struct nf_conntrack_tuple *tuple,
 102                enum nf_nat_manip_type maniptype)
 103 {
 104 #ifdef CONFIG_NF_CT_PROTO_SCTP
 105         struct sctphdr *hdr;
 106         int hdrsize = 8;
 107
 108         /* This could be an inner header returned in imcp packet; in such
 109          * cases we cannot update the checksum field since it is outside
 110          * of the 8 bytes of transport layer headers we are guaranteed.
 111          */
 112         if (skb->len >= hdroff + sizeof(*hdr))
 113                 hdrsize = sizeof(*hdr);
 114
 115         if (skb_ensure_writable(skb, hdroff + hdrsize))
 116                 return false;
 117
 118         hdr = (struct sctphdr *)(skb->data + hdroff);
 119
 120         if (maniptype == NF_NAT_MANIP_SRC) {
 121                 /* Get rid of src port */
 122                 hdr->source = tuple->src.u.sctp.port;
 123         } else {
 124                 /* Get rid of dst port */
 125                 hdr->dest = tuple->dst.u.sctp.port;
 126         }
 127
 128         if (hdrsize < sizeof(*hdr))
 129                 return true;
 130
 131         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 132                 hdr->checksum = sctp_compute_cksum(skb, hdroff);
 133                 skb->ip_summed = CHECKSUM_NONE;
 134         }
 135
 136 #endif
 137         return true;
 138 }
 139
 140 static bool
 141 tcp_manip_pkt(struct sk_buff *skb,
 142               unsigned int iphdroff, unsigned int hdroff,
 143               const struct nf_conntrack_tuple *tuple,
 144               enum nf_nat_manip_type maniptype)
 145 {
 146         struct tcphdr *hdr;
 147         __be16 *portptr, newport, oldport;
 148         int hdrsize = 8; /* TCP connection tracking guarantees this much */
 149
 150         /* this could be a inner header returned in icmp packet; in such
 151            cases we cannot update the checksum field since it is outside of
 152            the 8 bytes of transport layer headers we are guaranteed */
 153         if (skb->len >= hdroff + sizeof(struct tcphdr))
 154                 hdrsize = sizeof(struct tcphdr);
 155
 156         if (skb_ensure_writable(skb, hdroff + hdrsize))
 157                 return false;
 158
 159         hdr = (struct tcphdr *)(skb->data + hdroff);
 160
 161         if (maniptype == NF_NAT_MANIP_SRC) {
 162                 /* Get rid of src port */
 163                 newport = tuple->src.u.tcp.port;
 164                 portptr = &hdr->source;
 165         } else {
 166                 /* Get rid of dst port */
 167                 newport = tuple->dst.u.tcp.port;
 168                 portptr = &hdr->dest;
 169         }
 170
 171         oldport = *portptr;
 172         *portptr = newport;
 173
 174         if (hdrsize < sizeof(*hdr))
 175                 return true;
 176
 177         nf_csum_update(skb, iphdroff, &hdr->check, tuple, maniptype);
 178         inet_proto_csum_replace2(&hdr->check, skb, oldport, newport, false);
 179         return true;
 180 }
 181
 182 static bool
 183 dccp_manip_pkt(struct sk_buff *skb,
 184                unsigned int iphdroff, unsigned int hdroff,
 185                const struct nf_conntrack_tuple *tuple,
 186                enum nf_nat_manip_type maniptype)
 187 {
 188 #ifdef CONFIG_NF_CT_PROTO_DCCP
 189         struct dccp_hdr *hdr;
 190         __be16 *portptr, oldport, newport;
 191         int hdrsize = 8; /* DCCP connection tracking guarantees this much */
 192
 193         if (skb->len >= hdroff + sizeof(struct dccp_hdr))
 194                 hdrsize = sizeof(struct dccp_hdr);
 195
 196         if (skb_ensure_writable(skb, hdroff + hdrsize))
 197                 return false;
 198
 199         hdr = (struct dccp_hdr *)(skb->data + hdroff);
 200
 201         if (maniptype == NF_NAT_MANIP_SRC) {
 202                 newport = tuple->src.u.dccp.port;
 203                 portptr = &hdr->dccph_sport;
 204         } else {
 205                 newport = tuple->dst.u.dccp.port;
 206                 portptr = &hdr->dccph_dport;
 207         }
 208
 209         oldport = *portptr;
 210         *portptr = newport;
 211
 212         if (hdrsize < sizeof(*hdr))
 213                 return true;
 214
 215         nf_csum_update(skb, iphdroff, &hdr->dccph_checksum, tuple, maniptype);
 216         inet_proto_csum_replace2(&hdr->dccph_checksum, skb, oldport, newport,
 217                                  false);
 218 #endif
 219         return true;
 220 }
 221
 222 static bool
 223 icmp_manip_pkt(struct sk_buff *skb,
 224                unsigned int iphdroff, unsigned int hdroff,
 225                const struct nf_conntrack_tuple *tuple,
 226                enum nf_nat_manip_type maniptype)
 227 {
 228         struct icmphdr *hdr;
 229
 230         if (skb_ensure_writable(skb, hdroff + sizeof(*hdr)))
 231                 return false;
 232
 233         hdr = (struct icmphdr *)(skb->data + hdroff);
 234         switch (hdr->type) {
 235         case ICMP_ECHO:
 236         case ICMP_ECHOREPLY:
 237         case ICMP_TIMESTAMP:
 238         case ICMP_TIMESTAMPREPLY:
 239         case ICMP_INFO_REQUEST:
 240         case ICMP_INFO_REPLY:
 241         case ICMP_ADDRESS:
 242         case ICMP_ADDRESSREPLY:
 243                 break;
 244         default:
 245                 return true;
 246         }
 247         inet_proto_csum_replace2(&hdr->checksum, skb,
 248                                  hdr->un.echo.id, tuple->src.u.icmp.id, false);
 249         hdr->un.echo.id = tuple->src.u.icmp.id;
 250         return true;
 251 }
 252
 253 static bool
 254 icmpv6_manip_pkt(struct sk_buff *skb,
 255                  unsigned int iphdroff, unsigned int hdroff,
 256                  const struct nf_conntrack_tuple *tuple,
 257                  enum nf_nat_manip_type maniptype)
 258 {
 259         struct icmp6hdr *hdr;
 260
 261         if (skb_ensure_writable(skb, hdroff + sizeof(*hdr)))
 262                 return false;
 263
 264         hdr = (struct icmp6hdr *)(skb->data + hdroff);
 265         nf_csum_update(skb, iphdroff, &hdr->icmp6_cksum, tuple, maniptype);
 266         if (hdr->icmp6_type == ICMPV6_ECHO_REQUEST ||
 267             hdr->icmp6_type == ICMPV6_ECHO_REPLY) {
 268                 inet_proto_csum_replace2(&hdr->icmp6_cksum, skb,
 269                                          hdr->icmp6_identifier,
 270                                          tuple->src.u.icmp.id, false);
 271                 hdr->icmp6_identifier = tuple->src.u.icmp.id;
 272         }
 273         return true;
 274 }
 275
 276 /* manipulate a GRE packet according to maniptype */
 277 static bool
 278 gre_manip_pkt(struct sk_buff *skb,
 279               unsigned int iphdroff, unsigned int hdroff,
 280               const struct nf_conntrack_tuple *tuple,
 281               enum nf_nat_manip_type maniptype)
 282 {
 283 #if IS_ENABLED(CONFIG_NF_CT_PROTO_GRE)
 284         const struct gre_base_hdr *greh;
 285         struct pptp_gre_header *pgreh;
 286
 287         /* pgreh includes two optional 32bit fields which are not required
 288          * to be there.  That's where the magic '8' comes from */
 289         if (skb_ensure_writable(skb, hdroff + sizeof(*pgreh) - 8))
 290                 return false;
 291
 292         greh = (void *)skb->data + hdroff;
 293         pgreh = (struct pptp_gre_header *)greh;
 294
 295         /* we only have destination manip of a packet, since 'source key'
 296          * is not present in the packet itself */
 297         if (maniptype != NF_NAT_MANIP_DST)
 298                 return true;
 299
 300         switch (greh->flags & GRE_VERSION) {
 301         case GRE_VERSION_0:
 302                 /* We do not currently NAT any GREv0 packets.
 303                  * Try to behave like "nf_nat_proto_unknown" */
 304                 break;
 305         case GRE_VERSION_1:
 306                 pr_debug("call_id -> 0x%04x\n", ntohs(tuple->dst.u.gre.key));
 307                 pgreh->call_id = tuple->dst.u.gre.key;
 308                 break;
 309         default:
 310                 pr_debug("can't nat unknown GRE version\n");
 311                 return false;
 312         }
 313 #endif
 314         return true;
 315 }
 316
 317 static bool l4proto_manip_pkt(struct sk_buff *skb,
 318                               unsigned int iphdroff, unsigned int hdroff,
 319                               const struct nf_conntrack_tuple *tuple,
 320                               enum nf_nat_manip_type maniptype)
 321 {
 322         switch (tuple->dst.protonum) {
 323         case IPPROTO_TCP:
 324                 return tcp_manip_pkt(skb, iphdroff, hdroff,
 325                                      tuple, maniptype);
 326         case IPPROTO_UDP:
 327                 return udp_manip_pkt(skb, iphdroff, hdroff,
 328                                      tuple, maniptype);
 329         case IPPROTO_UDPLITE:
 330                 return udplite_manip_pkt(skb, iphdroff, hdroff,
 331                                          tuple, maniptype);
 332         case IPPROTO_SCTP:
 333                 return sctp_manip_pkt(skb, iphdroff, hdroff,
 334                                       tuple, maniptype);
 335         case IPPROTO_ICMP:
 336                 return icmp_manip_pkt(skb, iphdroff, hdroff,
 337                                       tuple, maniptype);
 338         case IPPROTO_ICMPV6:
 339                 return icmpv6_manip_pkt(skb, iphdroff, hdroff,
 340                                         tuple, maniptype);
 341         case IPPROTO_DCCP:
 342                 return dccp_manip_pkt(skb, iphdroff, hdroff,
 343                                       tuple, maniptype);
 344         case IPPROTO_GRE:
 345                 return gre_manip_pkt(skb, iphdroff, hdroff,
 346                                      tuple, maniptype);
 347         }
 348
 349         /* If we don't know protocol -- no error, pass it unmodified. */
 350         return true;
 351 }
 352
 353 static bool nf_nat_ipv4_manip_pkt(struct sk_buff *skb,
 354                                   unsigned int iphdroff,
 355                                   const struct nf_conntrack_tuple *target,
 356                                   enum nf_nat_manip_type maniptype)
 357 {
 358         struct iphdr *iph;
 359         unsigned int hdroff;
 360
 361         if (skb_ensure_writable(skb, iphdroff + sizeof(*iph)))
 362                 return false;
 363
 364         iph = (void *)skb->data + iphdroff;
 365         hdroff = iphdroff + iph->ihl * 4;
 366
 367         if (!l4proto_manip_pkt(skb, iphdroff, hdroff, target, maniptype))
 368                 return false;
 369         iph = (void *)skb->data + iphdroff;
 370
 371         if (maniptype == NF_NAT_MANIP_SRC) {
 372                 csum_replace4(&iph->check, iph->saddr, target->src.u3.ip);
 373                 iph->saddr = target->src.u3.ip;
 374         } else {
 375                 csum_replace4(&iph->check, iph->daddr, target->dst.u3.ip);
 376                 iph->daddr = target->dst.u3.ip;
 377         }
 378         return true;
 379 }
 380
 381 static bool nf_nat_ipv6_manip_pkt(struct sk_buff *skb,
 382                                   unsigned int iphdroff,
 383                                   const struct nf_conntrack_tuple *target,
 384                                   enum nf_nat_manip_type maniptype)
 385 {
 386 #if IS_ENABLED(CONFIG_IPV6)
 387         struct ipv6hdr *ipv6h;
 388         __be16 frag_off;
 389         int hdroff;
 390         u8 nexthdr;
 391
 392         if (skb_ensure_writable(skb, iphdroff + sizeof(*ipv6h)))
 393                 return false;
 394
 395         ipv6h = (void *)skb->data + iphdroff;
 396         nexthdr = ipv6h->nexthdr;
 397         hdroff = ipv6_skip_exthdr(skb, iphdroff + sizeof(*ipv6h),
 398                                   &nexthdr, &frag_off);
 399         if (hdroff < 0)
 400                 goto manip_addr;
 401
 402         if ((frag_off & htons(~0x7)) == 0 &&
 403             !l4proto_manip_pkt(skb, iphdroff, hdroff, target, maniptype))
 404                 return false;
 405
 406         /* must reload, offset might have changed */
 407         ipv6h = (void *)skb->data + iphdroff;
 408
 409 manip_addr:
 410         if (maniptype == NF_NAT_MANIP_SRC)
 411                 ipv6h->saddr = target->src.u3.in6;
 412         else
 413                 ipv6h->daddr = target->dst.u3.in6;
 414
 415 #endif
 416         return true;
 417 }
 418
 419 unsigned int nf_nat_manip_pkt(struct sk_buff *skb, struct nf_conn *ct,
 420                               enum nf_nat_manip_type mtype,
 421                               enum ip_conntrack_dir dir)
 422 {
 423         struct nf_conntrack_tuple target;
 424
 425         /* We are aiming to look like inverse of other direction. */
 426         nf_ct_invert_tuple(&target, &ct->tuplehash[!dir].tuple);
 427
 428         switch (target.src.l3num) {
 429         case NFPROTO_IPV6:
 430                 if (nf_nat_ipv6_manip_pkt(skb, 0, &target, mtype))
 431                         return NF_ACCEPT;
 432                 break;
 433         case NFPROTO_IPV4:
 434                 if (nf_nat_ipv4_manip_pkt(skb, 0, &target, mtype))
 435                         return NF_ACCEPT;
 436                 break;
 437         default:
 438                 WARN_ON_ONCE(1);
 439                 break;
 440         }
 441
 442         return NF_DROP;
 443 }
 444
 445 static void nf_nat_ipv4_csum_update(struct sk_buff *skb,
 446                                     unsigned int iphdroff, __sum16 *check,
 447                                     const struct nf_conntrack_tuple *t,
 448                                     enum nf_nat_manip_type maniptype)
 449 {
 450         struct iphdr *iph = (struct iphdr *)(skb->data + iphdroff);
 451         __be32 oldip, newip;
 452
 453         if (maniptype == NF_NAT_MANIP_SRC) {
 454                 oldip = iph->saddr;
 455                 newip = t->src.u3.ip;
 456         } else {
 457                 oldip = iph->daddr;
 458                 newip = t->dst.u3.ip;
 459         }
 460         inet_proto_csum_replace4(check, skb, oldip, newip, true);
 461 }
 462
 463 static void nf_nat_ipv6_csum_update(struct sk_buff *skb,
 464                                     unsigned int iphdroff, __sum16 *check,
 465                                     const struct nf_conntrack_tuple *t,
 466                                     enum nf_nat_manip_type maniptype)
 467 {
 468 #if IS_ENABLED(CONFIG_IPV6)
 469         const struct ipv6hdr *ipv6h = (struct ipv6hdr *)(skb->data + iphdroff);
 470         const struct in6_addr *oldip, *newip;
 471
 472         if (maniptype == NF_NAT_MANIP_SRC) {
 473                 oldip = &ipv6h->saddr;
 474                 newip = &t->src.u3.in6;
 475         } else {
 476                 oldip = &ipv6h->daddr;
 477                 newip = &t->dst.u3.in6;
 478         }
 479         inet_proto_csum_replace16(check, skb, oldip->s6_addr32,
 480                                   newip->s6_addr32, true);
 481 #endif
 482 }
 483
 484 static void nf_csum_update(struct sk_buff *skb,
 485                            unsigned int iphdroff, __sum16 *check,
 486                            const struct nf_conntrack_tuple *t,
 487                            enum nf_nat_manip_type maniptype)
 488 {
 489         switch (t->src.l3num) {
 490         case NFPROTO_IPV4:
 491                 nf_nat_ipv4_csum_update(skb, iphdroff, check, t, maniptype);
 492                 return;
 493         case NFPROTO_IPV6:
 494                 nf_nat_ipv6_csum_update(skb, iphdroff, check, t, maniptype);
 495                 return;
 496         }
 497 }
 498
 499 static void nf_nat_ipv4_csum_recalc(struct sk_buff *skb,
 500                                     u8 proto, void *data, __sum16 *check,
 501                                     int datalen, int oldlen)
 502 {
 503         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 504                 const struct iphdr *iph = ip_hdr(skb);
 505
 506                 skb->ip_summed = CHECKSUM_PARTIAL;
 507                 skb->csum_start = skb_headroom(skb) + skb_network_offset(skb) +
 508                         ip_hdrlen(skb);
 509                 skb->csum_offset = (void *)check - data;
 510                 *check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, datalen,
 511                                             proto, 0);
 512         } else {
 513                 inet_proto_csum_replace2(check, skb,
 514                                          htons(oldlen), htons(datalen), true);
 515         }
 516 }
 517
 518 #if IS_ENABLED(CONFIG_IPV6)
 519 static void nf_nat_ipv6_csum_recalc(struct sk_buff *skb,
 520                                     u8 proto, void *data, __sum16 *check,
 521                                     int datalen, int oldlen)
 522 {
 523         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 524                 const struct ipv6hdr *ipv6h = ipv6_hdr(skb);
 525
 526                 skb->ip_summed = CHECKSUM_PARTIAL;
 527                 skb->csum_start = skb_headroom(skb) + skb_network_offset(skb) +
 528                         (data - (void *)skb->data);
 529                 skb->csum_offset = (void *)check - data;
 530                 *check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
 531                                           datalen, proto, 0);
 532         } else {
 533                 inet_proto_csum_replace2(check, skb,
 534                                          htons(oldlen), htons(datalen), true);
 535         }
 536 }
 537 #endif
 538
 539 void nf_nat_csum_recalc(struct sk_buff *skb,
 540                         u8 nfproto, u8 proto, void *data, __sum16 *check,
 541                         int datalen, int oldlen)
 542 {
 543         switch (nfproto) {
 544         case NFPROTO_IPV4:
 545                 nf_nat_ipv4_csum_recalc(skb, proto, data, check,
 546                                         datalen, oldlen);
 547                 return;
 548 #if IS_ENABLED(CONFIG_IPV6)
 549         case NFPROTO_IPV6:
 550                 nf_nat_ipv6_csum_recalc(skb, proto, data, check,
 551                                         datalen, oldlen);
 552                 return;
 553 #endif
 554         }
 555
 556         WARN_ON_ONCE(1);
 557 }
 558
 559 int nf_nat_icmp_reply_translation(struct sk_buff *skb,
 560                                   struct nf_conn *ct,
 561                                   enum ip_conntrack_info ctinfo,
 562                                   unsigned int hooknum)
 563 {
 564         struct {
 565                 struct icmphdr  icmp;
 566                 struct iphdr    ip;
 567         } *inside;
 568         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 569         enum nf_nat_manip_type manip = HOOK2MANIP(hooknum);
 570         unsigned int hdrlen = ip_hdrlen(skb);
 571         struct nf_conntrack_tuple target;
 572         unsigned long statusbit;
 573
 574         WARN_ON(ctinfo != IP_CT_RELATED && ctinfo != IP_CT_RELATED_REPLY);
 575
 576         if (skb_ensure_writable(skb, hdrlen + sizeof(*inside)))
 577                 return 0;
 578         if (nf_ip_checksum(skb, hooknum, hdrlen, IPPROTO_ICMP))
 579                 return 0;
 580
 581         inside = (void *)skb->data + hdrlen;
 582         if (inside->icmp.type == ICMP_REDIRECT) {
 583                 if ((ct->status & IPS_NAT_DONE_MASK) != IPS_NAT_DONE_MASK)
 584                         return 0;
 585                 if (ct->status & IPS_NAT_MASK)
 586                         return 0;
 587         }
 588
 589         if (manip == NF_NAT_MANIP_SRC)
 590                 statusbit = IPS_SRC_NAT;
 591         else
 592                 statusbit = IPS_DST_NAT;
 593
 594         /* Invert if this is reply direction */
 595         if (dir == IP_CT_DIR_REPLY)
 596                 statusbit ^= IPS_NAT_MASK;
 597
 598         if (!(ct->status & statusbit))
 599                 return 1;
 600
 601         if (!nf_nat_ipv4_manip_pkt(skb, hdrlen + sizeof(inside->icmp),
 602                                    &ct->tuplehash[!dir].tuple, !manip))
 603                 return 0;
 604
 605         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 606                 /* Reloading "inside" here since manip_pkt may reallocate */
 607                 inside = (void *)skb->data + hdrlen;
 608                 inside->icmp.checksum = 0;
 609                 inside->icmp.checksum =
 610                         csum_fold(skb_checksum(skb, hdrlen,
 611                                                skb->len - hdrlen, 0));
 612         }
 613
 614         /* Change outer to look like the reply to an incoming packet */
 615         nf_ct_invert_tuple(&target, &ct->tuplehash[!dir].tuple);
 616         target.dst.protonum = IPPROTO_ICMP;
 617         if (!nf_nat_ipv4_manip_pkt(skb, 0, &target, manip))
 618                 return 0;
 619
 620         return 1;
 621 }
 622 EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation);
 623
 624 static unsigned int
 625 nf_nat_ipv4_fn(void *priv, struct sk_buff *skb,
 626                const struct nf_hook_state *state)
 627 {
 628         struct nf_conn *ct;
 629         enum ip_conntrack_info ctinfo;
 630
 631         ct = nf_ct_get(skb, &ctinfo);
 632         if (!ct)
 633                 return NF_ACCEPT;
 634
 635         if (ctinfo == IP_CT_RELATED || ctinfo == IP_CT_RELATED_REPLY) {
 636                 if (ip_hdr(skb)->protocol == IPPROTO_ICMP) {
 637                         if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
 638                                                            state->hook))
 639                                 return NF_DROP;
 640                         else
 641                                 return NF_ACCEPT;
 642                 }
 643         }
 644
 645         return nf_nat_inet_fn(priv, skb, state);
 646 }
 647
 648 static unsigned int
 649 nf_nat_ipv4_pre_routing(void *priv, struct sk_buff *skb,
 650                         const struct nf_hook_state *state)
 651 {
 652         unsigned int ret;
 653         __be32 daddr = ip_hdr(skb)->daddr;
 654
 655         ret = nf_nat_ipv4_fn(priv, skb, state);
 656         if (ret == NF_ACCEPT && daddr != ip_hdr(skb)->daddr)
 657                 skb_dst_drop(skb);
 658
 659         return ret;
 660 }
 661
 662 #ifdef CONFIG_XFRM
 663 static int nf_xfrm_me_harder(struct net *net, struct sk_buff *skb, unsigned int family)
 664 {
 665         struct sock *sk = skb->sk;
 666         struct dst_entry *dst;
 667         unsigned int hh_len;
 668         struct flowi fl;
 669         int err;
 670
 671         err = xfrm_decode_session(skb, &fl, family);
 672         if (err < 0)
 673                 return err;
 674
 675         dst = skb_dst(skb);
 676         if (dst->xfrm)
 677                 dst = ((struct xfrm_dst *)dst)->route;
 678         if (!dst_hold_safe(dst))
 679                 return -EHOSTUNREACH;
 680
 681         if (sk && !net_eq(net, sock_net(sk)))
 682                 sk = NULL;
 683
 684         dst = xfrm_lookup(net, dst, &fl, sk, 0);
 685         if (IS_ERR(dst))
 686                 return PTR_ERR(dst);
 687
 688         skb_dst_drop(skb);
 689         skb_dst_set(skb, dst);
 690
 691         /* Change in oif may mean change in hh_len. */
 692         hh_len = skb_dst(skb)->dev->hard_header_len;
 693         if (skb_headroom(skb) < hh_len &&
 694             pskb_expand_head(skb, hh_len - skb_headroom(skb), 0, GFP_ATOMIC))
 695                 return -ENOMEM;
 696         return 0;
 697 }
 698 #endif
 699
 700 static unsigned int
 701 nf_nat_ipv4_local_in(void *priv, struct sk_buff *skb,
 702                      const struct nf_hook_state *state)
 703 {
 704         __be32 saddr = ip_hdr(skb)->saddr;
 705         struct sock *sk = skb->sk;
 706         unsigned int ret;
 707
 708         ret = nf_nat_ipv4_fn(priv, skb, state);
 709
 710         if (ret == NF_ACCEPT && sk && saddr != ip_hdr(skb)->saddr &&
 711             !inet_sk_transparent(sk))
 712                 skb_orphan(skb); /* TCP edemux obtained wrong socket */
 713
 714         return ret;
 715 }
 716
 717 static unsigned int
 718 nf_nat_ipv4_out(void *priv, struct sk_buff *skb,
 719                 const struct nf_hook_state *state)
 720 {
 721 #ifdef CONFIG_XFRM
 722         const struct nf_conn *ct;
 723         enum ip_conntrack_info ctinfo;
 724         int err;
 725 #endif
 726         unsigned int ret;
 727
 728         ret = nf_nat_ipv4_fn(priv, skb, state);
 729 #ifdef CONFIG_XFRM
 730         if (ret != NF_ACCEPT)
 731                 return ret;
 732
 733         if (IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED)
 734                 return ret;
 735
 736         ct = nf_ct_get(skb, &ctinfo);
 737         if (ct) {
 738                 enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 739
 740                 if (ct->tuplehash[dir].tuple.src.u3.ip !=
 741                      ct->tuplehash[!dir].tuple.dst.u3.ip ||
 742                     (ct->tuplehash[dir].tuple.dst.protonum != IPPROTO_ICMP &&
 743                      ct->tuplehash[dir].tuple.src.u.all !=
 744                      ct->tuplehash[!dir].tuple.dst.u.all)) {
 745                         err = nf_xfrm_me_harder(state->net, skb, AF_INET);
 746                         if (err < 0)
 747                                 ret = NF_DROP_ERR(err);
 748                 }
 749         }
 750 #endif
 751         return ret;
 752 }
 753
 754 static unsigned int
 755 nf_nat_ipv4_local_fn(void *priv, struct sk_buff *skb,
 756                      const struct nf_hook_state *state)
 757 {
 758         const struct nf_conn *ct;
 759         enum ip_conntrack_info ctinfo;
 760         unsigned int ret;
 761         int err;
 762
 763         ret = nf_nat_ipv4_fn(priv, skb, state);
 764         if (ret != NF_ACCEPT)
 765                 return ret;
 766
 767         ct = nf_ct_get(skb, &ctinfo);
 768         if (ct) {
 769                 enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 770
 771                 if (ct->tuplehash[dir].tuple.dst.u3.ip !=
 772                     ct->tuplehash[!dir].tuple.src.u3.ip) {
 773                         err = ip_route_me_harder(state->net, state->sk, skb, RTN_UNSPEC);
 774                         if (err < 0)
 775                                 ret = NF_DROP_ERR(err);
 776                 }
 777 #ifdef CONFIG_XFRM
 778                 else if (!(IPCB(skb)->flags & IPSKB_XFRM_TRANSFORMED) &&
 779                          ct->tuplehash[dir].tuple.dst.protonum != IPPROTO_ICMP &&
 780                          ct->tuplehash[dir].tuple.dst.u.all !=
 781                          ct->tuplehash[!dir].tuple.src.u.all) {
 782                         err = nf_xfrm_me_harder(state->net, skb, AF_INET);
 783                         if (err < 0)
 784                                 ret = NF_DROP_ERR(err);
 785                 }
 786 #endif
 787         }
 788         return ret;
 789 }
 790
 791 static const struct nf_hook_ops nf_nat_ipv4_ops[] = {
 792         /* Before packet filtering, change destination */
 793         {
 794                 .hook           = nf_nat_ipv4_pre_routing,
 795                 .pf             = NFPROTO_IPV4,
 796                 .hooknum        = NF_INET_PRE_ROUTING,
 797                 .priority       = NF_IP_PRI_NAT_DST,
 798         },
 799         /* After packet filtering, change source */
 800         {
 801                 .hook           = nf_nat_ipv4_out,
 802                 .pf             = NFPROTO_IPV4,
 803                 .hooknum        = NF_INET_POST_ROUTING,
 804                 .priority       = NF_IP_PRI_NAT_SRC,
 805         },
 806         /* Before packet filtering, change destination */
 807         {
 808                 .hook           = nf_nat_ipv4_local_fn,
 809                 .pf             = NFPROTO_IPV4,
 810                 .hooknum        = NF_INET_LOCAL_OUT,
 811                 .priority       = NF_IP_PRI_NAT_DST,
 812         },
 813         /* After packet filtering, change source */
 814         {
 815                 .hook           = nf_nat_ipv4_local_in,
 816                 .pf             = NFPROTO_IPV4,
 817                 .hooknum        = NF_INET_LOCAL_IN,
 818                 .priority       = NF_IP_PRI_NAT_SRC,
 819         },
 820 };
 821
 822 int nf_nat_ipv4_register_fn(struct net *net, const struct nf_hook_ops *ops)
 823 {
 824         return nf_nat_register_fn(net, ops->pf, ops, nf_nat_ipv4_ops,
 825                                   ARRAY_SIZE(nf_nat_ipv4_ops));
 826 }
 827 EXPORT_SYMBOL_GPL(nf_nat_ipv4_register_fn);
 828
 829 void nf_nat_ipv4_unregister_fn(struct net *net, const struct nf_hook_ops *ops)
 830 {
 831         nf_nat_unregister_fn(net, ops->pf, ops, ARRAY_SIZE(nf_nat_ipv4_ops));
 832 }
 833 EXPORT_SYMBOL_GPL(nf_nat_ipv4_unregister_fn);
 834
 835 #if IS_ENABLED(CONFIG_IPV6)
 836 int nf_nat_icmpv6_reply_translation(struct sk_buff *skb,
 837                                     struct nf_conn *ct,
 838                                     enum ip_conntrack_info ctinfo,
 839                                     unsigned int hooknum,
 840                                     unsigned int hdrlen)
 841 {
 842         struct {
 843                 struct icmp6hdr icmp6;
 844                 struct ipv6hdr  ip6;
 845         } *inside;
 846         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 847         enum nf_nat_manip_type manip = HOOK2MANIP(hooknum);
 848         struct nf_conntrack_tuple target;
 849         unsigned long statusbit;
 850
 851         WARN_ON(ctinfo != IP_CT_RELATED && ctinfo != IP_CT_RELATED_REPLY);
 852
 853         if (skb_ensure_writable(skb, hdrlen + sizeof(*inside)))
 854                 return 0;
 855         if (nf_ip6_checksum(skb, hooknum, hdrlen, IPPROTO_ICMPV6))
 856                 return 0;
 857
 858         inside = (void *)skb->data + hdrlen;
 859         if (inside->icmp6.icmp6_type == NDISC_REDIRECT) {
 860                 if ((ct->status & IPS_NAT_DONE_MASK) != IPS_NAT_DONE_MASK)
 861                         return 0;
 862                 if (ct->status & IPS_NAT_MASK)
 863                         return 0;
 864         }
 865
 866         if (manip == NF_NAT_MANIP_SRC)
 867                 statusbit = IPS_SRC_NAT;
 868         else
 869                 statusbit = IPS_DST_NAT;
 870
 871         /* Invert if this is reply direction */
 872         if (dir == IP_CT_DIR_REPLY)
 873                 statusbit ^= IPS_NAT_MASK;
 874
 875         if (!(ct->status & statusbit))
 876                 return 1;
 877
 878         if (!nf_nat_ipv6_manip_pkt(skb, hdrlen + sizeof(inside->icmp6),
 879                                    &ct->tuplehash[!dir].tuple, !manip))
 880                 return 0;
 881
 882         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 883                 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
 884
 885                 inside = (void *)skb->data + hdrlen;
 886                 inside->icmp6.icmp6_cksum = 0;
 887                 inside->icmp6.icmp6_cksum =
 888                         csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
 889                                         skb->len - hdrlen, IPPROTO_ICMPV6,
 890                                         skb_checksum(skb, hdrlen,
 891                                                      skb->len - hdrlen, 0));
 892         }
 893
 894         nf_ct_invert_tuple(&target, &ct->tuplehash[!dir].tuple);
 895         target.dst.protonum = IPPROTO_ICMPV6;
 896         if (!nf_nat_ipv6_manip_pkt(skb, 0, &target, manip))
 897                 return 0;
 898
 899         return 1;
 900 }
 901 EXPORT_SYMBOL_GPL(nf_nat_icmpv6_reply_translation);
 902
 903 static unsigned int
 904 nf_nat_ipv6_fn(void *priv, struct sk_buff *skb,
 905                const struct nf_hook_state *state)
 906 {
 907         struct nf_conn *ct;
 908         enum ip_conntrack_info ctinfo;
 909         __be16 frag_off;
 910         int hdrlen;
 911         u8 nexthdr;
 912
 913         ct = nf_ct_get(skb, &ctinfo);
 914         /* Can't track?  It's not due to stress, or conntrack would
 915          * have dropped it.  Hence it's the user's responsibilty to
 916          * packet filter it out, or implement conntrack/NAT for that
 917          * protocol. 8) --RR
 918          */
 919         if (!ct)
 920                 return NF_ACCEPT;
 921
 922         if (ctinfo == IP_CT_RELATED || ctinfo == IP_CT_RELATED_REPLY) {
 923                 nexthdr = ipv6_hdr(skb)->nexthdr;
 924                 hdrlen = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr),
 925                                           &nexthdr, &frag_off);
 926
 927                 if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 928                         if (!nf_nat_icmpv6_reply_translation(skb, ct, ctinfo,
 929                                                              state->hook,
 930                                                              hdrlen))
 931                                 return NF_DROP;
 932                         else
 933                                 return NF_ACCEPT;
 934                 }
 935         }
 936
 937         return nf_nat_inet_fn(priv, skb, state);
 938 }
 939
 940 static unsigned int
 941 nf_nat_ipv6_in(void *priv, struct sk_buff *skb,
 942                const struct nf_hook_state *state)
 943 {
 944         unsigned int ret;
 945         struct in6_addr daddr = ipv6_hdr(skb)->daddr;
 946
 947         ret = nf_nat_ipv6_fn(priv, skb, state);
 948         if (ret != NF_DROP && ret != NF_STOLEN &&
 949             ipv6_addr_cmp(&daddr, &ipv6_hdr(skb)->daddr))
 950                 skb_dst_drop(skb);
 951
 952         return ret;
 953 }
 954
 955 static unsigned int
 956 nf_nat_ipv6_out(void *priv, struct sk_buff *skb,
 957                 const struct nf_hook_state *state)
 958 {
 959 #ifdef CONFIG_XFRM
 960         const struct nf_conn *ct;
 961         enum ip_conntrack_info ctinfo;
 962         int err;
 963 #endif
 964         unsigned int ret;
 965
 966         ret = nf_nat_ipv6_fn(priv, skb, state);
 967 #ifdef CONFIG_XFRM
 968         if (ret != NF_ACCEPT)
 969                 return ret;
 970
 971         if (IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED)
 972                 return ret;
 973         ct = nf_ct_get(skb, &ctinfo);
 974         if (ct) {
 975                 enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 976
 977                 if (!nf_inet_addr_cmp(&ct->tuplehash[dir].tuple.src.u3,
 978                                       &ct->tuplehash[!dir].tuple.dst.u3) ||
 979                     (ct->tuplehash[dir].tuple.dst.protonum != IPPROTO_ICMPV6 &&
 980                      ct->tuplehash[dir].tuple.src.u.all !=
 981                      ct->tuplehash[!dir].tuple.dst.u.all)) {
 982                         err = nf_xfrm_me_harder(state->net, skb, AF_INET6);
 983                         if (err < 0)
 984                                 ret = NF_DROP_ERR(err);
 985                 }
 986         }
 987 #endif
 988
 989         return ret;
 990 }
 991
 992 static unsigned int
 993 nf_nat_ipv6_local_fn(void *priv, struct sk_buff *skb,
 994                      const struct nf_hook_state *state)
 995 {
 996         const struct nf_conn *ct;
 997         enum ip_conntrack_info ctinfo;
 998         unsigned int ret;
 999         int err;
1000
1001         ret = nf_nat_ipv6_fn(priv, skb, state);
1002         if (ret != NF_ACCEPT)
1003                 return ret;
1004
1005         ct = nf_ct_get(skb, &ctinfo);
1006         if (ct) {
1007                 enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
1008
1009                 if (!nf_inet_addr_cmp(&ct->tuplehash[dir].tuple.dst.u3,
1010                                       &ct->tuplehash[!dir].tuple.src.u3)) {
1011                         err = nf_ip6_route_me_harder(state->net, state->sk, skb);
1012                         if (err < 0)
1013                                 ret = NF_DROP_ERR(err);
1014                 }
1015 #ifdef CONFIG_XFRM
1016                 else if (!(IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED) &&
1017                          ct->tuplehash[dir].tuple.dst.protonum != IPPROTO_ICMPV6 &&
1018                          ct->tuplehash[dir].tuple.dst.u.all !=
1019                          ct->tuplehash[!dir].tuple.src.u.all) {
1020                         err = nf_xfrm_me_harder(state->net, skb, AF_INET6);
1021                         if (err < 0)
1022                                 ret = NF_DROP_ERR(err);
1023                 }
1024 #endif
1025         }
1026
1027         return ret;
1028 }
1029
1030 static const struct nf_hook_ops nf_nat_ipv6_ops[] = {
1031         /* Before packet filtering, change destination */
1032         {
1033                 .hook           = nf_nat_ipv6_in,
1034                 .pf             = NFPROTO_IPV6,
1035                 .hooknum        = NF_INET_PRE_ROUTING,
1036                 .priority       = NF_IP6_PRI_NAT_DST,
1037         },
1038         /* After packet filtering, change source */
1039         {
1040                 .hook           = nf_nat_ipv6_out,
1041                 .pf             = NFPROTO_IPV6,
1042                 .hooknum        = NF_INET_POST_ROUTING,
1043                 .priority       = NF_IP6_PRI_NAT_SRC,
1044         },
1045         /* Before packet filtering, change destination */
1046         {
1047                 .hook           = nf_nat_ipv6_local_fn,
1048                 .pf             = NFPROTO_IPV6,
1049                 .hooknum        = NF_INET_LOCAL_OUT,
1050                 .priority       = NF_IP6_PRI_NAT_DST,
1051         },
1052         /* After packet filtering, change source */
1053         {
1054                 .hook           = nf_nat_ipv6_fn,
1055                 .pf             = NFPROTO_IPV6,
1056                 .hooknum        = NF_INET_LOCAL_IN,
1057                 .priority       = NF_IP6_PRI_NAT_SRC,
1058         },
1059 };
1060
1061 int nf_nat_ipv6_register_fn(struct net *net, const struct nf_hook_ops *ops)
1062 {
1063         return nf_nat_register_fn(net, ops->pf, ops, nf_nat_ipv6_ops,
1064                                   ARRAY_SIZE(nf_nat_ipv6_ops));
1065 }
1066 EXPORT_SYMBOL_GPL(nf_nat_ipv6_register_fn);
1067
1068 void nf_nat_ipv6_unregister_fn(struct net *net, const struct nf_hook_ops *ops)
1069 {
1070         nf_nat_unregister_fn(net, ops->pf, ops, ARRAY_SIZE(nf_nat_ipv6_ops));
1071 }
1072 EXPORT_SYMBOL_GPL(nf_nat_ipv6_unregister_fn);
1073 #endif /* CONFIG_IPV6 */
1074
1075 #if defined(CONFIG_NF_TABLES_INET) && IS_ENABLED(CONFIG_NFT_NAT)
1076 int nf_nat_inet_register_fn(struct net *net, const struct nf_hook_ops *ops)
1077 {
1078         int ret;
1079
1080         if (WARN_ON_ONCE(ops->pf != NFPROTO_INET))
1081                 return -EINVAL;
1082
1083         ret = nf_nat_register_fn(net, NFPROTO_IPV6, ops, nf_nat_ipv6_ops,
1084                                  ARRAY_SIZE(nf_nat_ipv6_ops));
1085         if (ret)
1086                 return ret;
1087
1088         ret = nf_nat_register_fn(net, NFPROTO_IPV4, ops, nf_nat_ipv4_ops,
1089                                  ARRAY_SIZE(nf_nat_ipv4_ops));
1090         if (ret)
1091                 nf_nat_unregister_fn(net, NFPROTO_IPV6, ops,
1092                                         ARRAY_SIZE(nf_nat_ipv6_ops));
1093         return ret;
1094 }
1095 EXPORT_SYMBOL_GPL(nf_nat_inet_register_fn);
1096
1097 void nf_nat_inet_unregister_fn(struct net *net, const struct nf_hook_ops *ops)
1098 {
1099         nf_nat_unregister_fn(net, NFPROTO_IPV4, ops, ARRAY_SIZE(nf_nat_ipv4_ops));
1100         nf_nat_unregister_fn(net, NFPROTO_IPV6, ops, ARRAY_SIZE(nf_nat_ipv6_ops));
1101 }
1102 EXPORT_SYMBOL_GPL(nf_nat_inet_unregister_fn);
1103 #endif /* NFT INET NAT */