net/ipv4/netfilter/nf_nat_helper.c

   1 /* ip_nat_helper.c - generic support functions for NAT helpers
   2  *
   3  * (C) 2000-2002 Harald Welte <laforge@netfilter.org>
   4  * (C) 2003-2006 Netfilter Core Team <coreteam@netfilter.org>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10 #include <linux/module.h>
  11 #include <linux/kmod.h>
  12 #include <linux/types.h>
  13 #include <linux/timer.h>
  14 #include <linux/skbuff.h>
  15 #include <linux/tcp.h>
  16 #include <linux/udp.h>
  17 #include <net/checksum.h>
  18 #include <net/tcp.h>
  19
  20 #include <linux/netfilter_ipv4.h>
  21 #include <net/netfilter/nf_conntrack.h>
  22 #include <net/netfilter/nf_conntrack_helper.h>
  23 #include <net/netfilter/nf_conntrack_expect.h>
  24 #include <net/netfilter/nf_nat.h>
  25 #include <net/netfilter/nf_nat_protocol.h>
  26 #include <net/netfilter/nf_nat_core.h>
  27 #include <net/netfilter/nf_nat_helper.h>
  28
  29 #define DUMP_OFFSET(x) \
  30         pr_debug("offset_before=%d, offset_after=%d, correction_pos=%u\n", \
  31                  x->offset_before, x->offset_after, x->correction_pos);
  32
  33 static DEFINE_SPINLOCK(nf_nat_seqofs_lock);
  34
  35 /* Setup TCP sequence correction given this change at this sequence */
  36 static inline void
  37 adjust_tcp_sequence(u32 seq,
  38                     int sizediff,
  39                     struct nf_conn *ct,
  40                     enum ip_conntrack_info ctinfo)
  41 {
  42         int dir;
  43         struct nf_nat_seq *this_way, *other_way;
  44         struct nf_conn_nat *nat = nfct_nat(ct);
  45
  46         pr_debug("adjust_tcp_sequence: seq = %u, sizediff = %d\n",
  47                  ntohl(seq), seq);
  48
  49         dir = CTINFO2DIR(ctinfo);
  50
  51         this_way = &nat->seq[dir];
  52         other_way = &nat->seq[!dir];
  53
  54         pr_debug("nf_nat_resize_packet: Seq_offset before: ");
  55         DUMP_OFFSET(this_way);
  56
  57         spin_lock_bh(&nf_nat_seqofs_lock);
  58
  59         /* SYN adjust. If it's uninitialized, or this is after last
  60          * correction, record it: we don't handle more than one
  61          * adjustment in the window, but do deal with common case of a
  62          * retransmit */
  63         if (this_way->offset_before == this_way->offset_after ||
  64             before(this_way->correction_pos, seq)) {
  65                    this_way->correction_pos = seq;
  66                    this_way->offset_before = this_way->offset_after;
  67                    this_way->offset_after += sizediff;
  68         }
  69         spin_unlock_bh(&nf_nat_seqofs_lock);
  70
  71         pr_debug("nf_nat_resize_packet: Seq_offset after: ");
  72         DUMP_OFFSET(this_way);
  73 }
  74
  75 /* Frobs data inside this packet, which is linear. */
  76 static void mangle_contents(struct sk_buff *skb,
  77                             unsigned int dataoff,
  78                             unsigned int match_offset,
  79                             unsigned int match_len,
  80                             const char *rep_buffer,
  81                             unsigned int rep_len)
  82 {
  83         unsigned char *data;
  84
  85         BUG_ON(skb_is_nonlinear(skb));
  86         data = skb_network_header(skb) + dataoff;
  87
  88         /* move post-replacement */
  89         memmove(data + match_offset + rep_len,
  90                 data + match_offset + match_len,
  91                 skb->tail - (skb->network_header + dataoff +
  92                              match_offset + match_len));
  93
  94         /* insert data from buffer */
  95         memcpy(data + match_offset, rep_buffer, rep_len);
  96
  97         /* update skb info */
  98         if (rep_len > match_len) {
  99                 pr_debug("nf_nat_mangle_packet: Extending packet by "
 100                          "%u from %u bytes\n", rep_len - match_len, skb->len);
 101                 skb_put(skb, rep_len - match_len);
 102         } else {
 103                 pr_debug("nf_nat_mangle_packet: Shrinking packet from "
 104                          "%u from %u bytes\n", match_len - rep_len, skb->len);
 105                 __skb_trim(skb, skb->len + rep_len - match_len);
 106         }
 107
 108         /* fix IP hdr checksum information */
 109         ip_hdr(skb)->tot_len = htons(skb->len);
 110         ip_send_check(ip_hdr(skb));
 111 }
 112
 113 /* Unusual, but possible case. */
 114 static int enlarge_skb(struct sk_buff **pskb, unsigned int extra)
 115 {
 116         struct sk_buff *nskb;
 117
 118         if ((*pskb)->len + extra > 65535)
 119                 return 0;
 120
 121         nskb = skb_copy_expand(*pskb, skb_headroom(*pskb), extra, GFP_ATOMIC);
 122         if (!nskb)
 123                 return 0;
 124
 125         /* Transfer socket to new skb. */
 126         if ((*pskb)->sk)
 127                 skb_set_owner_w(nskb, (*pskb)->sk);
 128         kfree_skb(*pskb);
 129         *pskb = nskb;
 130         return 1;
 131 }
 132
 133 /* Generic function for mangling variable-length address changes inside
 134  * NATed TCP connections (like the PORT XXX,XXX,XXX,XXX,XXX,XXX
 135  * command in FTP).
 136  *
 137  * Takes care about all the nasty sequence number changes, checksumming,
 138  * skb enlargement, ...
 139  *
 140  * */
 141 int
 142 nf_nat_mangle_tcp_packet(struct sk_buff **pskb,
 143                          struct nf_conn *ct,
 144                          enum ip_conntrack_info ctinfo,
 145                          unsigned int match_offset,
 146                          unsigned int match_len,
 147                          const char *rep_buffer,
 148                          unsigned int rep_len)
 149 {
 150         struct rtable *rt = (struct rtable *)(*pskb)->dst;
 151         struct iphdr *iph;
 152         struct tcphdr *tcph;
 153         int oldlen, datalen;
 154
 155         if (!skb_make_writable(pskb, (*pskb)->len))
 156                 return 0;
 157
 158         if (rep_len > match_len &&
 159             rep_len - match_len > skb_tailroom(*pskb) &&
 160             !enlarge_skb(pskb, rep_len - match_len))
 161                 return 0;
 162
 163         SKB_LINEAR_ASSERT(*pskb);
 164
 165         iph = ip_hdr(*pskb);
 166         tcph = (void *)iph + iph->ihl*4;
 167
 168         oldlen = (*pskb)->len - iph->ihl*4;
 169         mangle_contents(*pskb, iph->ihl*4 + tcph->doff*4,
 170                         match_offset, match_len, rep_buffer, rep_len);
 171
 172         datalen = (*pskb)->len - iph->ihl*4;
 173         if ((*pskb)->ip_summed != CHECKSUM_PARTIAL) {
 174                 if (!(rt->rt_flags & RTCF_LOCAL) &&
 175                     (*pskb)->dev->features & NETIF_F_V4_CSUM) {
 176                         (*pskb)->ip_summed = CHECKSUM_PARTIAL;
 177                         (*pskb)->csum_start = skb_headroom(*pskb) +
 178                                               skb_network_offset(*pskb) +
 179                                               iph->ihl * 4;
 180                         (*pskb)->csum_offset = offsetof(struct tcphdr, check);
 181                         tcph->check = ~tcp_v4_check(datalen,
 182                                                     iph->saddr, iph->daddr, 0);
 183                 } else {
 184                         tcph->check = 0;
 185                         tcph->check = tcp_v4_check(datalen,
 186                                                    iph->saddr, iph->daddr,
 187                                                    csum_partial(tcph,
 188                                                                 datalen, 0));
 189                 }
 190         } else
 191                 nf_proto_csum_replace2(&tcph->check, *pskb,
 192                                        htons(oldlen), htons(datalen), 1);
 193
 194         if (rep_len != match_len) {
 195                 set_bit(IPS_SEQ_ADJUST_BIT, &ct->status);
 196                 adjust_tcp_sequence(ntohl(tcph->seq),
 197                                     (int)rep_len - (int)match_len,
 198                                     ct, ctinfo);
 199                 /* Tell TCP window tracking about seq change */
 200                 nf_conntrack_tcp_update(*pskb, ip_hdrlen(*pskb),
 201                                         ct, CTINFO2DIR(ctinfo));
 202         }
 203         return 1;
 204 }
 205 EXPORT_SYMBOL(nf_nat_mangle_tcp_packet);
 206
 207 /* Generic function for mangling variable-length address changes inside
 208  * NATed UDP connections (like the CONNECT DATA XXXXX MESG XXXXX INDEX XXXXX
 209  * command in the Amanda protocol)
 210  *
 211  * Takes care about all the nasty sequence number changes, checksumming,
 212  * skb enlargement, ...
 213  *
 214  * XXX - This function could be merged with nf_nat_mangle_tcp_packet which
 215  *       should be fairly easy to do.
 216  */
 217 int
 218 nf_nat_mangle_udp_packet(struct sk_buff **pskb,
 219                          struct nf_conn *ct,
 220                          enum ip_conntrack_info ctinfo,
 221                          unsigned int match_offset,
 222                          unsigned int match_len,
 223                          const char *rep_buffer,
 224                          unsigned int rep_len)
 225 {
 226         struct rtable *rt = (struct rtable *)(*pskb)->dst;
 227         struct iphdr *iph;
 228         struct udphdr *udph;
 229         int datalen, oldlen;
 230
 231         /* UDP helpers might accidentally mangle the wrong packet */
 232         iph = ip_hdr(*pskb);
 233         if ((*pskb)->len < iph->ihl*4 + sizeof(*udph) +
 234                                match_offset + match_len)
 235                 return 0;
 236
 237         if (!skb_make_writable(pskb, (*pskb)->len))
 238                 return 0;
 239
 240         if (rep_len > match_len &&
 241             rep_len - match_len > skb_tailroom(*pskb) &&
 242             !enlarge_skb(pskb, rep_len - match_len))
 243                 return 0;
 244
 245         iph = ip_hdr(*pskb);
 246         udph = (void *)iph + iph->ihl*4;
 247
 248         oldlen = (*pskb)->len - iph->ihl*4;
 249         mangle_contents(*pskb, iph->ihl*4 + sizeof(*udph),
 250                         match_offset, match_len, rep_buffer, rep_len);
 251
 252         /* update the length of the UDP packet */
 253         datalen = (*pskb)->len - iph->ihl*4;
 254         udph->len = htons(datalen);
 255
 256         /* fix udp checksum if udp checksum was previously calculated */
 257         if (!udph->check && (*pskb)->ip_summed != CHECKSUM_PARTIAL)
 258                 return 1;
 259
 260         if ((*pskb)->ip_summed != CHECKSUM_PARTIAL) {
 261                 if (!(rt->rt_flags & RTCF_LOCAL) &&
 262                     (*pskb)->dev->features & NETIF_F_V4_CSUM) {
 263                         (*pskb)->ip_summed = CHECKSUM_PARTIAL;
 264                         (*pskb)->csum_start = skb_headroom(*pskb) +
 265                                               skb_network_offset(*pskb) +
 266                                               iph->ihl * 4;
 267                         (*pskb)->csum_offset = offsetof(struct udphdr, check);
 268                         udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
 269                                                          datalen, IPPROTO_UDP,
 270                                                          0);
 271                 } else {
 272                         udph->check = 0;
 273                         udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 274                                                         datalen, IPPROTO_UDP,
 275                                                         csum_partial(udph,
 276                                                                      datalen, 0));
 277                         if (!udph->check)
 278                                 udph->check = CSUM_MANGLED_0;
 279                 }
 280         } else
 281                 nf_proto_csum_replace2(&udph->check, *pskb,
 282                                        htons(oldlen), htons(datalen), 1);
 283
 284         return 1;
 285 }
 286 EXPORT_SYMBOL(nf_nat_mangle_udp_packet);
 287
 288 /* Adjust one found SACK option including checksum correction */
 289 static void
 290 sack_adjust(struct sk_buff *skb,
 291             struct tcphdr *tcph,
 292             unsigned int sackoff,
 293             unsigned int sackend,
 294             struct nf_nat_seq *natseq)
 295 {
 296         while (sackoff < sackend) {
 297                 struct tcp_sack_block_wire *sack;
 298                 __be32 new_start_seq, new_end_seq;
 299
 300                 sack = (void *)skb->data + sackoff;
 301                 if (after(ntohl(sack->start_seq) - natseq->offset_before,
 302                           natseq->correction_pos))
 303                         new_start_seq = htonl(ntohl(sack->start_seq)
 304                                         - natseq->offset_after);
 305                 else
 306                         new_start_seq = htonl(ntohl(sack->start_seq)
 307                                         - natseq->offset_before);
 308
 309                 if (after(ntohl(sack->end_seq) - natseq->offset_before,
 310                           natseq->correction_pos))
 311                         new_end_seq = htonl(ntohl(sack->end_seq)
 312                                       - natseq->offset_after);
 313                 else
 314                         new_end_seq = htonl(ntohl(sack->end_seq)
 315                                       - natseq->offset_before);
 316
 317                 pr_debug("sack_adjust: start_seq: %d->%d, end_seq: %d->%d\n",
 318                          ntohl(sack->start_seq), new_start_seq,
 319                          ntohl(sack->end_seq), new_end_seq);
 320
 321                 nf_proto_csum_replace4(&tcph->check, skb,
 322                                        sack->start_seq, new_start_seq, 0);
 323                 nf_proto_csum_replace4(&tcph->check, skb,
 324                                        sack->end_seq, new_end_seq, 0);
 325                 sack->start_seq = new_start_seq;
 326                 sack->end_seq = new_end_seq;
 327                 sackoff += sizeof(*sack);
 328         }
 329 }
 330
 331 /* TCP SACK sequence number adjustment */
 332 static inline unsigned int
 333 nf_nat_sack_adjust(struct sk_buff **pskb,
 334                    struct tcphdr *tcph,
 335                    struct nf_conn *ct,
 336                    enum ip_conntrack_info ctinfo)
 337 {
 338         unsigned int dir, optoff, optend;
 339         struct nf_conn_nat *nat = nfct_nat(ct);
 340
 341         optoff = ip_hdrlen(*pskb) + sizeof(struct tcphdr);
 342         optend = ip_hdrlen(*pskb) + tcph->doff * 4;
 343
 344         if (!skb_make_writable(pskb, optend))
 345                 return 0;
 346
 347         dir = CTINFO2DIR(ctinfo);
 348
 349         while (optoff < optend) {
 350                 /* Usually: option, length. */
 351                 unsigned char *op = (*pskb)->data + optoff;
 352
 353                 switch (op[0]) {
 354                 case TCPOPT_EOL:
 355                         return 1;
 356                 case TCPOPT_NOP:
 357                         optoff++;
 358                         continue;
 359                 default:
 360                         /* no partial options */
 361                         if (optoff + 1 == optend ||
 362                             optoff + op[1] > optend ||
 363                             op[1] < 2)
 364                                 return 0;
 365                         if (op[0] == TCPOPT_SACK &&
 366                             op[1] >= 2+TCPOLEN_SACK_PERBLOCK &&
 367                             ((op[1] - 2) % TCPOLEN_SACK_PERBLOCK) == 0)
 368                                 sack_adjust(*pskb, tcph, optoff+2,
 369                                             optoff+op[1], &nat->seq[!dir]);
 370                         optoff += op[1];
 371                 }
 372         }
 373         return 1;
 374 }
 375
 376 /* TCP sequence number adjustment.  Returns 1 on success, 0 on failure */
 377 int
 378 nf_nat_seq_adjust(struct sk_buff **pskb,
 379                   struct nf_conn *ct,
 380                   enum ip_conntrack_info ctinfo)
 381 {
 382         struct tcphdr *tcph;
 383         int dir;
 384         __be32 newseq, newack;
 385         struct nf_conn_nat *nat = nfct_nat(ct);
 386         struct nf_nat_seq *this_way, *other_way;
 387
 388         dir = CTINFO2DIR(ctinfo);
 389
 390         this_way = &nat->seq[dir];
 391         other_way = &nat->seq[!dir];
 392
 393         if (!skb_make_writable(pskb, ip_hdrlen(*pskb) + sizeof(*tcph)))
 394                 return 0;
 395
 396         tcph = (void *)(*pskb)->data + ip_hdrlen(*pskb);
 397         if (after(ntohl(tcph->seq), this_way->correction_pos))
 398                 newseq = htonl(ntohl(tcph->seq) + this_way->offset_after);
 399         else
 400                 newseq = htonl(ntohl(tcph->seq) + this_way->offset_before);
 401
 402         if (after(ntohl(tcph->ack_seq) - other_way->offset_before,
 403                   other_way->correction_pos))
 404                 newack = htonl(ntohl(tcph->ack_seq) - other_way->offset_after);
 405         else
 406                 newack = htonl(ntohl(tcph->ack_seq) - other_way->offset_before);
 407
 408         nf_proto_csum_replace4(&tcph->check, *pskb, tcph->seq, newseq, 0);
 409         nf_proto_csum_replace4(&tcph->check, *pskb, tcph->ack_seq, newack, 0);
 410
 411         pr_debug("Adjusting sequence number from %u->%u, ack from %u->%u\n",
 412                  ntohl(tcph->seq), ntohl(newseq), ntohl(tcph->ack_seq),
 413                  ntohl(newack));
 414
 415         tcph->seq = newseq;
 416         tcph->ack_seq = newack;
 417
 418         if (!nf_nat_sack_adjust(pskb, tcph, ct, ctinfo))
 419                 return 0;
 420
 421         nf_conntrack_tcp_update(*pskb, ip_hdrlen(*pskb), ct, dir);
 422
 423         return 1;
 424 }
 425 EXPORT_SYMBOL(nf_nat_seq_adjust);
 426
 427 /* Setup NAT on this expected conntrack so it follows master. */
 428 /* If we fail to get a free NAT slot, we'll get dropped on confirm */
 429 void nf_nat_follow_master(struct nf_conn *ct,
 430                           struct nf_conntrack_expect *exp)
 431 {
 432         struct nf_nat_range range;
 433
 434         /* This must be a fresh one. */
 435         BUG_ON(ct->status & IPS_NAT_DONE_MASK);
 436
 437         /* Change src to where master sends to */
 438         range.flags = IP_NAT_RANGE_MAP_IPS;
 439         range.min_ip = range.max_ip
 440                 = ct->master->tuplehash[!exp->dir].tuple.dst.u3.ip;
 441         /* hook doesn't matter, but it has to do source manip */
 442         nf_nat_setup_info(ct, &range, NF_IP_POST_ROUTING);
 443
 444         /* For DST manip, map port here to where it's expected. */
 445         range.flags = (IP_NAT_RANGE_MAP_IPS | IP_NAT_RANGE_PROTO_SPECIFIED);
 446         range.min = range.max = exp->saved_proto;
 447         range.min_ip = range.max_ip
 448                 = ct->master->tuplehash[!exp->dir].tuple.src.u3.ip;
 449         /* hook doesn't matter, but it has to do destination manip */
 450         nf_nat_setup_info(ct, &range, NF_IP_PRE_ROUTING);
 451 }
 452 EXPORT_SYMBOL(nf_nat_follow_master);