net/ipv4/netfilter/ip_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) 2003-2004 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  *      14 Jan 2002 Harald Welte <laforge@gnumonks.org>:
  11  *              - add support for SACK adjustment
  12  *      14 Mar 2002 Harald Welte <laforge@gnumonks.org>:
  13  *              - merge SACK support into newnat API
  14  *      16 Aug 2002 Brian J. Murrell <netfilter@interlinx.bc.ca>:
  15  *              - make ip_nat_resize_packet more generic (TCP and UDP)
  16  *              - add ip_nat_mangle_udp_packet
  17  */
  18 #include <linux/module.h>
  19 #include <linux/kmod.h>
  20 #include <linux/types.h>
  21 #include <linux/timer.h>
  22 #include <linux/skbuff.h>
  23 #include <linux/netfilter_ipv4.h>
  24 #include <net/checksum.h>
  25 #include <net/icmp.h>
  26 #include <net/ip.h>
  27 #include <net/tcp.h>
  28 #include <net/udp.h>
  29
  30 #define ASSERT_READ_LOCK(x)
  31 #define ASSERT_WRITE_LOCK(x)
  32
  33 #include <linux/netfilter_ipv4/ip_conntrack.h>
  34 #include <linux/netfilter_ipv4/ip_conntrack_helper.h>
  35 #include <linux/netfilter_ipv4/ip_nat.h>
  36 #include <linux/netfilter_ipv4/ip_nat_protocol.h>
  37 #include <linux/netfilter_ipv4/ip_nat_core.h>
  38 #include <linux/netfilter_ipv4/ip_nat_helper.h>
  39 #include <linux/netfilter_ipv4/listhelp.h>
  40
  41 #if 0
  42 #define DEBUGP printk
  43 #define DUMP_OFFSET(x)  printk("offset_before=%d, offset_after=%d, correction_pos=%u\n", x->offset_before, x->offset_after, x->correction_pos);
  44 #else
  45 #define DEBUGP(format, args...)
  46 #define DUMP_OFFSET(x)
  47 #endif
  48
  49 static DEFINE_SPINLOCK(ip_nat_seqofs_lock);
  50
  51 /* Setup TCP sequence correction given this change at this sequence */
  52 static inline void
  53 adjust_tcp_sequence(u32 seq,
  54                     int sizediff,
  55                     struct ip_conntrack *ct,
  56                     enum ip_conntrack_info ctinfo)
  57 {
  58         int dir;
  59         struct ip_nat_seq *this_way, *other_way;
  60
  61         DEBUGP("ip_nat_resize_packet: old_size = %u, new_size = %u\n",
  62                 (*skb)->len, new_size);
  63
  64         dir = CTINFO2DIR(ctinfo);
  65
  66         this_way = &ct->nat.info.seq[dir];
  67         other_way = &ct->nat.info.seq[!dir];
  68
  69         DEBUGP("ip_nat_resize_packet: Seq_offset before: ");
  70         DUMP_OFFSET(this_way);
  71
  72         spin_lock_bh(&ip_nat_seqofs_lock);
  73
  74         /* SYN adjust. If it's uninitialized, or this is after last
  75          * correction, record it: we don't handle more than one
  76          * adjustment in the window, but do deal with common case of a
  77          * retransmit */
  78         if (this_way->offset_before == this_way->offset_after
  79             || before(this_way->correction_pos, seq)) {
  80                     this_way->correction_pos = seq;
  81                     this_way->offset_before = this_way->offset_after;
  82                     this_way->offset_after += sizediff;
  83         }
  84         spin_unlock_bh(&ip_nat_seqofs_lock);
  85
  86         DEBUGP("ip_nat_resize_packet: Seq_offset after: ");
  87         DUMP_OFFSET(this_way);
  88 }
  89
  90 /* Frobs data inside this packet, which is linear. */
  91 static void mangle_contents(struct sk_buff *skb,
  92                             unsigned int dataoff,
  93                             unsigned int match_offset,
  94                             unsigned int match_len,
  95                             const char *rep_buffer,
  96                             unsigned int rep_len)
  97 {
  98         unsigned char *data;
  99
 100         BUG_ON(skb_is_nonlinear(skb));
 101         data = (unsigned char *)skb->nh.iph + dataoff;
 102
 103         /* move post-replacement */
 104         memmove(data + match_offset + rep_len,
 105                 data + match_offset + match_len,
 106                 skb->tail - (data + match_offset + match_len));
 107
 108         /* insert data from buffer */
 109         memcpy(data + match_offset, rep_buffer, rep_len);
 110
 111         /* update skb info */
 112         if (rep_len > match_len) {
 113                 DEBUGP("ip_nat_mangle_packet: Extending packet by "
 114                         "%u from %u bytes\n", rep_len - match_len,
 115                        skb->len);
 116                 skb_put(skb, rep_len - match_len);
 117         } else {
 118                 DEBUGP("ip_nat_mangle_packet: Shrinking packet from "
 119                         "%u from %u bytes\n", match_len - rep_len,
 120                        skb->len);
 121                 __skb_trim(skb, skb->len + rep_len - match_len);
 122         }
 123
 124         /* fix IP hdr checksum information */
 125         skb->nh.iph->tot_len = htons(skb->len);
 126         ip_send_check(skb->nh.iph);
 127 }
 128
 129 /* Unusual, but possible case. */
 130 static int enlarge_skb(struct sk_buff **pskb, unsigned int extra)
 131 {
 132         struct sk_buff *nskb;
 133
 134         if ((*pskb)->len + extra > 65535)
 135                 return 0;
 136
 137         nskb = skb_copy_expand(*pskb, skb_headroom(*pskb), extra, GFP_ATOMIC);
 138         if (!nskb)
 139                 return 0;
 140
 141         /* Transfer socket to new skb. */
 142         if ((*pskb)->sk)
 143                 skb_set_owner_w(nskb, (*pskb)->sk);
 144         kfree_skb(*pskb);
 145         *pskb = nskb;
 146         return 1;
 147 }
 148
 149 /* Generic function for mangling variable-length address changes inside
 150  * NATed TCP connections (like the PORT XXX,XXX,XXX,XXX,XXX,XXX
 151  * command in FTP).
 152  *
 153  * Takes care about all the nasty sequence number changes, checksumming,
 154  * skb enlargement, ...
 155  *
 156  * */
 157 int
 158 ip_nat_mangle_tcp_packet(struct sk_buff **pskb,
 159                          struct ip_conntrack *ct,
 160                          enum ip_conntrack_info ctinfo,
 161                          unsigned int match_offset,
 162                          unsigned int match_len,
 163                          const char *rep_buffer,
 164                          unsigned int rep_len)
 165 {
 166         struct iphdr *iph;
 167         struct tcphdr *tcph;
 168         int oldlen, datalen;
 169
 170         if (!skb_make_writable(pskb, (*pskb)->len))
 171                 return 0;
 172
 173         if (rep_len > match_len
 174             && rep_len - match_len > skb_tailroom(*pskb)
 175             && !enlarge_skb(pskb, rep_len - match_len))
 176                 return 0;
 177
 178         SKB_LINEAR_ASSERT(*pskb);
 179
 180         iph = (*pskb)->nh.iph;
 181         tcph = (void *)iph + iph->ihl*4;
 182
 183         oldlen = (*pskb)->len - iph->ihl*4;
 184         mangle_contents(*pskb, iph->ihl*4 + tcph->doff*4,
 185                         match_offset, match_len, rep_buffer, rep_len);
 186
 187         datalen = (*pskb)->len - iph->ihl*4;
 188         if ((*pskb)->ip_summed != CHECKSUM_PARTIAL) {
 189                 tcph->check = 0;
 190                 tcph->check = tcp_v4_check(tcph, datalen,
 191                                            iph->saddr, iph->daddr,
 192                                            csum_partial((char *)tcph,
 193                                                         datalen, 0));
 194         } else
 195                 tcph->check = nf_proto_csum_update(*pskb,
 196                                                    htons(oldlen) ^ 0xFFFF,
 197                                                    htons(datalen),
 198                                                    tcph->check, 1);
 199
 200         if (rep_len != match_len) {
 201                 set_bit(IPS_SEQ_ADJUST_BIT, &ct->status);
 202                 adjust_tcp_sequence(ntohl(tcph->seq),
 203                                     (int)rep_len - (int)match_len,
 204                                     ct, ctinfo);
 205                 /* Tell TCP window tracking about seq change */
 206                 ip_conntrack_tcp_update(*pskb, ct, CTINFO2DIR(ctinfo));
 207         }
 208         return 1;
 209 }
 210 EXPORT_SYMBOL(ip_nat_mangle_tcp_packet);
 211
 212 /* Generic function for mangling variable-length address changes inside
 213  * NATed UDP connections (like the CONNECT DATA XXXXX MESG XXXXX INDEX XXXXX
 214  * command in the Amanda protocol)
 215  *
 216  * Takes care about all the nasty sequence number changes, checksumming,
 217  * skb enlargement, ...
 218  *
 219  * XXX - This function could be merged with ip_nat_mangle_tcp_packet which
 220  *       should be fairly easy to do.
 221  */
 222 int
 223 ip_nat_mangle_udp_packet(struct sk_buff **pskb,
 224                          struct ip_conntrack *ct,
 225                          enum ip_conntrack_info ctinfo,
 226                          unsigned int match_offset,
 227                          unsigned int match_len,
 228                          const char *rep_buffer,
 229                          unsigned int rep_len)
 230 {
 231         struct iphdr *iph;
 232         struct udphdr *udph;
 233         int datalen, oldlen;
 234
 235         /* UDP helpers might accidentally mangle the wrong packet */
 236         iph = (*pskb)->nh.iph;
 237         if ((*pskb)->len < iph->ihl*4 + sizeof(*udph) +
 238                                match_offset + match_len)
 239                 return 0;
 240
 241         if (!skb_make_writable(pskb, (*pskb)->len))
 242                 return 0;
 243
 244         if (rep_len > match_len
 245             && rep_len - match_len > skb_tailroom(*pskb)
 246             && !enlarge_skb(pskb, rep_len - match_len))
 247                 return 0;
 248
 249         iph = (*pskb)->nh.iph;
 250         udph = (void *)iph + iph->ihl*4;
 251
 252         oldlen = (*pskb)->len - iph->ihl*4;
 253         mangle_contents(*pskb, iph->ihl*4 + sizeof(*udph),
 254                         match_offset, match_len, rep_buffer, rep_len);
 255
 256         /* update the length of the UDP packet */
 257         datalen = (*pskb)->len - iph->ihl*4;
 258         udph->len = htons(datalen);
 259
 260         /* fix udp checksum if udp checksum was previously calculated */
 261         if (!udph->check && (*pskb)->ip_summed != CHECKSUM_PARTIAL)
 262                 return 1;
 263
 264         if ((*pskb)->ip_summed != CHECKSUM_PARTIAL) {
 265                 udph->check = 0;
 266                 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 267                                                 datalen, IPPROTO_UDP,
 268                                                 csum_partial((char *)udph,
 269                                                              datalen, 0));
 270                 if (!udph->check)
 271                         udph->check = -1;
 272         } else
 273                 udph->check = nf_proto_csum_update(*pskb,
 274                                                    htons(oldlen) ^ 0xFFFF,
 275                                                    htons(datalen),
 276                                                    udph->check, 1);
 277         return 1;
 278 }
 279 EXPORT_SYMBOL(ip_nat_mangle_udp_packet);
 280
 281 /* Adjust one found SACK option including checksum correction */
 282 static void
 283 sack_adjust(struct sk_buff *skb,
 284             struct tcphdr *tcph,
 285             unsigned int sackoff,
 286             unsigned int sackend,
 287             struct ip_nat_seq *natseq)
 288 {
 289         while (sackoff < sackend) {
 290                 struct tcp_sack_block *sack;
 291                 u_int32_t new_start_seq, new_end_seq;
 292
 293                 sack = (void *)skb->data + sackoff;
 294                 if (after(ntohl(sack->start_seq) - natseq->offset_before,
 295                           natseq->correction_pos))
 296                         new_start_seq = ntohl(sack->start_seq)
 297                                         - natseq->offset_after;
 298                 else
 299                         new_start_seq = ntohl(sack->start_seq)
 300                                         - natseq->offset_before;
 301                 new_start_seq = htonl(new_start_seq);
 302
 303                 if (after(ntohl(sack->end_seq) - natseq->offset_before,
 304                           natseq->correction_pos))
 305                         new_end_seq = ntohl(sack->end_seq)
 306                                       - natseq->offset_after;
 307                 else
 308                         new_end_seq = ntohl(sack->end_seq)
 309                                       - natseq->offset_before;
 310                 new_end_seq = htonl(new_end_seq);
 311
 312                 DEBUGP("sack_adjust: start_seq: %d->%d, end_seq: %d->%d\n",
 313                         ntohl(sack->start_seq), new_start_seq,
 314                         ntohl(sack->end_seq), new_end_seq);
 315
 316                 tcph->check = nf_proto_csum_update(skb,
 317                                                    ~sack->start_seq,
 318                                                    new_start_seq,
 319                                                    tcph->check, 0);
 320                 tcph->check = nf_proto_csum_update(skb,
 321                                                    ~sack->end_seq,
 322                                                    new_end_seq,
 323                                                    tcph->check, 0);
 324                 sack->start_seq = new_start_seq;
 325                 sack->end_seq = new_end_seq;
 326                 sackoff += sizeof(*sack);
 327         }
 328 }
 329
 330 /* TCP SACK sequence number adjustment */
 331 static inline unsigned int
 332 ip_nat_sack_adjust(struct sk_buff **pskb,
 333                    struct tcphdr *tcph,
 334                    struct ip_conntrack *ct,
 335                    enum ip_conntrack_info ctinfo)
 336 {
 337         unsigned int dir, optoff, optend;
 338
 339         optoff = (*pskb)->nh.iph->ihl*4 + sizeof(struct tcphdr);
 340         optend = (*pskb)->nh.iph->ihl*4 + tcph->doff*4;
 341
 342         if (!skb_make_writable(pskb, optend))
 343                 return 0;
 344
 345         dir = CTINFO2DIR(ctinfo);
 346
 347         while (optoff < optend) {
 348                 /* Usually: option, length. */
 349                 unsigned char *op = (*pskb)->data + optoff;
 350
 351                 switch (op[0]) {
 352                 case TCPOPT_EOL:
 353                         return 1;
 354                 case TCPOPT_NOP:
 355                         optoff++;
 356                         continue;
 357                 default:
 358                         /* no partial options */
 359                         if (optoff + 1 == optend
 360                             || optoff + op[1] > optend
 361                             || op[1] < 2)
 362                                 return 0;
 363                         if (op[0] == TCPOPT_SACK
 364                             && op[1] >= 2+TCPOLEN_SACK_PERBLOCK
 365                             && ((op[1] - 2) % TCPOLEN_SACK_PERBLOCK) == 0)
 366                                 sack_adjust(*pskb, tcph, optoff+2,
 367                                             optoff+op[1],
 368                                             &ct->nat.info.seq[!dir]);
 369                         optoff += op[1];
 370                 }
 371         }
 372         return 1;
 373 }
 374
 375 /* TCP sequence number adjustment.  Returns 1 on success, 0 on failure */
 376 int
 377 ip_nat_seq_adjust(struct sk_buff **pskb,
 378                   struct ip_conntrack *ct,
 379                   enum ip_conntrack_info ctinfo)
 380 {
 381         struct tcphdr *tcph;
 382         int dir, newseq, newack;
 383         struct ip_nat_seq *this_way, *other_way;
 384
 385         dir = CTINFO2DIR(ctinfo);
 386
 387         this_way = &ct->nat.info.seq[dir];
 388         other_way = &ct->nat.info.seq[!dir];
 389
 390         if (!skb_make_writable(pskb, (*pskb)->nh.iph->ihl*4+sizeof(*tcph)))
 391                 return 0;
 392
 393         tcph = (void *)(*pskb)->data + (*pskb)->nh.iph->ihl*4;
 394         if (after(ntohl(tcph->seq), this_way->correction_pos))
 395                 newseq = ntohl(tcph->seq) + this_way->offset_after;
 396         else
 397                 newseq = ntohl(tcph->seq) + this_way->offset_before;
 398         newseq = htonl(newseq);
 399
 400         if (after(ntohl(tcph->ack_seq) - other_way->offset_before,
 401                   other_way->correction_pos))
 402                 newack = ntohl(tcph->ack_seq) - other_way->offset_after;
 403         else
 404                 newack = ntohl(tcph->ack_seq) - other_way->offset_before;
 405         newack = htonl(newack);
 406
 407         tcph->check = nf_proto_csum_update(*pskb, ~tcph->seq, newseq,
 408                                            tcph->check, 0);
 409         tcph->check = nf_proto_csum_update(*pskb, ~tcph->ack_seq, newack,
 410                                            tcph->check, 0);
 411
 412         DEBUGP("Adjusting sequence number from %u->%u, ack from %u->%u\n",
 413                 ntohl(tcph->seq), ntohl(newseq), ntohl(tcph->ack_seq),
 414                 ntohl(newack));
 415
 416         tcph->seq = newseq;
 417         tcph->ack_seq = newack;
 418
 419         if (!ip_nat_sack_adjust(pskb, tcph, ct, ctinfo))
 420                 return 0;
 421
 422         ip_conntrack_tcp_update(*pskb, ct, dir);
 423
 424         return 1;
 425 }
 426 EXPORT_SYMBOL(ip_nat_seq_adjust);
 427
 428 /* Setup NAT on this expected conntrack so it follows master. */
 429 /* If we fail to get a free NAT slot, we'll get dropped on confirm */
 430 void ip_nat_follow_master(struct ip_conntrack *ct,
 431                           struct ip_conntrack_expect *exp)
 432 {
 433         struct ip_nat_range range;
 434
 435         /* This must be a fresh one. */
 436         BUG_ON(ct->status & IPS_NAT_DONE_MASK);
 437
 438         /* Change src to where master sends to */
 439         range.flags = IP_NAT_RANGE_MAP_IPS;
 440         range.min_ip = range.max_ip
 441                 = ct->master->tuplehash[!exp->dir].tuple.dst.ip;
 442         /* hook doesn't matter, but it has to do source manip */
 443         ip_nat_setup_info(ct, &range, NF_IP_POST_ROUTING);
 444
 445         /* For DST manip, map port here to where it's expected. */
 446         range.flags = (IP_NAT_RANGE_MAP_IPS | IP_NAT_RANGE_PROTO_SPECIFIED);
 447         range.min = range.max = exp->saved_proto;
 448         range.min_ip = range.max_ip
 449                 = ct->master->tuplehash[!exp->dir].tuple.src.ip;
 450         /* hook doesn't matter, but it has to do destination manip */
 451         ip_nat_setup_info(ct, &range, NF_IP_PRE_ROUTING);
 452 }
 453 EXPORT_SYMBOL(ip_nat_follow_master);