git.oblomov.eu Git - linux-2.6/blob - 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 #include <linux/netfilter_ipv4/ip_conntrack.h>
  31 #include <linux/netfilter_ipv4/ip_conntrack_helper.h>
  32 #include <linux/netfilter_ipv4/ip_nat.h>
  33 #include <linux/netfilter_ipv4/ip_nat_protocol.h>
  34 #include <linux/netfilter_ipv4/ip_nat_core.h>
  35 #include <linux/netfilter_ipv4/ip_nat_helper.h>
  36
  37 #if 0
  38 #define DEBUGP printk
  39 #define DUMP_OFFSET(x)  printk("offset_before=%d, offset_after=%d, correction_pos=%u\n", x->offset_before, x->offset_after, x->correction_pos);
  40 #else
  41 #define DEBUGP(format, args...)
  42 #define DUMP_OFFSET(x)
  43 #endif
  44
  45 static DEFINE_SPINLOCK(ip_nat_seqofs_lock);
  46
  47 /* Setup TCP sequence correction given this change at this sequence */
  48 static inline void
  49 adjust_tcp_sequence(u32 seq,
  50                     int sizediff,
  51                     struct ip_conntrack *ct,
  52                     enum ip_conntrack_info ctinfo)
  53 {
  54         int dir;
  55         struct ip_nat_seq *this_way, *other_way;
  56
  57         DEBUGP("ip_nat_resize_packet: old_size = %u, new_size = %u\n",
  58                 (*skb)->len, new_size);
  59
  60         dir = CTINFO2DIR(ctinfo);
  61
  62         this_way = &ct->nat.info.seq[dir];
  63         other_way = &ct->nat.info.seq[!dir];
  64
  65         DEBUGP("ip_nat_resize_packet: Seq_offset before: ");
  66         DUMP_OFFSET(this_way);
  67
  68         spin_lock_bh(&ip_nat_seqofs_lock);
  69
  70         /* SYN adjust. If it's uninitialized, or this is after last
  71          * correction, record it: we don't handle more than one
  72          * adjustment in the window, but do deal with common case of a
  73          * retransmit */
  74         if (this_way->offset_before == this_way->offset_after
  75             || before(this_way->correction_pos, seq)) {
  76                     this_way->correction_pos = seq;
  77                     this_way->offset_before = this_way->offset_after;
  78                     this_way->offset_after += sizediff;
  79         }
  80         spin_unlock_bh(&ip_nat_seqofs_lock);
  81
  82         DEBUGP("ip_nat_resize_packet: Seq_offset after: ");
  83         DUMP_OFFSET(this_way);
  84 }
  85
  86 /* Frobs data inside this packet, which is linear. */
  87 static void mangle_contents(struct sk_buff *skb,
  88                             unsigned int dataoff,
  89                             unsigned int match_offset,
  90                             unsigned int match_len,
  91                             const char *rep_buffer,
  92                             unsigned int rep_len)
  93 {
  94         unsigned char *data;
  95
  96         BUG_ON(skb_is_nonlinear(skb));
  97         data = (unsigned char *)skb->nh.iph + dataoff;
  98
  99         /* move post-replacement */
 100         memmove(data + match_offset + rep_len,
 101                 data + match_offset + match_len,
 102                 skb->tail - (data + match_offset + match_len));
 103
 104         /* insert data from buffer */
 105         memcpy(data + match_offset, rep_buffer, rep_len);
 106
 107         /* update skb info */
 108         if (rep_len > match_len) {
 109                 DEBUGP("ip_nat_mangle_packet: Extending packet by "
 110                         "%u from %u bytes\n", rep_len - match_len,
 111                        skb->len);
 112                 skb_put(skb, rep_len - match_len);
 113         } else {
 114                 DEBUGP("ip_nat_mangle_packet: Shrinking packet from "
 115                         "%u from %u bytes\n", match_len - rep_len,
 116                        skb->len);
 117                 __skb_trim(skb, skb->len + rep_len - match_len);
 118         }
 119
 120         /* fix IP hdr checksum information */
 121         skb->nh.iph->tot_len = htons(skb->len);
 122         ip_send_check(skb->nh.iph);
 123 }
 124
 125 /* Unusual, but possible case. */
 126 static int enlarge_skb(struct sk_buff **pskb, unsigned int extra)
 127 {
 128         struct sk_buff *nskb;
 129
 130         if ((*pskb)->len + extra > 65535)
 131                 return 0;
 132
 133         nskb = skb_copy_expand(*pskb, skb_headroom(*pskb), extra, GFP_ATOMIC);
 134         if (!nskb)
 135                 return 0;
 136
 137         /* Transfer socket to new skb. */
 138         if ((*pskb)->sk)
 139                 skb_set_owner_w(nskb, (*pskb)->sk);
 140         kfree_skb(*pskb);
 141         *pskb = nskb;
 142         return 1;
 143 }
 144
 145 /* Generic function for mangling variable-length address changes inside
 146  * NATed TCP connections (like the PORT XXX,XXX,XXX,XXX,XXX,XXX
 147  * command in FTP).
 148  *
 149  * Takes care about all the nasty sequence number changes, checksumming,
 150  * skb enlargement, ...
 151  *
 152  * */
 153 int
 154 ip_nat_mangle_tcp_packet(struct sk_buff **pskb,
 155                          struct ip_conntrack *ct,
 156                          enum ip_conntrack_info ctinfo,
 157                          unsigned int match_offset,
 158                          unsigned int match_len,
 159                          const char *rep_buffer,
 160                          unsigned int rep_len)
 161 {
 162         struct iphdr *iph;
 163         struct tcphdr *tcph;
 164         int oldlen, datalen;
 165
 166         if (!skb_make_writable(pskb, (*pskb)->len))
 167                 return 0;
 168
 169         if (rep_len > match_len
 170             && rep_len - match_len > skb_tailroom(*pskb)
 171             && !enlarge_skb(pskb, rep_len - match_len))
 172                 return 0;
 173
 174         SKB_LINEAR_ASSERT(*pskb);
 175
 176         iph = (*pskb)->nh.iph;
 177         tcph = (void *)iph + iph->ihl*4;
 178
 179         oldlen = (*pskb)->len - iph->ihl*4;
 180         mangle_contents(*pskb, iph->ihl*4 + tcph->doff*4,
 181                         match_offset, match_len, rep_buffer, rep_len);
 182
 183         datalen = (*pskb)->len - iph->ihl*4;
 184         if ((*pskb)->ip_summed != CHECKSUM_PARTIAL) {
 185                 tcph->check = 0;
 186                 tcph->check = tcp_v4_check(tcph, datalen,
 187                                            iph->saddr, iph->daddr,
 188                                            csum_partial((char *)tcph,
 189                                                         datalen, 0));
 190         } else
 191                 tcph->check = nf_proto_csum_update(*pskb,
 192                                                    htons(oldlen) ^ htons(0xFFFF),
 193                                                    htons(datalen),
 194                                                    tcph->check, 1);
 195
 196         if (rep_len != match_len) {
 197                 set_bit(IPS_SEQ_ADJUST_BIT, &ct->status);
 198                 adjust_tcp_sequence(ntohl(tcph->seq),
 199                                     (int)rep_len - (int)match_len,
 200                                     ct, ctinfo);
 201                 /* Tell TCP window tracking about seq change */
 202                 ip_conntrack_tcp_update(*pskb, ct, CTINFO2DIR(ctinfo));
 203         }
 204         return 1;
 205 }
 206 EXPORT_SYMBOL(ip_nat_mangle_tcp_packet);
 207
 208 /* Generic function for mangling variable-length address changes inside
 209  * NATed UDP connections (like the CONNECT DATA XXXXX MESG XXXXX INDEX XXXXX
 210  * command in the Amanda protocol)
 211  *
 212  * Takes care about all the nasty sequence number changes, checksumming,
 213  * skb enlargement, ...
 214  *
 215  * XXX - This function could be merged with ip_nat_mangle_tcp_packet which
 216  *       should be fairly easy to do.
 217  */
 218 int
 219 ip_nat_mangle_udp_packet(struct sk_buff **pskb,
 220                          struct ip_conntrack *ct,
 221                          enum ip_conntrack_info ctinfo,
 222                          unsigned int match_offset,
 223                          unsigned int match_len,
 224                          const char *rep_buffer,
 225                          unsigned int rep_len)
 226 {
 227         struct iphdr *iph;
 228         struct udphdr *udph;
 229         int datalen, oldlen;
 230
 231         /* UDP helpers might accidentally mangle the wrong packet */
 232         iph = (*pskb)->nh.iph;
 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 = (*pskb)->nh.iph;
 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                 udph->check = 0;
 262                 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 263                                                 datalen, IPPROTO_UDP,
 264                                                 csum_partial((char *)udph,
 265                                                              datalen, 0));
 266                 if (!udph->check)
 267                         udph->check = -1;
 268         } else
 269                 udph->check = nf_proto_csum_update(*pskb,
 270                                                    htons(oldlen) ^ htons(0xFFFF),
 271                                                    htons(datalen),
 272                                                    udph->check, 1);
 273         return 1;
 274 }
 275 EXPORT_SYMBOL(ip_nat_mangle_udp_packet);
 276
 277 /* Adjust one found SACK option including checksum correction */
 278 static void
 279 sack_adjust(struct sk_buff *skb,
 280             struct tcphdr *tcph,
 281             unsigned int sackoff,
 282             unsigned int sackend,
 283             struct ip_nat_seq *natseq)
 284 {
 285         while (sackoff < sackend) {
 286                 struct tcp_sack_block_wire *sack;
 287                 __be32 new_start_seq, new_end_seq;
 288
 289                 sack = (void *)skb->data + sackoff;
 290                 if (after(ntohl(sack->start_seq) - natseq->offset_before,
 291                           natseq->correction_pos))
 292                         new_start_seq = htonl(ntohl(sack->start_seq)
 293                                         - natseq->offset_after);
 294                 else
 295                         new_start_seq = htonl(ntohl(sack->start_seq)
 296                                         - natseq->offset_before);
 297
 298                 if (after(ntohl(sack->end_seq) - natseq->offset_before,
 299                           natseq->correction_pos))
 300                         new_end_seq = htonl(ntohl(sack->end_seq)
 301                                       - natseq->offset_after);
 302                 else
 303                         new_end_seq = htonl(ntohl(sack->end_seq)
 304                                       - natseq->offset_before);
 305
 306                 DEBUGP("sack_adjust: start_seq: %d->%d, end_seq: %d->%d\n",
 307                         ntohl(sack->start_seq), new_start_seq,
 308                         ntohl(sack->end_seq), new_end_seq);
 309
 310                 tcph->check = nf_proto_csum_update(skb,
 311                                                    ~sack->start_seq,
 312                                                    new_start_seq,
 313                                                    tcph->check, 0);
 314                 tcph->check = nf_proto_csum_update(skb,
 315                                                    ~sack->end_seq,
 316                                                    new_end_seq,
 317                                                    tcph->check, 0);
 318                 sack->start_seq = new_start_seq;
 319                 sack->end_seq = new_end_seq;
 320                 sackoff += sizeof(*sack);
 321         }
 322 }
 323
 324 /* TCP SACK sequence number adjustment */
 325 static inline unsigned int
 326 ip_nat_sack_adjust(struct sk_buff **pskb,
 327                    struct tcphdr *tcph,
 328                    struct ip_conntrack *ct,
 329                    enum ip_conntrack_info ctinfo)
 330 {
 331         unsigned int dir, optoff, optend;
 332
 333         optoff = (*pskb)->nh.iph->ihl*4 + sizeof(struct tcphdr);
 334         optend = (*pskb)->nh.iph->ihl*4 + tcph->doff*4;
 335
 336         if (!skb_make_writable(pskb, optend))
 337                 return 0;
 338
 339         dir = CTINFO2DIR(ctinfo);
 340
 341         while (optoff < optend) {
 342                 /* Usually: option, length. */
 343                 unsigned char *op = (*pskb)->data + optoff;
 344
 345                 switch (op[0]) {
 346                 case TCPOPT_EOL:
 347                         return 1;
 348                 case TCPOPT_NOP:
 349                         optoff++;
 350                         continue;
 351                 default:
 352                         /* no partial options */
 353                         if (optoff + 1 == optend
 354                             || optoff + op[1] > optend
 355                             || op[1] < 2)
 356                                 return 0;
 357                         if (op[0] == TCPOPT_SACK
 358                             && op[1] >= 2+TCPOLEN_SACK_PERBLOCK
 359                             && ((op[1] - 2) % TCPOLEN_SACK_PERBLOCK) == 0)
 360                                 sack_adjust(*pskb, tcph, optoff+2,
 361                                             optoff+op[1],
 362                                             &ct->nat.info.seq[!dir]);
 363                         optoff += op[1];
 364                 }
 365         }
 366         return 1;
 367 }
 368
 369 /* TCP sequence number adjustment.  Returns 1 on success, 0 on failure */
 370 int
 371 ip_nat_seq_adjust(struct sk_buff **pskb,
 372                   struct ip_conntrack *ct,
 373                   enum ip_conntrack_info ctinfo)
 374 {
 375         struct tcphdr *tcph;
 376         int dir;
 377         __be32 newseq, newack;
 378         struct ip_nat_seq *this_way, *other_way;
 379
 380         dir = CTINFO2DIR(ctinfo);
 381
 382         this_way = &ct->nat.info.seq[dir];
 383         other_way = &ct->nat.info.seq[!dir];
 384
 385         if (!skb_make_writable(pskb, (*pskb)->nh.iph->ihl*4+sizeof(*tcph)))
 386                 return 0;
 387
 388         tcph = (void *)(*pskb)->data + (*pskb)->nh.iph->ihl*4;
 389         if (after(ntohl(tcph->seq), this_way->correction_pos))
 390                 newseq = htonl(ntohl(tcph->seq) + this_way->offset_after);
 391         else
 392                 newseq = htonl(ntohl(tcph->seq) + this_way->offset_before);
 393
 394         if (after(ntohl(tcph->ack_seq) - other_way->offset_before,
 395                   other_way->correction_pos))
 396                 newack = htonl(ntohl(tcph->ack_seq) - other_way->offset_after);
 397         else
 398                 newack = htonl(ntohl(tcph->ack_seq) - other_way->offset_before);
 399
 400         tcph->check = nf_proto_csum_update(*pskb, ~tcph->seq, newseq,
 401                                            tcph->check, 0);
 402         tcph->check = nf_proto_csum_update(*pskb, ~tcph->ack_seq, newack,
 403                                            tcph->check, 0);
 404
 405         DEBUGP("Adjusting sequence number from %u->%u, ack from %u->%u\n",
 406                 ntohl(tcph->seq), ntohl(newseq), ntohl(tcph->ack_seq),
 407                 ntohl(newack));
 408
 409         tcph->seq = newseq;
 410         tcph->ack_seq = newack;
 411
 412         if (!ip_nat_sack_adjust(pskb, tcph, ct, ctinfo))
 413                 return 0;
 414
 415         ip_conntrack_tcp_update(*pskb, ct, dir);
 416
 417         return 1;
 418 }
 419 EXPORT_SYMBOL(ip_nat_seq_adjust);
 420
 421 /* Setup NAT on this expected conntrack so it follows master. */
 422 /* If we fail to get a free NAT slot, we'll get dropped on confirm */
 423 void ip_nat_follow_master(struct ip_conntrack *ct,
 424                           struct ip_conntrack_expect *exp)
 425 {
 426         struct ip_nat_range range;
 427
 428         /* This must be a fresh one. */
 429         BUG_ON(ct->status & IPS_NAT_DONE_MASK);
 430
 431         /* Change src to where master sends to */
 432         range.flags = IP_NAT_RANGE_MAP_IPS;
 433         range.min_ip = range.max_ip
 434                 = ct->master->tuplehash[!exp->dir].tuple.dst.ip;
 435         /* hook doesn't matter, but it has to do source manip */
 436         ip_nat_setup_info(ct, &range, NF_IP_POST_ROUTING);
 437
 438         /* For DST manip, map port here to where it's expected. */
 439         range.flags = (IP_NAT_RANGE_MAP_IPS | IP_NAT_RANGE_PROTO_SPECIFIED);
 440         range.min = range.max = exp->saved_proto;
 441         range.min_ip = range.max_ip
 442                 = ct->master->tuplehash[!exp->dir].tuple.src.ip;
 443         /* hook doesn't matter, but it has to do destination manip */
 444         ip_nat_setup_info(ct, &range, NF_IP_PRE_ROUTING);
 445 }
 446 EXPORT_SYMBOL(ip_nat_follow_master);