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 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         mangle_contents(*pskb, iph->ihl*4 + tcph->doff*4,
 184                         match_offset, match_len, rep_buffer, rep_len);
 185
 186         datalen = (*pskb)->len - iph->ihl*4;
 187         tcph->check = 0;
 188         tcph->check = tcp_v4_check(tcph, datalen, iph->saddr, iph->daddr,
 189                                    csum_partial((char *)tcph, datalen, 0));
 190
 191         if (rep_len != match_len) {
 192                 set_bit(IPS_SEQ_ADJUST_BIT, &ct->status);
 193                 adjust_tcp_sequence(ntohl(tcph->seq),
 194                                     (int)rep_len - (int)match_len,
 195                                     ct, ctinfo);
 196                 /* Tell TCP window tracking about seq change */
 197                 ip_conntrack_tcp_update(*pskb, ct, CTINFO2DIR(ctinfo));
 198         }
 199         return 1;
 200 }
 201 EXPORT_SYMBOL(ip_nat_mangle_tcp_packet);
 202
 203 /* Generic function for mangling variable-length address changes inside
 204  * NATed UDP connections (like the CONNECT DATA XXXXX MESG XXXXX INDEX XXXXX
 205  * command in the Amanda protocol)
 206  *
 207  * Takes care about all the nasty sequence number changes, checksumming,
 208  * skb enlargement, ...
 209  *
 210  * XXX - This function could be merged with ip_nat_mangle_tcp_packet which
 211  *       should be fairly easy to do.
 212  */
 213 int
 214 ip_nat_mangle_udp_packet(struct sk_buff **pskb,
 215                          struct ip_conntrack *ct,
 216                          enum ip_conntrack_info ctinfo,
 217                          unsigned int match_offset,
 218                          unsigned int match_len,
 219                          const char *rep_buffer,
 220                          unsigned int rep_len)
 221 {
 222         struct iphdr *iph;
 223         struct udphdr *udph;
 224
 225         /* UDP helpers might accidentally mangle the wrong packet */
 226         iph = (*pskb)->nh.iph;
 227         if ((*pskb)->len < iph->ihl*4 + sizeof(*udph) +
 228                                match_offset + match_len)
 229                 return 0;
 230
 231         if (!skb_make_writable(pskb, (*pskb)->len))
 232                 return 0;
 233
 234         if (rep_len > match_len
 235             && rep_len - match_len > skb_tailroom(*pskb)
 236             && !enlarge_skb(pskb, rep_len - match_len))
 237                 return 0;
 238
 239         iph = (*pskb)->nh.iph;
 240         udph = (void *)iph + iph->ihl*4;
 241         mangle_contents(*pskb, iph->ihl*4 + sizeof(*udph),
 242                         match_offset, match_len, rep_buffer, rep_len);
 243
 244         /* update the length of the UDP packet */
 245         udph->len = htons((*pskb)->len - iph->ihl*4);
 246
 247         /* fix udp checksum if udp checksum was previously calculated */
 248         if (udph->check) {
 249                 int datalen = (*pskb)->len - iph->ihl * 4;
 250                 udph->check = 0;
 251                 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 252                                                 datalen, IPPROTO_UDP,
 253                                                 csum_partial((char *)udph,
 254                                                              datalen, 0));
 255         }
 256
 257         return 1;
 258 }
 259 EXPORT_SYMBOL(ip_nat_mangle_udp_packet);
 260
 261 /* Adjust one found SACK option including checksum correction */
 262 static void
 263 sack_adjust(struct sk_buff *skb,
 264             struct tcphdr *tcph,
 265             unsigned int sackoff,
 266             unsigned int sackend,
 267             struct ip_nat_seq *natseq)
 268 {
 269         while (sackoff < sackend) {
 270                 struct tcp_sack_block *sack;
 271                 u_int32_t new_start_seq, new_end_seq;
 272
 273                 sack = (void *)skb->data + sackoff;
 274                 if (after(ntohl(sack->start_seq) - natseq->offset_before,
 275                           natseq->correction_pos))
 276                         new_start_seq = ntohl(sack->start_seq)
 277                                         - natseq->offset_after;
 278                 else
 279                         new_start_seq = ntohl(sack->start_seq)
 280                                         - natseq->offset_before;
 281                 new_start_seq = htonl(new_start_seq);
 282
 283                 if (after(ntohl(sack->end_seq) - natseq->offset_before,
 284                           natseq->correction_pos))
 285                         new_end_seq = ntohl(sack->end_seq)
 286                                       - natseq->offset_after;
 287                 else
 288                         new_end_seq = ntohl(sack->end_seq)
 289                                       - natseq->offset_before;
 290                 new_end_seq = htonl(new_end_seq);
 291
 292                 DEBUGP("sack_adjust: start_seq: %d->%d, end_seq: %d->%d\n",
 293                         ntohl(sack->start_seq), new_start_seq,
 294                         ntohl(sack->end_seq), new_end_seq);
 295
 296                 tcph->check =
 297                         ip_nat_cheat_check(~sack->start_seq, new_start_seq,
 298                                            ip_nat_cheat_check(~sack->end_seq,
 299                                                               new_end_seq,
 300                                                               tcph->check));
 301                 sack->start_seq = new_start_seq;
 302                 sack->end_seq = new_end_seq;
 303                 sackoff += sizeof(*sack);
 304         }
 305 }
 306
 307 /* TCP SACK sequence number adjustment */
 308 static inline unsigned int
 309 ip_nat_sack_adjust(struct sk_buff **pskb,
 310                    struct tcphdr *tcph,
 311                    struct ip_conntrack *ct,
 312                    enum ip_conntrack_info ctinfo)
 313 {
 314         unsigned int dir, optoff, optend;
 315
 316         optoff = (*pskb)->nh.iph->ihl*4 + sizeof(struct tcphdr);
 317         optend = (*pskb)->nh.iph->ihl*4 + tcph->doff*4;
 318
 319         if (!skb_make_writable(pskb, optend))
 320                 return 0;
 321
 322         dir = CTINFO2DIR(ctinfo);
 323
 324         while (optoff < optend) {
 325                 /* Usually: option, length. */
 326                 unsigned char *op = (*pskb)->data + optoff;
 327
 328                 switch (op[0]) {
 329                 case TCPOPT_EOL:
 330                         return 1;
 331                 case TCPOPT_NOP:
 332                         optoff++;
 333                         continue;
 334                 default:
 335                         /* no partial options */
 336                         if (optoff + 1 == optend
 337                             || optoff + op[1] > optend
 338                             || op[1] < 2)
 339                                 return 0;
 340                         if (op[0] == TCPOPT_SACK
 341                             && op[1] >= 2+TCPOLEN_SACK_PERBLOCK
 342                             && ((op[1] - 2) % TCPOLEN_SACK_PERBLOCK) == 0)
 343                                 sack_adjust(*pskb, tcph, optoff+2,
 344                                             optoff+op[1],
 345                                             &ct->nat.info.seq[!dir]);
 346                         optoff += op[1];
 347                 }
 348         }
 349         return 1;
 350 }
 351
 352 /* TCP sequence number adjustment.  Returns 1 on success, 0 on failure */
 353 int
 354 ip_nat_seq_adjust(struct sk_buff **pskb,
 355                   struct ip_conntrack *ct,
 356                   enum ip_conntrack_info ctinfo)
 357 {
 358         struct tcphdr *tcph;
 359         int dir, newseq, newack;
 360         struct ip_nat_seq *this_way, *other_way;
 361
 362         dir = CTINFO2DIR(ctinfo);
 363
 364         this_way = &ct->nat.info.seq[dir];
 365         other_way = &ct->nat.info.seq[!dir];
 366
 367         if (!skb_make_writable(pskb, (*pskb)->nh.iph->ihl*4+sizeof(*tcph)))
 368                 return 0;
 369
 370         tcph = (void *)(*pskb)->data + (*pskb)->nh.iph->ihl*4;
 371         if (after(ntohl(tcph->seq), this_way->correction_pos))
 372                 newseq = ntohl(tcph->seq) + this_way->offset_after;
 373         else
 374                 newseq = ntohl(tcph->seq) + this_way->offset_before;
 375         newseq = htonl(newseq);
 376
 377         if (after(ntohl(tcph->ack_seq) - other_way->offset_before,
 378                   other_way->correction_pos))
 379                 newack = ntohl(tcph->ack_seq) - other_way->offset_after;
 380         else
 381                 newack = ntohl(tcph->ack_seq) - other_way->offset_before;
 382         newack = htonl(newack);
 383
 384         tcph->check = ip_nat_cheat_check(~tcph->seq, newseq,
 385                                          ip_nat_cheat_check(~tcph->ack_seq,
 386                                                             newack,
 387                                                             tcph->check));
 388
 389         DEBUGP("Adjusting sequence number from %u->%u, ack from %u->%u\n",
 390                 ntohl(tcph->seq), ntohl(newseq), ntohl(tcph->ack_seq),
 391                 ntohl(newack));
 392
 393         tcph->seq = newseq;
 394         tcph->ack_seq = newack;
 395
 396         if (!ip_nat_sack_adjust(pskb, tcph, ct, ctinfo))
 397                 return 0;
 398
 399         ip_conntrack_tcp_update(*pskb, ct, dir);
 400
 401         return 1;
 402 }
 403 EXPORT_SYMBOL(ip_nat_seq_adjust);
 404
 405 /* Setup NAT on this expected conntrack so it follows master. */
 406 /* If we fail to get a free NAT slot, we'll get dropped on confirm */
 407 void ip_nat_follow_master(struct ip_conntrack *ct,
 408                           struct ip_conntrack_expect *exp)
 409 {
 410         struct ip_nat_range range;
 411
 412         /* This must be a fresh one. */
 413         BUG_ON(ct->status & IPS_NAT_DONE_MASK);
 414
 415         /* Change src to where master sends to */
 416         range.flags = IP_NAT_RANGE_MAP_IPS;
 417         range.min_ip = range.max_ip
 418                 = ct->master->tuplehash[!exp->dir].tuple.dst.ip;
 419         /* hook doesn't matter, but it has to do source manip */
 420         ip_nat_setup_info(ct, &range, NF_IP_POST_ROUTING);
 421
 422         /* For DST manip, map port here to where it's expected. */
 423         range.flags = (IP_NAT_RANGE_MAP_IPS | IP_NAT_RANGE_PROTO_SPECIFIED);
 424         range.min = range.max = exp->saved_proto;
 425         range.min_ip = range.max_ip
 426                 = ct->master->tuplehash[!exp->dir].tuple.src.ip;
 427         /* hook doesn't matter, but it has to do destination manip */
 428         ip_nat_setup_info(ct, &range, NF_IP_PRE_ROUTING);
 429 }
 430 EXPORT_SYMBOL(ip_nat_follow_master);