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 #if 0
  30 #define DEBUGP printk
  31 #define DUMP_OFFSET(x)  printk("offset_before=%d, offset_after=%d, correction_pos=%u\n", x->offset_before, x->offset_after, x->correction_pos);
  32 #else
  33 #define DEBUGP(format, args...)
  34 #define DUMP_OFFSET(x)
  35 #endif
  36
  37 static DEFINE_SPINLOCK(nf_nat_seqofs_lock);
  38
  39 /* Setup TCP sequence correction given this change at this sequence */
  40 static inline void
  41 adjust_tcp_sequence(u32 seq,
  42                     int sizediff,
  43                     struct nf_conn *ct,
  44                     enum ip_conntrack_info ctinfo)
  45 {
  46         int dir;
  47         struct nf_nat_seq *this_way, *other_way;
  48         struct nf_conn_nat *nat = nfct_nat(ct);
  49
  50         DEBUGP("nf_nat_resize_packet: old_size = %u, new_size = %u\n",
  51                 (*skb)->len, new_size);
  52
  53         dir = CTINFO2DIR(ctinfo);
  54
  55         this_way = &nat->info.seq[dir];
  56         other_way = &nat->info.seq[!dir];
  57
  58         DEBUGP("nf_nat_resize_packet: Seq_offset before: ");
  59         DUMP_OFFSET(this_way);
  60
  61         spin_lock_bh(&nf_nat_seqofs_lock);
  62
  63         /* SYN adjust. If it's uninitialized, or this is after last
  64          * correction, record it: we don't handle more than one
  65          * adjustment in the window, but do deal with common case of a
  66          * retransmit */
  67         if (this_way->offset_before == this_way->offset_after ||
  68             before(this_way->correction_pos, seq)) {
  69                    this_way->correction_pos = seq;
  70                    this_way->offset_before = this_way->offset_after;
  71                    this_way->offset_after += sizediff;
  72         }
  73         spin_unlock_bh(&nf_nat_seqofs_lock);
  74
  75         DEBUGP("nf_nat_resize_packet: Seq_offset after: ");
  76         DUMP_OFFSET(this_way);
  77 }
  78
  79 /* Frobs data inside this packet, which is linear. */
  80 static void mangle_contents(struct sk_buff *skb,
  81                             unsigned int dataoff,
  82                             unsigned int match_offset,
  83                             unsigned int match_len,
  84                             const char *rep_buffer,
  85                             unsigned int rep_len)
  86 {
  87         unsigned char *data;
  88
  89         BUG_ON(skb_is_nonlinear(skb));
  90         data = skb_network_header(skb) + dataoff;
  91
  92         /* move post-replacement */
  93         memmove(data + match_offset + rep_len,
  94                 data + match_offset + match_len,
  95                 skb->tail - (skb->network_header + dataoff +
  96                              match_offset + match_len));
  97
  98         /* insert data from buffer */
  99         memcpy(data + match_offset, rep_buffer, rep_len);
 100
 101         /* update skb info */
 102         if (rep_len > match_len) {
 103                 DEBUGP("nf_nat_mangle_packet: Extending packet by "
 104                        "%u from %u bytes\n", rep_len - match_len,
 105                        skb->len);
 106                 skb_put(skb, rep_len - match_len);
 107         } else {
 108                 DEBUGP("nf_nat_mangle_packet: Shrinking packet from "
 109                        "%u from %u bytes\n", match_len - rep_len,
 110                        skb->len);
 111                 __skb_trim(skb, skb->len + rep_len - match_len);
 112         }
 113
 114         /* fix IP hdr checksum information */
 115         ip_hdr(skb)->tot_len = htons(skb->len);
 116         ip_send_check(ip_hdr(skb));
 117 }
 118
 119 /* Unusual, but possible case. */
 120 static int enlarge_skb(struct sk_buff **pskb, unsigned int extra)
 121 {
 122         struct sk_buff *nskb;
 123
 124         if ((*pskb)->len + extra > 65535)
 125                 return 0;
 126
 127         nskb = skb_copy_expand(*pskb, skb_headroom(*pskb), extra, GFP_ATOMIC);
 128         if (!nskb)
 129                 return 0;
 130
 131         /* Transfer socket to new skb. */
 132         if ((*pskb)->sk)
 133                 skb_set_owner_w(nskb, (*pskb)->sk);
 134         kfree_skb(*pskb);
 135         *pskb = nskb;
 136         return 1;
 137 }
 138
 139 /* Generic function for mangling variable-length address changes inside
 140  * NATed TCP connections (like the PORT XXX,XXX,XXX,XXX,XXX,XXX
 141  * command in FTP).
 142  *
 143  * Takes care about all the nasty sequence number changes, checksumming,
 144  * skb enlargement, ...
 145  *
 146  * */
 147 int
 148 nf_nat_mangle_tcp_packet(struct sk_buff **pskb,
 149                          struct nf_conn *ct,
 150                          enum ip_conntrack_info ctinfo,
 151                          unsigned int match_offset,
 152                          unsigned int match_len,
 153                          const char *rep_buffer,
 154                          unsigned int rep_len)
 155 {
 156         struct rtable *rt = (struct rtable *)(*pskb)->dst;
 157         struct iphdr *iph;
 158         struct tcphdr *tcph;
 159         int oldlen, datalen;
 160
 161         if (!skb_make_writable(pskb, (*pskb)->len))
 162                 return 0;
 163
 164         if (rep_len > match_len &&
 165             rep_len - match_len > skb_tailroom(*pskb) &&
 166             !enlarge_skb(pskb, rep_len - match_len))
 167                 return 0;
 168
 169         SKB_LINEAR_ASSERT(*pskb);
 170
 171         iph = ip_hdr(*pskb);
 172         tcph = (void *)iph + iph->ihl*4;
 173
 174         oldlen = (*pskb)->len - iph->ihl*4;
 175         mangle_contents(*pskb, iph->ihl*4 + tcph->doff*4,
 176                         match_offset, match_len, rep_buffer, rep_len);
 177
 178         datalen = (*pskb)->len - iph->ihl*4;
 179         if ((*pskb)->ip_summed != CHECKSUM_PARTIAL) {
 180                 if (!(rt->rt_flags & RTCF_LOCAL) &&
 181                     (*pskb)->dev->features & NETIF_F_ALL_CSUM) {
 182                         (*pskb)->ip_summed = CHECKSUM_PARTIAL;
 183                         (*pskb)->csum_start = skb_headroom(*pskb) +
 184                                               skb_network_offset(*pskb) +
 185                                               iph->ihl * 4;
 186                         (*pskb)->csum_offset = offsetof(struct tcphdr, check);
 187                         tcph->check = ~tcp_v4_check(datalen,
 188                                                     iph->saddr, iph->daddr, 0);
 189                 } else {
 190                         tcph->check = 0;
 191                         tcph->check = tcp_v4_check(datalen,
 192                                                    iph->saddr, iph->daddr,
 193                                                    csum_partial((char *)tcph,
 194                                                                 datalen, 0));
 195                 }
 196         } else
 197                 nf_proto_csum_replace2(&tcph->check, *pskb,
 198                                        htons(oldlen), htons(datalen), 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                 nf_conntrack_tcp_update(*pskb, ip_hdrlen(*pskb),
 207                                         ct, CTINFO2DIR(ctinfo));
 208         }
 209         return 1;
 210 }
 211 EXPORT_SYMBOL(nf_nat_mangle_tcp_packet);
 212
 213 /* Generic function for mangling variable-length address changes inside
 214  * NATed UDP connections (like the CONNECT DATA XXXXX MESG XXXXX INDEX XXXXX
 215  * command in the Amanda protocol)
 216  *
 217  * Takes care about all the nasty sequence number changes, checksumming,
 218  * skb enlargement, ...
 219  *
 220  * XXX - This function could be merged with nf_nat_mangle_tcp_packet which
 221  *       should be fairly easy to do.
 222  */
 223 int
 224 nf_nat_mangle_udp_packet(struct sk_buff **pskb,
 225                          struct nf_conn *ct,
 226                          enum ip_conntrack_info ctinfo,
 227                          unsigned int match_offset,
 228                          unsigned int match_len,
 229                          const char *rep_buffer,
 230                          unsigned int rep_len)
 231 {
 232         struct rtable *rt = (struct rtable *)(*pskb)->dst;
 233         struct iphdr *iph;
 234         struct udphdr *udph;
 235         int datalen, oldlen;
 236
 237         /* UDP helpers might accidentally mangle the wrong packet */
 238         iph = ip_hdr(*pskb);
 239         if ((*pskb)->len < iph->ihl*4 + sizeof(*udph) +
 240                                match_offset + match_len)
 241                 return 0;
 242
 243         if (!skb_make_writable(pskb, (*pskb)->len))
 244                 return 0;
 245
 246         if (rep_len > match_len &&
 247             rep_len - match_len > skb_tailroom(*pskb) &&
 248             !enlarge_skb(pskb, rep_len - match_len))
 249                 return 0;
 250
 251         iph = ip_hdr(*pskb);
 252         udph = (void *)iph + iph->ihl*4;
 253
 254         oldlen = (*pskb)->len - iph->ihl*4;
 255         mangle_contents(*pskb, iph->ihl*4 + sizeof(*udph),
 256                         match_offset, match_len, rep_buffer, rep_len);
 257
 258         /* update the length of the UDP packet */
 259         datalen = (*pskb)->len - iph->ihl*4;
 260         udph->len = htons(datalen);
 261
 262         /* fix udp checksum if udp checksum was previously calculated */
 263         if (!udph->check && (*pskb)->ip_summed != CHECKSUM_PARTIAL)
 264                 return 1;
 265
 266         if ((*pskb)->ip_summed != CHECKSUM_PARTIAL) {
 267                 if (!(rt->rt_flags & RTCF_LOCAL) &&
 268                     (*pskb)->dev->features & NETIF_F_ALL_CSUM) {
 269                         (*pskb)->ip_summed = CHECKSUM_PARTIAL;
 270                         (*pskb)->csum_start = skb_headroom(*pskb) +
 271                                               skb_network_offset(*pskb) +
 272                                               iph->ihl * 4;
 273                         (*pskb)->csum_offset = offsetof(struct udphdr, check);
 274                         udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
 275                                                          datalen, IPPROTO_UDP,
 276                                                          0);
 277                 } else {
 278                         udph->check = 0;
 279                         udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 280                                                         datalen, IPPROTO_UDP,
 281                                                         csum_partial((char *)udph,
 282                                                                      datalen, 0));
 283                         if (!udph->check)
 284                                 udph->check = CSUM_MANGLED_0;
 285                 }
 286         } else
 287                 nf_proto_csum_replace2(&udph->check, *pskb,
 288                                        htons(oldlen), htons(datalen), 1);
 289
 290         return 1;
 291 }
 292 EXPORT_SYMBOL(nf_nat_mangle_udp_packet);
 293
 294 /* Adjust one found SACK option including checksum correction */
 295 static void
 296 sack_adjust(struct sk_buff *skb,
 297             struct tcphdr *tcph,
 298             unsigned int sackoff,
 299             unsigned int sackend,
 300             struct nf_nat_seq *natseq)
 301 {
 302         while (sackoff < sackend) {
 303                 struct tcp_sack_block_wire *sack;
 304                 __be32 new_start_seq, new_end_seq;
 305
 306                 sack = (void *)skb->data + sackoff;
 307                 if (after(ntohl(sack->start_seq) - natseq->offset_before,
 308                           natseq->correction_pos))
 309                         new_start_seq = htonl(ntohl(sack->start_seq)
 310                                         - natseq->offset_after);
 311                 else
 312                         new_start_seq = htonl(ntohl(sack->start_seq)
 313                                         - natseq->offset_before);
 314
 315                 if (after(ntohl(sack->end_seq) - natseq->offset_before,
 316                           natseq->correction_pos))
 317                         new_end_seq = htonl(ntohl(sack->end_seq)
 318                                       - natseq->offset_after);
 319                 else
 320                         new_end_seq = htonl(ntohl(sack->end_seq)
 321                                       - natseq->offset_before);
 322
 323                 DEBUGP("sack_adjust: start_seq: %d->%d, end_seq: %d->%d\n",
 324                         ntohl(sack->start_seq), new_start_seq,
 325                         ntohl(sack->end_seq), new_end_seq);
 326
 327                 nf_proto_csum_replace4(&tcph->check, skb,
 328                                        sack->start_seq, new_start_seq, 0);
 329                 nf_proto_csum_replace4(&tcph->check, skb,
 330                                        sack->end_seq, new_end_seq, 0);
 331                 sack->start_seq = new_start_seq;
 332                 sack->end_seq = new_end_seq;
 333                 sackoff += sizeof(*sack);
 334         }
 335 }
 336
 337 /* TCP SACK sequence number adjustment */
 338 static inline unsigned int
 339 nf_nat_sack_adjust(struct sk_buff **pskb,
 340                    struct tcphdr *tcph,
 341                    struct nf_conn *ct,
 342                    enum ip_conntrack_info ctinfo)
 343 {
 344         unsigned int dir, optoff, optend;
 345         struct nf_conn_nat *nat = nfct_nat(ct);
 346
 347         optoff = ip_hdrlen(*pskb) + sizeof(struct tcphdr);
 348         optend = ip_hdrlen(*pskb) + tcph->doff * 4;
 349
 350         if (!skb_make_writable(pskb, optend))
 351                 return 0;
 352
 353         dir = CTINFO2DIR(ctinfo);
 354
 355         while (optoff < optend) {
 356                 /* Usually: option, length. */
 357                 unsigned char *op = (*pskb)->data + optoff;
 358
 359                 switch (op[0]) {
 360                 case TCPOPT_EOL:
 361                         return 1;
 362                 case TCPOPT_NOP:
 363                         optoff++;
 364                         continue;
 365                 default:
 366                         /* no partial options */
 367                         if (optoff + 1 == optend ||
 368                             optoff + op[1] > optend ||
 369                             op[1] < 2)
 370                                 return 0;
 371                         if (op[0] == TCPOPT_SACK &&
 372                             op[1] >= 2+TCPOLEN_SACK_PERBLOCK &&
 373                             ((op[1] - 2) % TCPOLEN_SACK_PERBLOCK) == 0)
 374                                 sack_adjust(*pskb, tcph, optoff+2,
 375                                             optoff+op[1],
 376                                             &nat->info.seq[!dir]);
 377                         optoff += op[1];
 378                 }
 379         }
 380         return 1;
 381 }
 382
 383 /* TCP sequence number adjustment.  Returns 1 on success, 0 on failure */
 384 int
 385 nf_nat_seq_adjust(struct sk_buff **pskb,
 386                   struct nf_conn *ct,
 387                   enum ip_conntrack_info ctinfo)
 388 {
 389         struct tcphdr *tcph;
 390         int dir;
 391         __be32 newseq, newack;
 392         struct nf_conn_nat *nat = nfct_nat(ct);
 393         struct nf_nat_seq *this_way, *other_way;
 394
 395         dir = CTINFO2DIR(ctinfo);
 396
 397         this_way = &nat->info.seq[dir];
 398         other_way = &nat->info.seq[!dir];
 399
 400         if (!skb_make_writable(pskb, ip_hdrlen(*pskb) + sizeof(*tcph)))
 401                 return 0;
 402
 403         tcph = (void *)(*pskb)->data + ip_hdrlen(*pskb);
 404         if (after(ntohl(tcph->seq), this_way->correction_pos))
 405                 newseq = htonl(ntohl(tcph->seq) + this_way->offset_after);
 406         else
 407                 newseq = htonl(ntohl(tcph->seq) + this_way->offset_before);
 408
 409         if (after(ntohl(tcph->ack_seq) - other_way->offset_before,
 410                   other_way->correction_pos))
 411                 newack = htonl(ntohl(tcph->ack_seq) - other_way->offset_after);
 412         else
 413                 newack = htonl(ntohl(tcph->ack_seq) - other_way->offset_before);
 414
 415         nf_proto_csum_replace4(&tcph->check, *pskb, tcph->seq, newseq, 0);
 416         nf_proto_csum_replace4(&tcph->check, *pskb, tcph->ack_seq, newack, 0);
 417
 418         DEBUGP("Adjusting sequence number from %u->%u, ack from %u->%u\n",
 419                 ntohl(tcph->seq), ntohl(newseq), ntohl(tcph->ack_seq),
 420                 ntohl(newack));
 421
 422         tcph->seq = newseq;
 423         tcph->ack_seq = newack;
 424
 425         if (!nf_nat_sack_adjust(pskb, tcph, ct, ctinfo))
 426                 return 0;
 427
 428         nf_conntrack_tcp_update(*pskb, ip_hdrlen(*pskb), ct, dir);
 429
 430         return 1;
 431 }
 432 EXPORT_SYMBOL(nf_nat_seq_adjust);
 433
 434 /* Setup NAT on this expected conntrack so it follows master. */
 435 /* If we fail to get a free NAT slot, we'll get dropped on confirm */
 436 void nf_nat_follow_master(struct nf_conn *ct,
 437                           struct nf_conntrack_expect *exp)
 438 {
 439         struct nf_nat_range range;
 440
 441         /* This must be a fresh one. */
 442         BUG_ON(ct->status & IPS_NAT_DONE_MASK);
 443
 444         /* Change src to where master sends to */
 445         range.flags = IP_NAT_RANGE_MAP_IPS;
 446         range.min_ip = range.max_ip
 447                 = ct->master->tuplehash[!exp->dir].tuple.dst.u3.ip;
 448         /* hook doesn't matter, but it has to do source manip */
 449         nf_nat_setup_info(ct, &range, NF_IP_POST_ROUTING);
 450
 451         /* For DST manip, map port here to where it's expected. */
 452         range.flags = (IP_NAT_RANGE_MAP_IPS | IP_NAT_RANGE_PROTO_SPECIFIED);
 453         range.min = range.max = exp->saved_proto;
 454         range.min_ip = range.max_ip
 455                 = ct->master->tuplehash[!exp->dir].tuple.src.u3.ip;
 456         /* hook doesn't matter, but it has to do destination manip */
 457         nf_nat_setup_info(ct, &range, NF_IP_PRE_ROUTING);
 458 }
 459 EXPORT_SYMBOL(nf_nat_follow_master);