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 = (unsigned char *)skb->nh.iph + dataoff;
  91
  92         /* move post-replacement */
  93         memmove(data + match_offset + rep_len,
  94                 data + match_offset + match_len,
  95                 skb->tail - (data + match_offset + match_len));
  96
  97         /* insert data from buffer */
  98         memcpy(data + match_offset, rep_buffer, rep_len);
  99
 100         /* update skb info */
 101         if (rep_len > match_len) {
 102                 DEBUGP("nf_nat_mangle_packet: Extending packet by "
 103                        "%u from %u bytes\n", rep_len - match_len,
 104                        skb->len);
 105                 skb_put(skb, rep_len - match_len);
 106         } else {
 107                 DEBUGP("nf_nat_mangle_packet: Shrinking packet from "
 108                        "%u from %u bytes\n", match_len - rep_len,
 109                        skb->len);
 110                 __skb_trim(skb, skb->len + rep_len - match_len);
 111         }
 112
 113         /* fix IP hdr checksum information */
 114         skb->nh.iph->tot_len = htons(skb->len);
 115         ip_send_check(skb->nh.iph);
 116 }
 117
 118 /* Unusual, but possible case. */
 119 static int enlarge_skb(struct sk_buff **pskb, unsigned int extra)
 120 {
 121         struct sk_buff *nskb;
 122
 123         if ((*pskb)->len + extra > 65535)
 124                 return 0;
 125
 126         nskb = skb_copy_expand(*pskb, skb_headroom(*pskb), extra, GFP_ATOMIC);
 127         if (!nskb)
 128                 return 0;
 129
 130         /* Transfer socket to new skb. */
 131         if ((*pskb)->sk)
 132                 skb_set_owner_w(nskb, (*pskb)->sk);
 133         kfree_skb(*pskb);
 134         *pskb = nskb;
 135         return 1;
 136 }
 137
 138 /* Generic function for mangling variable-length address changes inside
 139  * NATed TCP connections (like the PORT XXX,XXX,XXX,XXX,XXX,XXX
 140  * command in FTP).
 141  *
 142  * Takes care about all the nasty sequence number changes, checksumming,
 143  * skb enlargement, ...
 144  *
 145  * */
 146 int
 147 nf_nat_mangle_tcp_packet(struct sk_buff **pskb,
 148                          struct nf_conn *ct,
 149                          enum ip_conntrack_info ctinfo,
 150                          unsigned int match_offset,
 151                          unsigned int match_len,
 152                          const char *rep_buffer,
 153                          unsigned int rep_len)
 154 {
 155         struct iphdr *iph;
 156         struct tcphdr *tcph;
 157         int oldlen, datalen;
 158
 159         if (!skb_make_writable(pskb, (*pskb)->len))
 160                 return 0;
 161
 162         if (rep_len > match_len &&
 163             rep_len - match_len > skb_tailroom(*pskb) &&
 164             !enlarge_skb(pskb, rep_len - match_len))
 165                 return 0;
 166
 167         SKB_LINEAR_ASSERT(*pskb);
 168
 169         iph = (*pskb)->nh.iph;
 170         tcph = (void *)iph + iph->ihl*4;
 171
 172         oldlen = (*pskb)->len - iph->ihl*4;
 173         mangle_contents(*pskb, iph->ihl*4 + tcph->doff*4,
 174                         match_offset, match_len, rep_buffer, rep_len);
 175
 176         datalen = (*pskb)->len - iph->ihl*4;
 177         if ((*pskb)->ip_summed != CHECKSUM_PARTIAL) {
 178                 tcph->check = 0;
 179                 tcph->check = tcp_v4_check(datalen,
 180                                            iph->saddr, iph->daddr,
 181                                            csum_partial((char *)tcph,
 182                                                         datalen, 0));
 183         } else
 184                 nf_proto_csum_replace2(&tcph->check, *pskb,
 185                                        htons(oldlen), htons(datalen), 1);
 186
 187         if (rep_len != match_len) {
 188                 set_bit(IPS_SEQ_ADJUST_BIT, &ct->status);
 189                 adjust_tcp_sequence(ntohl(tcph->seq),
 190                                     (int)rep_len - (int)match_len,
 191                                     ct, ctinfo);
 192                 /* Tell TCP window tracking about seq change */
 193                 nf_conntrack_tcp_update(*pskb, (*pskb)->nh.iph->ihl*4,
 194                                         ct, CTINFO2DIR(ctinfo));
 195         }
 196         return 1;
 197 }
 198 EXPORT_SYMBOL(nf_nat_mangle_tcp_packet);
 199
 200 /* Generic function for mangling variable-length address changes inside
 201  * NATed UDP connections (like the CONNECT DATA XXXXX MESG XXXXX INDEX XXXXX
 202  * command in the Amanda protocol)
 203  *
 204  * Takes care about all the nasty sequence number changes, checksumming,
 205  * skb enlargement, ...
 206  *
 207  * XXX - This function could be merged with nf_nat_mangle_tcp_packet which
 208  *       should be fairly easy to do.
 209  */
 210 int
 211 nf_nat_mangle_udp_packet(struct sk_buff **pskb,
 212                          struct nf_conn *ct,
 213                          enum ip_conntrack_info ctinfo,
 214                          unsigned int match_offset,
 215                          unsigned int match_len,
 216                          const char *rep_buffer,
 217                          unsigned int rep_len)
 218 {
 219         struct iphdr *iph;
 220         struct udphdr *udph;
 221         int datalen, oldlen;
 222
 223         /* UDP helpers might accidentally mangle the wrong packet */
 224         iph = (*pskb)->nh.iph;
 225         if ((*pskb)->len < iph->ihl*4 + sizeof(*udph) +
 226                                match_offset + match_len)
 227                 return 0;
 228
 229         if (!skb_make_writable(pskb, (*pskb)->len))
 230                 return 0;
 231
 232         if (rep_len > match_len &&
 233             rep_len - match_len > skb_tailroom(*pskb) &&
 234             !enlarge_skb(pskb, rep_len - match_len))
 235                 return 0;
 236
 237         iph = (*pskb)->nh.iph;
 238         udph = (void *)iph + iph->ihl*4;
 239
 240         oldlen = (*pskb)->len - 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         datalen = (*pskb)->len - iph->ihl*4;
 246         udph->len = htons(datalen);
 247
 248         /* fix udp checksum if udp checksum was previously calculated */
 249         if (!udph->check && (*pskb)->ip_summed != CHECKSUM_PARTIAL)
 250                 return 1;
 251
 252         if ((*pskb)->ip_summed != CHECKSUM_PARTIAL) {
 253                 udph->check = 0;
 254                 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 255                                                 datalen, IPPROTO_UDP,
 256                                                 csum_partial((char *)udph,
 257                                                              datalen, 0));
 258                 if (!udph->check)
 259                         udph->check = CSUM_MANGLED_0;
 260         } else
 261                 nf_proto_csum_replace2(&udph->check, *pskb,
 262                                        htons(oldlen), htons(datalen), 1);
 263
 264         return 1;
 265 }
 266 EXPORT_SYMBOL(nf_nat_mangle_udp_packet);
 267
 268 /* Adjust one found SACK option including checksum correction */
 269 static void
 270 sack_adjust(struct sk_buff *skb,
 271             struct tcphdr *tcph,
 272             unsigned int sackoff,
 273             unsigned int sackend,
 274             struct nf_nat_seq *natseq)
 275 {
 276         while (sackoff < sackend) {
 277                 struct tcp_sack_block_wire *sack;
 278                 __be32 new_start_seq, new_end_seq;
 279
 280                 sack = (void *)skb->data + sackoff;
 281                 if (after(ntohl(sack->start_seq) - natseq->offset_before,
 282                           natseq->correction_pos))
 283                         new_start_seq = htonl(ntohl(sack->start_seq)
 284                                         - natseq->offset_after);
 285                 else
 286                         new_start_seq = htonl(ntohl(sack->start_seq)
 287                                         - natseq->offset_before);
 288
 289                 if (after(ntohl(sack->end_seq) - natseq->offset_before,
 290                           natseq->correction_pos))
 291                         new_end_seq = htonl(ntohl(sack->end_seq)
 292                                       - natseq->offset_after);
 293                 else
 294                         new_end_seq = htonl(ntohl(sack->end_seq)
 295                                       - natseq->offset_before);
 296
 297                 DEBUGP("sack_adjust: start_seq: %d->%d, end_seq: %d->%d\n",
 298                         ntohl(sack->start_seq), new_start_seq,
 299                         ntohl(sack->end_seq), new_end_seq);
 300
 301                 nf_proto_csum_replace4(&tcph->check, skb,
 302                                        sack->start_seq, new_start_seq, 0);
 303                 nf_proto_csum_replace4(&tcph->check, skb,
 304                                        sack->end_seq, new_end_seq, 0);
 305                 sack->start_seq = new_start_seq;
 306                 sack->end_seq = new_end_seq;
 307                 sackoff += sizeof(*sack);
 308         }
 309 }
 310
 311 /* TCP SACK sequence number adjustment */
 312 static inline unsigned int
 313 nf_nat_sack_adjust(struct sk_buff **pskb,
 314                    struct tcphdr *tcph,
 315                    struct nf_conn *ct,
 316                    enum ip_conntrack_info ctinfo)
 317 {
 318         unsigned int dir, optoff, optend;
 319         struct nf_conn_nat *nat = nfct_nat(ct);
 320
 321         optoff = (*pskb)->nh.iph->ihl*4 + sizeof(struct tcphdr);
 322         optend = (*pskb)->nh.iph->ihl*4 + tcph->doff*4;
 323
 324         if (!skb_make_writable(pskb, optend))
 325                 return 0;
 326
 327         dir = CTINFO2DIR(ctinfo);
 328
 329         while (optoff < optend) {
 330                 /* Usually: option, length. */
 331                 unsigned char *op = (*pskb)->data + optoff;
 332
 333                 switch (op[0]) {
 334                 case TCPOPT_EOL:
 335                         return 1;
 336                 case TCPOPT_NOP:
 337                         optoff++;
 338                         continue;
 339                 default:
 340                         /* no partial options */
 341                         if (optoff + 1 == optend ||
 342                             optoff + op[1] > optend ||
 343                             op[1] < 2)
 344                                 return 0;
 345                         if (op[0] == TCPOPT_SACK &&
 346                             op[1] >= 2+TCPOLEN_SACK_PERBLOCK &&
 347                             ((op[1] - 2) % TCPOLEN_SACK_PERBLOCK) == 0)
 348                                 sack_adjust(*pskb, tcph, optoff+2,
 349                                             optoff+op[1],
 350                                             &nat->info.seq[!dir]);
 351                         optoff += op[1];
 352                 }
 353         }
 354         return 1;
 355 }
 356
 357 /* TCP sequence number adjustment.  Returns 1 on success, 0 on failure */
 358 int
 359 nf_nat_seq_adjust(struct sk_buff **pskb,
 360                   struct nf_conn *ct,
 361                   enum ip_conntrack_info ctinfo)
 362 {
 363         struct tcphdr *tcph;
 364         int dir;
 365         __be32 newseq, newack;
 366         struct nf_conn_nat *nat = nfct_nat(ct);
 367         struct nf_nat_seq *this_way, *other_way;
 368
 369         dir = CTINFO2DIR(ctinfo);
 370
 371         this_way = &nat->info.seq[dir];
 372         other_way = &nat->info.seq[!dir];
 373
 374         if (!skb_make_writable(pskb, (*pskb)->nh.iph->ihl*4+sizeof(*tcph)))
 375                 return 0;
 376
 377         tcph = (void *)(*pskb)->data + (*pskb)->nh.iph->ihl*4;
 378         if (after(ntohl(tcph->seq), this_way->correction_pos))
 379                 newseq = htonl(ntohl(tcph->seq) + this_way->offset_after);
 380         else
 381                 newseq = htonl(ntohl(tcph->seq) + this_way->offset_before);
 382
 383         if (after(ntohl(tcph->ack_seq) - other_way->offset_before,
 384                   other_way->correction_pos))
 385                 newack = htonl(ntohl(tcph->ack_seq) - other_way->offset_after);
 386         else
 387                 newack = htonl(ntohl(tcph->ack_seq) - other_way->offset_before);
 388
 389         nf_proto_csum_replace4(&tcph->check, *pskb, tcph->seq, newseq, 0);
 390         nf_proto_csum_replace4(&tcph->check, *pskb, tcph->ack_seq, newack, 0);
 391
 392         DEBUGP("Adjusting sequence number from %u->%u, ack from %u->%u\n",
 393                 ntohl(tcph->seq), ntohl(newseq), ntohl(tcph->ack_seq),
 394                 ntohl(newack));
 395
 396         tcph->seq = newseq;
 397         tcph->ack_seq = newack;
 398
 399         if (!nf_nat_sack_adjust(pskb, tcph, ct, ctinfo))
 400                 return 0;
 401
 402         nf_conntrack_tcp_update(*pskb, (*pskb)->nh.iph->ihl*4, ct, dir);
 403
 404         return 1;
 405 }
 406 EXPORT_SYMBOL(nf_nat_seq_adjust);
 407
 408 /* Setup NAT on this expected conntrack so it follows master. */
 409 /* If we fail to get a free NAT slot, we'll get dropped on confirm */
 410 void nf_nat_follow_master(struct nf_conn *ct,
 411                           struct nf_conntrack_expect *exp)
 412 {
 413         struct nf_nat_range range;
 414
 415         /* This must be a fresh one. */
 416         BUG_ON(ct->status & IPS_NAT_DONE_MASK);
 417
 418         /* Change src to where master sends to */
 419         range.flags = IP_NAT_RANGE_MAP_IPS;
 420         range.min_ip = range.max_ip
 421                 = ct->master->tuplehash[!exp->dir].tuple.dst.u3.ip;
 422         /* hook doesn't matter, but it has to do source manip */
 423         nf_nat_setup_info(ct, &range, NF_IP_POST_ROUTING);
 424
 425         /* For DST manip, map port here to where it's expected. */
 426         range.flags = (IP_NAT_RANGE_MAP_IPS | IP_NAT_RANGE_PROTO_SPECIFIED);
 427         range.min = range.max = exp->saved_proto;
 428         range.min_ip = range.max_ip
 429                 = ct->master->tuplehash[!exp->dir].tuple.src.u3.ip;
 430         /* hook doesn't matter, but it has to do destination manip */
 431         nf_nat_setup_info(ct, &range, NF_IP_PRE_ROUTING);
 432 }
 433 EXPORT_SYMBOL(nf_nat_follow_master);