SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 iphdr *iph;
 157         struct tcphdr *tcph;
 158         int oldlen, datalen;
 159
 160         if (!skb_make_writable(pskb, (*pskb)->len))
 161                 return 0;
 162
 163         if (rep_len > match_len &&
 164             rep_len - match_len > skb_tailroom(*pskb) &&
 165             !enlarge_skb(pskb, rep_len - match_len))
 166                 return 0;
 167
 168         SKB_LINEAR_ASSERT(*pskb);
 169
 170         iph = ip_hdr(*pskb);
 171         tcph = (void *)iph + iph->ihl*4;
 172
 173         oldlen = (*pskb)->len - iph->ihl*4;
 174         mangle_contents(*pskb, iph->ihl*4 + tcph->doff*4,
 175                         match_offset, match_len, rep_buffer, rep_len);
 176
 177         datalen = (*pskb)->len - iph->ihl*4;
 178         if ((*pskb)->ip_summed != CHECKSUM_PARTIAL) {
 179                 tcph->check = 0;
 180                 tcph->check = tcp_v4_check(datalen,
 181                                            iph->saddr, iph->daddr,
 182                                            csum_partial((char *)tcph,
 183                                                         datalen, 0));
 184         } else
 185                 nf_proto_csum_replace2(&tcph->check, *pskb,
 186                                        htons(oldlen), htons(datalen), 1);
 187
 188         if (rep_len != match_len) {
 189                 set_bit(IPS_SEQ_ADJUST_BIT, &ct->status);
 190                 adjust_tcp_sequence(ntohl(tcph->seq),
 191                                     (int)rep_len - (int)match_len,
 192                                     ct, ctinfo);
 193                 /* Tell TCP window tracking about seq change */
 194                 nf_conntrack_tcp_update(*pskb, ip_hdrlen(*pskb),
 195                                         ct, CTINFO2DIR(ctinfo));
 196         }
 197         return 1;
 198 }
 199 EXPORT_SYMBOL(nf_nat_mangle_tcp_packet);
 200
 201 /* Generic function for mangling variable-length address changes inside
 202  * NATed UDP connections (like the CONNECT DATA XXXXX MESG XXXXX INDEX XXXXX
 203  * command in the Amanda protocol)
 204  *
 205  * Takes care about all the nasty sequence number changes, checksumming,
 206  * skb enlargement, ...
 207  *
 208  * XXX - This function could be merged with nf_nat_mangle_tcp_packet which
 209  *       should be fairly easy to do.
 210  */
 211 int
 212 nf_nat_mangle_udp_packet(struct sk_buff **pskb,
 213                          struct nf_conn *ct,
 214                          enum ip_conntrack_info ctinfo,
 215                          unsigned int match_offset,
 216                          unsigned int match_len,
 217                          const char *rep_buffer,
 218                          unsigned int rep_len)
 219 {
 220         struct iphdr *iph;
 221         struct udphdr *udph;
 222         int datalen, oldlen;
 223
 224         /* UDP helpers might accidentally mangle the wrong packet */
 225         iph = ip_hdr(*pskb);
 226         if ((*pskb)->len < iph->ihl*4 + sizeof(*udph) +
 227                                match_offset + match_len)
 228                 return 0;
 229
 230         if (!skb_make_writable(pskb, (*pskb)->len))
 231                 return 0;
 232
 233         if (rep_len > match_len &&
 234             rep_len - match_len > skb_tailroom(*pskb) &&
 235             !enlarge_skb(pskb, rep_len - match_len))
 236                 return 0;
 237
 238         iph = ip_hdr(*pskb);
 239         udph = (void *)iph + iph->ihl*4;
 240
 241         oldlen = (*pskb)->len - iph->ihl*4;
 242         mangle_contents(*pskb, iph->ihl*4 + sizeof(*udph),
 243                         match_offset, match_len, rep_buffer, rep_len);
 244
 245         /* update the length of the UDP packet */
 246         datalen = (*pskb)->len - iph->ihl*4;
 247         udph->len = htons(datalen);
 248
 249         /* fix udp checksum if udp checksum was previously calculated */
 250         if (!udph->check && (*pskb)->ip_summed != CHECKSUM_PARTIAL)
 251                 return 1;
 252
 253         if ((*pskb)->ip_summed != CHECKSUM_PARTIAL) {
 254                 udph->check = 0;
 255                 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 256                                                 datalen, IPPROTO_UDP,
 257                                                 csum_partial((char *)udph,
 258                                                              datalen, 0));
 259                 if (!udph->check)
 260                         udph->check = CSUM_MANGLED_0;
 261         } else
 262                 nf_proto_csum_replace2(&udph->check, *pskb,
 263                                        htons(oldlen), htons(datalen), 1);
 264
 265         return 1;
 266 }
 267 EXPORT_SYMBOL(nf_nat_mangle_udp_packet);
 268
 269 /* Adjust one found SACK option including checksum correction */
 270 static void
 271 sack_adjust(struct sk_buff *skb,
 272             struct tcphdr *tcph,
 273             unsigned int sackoff,
 274             unsigned int sackend,
 275             struct nf_nat_seq *natseq)
 276 {
 277         while (sackoff < sackend) {
 278                 struct tcp_sack_block_wire *sack;
 279                 __be32 new_start_seq, new_end_seq;
 280
 281                 sack = (void *)skb->data + sackoff;
 282                 if (after(ntohl(sack->start_seq) - natseq->offset_before,
 283                           natseq->correction_pos))
 284                         new_start_seq = htonl(ntohl(sack->start_seq)
 285                                         - natseq->offset_after);
 286                 else
 287                         new_start_seq = htonl(ntohl(sack->start_seq)
 288                                         - natseq->offset_before);
 289
 290                 if (after(ntohl(sack->end_seq) - natseq->offset_before,
 291                           natseq->correction_pos))
 292                         new_end_seq = htonl(ntohl(sack->end_seq)
 293                                       - natseq->offset_after);
 294                 else
 295                         new_end_seq = htonl(ntohl(sack->end_seq)
 296                                       - natseq->offset_before);
 297
 298                 DEBUGP("sack_adjust: start_seq: %d->%d, end_seq: %d->%d\n",
 299                         ntohl(sack->start_seq), new_start_seq,
 300                         ntohl(sack->end_seq), new_end_seq);
 301
 302                 nf_proto_csum_replace4(&tcph->check, skb,
 303                                        sack->start_seq, new_start_seq, 0);
 304                 nf_proto_csum_replace4(&tcph->check, skb,
 305                                        sack->end_seq, new_end_seq, 0);
 306                 sack->start_seq = new_start_seq;
 307                 sack->end_seq = new_end_seq;
 308                 sackoff += sizeof(*sack);
 309         }
 310 }
 311
 312 /* TCP SACK sequence number adjustment */
 313 static inline unsigned int
 314 nf_nat_sack_adjust(struct sk_buff **pskb,
 315                    struct tcphdr *tcph,
 316                    struct nf_conn *ct,
 317                    enum ip_conntrack_info ctinfo)
 318 {
 319         unsigned int dir, optoff, optend;
 320         struct nf_conn_nat *nat = nfct_nat(ct);
 321
 322         optoff = ip_hdrlen(*pskb) + sizeof(struct tcphdr);
 323         optend = ip_hdrlen(*pskb) + tcph->doff * 4;
 324
 325         if (!skb_make_writable(pskb, optend))
 326                 return 0;
 327
 328         dir = CTINFO2DIR(ctinfo);
 329
 330         while (optoff < optend) {
 331                 /* Usually: option, length. */
 332                 unsigned char *op = (*pskb)->data + optoff;
 333
 334                 switch (op[0]) {
 335                 case TCPOPT_EOL:
 336                         return 1;
 337                 case TCPOPT_NOP:
 338                         optoff++;
 339                         continue;
 340                 default:
 341                         /* no partial options */
 342                         if (optoff + 1 == optend ||
 343                             optoff + op[1] > optend ||
 344                             op[1] < 2)
 345                                 return 0;
 346                         if (op[0] == TCPOPT_SACK &&
 347                             op[1] >= 2+TCPOLEN_SACK_PERBLOCK &&
 348                             ((op[1] - 2) % TCPOLEN_SACK_PERBLOCK) == 0)
 349                                 sack_adjust(*pskb, tcph, optoff+2,
 350                                             optoff+op[1],
 351                                             &nat->info.seq[!dir]);
 352                         optoff += op[1];
 353                 }
 354         }
 355         return 1;
 356 }
 357
 358 /* TCP sequence number adjustment.  Returns 1 on success, 0 on failure */
 359 int
 360 nf_nat_seq_adjust(struct sk_buff **pskb,
 361                   struct nf_conn *ct,
 362                   enum ip_conntrack_info ctinfo)
 363 {
 364         struct tcphdr *tcph;
 365         int dir;
 366         __be32 newseq, newack;
 367         struct nf_conn_nat *nat = nfct_nat(ct);
 368         struct nf_nat_seq *this_way, *other_way;
 369
 370         dir = CTINFO2DIR(ctinfo);
 371
 372         this_way = &nat->info.seq[dir];
 373         other_way = &nat->info.seq[!dir];
 374
 375         if (!skb_make_writable(pskb, ip_hdrlen(*pskb) + sizeof(*tcph)))
 376                 return 0;
 377
 378         tcph = (void *)(*pskb)->data + ip_hdrlen(*pskb);
 379         if (after(ntohl(tcph->seq), this_way->correction_pos))
 380                 newseq = htonl(ntohl(tcph->seq) + this_way->offset_after);
 381         else
 382                 newseq = htonl(ntohl(tcph->seq) + this_way->offset_before);
 383
 384         if (after(ntohl(tcph->ack_seq) - other_way->offset_before,
 385                   other_way->correction_pos))
 386                 newack = htonl(ntohl(tcph->ack_seq) - other_way->offset_after);
 387         else
 388                 newack = htonl(ntohl(tcph->ack_seq) - other_way->offset_before);
 389
 390         nf_proto_csum_replace4(&tcph->check, *pskb, tcph->seq, newseq, 0);
 391         nf_proto_csum_replace4(&tcph->check, *pskb, tcph->ack_seq, newack, 0);
 392
 393         DEBUGP("Adjusting sequence number from %u->%u, ack from %u->%u\n",
 394                 ntohl(tcph->seq), ntohl(newseq), ntohl(tcph->ack_seq),
 395                 ntohl(newack));
 396
 397         tcph->seq = newseq;
 398         tcph->ack_seq = newack;
 399
 400         if (!nf_nat_sack_adjust(pskb, tcph, ct, ctinfo))
 401                 return 0;
 402
 403         nf_conntrack_tcp_update(*pskb, ip_hdrlen(*pskb), ct, dir);
 404
 405         return 1;
 406 }
 407 EXPORT_SYMBOL(nf_nat_seq_adjust);
 408
 409 /* Setup NAT on this expected conntrack so it follows master. */
 410 /* If we fail to get a free NAT slot, we'll get dropped on confirm */
 411 void nf_nat_follow_master(struct nf_conn *ct,
 412                           struct nf_conntrack_expect *exp)
 413 {
 414         struct nf_nat_range range;
 415
 416         /* This must be a fresh one. */
 417         BUG_ON(ct->status & IPS_NAT_DONE_MASK);
 418
 419         /* Change src to where master sends to */
 420         range.flags = IP_NAT_RANGE_MAP_IPS;
 421         range.min_ip = range.max_ip
 422                 = ct->master->tuplehash[!exp->dir].tuple.dst.u3.ip;
 423         /* hook doesn't matter, but it has to do source manip */
 424         nf_nat_setup_info(ct, &range, NF_IP_POST_ROUTING);
 425
 426         /* For DST manip, map port here to where it's expected. */
 427         range.flags = (IP_NAT_RANGE_MAP_IPS | IP_NAT_RANGE_PROTO_SPECIFIED);
 428         range.min = range.max = exp->saved_proto;
 429         range.min_ip = range.max_ip
 430                 = ct->master->tuplehash[!exp->dir].tuple.src.u3.ip;
 431         /* hook doesn't matter, but it has to do destination manip */
 432         nf_nat_setup_info(ct, &range, NF_IP_PRE_ROUTING);
 433 }
 434 EXPORT_SYMBOL(nf_nat_follow_master);