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 #include <net/route.h>
  20
  21 #include <linux/netfilter_ipv4.h>
  22 #include <net/netfilter/nf_conntrack.h>
  23 #include <net/netfilter/nf_conntrack_helper.h>
  24 #include <net/netfilter/nf_conntrack_ecache.h>
  25 #include <net/netfilter/nf_conntrack_expect.h>
  26 #include <net/netfilter/nf_nat.h>
  27 #include <net/netfilter/nf_nat_protocol.h>
  28 #include <net/netfilter/nf_nat_core.h>
  29 #include <net/netfilter/nf_nat_helper.h>
  30
  31 #define DUMP_OFFSET(x) \
  32         pr_debug("offset_before=%d, offset_after=%d, correction_pos=%u\n", \
  33                  x->offset_before, x->offset_after, x->correction_pos);
  34
  35 static DEFINE_SPINLOCK(nf_nat_seqofs_lock);
  36
  37 /* Setup TCP sequence correction given this change at this sequence */
  38 static inline void
  39 adjust_tcp_sequence(u32 seq,
  40                     int sizediff,
  41                     struct nf_conn *ct,
  42                     enum ip_conntrack_info ctinfo)
  43 {
  44         int dir;
  45         struct nf_nat_seq *this_way, *other_way;
  46         struct nf_conn_nat *nat = nfct_nat(ct);
  47
  48         pr_debug("adjust_tcp_sequence: seq = %u, sizediff = %d\n", seq, seq);
  49
  50         dir = CTINFO2DIR(ctinfo);
  51
  52         this_way = &nat->seq[dir];
  53         other_way = &nat->seq[!dir];
  54
  55         pr_debug("nf_nat_resize_packet: Seq_offset before: ");
  56         DUMP_OFFSET(this_way);
  57
  58         spin_lock_bh(&nf_nat_seqofs_lock);
  59
  60         /* SYN adjust. If it's uninitialized, or this is after last
  61          * correction, record it: we don't handle more than one
  62          * adjustment in the window, but do deal with common case of a
  63          * retransmit */
  64         if (this_way->offset_before == this_way->offset_after ||
  65             before(this_way->correction_pos, seq)) {
  66                    this_way->correction_pos = seq;
  67                    this_way->offset_before = this_way->offset_after;
  68                    this_way->offset_after += sizediff;
  69         }
  70         spin_unlock_bh(&nf_nat_seqofs_lock);
  71
  72         pr_debug("nf_nat_resize_packet: Seq_offset after: ");
  73         DUMP_OFFSET(this_way);
  74 }
  75
  76 /* Frobs data inside this packet, which is linear. */
  77 static void mangle_contents(struct sk_buff *skb,
  78                             unsigned int dataoff,
  79                             unsigned int match_offset,
  80                             unsigned int match_len,
  81                             const char *rep_buffer,
  82                             unsigned int rep_len)
  83 {
  84         unsigned char *data;
  85
  86         BUG_ON(skb_is_nonlinear(skb));
  87         data = skb_network_header(skb) + dataoff;
  88
  89         /* move post-replacement */
  90         memmove(data + match_offset + rep_len,
  91                 data + match_offset + match_len,
  92                 skb->tail - (skb->network_header + dataoff +
  93                              match_offset + match_len));
  94
  95         /* insert data from buffer */
  96         memcpy(data + match_offset, rep_buffer, rep_len);
  97
  98         /* update skb info */
  99         if (rep_len > match_len) {
 100                 pr_debug("nf_nat_mangle_packet: Extending packet by "
 101                          "%u from %u bytes\n", rep_len - match_len, skb->len);
 102                 skb_put(skb, rep_len - match_len);
 103         } else {
 104                 pr_debug("nf_nat_mangle_packet: Shrinking packet from "
 105                          "%u from %u bytes\n", match_len - rep_len, skb->len);
 106                 __skb_trim(skb, skb->len + rep_len - match_len);
 107         }
 108
 109         /* fix IP hdr checksum information */
 110         ip_hdr(skb)->tot_len = htons(skb->len);
 111         ip_send_check(ip_hdr(skb));
 112 }
 113
 114 /* Unusual, but possible case. */
 115 static int enlarge_skb(struct sk_buff *skb, unsigned int extra)
 116 {
 117         if (skb->len + extra > 65535)
 118                 return 0;
 119
 120         if (pskb_expand_head(skb, 0, extra - skb_tailroom(skb), GFP_ATOMIC))
 121                 return 0;
 122
 123         return 1;
 124 }
 125
 126 /* Generic function for mangling variable-length address changes inside
 127  * NATed TCP connections (like the PORT XXX,XXX,XXX,XXX,XXX,XXX
 128  * command in FTP).
 129  *
 130  * Takes care about all the nasty sequence number changes, checksumming,
 131  * skb enlargement, ...
 132  *
 133  * */
 134 int
 135 nf_nat_mangle_tcp_packet(struct sk_buff *skb,
 136                          struct nf_conn *ct,
 137                          enum ip_conntrack_info ctinfo,
 138                          unsigned int match_offset,
 139                          unsigned int match_len,
 140                          const char *rep_buffer,
 141                          unsigned int rep_len)
 142 {
 143         struct rtable *rt = skb_rtable(skb);
 144         struct iphdr *iph;
 145         struct tcphdr *tcph;
 146         int oldlen, datalen;
 147
 148         if (!skb_make_writable(skb, skb->len))
 149                 return 0;
 150
 151         if (rep_len > match_len &&
 152             rep_len - match_len > skb_tailroom(skb) &&
 153             !enlarge_skb(skb, rep_len - match_len))
 154                 return 0;
 155
 156         SKB_LINEAR_ASSERT(skb);
 157
 158         iph = ip_hdr(skb);
 159         tcph = (void *)iph + iph->ihl*4;
 160
 161         oldlen = skb->len - iph->ihl*4;
 162         mangle_contents(skb, iph->ihl*4 + tcph->doff*4,
 163                         match_offset, match_len, rep_buffer, rep_len);
 164
 165         datalen = skb->len - iph->ihl*4;
 166         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 167                 if (!(rt->rt_flags & RTCF_LOCAL) &&
 168                     skb->dev->features & NETIF_F_V4_CSUM) {
 169                         skb->ip_summed = CHECKSUM_PARTIAL;
 170                         skb->csum_start = skb_headroom(skb) +
 171                                           skb_network_offset(skb) +
 172                                           iph->ihl * 4;
 173                         skb->csum_offset = offsetof(struct tcphdr, check);
 174                         tcph->check = ~tcp_v4_check(datalen,
 175                                                     iph->saddr, iph->daddr, 0);
 176                 } else {
 177                         tcph->check = 0;
 178                         tcph->check = tcp_v4_check(datalen,
 179                                                    iph->saddr, iph->daddr,
 180                                                    csum_partial(tcph,
 181                                                                 datalen, 0));
 182                 }
 183         } else
 184                 inet_proto_csum_replace2(&tcph->check, skb,
 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(skb, ip_hdrlen(skb),
 194                                         ct, CTINFO2DIR(ctinfo));
 195
 196                 nf_conntrack_event_cache(IPCT_NATSEQADJ, ct);
 197         }
 198         return 1;
 199 }
 200 EXPORT_SYMBOL(nf_nat_mangle_tcp_packet);
 201
 202 /* Generic function for mangling variable-length address changes inside
 203  * NATed UDP connections (like the CONNECT DATA XXXXX MESG XXXXX INDEX XXXXX
 204  * command in the Amanda protocol)
 205  *
 206  * Takes care about all the nasty sequence number changes, checksumming,
 207  * skb enlargement, ...
 208  *
 209  * XXX - This function could be merged with nf_nat_mangle_tcp_packet which
 210  *       should be fairly easy to do.
 211  */
 212 int
 213 nf_nat_mangle_udp_packet(struct sk_buff *skb,
 214                          struct nf_conn *ct,
 215                          enum ip_conntrack_info ctinfo,
 216                          unsigned int match_offset,
 217                          unsigned int match_len,
 218                          const char *rep_buffer,
 219                          unsigned int rep_len)
 220 {
 221         struct rtable *rt = skb_rtable(skb);
 222         struct iphdr *iph;
 223         struct udphdr *udph;
 224         int datalen, oldlen;
 225
 226         /* UDP helpers might accidentally mangle the wrong packet */
 227         iph = ip_hdr(skb);
 228         if (skb->len < iph->ihl*4 + sizeof(*udph) +
 229                                match_offset + match_len)
 230                 return 0;
 231
 232         if (!skb_make_writable(skb, skb->len))
 233                 return 0;
 234
 235         if (rep_len > match_len &&
 236             rep_len - match_len > skb_tailroom(skb) &&
 237             !enlarge_skb(skb, rep_len - match_len))
 238                 return 0;
 239
 240         iph = ip_hdr(skb);
 241         udph = (void *)iph + iph->ihl*4;
 242
 243         oldlen = skb->len - iph->ihl*4;
 244         mangle_contents(skb, iph->ihl*4 + sizeof(*udph),
 245                         match_offset, match_len, rep_buffer, rep_len);
 246
 247         /* update the length of the UDP packet */
 248         datalen = skb->len - iph->ihl*4;
 249         udph->len = htons(datalen);
 250
 251         /* fix udp checksum if udp checksum was previously calculated */
 252         if (!udph->check && skb->ip_summed != CHECKSUM_PARTIAL)
 253                 return 1;
 254
 255         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 256                 if (!(rt->rt_flags & RTCF_LOCAL) &&
 257                     skb->dev->features & NETIF_F_V4_CSUM) {
 258                         skb->ip_summed = CHECKSUM_PARTIAL;
 259                         skb->csum_start = skb_headroom(skb) +
 260                                           skb_network_offset(skb) +
 261                                           iph->ihl * 4;
 262                         skb->csum_offset = offsetof(struct udphdr, check);
 263                         udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
 264                                                          datalen, IPPROTO_UDP,
 265                                                          0);
 266                 } else {
 267                         udph->check = 0;
 268                         udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 269                                                         datalen, IPPROTO_UDP,
 270                                                         csum_partial(udph,
 271                                                                      datalen, 0));
 272                         if (!udph->check)
 273                                 udph->check = CSUM_MANGLED_0;
 274                 }
 275         } else
 276                 inet_proto_csum_replace2(&udph->check, skb,
 277                                          htons(oldlen), htons(datalen), 1);
 278
 279         return 1;
 280 }
 281 EXPORT_SYMBOL(nf_nat_mangle_udp_packet);
 282
 283 /* Adjust one found SACK option including checksum correction */
 284 static void
 285 sack_adjust(struct sk_buff *skb,
 286             struct tcphdr *tcph,
 287             unsigned int sackoff,
 288             unsigned int sackend,
 289             struct nf_nat_seq *natseq)
 290 {
 291         while (sackoff < sackend) {
 292                 struct tcp_sack_block_wire *sack;
 293                 __be32 new_start_seq, new_end_seq;
 294
 295                 sack = (void *)skb->data + sackoff;
 296                 if (after(ntohl(sack->start_seq) - natseq->offset_before,
 297                           natseq->correction_pos))
 298                         new_start_seq = htonl(ntohl(sack->start_seq)
 299                                         - natseq->offset_after);
 300                 else
 301                         new_start_seq = htonl(ntohl(sack->start_seq)
 302                                         - natseq->offset_before);
 303
 304                 if (after(ntohl(sack->end_seq) - natseq->offset_before,
 305                           natseq->correction_pos))
 306                         new_end_seq = htonl(ntohl(sack->end_seq)
 307                                       - natseq->offset_after);
 308                 else
 309                         new_end_seq = htonl(ntohl(sack->end_seq)
 310                                       - natseq->offset_before);
 311
 312                 pr_debug("sack_adjust: start_seq: %d->%d, end_seq: %d->%d\n",
 313                          ntohl(sack->start_seq), new_start_seq,
 314                          ntohl(sack->end_seq), new_end_seq);
 315
 316                 inet_proto_csum_replace4(&tcph->check, skb,
 317                                          sack->start_seq, new_start_seq, 0);
 318                 inet_proto_csum_replace4(&tcph->check, skb,
 319                                          sack->end_seq, new_end_seq, 0);
 320                 sack->start_seq = new_start_seq;
 321                 sack->end_seq = new_end_seq;
 322                 sackoff += sizeof(*sack);
 323         }
 324 }
 325
 326 /* TCP SACK sequence number adjustment */
 327 static inline unsigned int
 328 nf_nat_sack_adjust(struct sk_buff *skb,
 329                    struct tcphdr *tcph,
 330                    struct nf_conn *ct,
 331                    enum ip_conntrack_info ctinfo)
 332 {
 333         unsigned int dir, optoff, optend;
 334         struct nf_conn_nat *nat = nfct_nat(ct);
 335
 336         optoff = ip_hdrlen(skb) + sizeof(struct tcphdr);
 337         optend = ip_hdrlen(skb) + tcph->doff * 4;
 338
 339         if (!skb_make_writable(skb, optend))
 340                 return 0;
 341
 342         dir = CTINFO2DIR(ctinfo);
 343
 344         while (optoff < optend) {
 345                 /* Usually: option, length. */
 346                 unsigned char *op = skb->data + optoff;
 347
 348                 switch (op[0]) {
 349                 case TCPOPT_EOL:
 350                         return 1;
 351                 case TCPOPT_NOP:
 352                         optoff++;
 353                         continue;
 354                 default:
 355                         /* no partial options */
 356                         if (optoff + 1 == optend ||
 357                             optoff + op[1] > optend ||
 358                             op[1] < 2)
 359                                 return 0;
 360                         if (op[0] == TCPOPT_SACK &&
 361                             op[1] >= 2+TCPOLEN_SACK_PERBLOCK &&
 362                             ((op[1] - 2) % TCPOLEN_SACK_PERBLOCK) == 0)
 363                                 sack_adjust(skb, tcph, optoff+2,
 364                                             optoff+op[1], &nat->seq[!dir]);
 365                         optoff += op[1];
 366                 }
 367         }
 368         return 1;
 369 }
 370
 371 /* TCP sequence number adjustment.  Returns 1 on success, 0 on failure */
 372 int
 373 nf_nat_seq_adjust(struct sk_buff *skb,
 374                   struct nf_conn *ct,
 375                   enum ip_conntrack_info ctinfo)
 376 {
 377         struct tcphdr *tcph;
 378         int dir;
 379         __be32 newseq, newack;
 380         struct nf_conn_nat *nat = nfct_nat(ct);
 381         struct nf_nat_seq *this_way, *other_way;
 382
 383         dir = CTINFO2DIR(ctinfo);
 384
 385         this_way = &nat->seq[dir];
 386         other_way = &nat->seq[!dir];
 387
 388         if (!skb_make_writable(skb, ip_hdrlen(skb) + sizeof(*tcph)))
 389                 return 0;
 390
 391         tcph = (void *)skb->data + ip_hdrlen(skb);
 392         if (after(ntohl(tcph->seq), this_way->correction_pos))
 393                 newseq = htonl(ntohl(tcph->seq) + this_way->offset_after);
 394         else
 395                 newseq = htonl(ntohl(tcph->seq) + this_way->offset_before);
 396
 397         if (after(ntohl(tcph->ack_seq) - other_way->offset_before,
 398                   other_way->correction_pos))
 399                 newack = htonl(ntohl(tcph->ack_seq) - other_way->offset_after);
 400         else
 401                 newack = htonl(ntohl(tcph->ack_seq) - other_way->offset_before);
 402
 403         inet_proto_csum_replace4(&tcph->check, skb, tcph->seq, newseq, 0);
 404         inet_proto_csum_replace4(&tcph->check, skb, tcph->ack_seq, newack, 0);
 405
 406         pr_debug("Adjusting sequence number from %u->%u, ack from %u->%u\n",
 407                  ntohl(tcph->seq), ntohl(newseq), ntohl(tcph->ack_seq),
 408                  ntohl(newack));
 409
 410         tcph->seq = newseq;
 411         tcph->ack_seq = newack;
 412
 413         if (!nf_nat_sack_adjust(skb, tcph, ct, ctinfo))
 414                 return 0;
 415
 416         nf_conntrack_tcp_update(skb, ip_hdrlen(skb), ct, dir);
 417
 418         return 1;
 419 }
 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 nf_nat_follow_master(struct nf_conn *ct,
 424                           struct nf_conntrack_expect *exp)
 425 {
 426         struct nf_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.u3.ip;
 435         nf_nat_setup_info(ct, &range, IP_NAT_MANIP_SRC);
 436
 437         /* For DST manip, map port here to where it's expected. */
 438         range.flags = (IP_NAT_RANGE_MAP_IPS | IP_NAT_RANGE_PROTO_SPECIFIED);
 439         range.min = range.max = exp->saved_proto;
 440         range.min_ip = range.max_ip
 441                 = ct->master->tuplehash[!exp->dir].tuple.src.u3.ip;
 442         nf_nat_setup_info(ct, &range, IP_NAT_MANIP_DST);
 443 }
 444 EXPORT_SYMBOL(nf_nat_follow_master);