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 /* Get the offset value, for conntrack */
  77 s16 nf_nat_get_offset(const struct nf_conn *ct,
  78                       enum ip_conntrack_dir dir,
  79                       u32 seq)
  80 {
  81         struct nf_conn_nat *nat = nfct_nat(ct);
  82         struct nf_nat_seq *this_way;
  83         s16 offset;
  84
  85         if (!nat)
  86                 return 0;
  87
  88         this_way = &nat->seq[dir];
  89         spin_lock_bh(&nf_nat_seqofs_lock);
  90         offset = after(seq, this_way->correction_pos)
  91                  ? this_way->offset_after : this_way->offset_before;
  92         spin_unlock_bh(&nf_nat_seqofs_lock);
  93
  94         return offset;
  95 }
  96 EXPORT_SYMBOL_GPL(nf_nat_get_offset);
  97
  98 /* Frobs data inside this packet, which is linear. */
  99 static void mangle_contents(struct sk_buff *skb,
 100                             unsigned int dataoff,
 101                             unsigned int match_offset,
 102                             unsigned int match_len,
 103                             const char *rep_buffer,
 104                             unsigned int rep_len)
 105 {
 106         unsigned char *data;
 107
 108         BUG_ON(skb_is_nonlinear(skb));
 109         data = skb_network_header(skb) + dataoff;
 110
 111         /* move post-replacement */
 112         memmove(data + match_offset + rep_len,
 113                 data + match_offset + match_len,
 114                 skb->tail - (skb->network_header + dataoff +
 115                              match_offset + match_len));
 116
 117         /* insert data from buffer */
 118         memcpy(data + match_offset, rep_buffer, rep_len);
 119
 120         /* update skb info */
 121         if (rep_len > match_len) {
 122                 pr_debug("nf_nat_mangle_packet: Extending packet by "
 123                          "%u from %u bytes\n", rep_len - match_len, skb->len);
 124                 skb_put(skb, rep_len - match_len);
 125         } else {
 126                 pr_debug("nf_nat_mangle_packet: Shrinking packet from "
 127                          "%u from %u bytes\n", match_len - rep_len, skb->len);
 128                 __skb_trim(skb, skb->len + rep_len - match_len);
 129         }
 130
 131         /* fix IP hdr checksum information */
 132         ip_hdr(skb)->tot_len = htons(skb->len);
 133         ip_send_check(ip_hdr(skb));
 134 }
 135
 136 /* Unusual, but possible case. */
 137 static int enlarge_skb(struct sk_buff *skb, unsigned int extra)
 138 {
 139         if (skb->len + extra > 65535)
 140                 return 0;
 141
 142         if (pskb_expand_head(skb, 0, extra - skb_tailroom(skb), GFP_ATOMIC))
 143                 return 0;
 144
 145         return 1;
 146 }
 147
 148 /* Generic function for mangling variable-length address changes inside
 149  * NATed TCP connections (like the PORT XXX,XXX,XXX,XXX,XXX,XXX
 150  * command in FTP).
 151  *
 152  * Takes care about all the nasty sequence number changes, checksumming,
 153  * skb enlargement, ...
 154  *
 155  * */
 156 int
 157 nf_nat_mangle_tcp_packet(struct sk_buff *skb,
 158                          struct nf_conn *ct,
 159                          enum ip_conntrack_info ctinfo,
 160                          unsigned int match_offset,
 161                          unsigned int match_len,
 162                          const char *rep_buffer,
 163                          unsigned int rep_len)
 164 {
 165         struct rtable *rt = skb_rtable(skb);
 166         struct iphdr *iph;
 167         struct tcphdr *tcph;
 168         int oldlen, datalen;
 169
 170         if (!skb_make_writable(skb, skb->len))
 171                 return 0;
 172
 173         if (rep_len > match_len &&
 174             rep_len - match_len > skb_tailroom(skb) &&
 175             !enlarge_skb(skb, rep_len - match_len))
 176                 return 0;
 177
 178         SKB_LINEAR_ASSERT(skb);
 179
 180         iph = ip_hdr(skb);
 181         tcph = (void *)iph + iph->ihl*4;
 182
 183         oldlen = skb->len - iph->ihl*4;
 184         mangle_contents(skb, iph->ihl*4 + tcph->doff*4,
 185                         match_offset, match_len, rep_buffer, rep_len);
 186
 187         datalen = skb->len - iph->ihl*4;
 188         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 189                 if (!(rt->rt_flags & RTCF_LOCAL) &&
 190                     skb->dev->features & NETIF_F_V4_CSUM) {
 191                         skb->ip_summed = CHECKSUM_PARTIAL;
 192                         skb->csum_start = skb_headroom(skb) +
 193                                           skb_network_offset(skb) +
 194                                           iph->ihl * 4;
 195                         skb->csum_offset = offsetof(struct tcphdr, check);
 196                         tcph->check = ~tcp_v4_check(datalen,
 197                                                     iph->saddr, iph->daddr, 0);
 198                 } else {
 199                         tcph->check = 0;
 200                         tcph->check = tcp_v4_check(datalen,
 201                                                    iph->saddr, iph->daddr,
 202                                                    csum_partial(tcph,
 203                                                                 datalen, 0));
 204                 }
 205         } else
 206                 inet_proto_csum_replace2(&tcph->check, skb,
 207                                          htons(oldlen), htons(datalen), 1);
 208
 209         if (rep_len != match_len) {
 210                 set_bit(IPS_SEQ_ADJUST_BIT, &ct->status);
 211                 adjust_tcp_sequence(ntohl(tcph->seq),
 212                                     (int)rep_len - (int)match_len,
 213                                     ct, ctinfo);
 214                 nf_conntrack_event_cache(IPCT_NATSEQADJ, ct);
 215         }
 216         return 1;
 217 }
 218 EXPORT_SYMBOL(nf_nat_mangle_tcp_packet);
 219
 220 /* Generic function for mangling variable-length address changes inside
 221  * NATed UDP connections (like the CONNECT DATA XXXXX MESG XXXXX INDEX XXXXX
 222  * command in the Amanda protocol)
 223  *
 224  * Takes care about all the nasty sequence number changes, checksumming,
 225  * skb enlargement, ...
 226  *
 227  * XXX - This function could be merged with nf_nat_mangle_tcp_packet which
 228  *       should be fairly easy to do.
 229  */
 230 int
 231 nf_nat_mangle_udp_packet(struct sk_buff *skb,
 232                          struct nf_conn *ct,
 233                          enum ip_conntrack_info ctinfo,
 234                          unsigned int match_offset,
 235                          unsigned int match_len,
 236                          const char *rep_buffer,
 237                          unsigned int rep_len)
 238 {
 239         struct rtable *rt = skb_rtable(skb);
 240         struct iphdr *iph;
 241         struct udphdr *udph;
 242         int datalen, oldlen;
 243
 244         /* UDP helpers might accidentally mangle the wrong packet */
 245         iph = ip_hdr(skb);
 246         if (skb->len < iph->ihl*4 + sizeof(*udph) +
 247                                match_offset + match_len)
 248                 return 0;
 249
 250         if (!skb_make_writable(skb, skb->len))
 251                 return 0;
 252
 253         if (rep_len > match_len &&
 254             rep_len - match_len > skb_tailroom(skb) &&
 255             !enlarge_skb(skb, rep_len - match_len))
 256                 return 0;
 257
 258         iph = ip_hdr(skb);
 259         udph = (void *)iph + iph->ihl*4;
 260
 261         oldlen = skb->len - iph->ihl*4;
 262         mangle_contents(skb, iph->ihl*4 + sizeof(*udph),
 263                         match_offset, match_len, rep_buffer, rep_len);
 264
 265         /* update the length of the UDP packet */
 266         datalen = skb->len - iph->ihl*4;
 267         udph->len = htons(datalen);
 268
 269         /* fix udp checksum if udp checksum was previously calculated */
 270         if (!udph->check && skb->ip_summed != CHECKSUM_PARTIAL)
 271                 return 1;
 272
 273         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 274                 if (!(rt->rt_flags & RTCF_LOCAL) &&
 275                     skb->dev->features & NETIF_F_V4_CSUM) {
 276                         skb->ip_summed = CHECKSUM_PARTIAL;
 277                         skb->csum_start = skb_headroom(skb) +
 278                                           skb_network_offset(skb) +
 279                                           iph->ihl * 4;
 280                         skb->csum_offset = offsetof(struct udphdr, check);
 281                         udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
 282                                                          datalen, IPPROTO_UDP,
 283                                                          0);
 284                 } else {
 285                         udph->check = 0;
 286                         udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 287                                                         datalen, IPPROTO_UDP,
 288                                                         csum_partial(udph,
 289                                                                      datalen, 0));
 290                         if (!udph->check)
 291                                 udph->check = CSUM_MANGLED_0;
 292                 }
 293         } else
 294                 inet_proto_csum_replace2(&udph->check, skb,
 295                                          htons(oldlen), htons(datalen), 1);
 296
 297         return 1;
 298 }
 299 EXPORT_SYMBOL(nf_nat_mangle_udp_packet);
 300
 301 /* Adjust one found SACK option including checksum correction */
 302 static void
 303 sack_adjust(struct sk_buff *skb,
 304             struct tcphdr *tcph,
 305             unsigned int sackoff,
 306             unsigned int sackend,
 307             struct nf_nat_seq *natseq)
 308 {
 309         while (sackoff < sackend) {
 310                 struct tcp_sack_block_wire *sack;
 311                 __be32 new_start_seq, new_end_seq;
 312
 313                 sack = (void *)skb->data + sackoff;
 314                 if (after(ntohl(sack->start_seq) - natseq->offset_before,
 315                           natseq->correction_pos))
 316                         new_start_seq = htonl(ntohl(sack->start_seq)
 317                                         - natseq->offset_after);
 318                 else
 319                         new_start_seq = htonl(ntohl(sack->start_seq)
 320                                         - natseq->offset_before);
 321
 322                 if (after(ntohl(sack->end_seq) - natseq->offset_before,
 323                           natseq->correction_pos))
 324                         new_end_seq = htonl(ntohl(sack->end_seq)
 325                                       - natseq->offset_after);
 326                 else
 327                         new_end_seq = htonl(ntohl(sack->end_seq)
 328                                       - natseq->offset_before);
 329
 330                 pr_debug("sack_adjust: start_seq: %d->%d, end_seq: %d->%d\n",
 331                          ntohl(sack->start_seq), new_start_seq,
 332                          ntohl(sack->end_seq), new_end_seq);
 333
 334                 inet_proto_csum_replace4(&tcph->check, skb,
 335                                          sack->start_seq, new_start_seq, 0);
 336                 inet_proto_csum_replace4(&tcph->check, skb,
 337                                          sack->end_seq, new_end_seq, 0);
 338                 sack->start_seq = new_start_seq;
 339                 sack->end_seq = new_end_seq;
 340                 sackoff += sizeof(*sack);
 341         }
 342 }
 343
 344 /* TCP SACK sequence number adjustment */
 345 static inline unsigned int
 346 nf_nat_sack_adjust(struct sk_buff *skb,
 347                    struct tcphdr *tcph,
 348                    struct nf_conn *ct,
 349                    enum ip_conntrack_info ctinfo)
 350 {
 351         unsigned int dir, optoff, optend;
 352         struct nf_conn_nat *nat = nfct_nat(ct);
 353
 354         optoff = ip_hdrlen(skb) + sizeof(struct tcphdr);
 355         optend = ip_hdrlen(skb) + tcph->doff * 4;
 356
 357         if (!skb_make_writable(skb, optend))
 358                 return 0;
 359
 360         dir = CTINFO2DIR(ctinfo);
 361
 362         while (optoff < optend) {
 363                 /* Usually: option, length. */
 364                 unsigned char *op = skb->data + optoff;
 365
 366                 switch (op[0]) {
 367                 case TCPOPT_EOL:
 368                         return 1;
 369                 case TCPOPT_NOP:
 370                         optoff++;
 371                         continue;
 372                 default:
 373                         /* no partial options */
 374                         if (optoff + 1 == optend ||
 375                             optoff + op[1] > optend ||
 376                             op[1] < 2)
 377                                 return 0;
 378                         if (op[0] == TCPOPT_SACK &&
 379                             op[1] >= 2+TCPOLEN_SACK_PERBLOCK &&
 380                             ((op[1] - 2) % TCPOLEN_SACK_PERBLOCK) == 0)
 381                                 sack_adjust(skb, tcph, optoff+2,
 382                                             optoff+op[1], &nat->seq[!dir]);
 383                         optoff += op[1];
 384                 }
 385         }
 386         return 1;
 387 }
 388
 389 /* TCP sequence number adjustment.  Returns 1 on success, 0 on failure */
 390 int
 391 nf_nat_seq_adjust(struct sk_buff *skb,
 392                   struct nf_conn *ct,
 393                   enum ip_conntrack_info ctinfo)
 394 {
 395         struct tcphdr *tcph;
 396         int dir;
 397         __be32 newseq, newack;
 398         s16 seqoff, ackoff;
 399         struct nf_conn_nat *nat = nfct_nat(ct);
 400         struct nf_nat_seq *this_way, *other_way;
 401
 402         dir = CTINFO2DIR(ctinfo);
 403
 404         this_way = &nat->seq[dir];
 405         other_way = &nat->seq[!dir];
 406
 407         if (!skb_make_writable(skb, ip_hdrlen(skb) + sizeof(*tcph)))
 408                 return 0;
 409
 410         tcph = (void *)skb->data + ip_hdrlen(skb);
 411         if (after(ntohl(tcph->seq), this_way->correction_pos))
 412                 seqoff = this_way->offset_after;
 413         else
 414                 seqoff = this_way->offset_before;
 415
 416         if (after(ntohl(tcph->ack_seq) - other_way->offset_before,
 417                   other_way->correction_pos))
 418                 ackoff = other_way->offset_after;
 419         else
 420                 ackoff = other_way->offset_before;
 421
 422         newseq = htonl(ntohl(tcph->seq) + seqoff);
 423         newack = htonl(ntohl(tcph->ack_seq) - ackoff);
 424
 425         inet_proto_csum_replace4(&tcph->check, skb, tcph->seq, newseq, 0);
 426         inet_proto_csum_replace4(&tcph->check, skb, tcph->ack_seq, newack, 0);
 427
 428         pr_debug("Adjusting sequence number from %u->%u, ack from %u->%u\n",
 429                  ntohl(tcph->seq), ntohl(newseq), ntohl(tcph->ack_seq),
 430                  ntohl(newack));
 431
 432         tcph->seq = newseq;
 433         tcph->ack_seq = newack;
 434
 435         return nf_nat_sack_adjust(skb, tcph, ct, ctinfo);
 436 }
 437
 438 /* Setup NAT on this expected conntrack so it follows master. */
 439 /* If we fail to get a free NAT slot, we'll get dropped on confirm */
 440 void nf_nat_follow_master(struct nf_conn *ct,
 441                           struct nf_conntrack_expect *exp)
 442 {
 443         struct nf_nat_range range;
 444
 445         /* This must be a fresh one. */
 446         BUG_ON(ct->status & IPS_NAT_DONE_MASK);
 447
 448         /* Change src to where master sends to */
 449         range.flags = IP_NAT_RANGE_MAP_IPS;
 450         range.min_ip = range.max_ip
 451                 = ct->master->tuplehash[!exp->dir].tuple.dst.u3.ip;
 452         nf_nat_setup_info(ct, &range, IP_NAT_MANIP_SRC);
 453
 454         /* For DST manip, map port here to where it's expected. */
 455         range.flags = (IP_NAT_RANGE_MAP_IPS | IP_NAT_RANGE_PROTO_SPECIFIED);
 456         range.min = range.max = exp->saved_proto;
 457         range.min_ip = range.max_ip
 458                 = ct->master->tuplehash[!exp->dir].tuple.src.u3.ip;
 459         nf_nat_setup_info(ct, &range, IP_NAT_MANIP_DST);
 460 }
 461 EXPORT_SYMBOL(nf_nat_follow_master);