net/netfilter/nf_nat_helper.c

   1 /* nf_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  * (C) 2007-2012 Patrick McHardy <kaber@trash.net>
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/module.h>
  12 #include <linux/gfp.h>
  13 #include <linux/types.h>
  14 #include <linux/skbuff.h>
  15 #include <linux/tcp.h>
  16 #include <linux/udp.h>
  17 #include <net/tcp.h>
  18
  19 #include <net/netfilter/nf_conntrack.h>
  20 #include <net/netfilter/nf_conntrack_helper.h>
  21 #include <net/netfilter/nf_conntrack_ecache.h>
  22 #include <net/netfilter/nf_conntrack_expect.h>
  23 #include <net/netfilter/nf_nat.h>
  24 #include <net/netfilter/nf_nat_l3proto.h>
  25 #include <net/netfilter/nf_nat_l4proto.h>
  26 #include <net/netfilter/nf_nat_core.h>
  27 #include <net/netfilter/nf_nat_helper.h>
  28
  29 #define DUMP_OFFSET(x) \
  30         pr_debug("offset_before=%d, offset_after=%d, correction_pos=%u\n", \
  31                  x->offset_before, x->offset_after, x->correction_pos);
  32
  33 /* Setup TCP sequence correction given this change at this sequence */
  34 static inline void
  35 adjust_tcp_sequence(u32 seq,
  36                     int sizediff,
  37                     struct nf_conn *ct,
  38                     enum ip_conntrack_info ctinfo)
  39 {
  40         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
  41         struct nf_conn_nat *nat = nfct_nat(ct);
  42         struct nf_nat_seq *this_way = &nat->seq[dir];
  43
  44         pr_debug("adjust_tcp_sequence: seq = %u, sizediff = %d\n",
  45                  seq, sizediff);
  46
  47         pr_debug("adjust_tcp_sequence: Seq_offset before: ");
  48         DUMP_OFFSET(this_way);
  49
  50         spin_lock_bh(&ct->lock);
  51
  52         /* SYN adjust. If it's uninitialized, or this is after last
  53          * correction, record it: we don't handle more than one
  54          * adjustment in the window, but do deal with common case of a
  55          * retransmit */
  56         if (this_way->offset_before == this_way->offset_after ||
  57             before(this_way->correction_pos, seq)) {
  58                 this_way->correction_pos = seq;
  59                 this_way->offset_before = this_way->offset_after;
  60                 this_way->offset_after += sizediff;
  61         }
  62         spin_unlock_bh(&ct->lock);
  63
  64         pr_debug("adjust_tcp_sequence: Seq_offset after: ");
  65         DUMP_OFFSET(this_way);
  66 }
  67
  68 /* Get the offset value, for conntrack. Caller must have the conntrack locked */
  69 s32 nf_nat_get_offset(const struct nf_conn *ct,
  70                       enum ip_conntrack_dir dir,
  71                       u32 seq)
  72 {
  73         struct nf_conn_nat *nat = nfct_nat(ct);
  74         struct nf_nat_seq *this_way;
  75
  76         if (!nat)
  77                 return 0;
  78
  79         this_way = &nat->seq[dir];
  80         return after(seq, this_way->correction_pos)
  81                  ? this_way->offset_after : this_way->offset_before;
  82 }
  83
  84 /* Frobs data inside this packet, which is linear. */
  85 static void mangle_contents(struct sk_buff *skb,
  86                             unsigned int dataoff,
  87                             unsigned int match_offset,
  88                             unsigned int match_len,
  89                             const char *rep_buffer,
  90                             unsigned int rep_len)
  91 {
  92         unsigned char *data;
  93
  94         BUG_ON(skb_is_nonlinear(skb));
  95         data = skb_network_header(skb) + dataoff;
  96
  97         /* move post-replacement */
  98         memmove(data + match_offset + rep_len,
  99                 data + match_offset + match_len,
 100                 skb_tail_pointer(skb) - (skb_network_header(skb) + dataoff +
 101                              match_offset + match_len));
 102
 103         /* insert data from buffer */
 104         memcpy(data + match_offset, rep_buffer, rep_len);
 105
 106         /* update skb info */
 107         if (rep_len > match_len) {
 108                 pr_debug("nf_nat_mangle_packet: Extending packet by "
 109                          "%u from %u bytes\n", rep_len - match_len, skb->len);
 110                 skb_put(skb, rep_len - match_len);
 111         } else {
 112                 pr_debug("nf_nat_mangle_packet: Shrinking packet from "
 113                          "%u from %u bytes\n", match_len - rep_len, skb->len);
 114                 __skb_trim(skb, skb->len + rep_len - match_len);
 115         }
 116
 117         if (nf_ct_l3num((struct nf_conn *)skb->nfct) == NFPROTO_IPV4) {
 118                 /* fix IP hdr checksum information */
 119                 ip_hdr(skb)->tot_len = htons(skb->len);
 120                 ip_send_check(ip_hdr(skb));
 121         } else
 122                 ipv6_hdr(skb)->payload_len =
 123                         htons(skb->len - sizeof(struct ipv6hdr));
 124 }
 125
 126 /* Unusual, but possible case. */
 127 static int enlarge_skb(struct sk_buff *skb, unsigned int extra)
 128 {
 129         if (skb->len + extra > 65535)
 130                 return 0;
 131
 132         if (pskb_expand_head(skb, 0, extra - skb_tailroom(skb), GFP_ATOMIC))
 133                 return 0;
 134
 135         return 1;
 136 }
 137
 138 void nf_nat_set_seq_adjust(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
 139                            __be32 seq, s32 off)
 140 {
 141         if (!off)
 142                 return;
 143         set_bit(IPS_SEQ_ADJUST_BIT, &ct->status);
 144         adjust_tcp_sequence(ntohl(seq), off, ct, ctinfo);
 145         nf_conntrack_event_cache(IPCT_NATSEQADJ, ct);
 146 }
 147 EXPORT_SYMBOL_GPL(nf_nat_set_seq_adjust);
 148
 149 void nf_nat_tcp_seq_adjust(struct sk_buff *skb, struct nf_conn *ct,
 150                            u32 ctinfo, int off)
 151 {
 152         const struct tcphdr *th;
 153
 154         if (nf_ct_protonum(ct) != IPPROTO_TCP)
 155                 return;
 156
 157         th = (struct tcphdr *)(skb_network_header(skb)+ ip_hdrlen(skb));
 158         nf_nat_set_seq_adjust(ct, ctinfo, th->seq, off);
 159 }
 160 EXPORT_SYMBOL_GPL(nf_nat_tcp_seq_adjust);
 161
 162 /* Generic function for mangling variable-length address changes inside
 163  * NATed TCP connections (like the PORT XXX,XXX,XXX,XXX,XXX,XXX
 164  * command in FTP).
 165  *
 166  * Takes care about all the nasty sequence number changes, checksumming,
 167  * skb enlargement, ...
 168  *
 169  * */
 170 int __nf_nat_mangle_tcp_packet(struct sk_buff *skb,
 171                                struct nf_conn *ct,
 172                                enum ip_conntrack_info ctinfo,
 173                                unsigned int protoff,
 174                                unsigned int match_offset,
 175                                unsigned int match_len,
 176                                const char *rep_buffer,
 177                                unsigned int rep_len, bool adjust)
 178 {
 179         const struct nf_nat_l3proto *l3proto;
 180         struct tcphdr *tcph;
 181         int oldlen, datalen;
 182
 183         if (!skb_make_writable(skb, skb->len))
 184                 return 0;
 185
 186         if (rep_len > match_len &&
 187             rep_len - match_len > skb_tailroom(skb) &&
 188             !enlarge_skb(skb, rep_len - match_len))
 189                 return 0;
 190
 191         SKB_LINEAR_ASSERT(skb);
 192
 193         tcph = (void *)skb->data + protoff;
 194
 195         oldlen = skb->len - protoff;
 196         mangle_contents(skb, protoff + tcph->doff*4,
 197                         match_offset, match_len, rep_buffer, rep_len);
 198
 199         datalen = skb->len - protoff;
 200
 201         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
 202         l3proto->csum_recalc(skb, IPPROTO_TCP, tcph, &tcph->check,
 203                              datalen, oldlen);
 204
 205         if (adjust && rep_len != match_len)
 206                 nf_nat_set_seq_adjust(ct, ctinfo, tcph->seq,
 207                                       (int)rep_len - (int)match_len);
 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 *skb,
 225                          struct nf_conn *ct,
 226                          enum ip_conntrack_info ctinfo,
 227                          unsigned int protoff,
 228                          unsigned int match_offset,
 229                          unsigned int match_len,
 230                          const char *rep_buffer,
 231                          unsigned int rep_len)
 232 {
 233         const struct nf_nat_l3proto *l3proto;
 234         struct udphdr *udph;
 235         int datalen, oldlen;
 236
 237         if (!skb_make_writable(skb, skb->len))
 238                 return 0;
 239
 240         if (rep_len > match_len &&
 241             rep_len - match_len > skb_tailroom(skb) &&
 242             !enlarge_skb(skb, rep_len - match_len))
 243                 return 0;
 244
 245         udph = (void *)skb->data + protoff;
 246
 247         oldlen = skb->len - protoff;
 248         mangle_contents(skb, protoff + sizeof(*udph),
 249                         match_offset, match_len, rep_buffer, rep_len);
 250
 251         /* update the length of the UDP packet */
 252         datalen = skb->len - protoff;
 253         udph->len = htons(datalen);
 254
 255         /* fix udp checksum if udp checksum was previously calculated */
 256         if (!udph->check && skb->ip_summed != CHECKSUM_PARTIAL)
 257                 return 1;
 258
 259         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
 260         l3proto->csum_recalc(skb, IPPROTO_UDP, udph, &udph->check,
 261                              datalen, oldlen);
 262
 263         return 1;
 264 }
 265 EXPORT_SYMBOL(nf_nat_mangle_udp_packet);
 266
 267 /* Adjust one found SACK option including checksum correction */
 268 static void
 269 sack_adjust(struct sk_buff *skb,
 270             struct tcphdr *tcph,
 271             unsigned int sackoff,
 272             unsigned int sackend,
 273             struct nf_nat_seq *natseq)
 274 {
 275         while (sackoff < sackend) {
 276                 struct tcp_sack_block_wire *sack;
 277                 __be32 new_start_seq, new_end_seq;
 278
 279                 sack = (void *)skb->data + sackoff;
 280                 if (after(ntohl(sack->start_seq) - natseq->offset_before,
 281                           natseq->correction_pos))
 282                         new_start_seq = htonl(ntohl(sack->start_seq)
 283                                         - natseq->offset_after);
 284                 else
 285                         new_start_seq = htonl(ntohl(sack->start_seq)
 286                                         - natseq->offset_before);
 287
 288                 if (after(ntohl(sack->end_seq) - natseq->offset_before,
 289                           natseq->correction_pos))
 290                         new_end_seq = htonl(ntohl(sack->end_seq)
 291                                       - natseq->offset_after);
 292                 else
 293                         new_end_seq = htonl(ntohl(sack->end_seq)
 294                                       - natseq->offset_before);
 295
 296                 pr_debug("sack_adjust: start_seq: %d->%d, end_seq: %d->%d\n",
 297                          ntohl(sack->start_seq), new_start_seq,
 298                          ntohl(sack->end_seq), new_end_seq);
 299
 300                 inet_proto_csum_replace4(&tcph->check, skb,
 301                                          sack->start_seq, new_start_seq, 0);
 302                 inet_proto_csum_replace4(&tcph->check, skb,
 303                                          sack->end_seq, new_end_seq, 0);
 304                 sack->start_seq = new_start_seq;
 305                 sack->end_seq = new_end_seq;
 306                 sackoff += sizeof(*sack);
 307         }
 308 }
 309
 310 /* TCP SACK sequence number adjustment */
 311 static inline unsigned int
 312 nf_nat_sack_adjust(struct sk_buff *skb,
 313                    unsigned int protoff,
 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 = protoff + sizeof(struct tcphdr);
 322         optend = protoff + tcph->doff * 4;
 323
 324         if (!skb_make_writable(skb, optend))
 325                 return 0;
 326
 327         dir = CTINFO2DIR(ctinfo);
 328
 329         while (optoff < optend) {
 330                 /* Usually: option, length. */
 331                 unsigned char *op = skb->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(skb, tcph, optoff+2,
 349                                             optoff+op[1], &nat->seq[!dir]);
 350                         optoff += op[1];
 351                 }
 352         }
 353         return 1;
 354 }
 355
 356 /* TCP sequence number adjustment.  Returns 1 on success, 0 on failure */
 357 int
 358 nf_nat_seq_adjust(struct sk_buff *skb,
 359                   struct nf_conn *ct,
 360                   enum ip_conntrack_info ctinfo,
 361                   unsigned int protoff)
 362 {
 363         struct tcphdr *tcph;
 364         int dir;
 365         __be32 newseq, newack;
 366         s32 seqoff, ackoff;
 367         struct nf_conn_nat *nat = nfct_nat(ct);
 368         struct nf_nat_seq *this_way, *other_way;
 369         int res;
 370
 371         dir = CTINFO2DIR(ctinfo);
 372
 373         this_way = &nat->seq[dir];
 374         other_way = &nat->seq[!dir];
 375
 376         if (!skb_make_writable(skb, protoff + sizeof(*tcph)))
 377                 return 0;
 378
 379         tcph = (void *)skb->data + protoff;
 380         spin_lock_bh(&ct->lock);
 381         if (after(ntohl(tcph->seq), this_way->correction_pos))
 382                 seqoff = this_way->offset_after;
 383         else
 384                 seqoff = this_way->offset_before;
 385
 386         if (after(ntohl(tcph->ack_seq) - other_way->offset_before,
 387                   other_way->correction_pos))
 388                 ackoff = other_way->offset_after;
 389         else
 390                 ackoff = other_way->offset_before;
 391
 392         newseq = htonl(ntohl(tcph->seq) + seqoff);
 393         newack = htonl(ntohl(tcph->ack_seq) - ackoff);
 394
 395         inet_proto_csum_replace4(&tcph->check, skb, tcph->seq, newseq, 0);
 396         inet_proto_csum_replace4(&tcph->check, skb, tcph->ack_seq, newack, 0);
 397
 398         pr_debug("Adjusting sequence number from %u->%u, ack from %u->%u\n",
 399                  ntohl(tcph->seq), ntohl(newseq), ntohl(tcph->ack_seq),
 400                  ntohl(newack));
 401
 402         tcph->seq = newseq;
 403         tcph->ack_seq = newack;
 404
 405         res = nf_nat_sack_adjust(skb, protoff, tcph, ct, ctinfo);
 406         spin_unlock_bh(&ct->lock);
 407
 408         return res;
 409 }
 410
 411 /* Setup NAT on this expected conntrack so it follows master. */
 412 /* If we fail to get a free NAT slot, we'll get dropped on confirm */
 413 void nf_nat_follow_master(struct nf_conn *ct,
 414                           struct nf_conntrack_expect *exp)
 415 {
 416         struct nf_nat_range range;
 417
 418         /* This must be a fresh one. */
 419         BUG_ON(ct->status & IPS_NAT_DONE_MASK);
 420
 421         /* Change src to where master sends to */
 422         range.flags = NF_NAT_RANGE_MAP_IPS;
 423         range.min_addr = range.max_addr
 424                 = ct->master->tuplehash[!exp->dir].tuple.dst.u3;
 425         nf_nat_setup_info(ct, &range, NF_NAT_MANIP_SRC);
 426
 427         /* For DST manip, map port here to where it's expected. */
 428         range.flags = (NF_NAT_RANGE_MAP_IPS | NF_NAT_RANGE_PROTO_SPECIFIED);
 429         range.min_proto = range.max_proto = exp->saved_proto;
 430         range.min_addr = range.max_addr
 431                 = ct->master->tuplehash[!exp->dir].tuple.src.u3;
 432         nf_nat_setup_info(ct, &range, NF_NAT_MANIP_DST);
 433 }
 434 EXPORT_SYMBOL(nf_nat_follow_master);