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 static DEFINE_SPINLOCK(nf_nat_seqofs_lock);
  34
  35 /* Setup TCP sequence correction given this change at this sequence */
  36 static inline void
  37 adjust_tcp_sequence(u32 seq,
  38                     int sizediff,
  39                     struct nf_conn *ct,
  40                     enum ip_conntrack_info ctinfo)
  41 {
  42         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
  43         struct nf_conn_nat *nat = nfct_nat(ct);
  44         struct nf_nat_seq *this_way = &nat->seq[dir];
  45
  46         pr_debug("adjust_tcp_sequence: seq = %u, sizediff = %d\n",
  47                  seq, sizediff);
  48
  49         pr_debug("adjust_tcp_sequence: Seq_offset before: ");
  50         DUMP_OFFSET(this_way);
  51
  52         spin_lock_bh(&nf_nat_seqofs_lock);
  53
  54         /* SYN adjust. If it's uninitialized, or this is after last
  55          * correction, record it: we don't handle more than one
  56          * adjustment in the window, but do deal with common case of a
  57          * retransmit */
  58         if (this_way->offset_before == this_way->offset_after ||
  59             before(this_way->correction_pos, seq)) {
  60                 this_way->correction_pos = seq;
  61                 this_way->offset_before = this_way->offset_after;
  62                 this_way->offset_after += sizediff;
  63         }
  64         spin_unlock_bh(&nf_nat_seqofs_lock);
  65
  66         pr_debug("adjust_tcp_sequence: Seq_offset after: ");
  67         DUMP_OFFSET(this_way);
  68 }
  69
  70 /* Get the offset value, for conntrack */
  71 s16 nf_nat_get_offset(const struct nf_conn *ct,
  72                       enum ip_conntrack_dir dir,
  73                       u32 seq)
  74 {
  75         struct nf_conn_nat *nat = nfct_nat(ct);
  76         struct nf_nat_seq *this_way;
  77         s16 offset;
  78
  79         if (!nat)
  80                 return 0;
  81
  82         this_way = &nat->seq[dir];
  83         spin_lock_bh(&nf_nat_seqofs_lock);
  84         offset = after(seq, this_way->correction_pos)
  85                  ? this_way->offset_after : this_way->offset_before;
  86         spin_unlock_bh(&nf_nat_seqofs_lock);
  87
  88         return offset;
  89 }
  90
  91 /* Frobs data inside this packet, which is linear. */
  92 static void mangle_contents(struct sk_buff *skb,
  93                             unsigned int dataoff,
  94                             unsigned int match_offset,
  95                             unsigned int match_len,
  96                             const char *rep_buffer,
  97                             unsigned int rep_len)
  98 {
  99         unsigned char *data;
 100
 101         BUG_ON(skb_is_nonlinear(skb));
 102         data = skb_network_header(skb) + dataoff;
 103
 104         /* move post-replacement */
 105         memmove(data + match_offset + rep_len,
 106                 data + match_offset + match_len,
 107                 skb_tail_pointer(skb) - (skb_network_header(skb) + dataoff +
 108                              match_offset + match_len));
 109
 110         /* insert data from buffer */
 111         memcpy(data + match_offset, rep_buffer, rep_len);
 112
 113         /* update skb info */
 114         if (rep_len > match_len) {
 115                 pr_debug("nf_nat_mangle_packet: Extending packet by "
 116                          "%u from %u bytes\n", rep_len - match_len, skb->len);
 117                 skb_put(skb, rep_len - match_len);
 118         } else {
 119                 pr_debug("nf_nat_mangle_packet: Shrinking packet from "
 120                          "%u from %u bytes\n", match_len - rep_len, skb->len);
 121                 __skb_trim(skb, skb->len + rep_len - match_len);
 122         }
 123
 124         if (nf_ct_l3num((struct nf_conn *)skb->nfct) == NFPROTO_IPV4) {
 125                 /* fix IP hdr checksum information */
 126                 ip_hdr(skb)->tot_len = htons(skb->len);
 127                 ip_send_check(ip_hdr(skb));
 128         } else
 129                 ipv6_hdr(skb)->payload_len =
 130                         htons(skb->len - sizeof(struct ipv6hdr));
 131 }
 132
 133 /* Unusual, but possible case. */
 134 static int enlarge_skb(struct sk_buff *skb, unsigned int extra)
 135 {
 136         if (skb->len + extra > 65535)
 137                 return 0;
 138
 139         if (pskb_expand_head(skb, 0, extra - skb_tailroom(skb), GFP_ATOMIC))
 140                 return 0;
 141
 142         return 1;
 143 }
 144
 145 void nf_nat_set_seq_adjust(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
 146                            __be32 seq, s16 off)
 147 {
 148         if (!off)
 149                 return;
 150         set_bit(IPS_SEQ_ADJUST_BIT, &ct->status);
 151         adjust_tcp_sequence(ntohl(seq), off, ct, ctinfo);
 152         nf_conntrack_event_cache(IPCT_NATSEQADJ, ct);
 153 }
 154 EXPORT_SYMBOL_GPL(nf_nat_set_seq_adjust);
 155
 156 void nf_nat_tcp_seq_adjust(struct sk_buff *skb, struct nf_conn *ct,
 157                            u32 ctinfo, int off)
 158 {
 159         const struct tcphdr *th;
 160
 161         if (nf_ct_protonum(ct) != IPPROTO_TCP)
 162                 return;
 163
 164         th = (struct tcphdr *)(skb_network_header(skb)+ ip_hdrlen(skb));
 165         nf_nat_set_seq_adjust(ct, ctinfo, th->seq, off);
 166 }
 167 EXPORT_SYMBOL_GPL(nf_nat_tcp_seq_adjust);
 168
 169 /* Generic function for mangling variable-length address changes inside
 170  * NATed TCP connections (like the PORT XXX,XXX,XXX,XXX,XXX,XXX
 171  * command in FTP).
 172  *
 173  * Takes care about all the nasty sequence number changes, checksumming,
 174  * skb enlargement, ...
 175  *
 176  * */
 177 int __nf_nat_mangle_tcp_packet(struct sk_buff *skb,
 178                                struct nf_conn *ct,
 179                                enum ip_conntrack_info ctinfo,
 180                                unsigned int protoff,
 181                                unsigned int match_offset,
 182                                unsigned int match_len,
 183                                const char *rep_buffer,
 184                                unsigned int rep_len, bool adjust)
 185 {
 186         const struct nf_nat_l3proto *l3proto;
 187         struct tcphdr *tcph;
 188         int oldlen, datalen;
 189
 190         if (!skb_make_writable(skb, skb->len))
 191                 return 0;
 192
 193         if (rep_len > match_len &&
 194             rep_len - match_len > skb_tailroom(skb) &&
 195             !enlarge_skb(skb, rep_len - match_len))
 196                 return 0;
 197
 198         SKB_LINEAR_ASSERT(skb);
 199
 200         tcph = (void *)skb->data + protoff;
 201
 202         oldlen = skb->len - protoff;
 203         mangle_contents(skb, protoff + tcph->doff*4,
 204                         match_offset, match_len, rep_buffer, rep_len);
 205
 206         datalen = skb->len - protoff;
 207
 208         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
 209         l3proto->csum_recalc(skb, IPPROTO_TCP, tcph, &tcph->check,
 210                              datalen, oldlen);
 211
 212         if (adjust && rep_len != match_len)
 213                 nf_nat_set_seq_adjust(ct, ctinfo, tcph->seq,
 214                                       (int)rep_len - (int)match_len);
 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 protoff,
 235                          unsigned int match_offset,
 236                          unsigned int match_len,
 237                          const char *rep_buffer,
 238                          unsigned int rep_len)
 239 {
 240         const struct nf_nat_l3proto *l3proto;
 241         struct udphdr *udph;
 242         int datalen, oldlen;
 243
 244         if (!skb_make_writable(skb, skb->len))
 245                 return 0;
 246
 247         if (rep_len > match_len &&
 248             rep_len - match_len > skb_tailroom(skb) &&
 249             !enlarge_skb(skb, rep_len - match_len))
 250                 return 0;
 251
 252         udph = (void *)skb->data + protoff;
 253
 254         oldlen = skb->len - protoff;
 255         mangle_contents(skb, protoff + sizeof(*udph),
 256                         match_offset, match_len, rep_buffer, rep_len);
 257
 258         /* update the length of the UDP packet */
 259         datalen = skb->len - protoff;
 260         udph->len = htons(datalen);
 261
 262         /* fix udp checksum if udp checksum was previously calculated */
 263         if (!udph->check && skb->ip_summed != CHECKSUM_PARTIAL)
 264                 return 1;
 265
 266         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
 267         l3proto->csum_recalc(skb, IPPROTO_UDP, udph, &udph->check,
 268                              datalen, oldlen);
 269
 270         return 1;
 271 }
 272 EXPORT_SYMBOL(nf_nat_mangle_udp_packet);
 273
 274 /* Adjust one found SACK option including checksum correction */
 275 static void
 276 sack_adjust(struct sk_buff *skb,
 277             struct tcphdr *tcph,
 278             unsigned int sackoff,
 279             unsigned int sackend,
 280             struct nf_nat_seq *natseq)
 281 {
 282         while (sackoff < sackend) {
 283                 struct tcp_sack_block_wire *sack;
 284                 __be32 new_start_seq, new_end_seq;
 285
 286                 sack = (void *)skb->data + sackoff;
 287                 if (after(ntohl(sack->start_seq) - natseq->offset_before,
 288                           natseq->correction_pos))
 289                         new_start_seq = htonl(ntohl(sack->start_seq)
 290                                         - natseq->offset_after);
 291                 else
 292                         new_start_seq = htonl(ntohl(sack->start_seq)
 293                                         - natseq->offset_before);
 294
 295                 if (after(ntohl(sack->end_seq) - natseq->offset_before,
 296                           natseq->correction_pos))
 297                         new_end_seq = htonl(ntohl(sack->end_seq)
 298                                       - natseq->offset_after);
 299                 else
 300                         new_end_seq = htonl(ntohl(sack->end_seq)
 301                                       - natseq->offset_before);
 302
 303                 pr_debug("sack_adjust: start_seq: %d->%d, end_seq: %d->%d\n",
 304                          ntohl(sack->start_seq), new_start_seq,
 305                          ntohl(sack->end_seq), new_end_seq);
 306
 307                 inet_proto_csum_replace4(&tcph->check, skb,
 308                                          sack->start_seq, new_start_seq, 0);
 309                 inet_proto_csum_replace4(&tcph->check, skb,
 310                                          sack->end_seq, new_end_seq, 0);
 311                 sack->start_seq = new_start_seq;
 312                 sack->end_seq = new_end_seq;
 313                 sackoff += sizeof(*sack);
 314         }
 315 }
 316
 317 /* TCP SACK sequence number adjustment */
 318 static inline unsigned int
 319 nf_nat_sack_adjust(struct sk_buff *skb,
 320                    unsigned int protoff,
 321                    struct tcphdr *tcph,
 322                    struct nf_conn *ct,
 323                    enum ip_conntrack_info ctinfo)
 324 {
 325         unsigned int dir, optoff, optend;
 326         struct nf_conn_nat *nat = nfct_nat(ct);
 327
 328         optoff = protoff + sizeof(struct tcphdr);
 329         optend = protoff + tcph->doff * 4;
 330
 331         if (!skb_make_writable(skb, optend))
 332                 return 0;
 333
 334         dir = CTINFO2DIR(ctinfo);
 335
 336         while (optoff < optend) {
 337                 /* Usually: option, length. */
 338                 unsigned char *op = skb->data + optoff;
 339
 340                 switch (op[0]) {
 341                 case TCPOPT_EOL:
 342                         return 1;
 343                 case TCPOPT_NOP:
 344                         optoff++;
 345                         continue;
 346                 default:
 347                         /* no partial options */
 348                         if (optoff + 1 == optend ||
 349                             optoff + op[1] > optend ||
 350                             op[1] < 2)
 351                                 return 0;
 352                         if (op[0] == TCPOPT_SACK &&
 353                             op[1] >= 2+TCPOLEN_SACK_PERBLOCK &&
 354                             ((op[1] - 2) % TCPOLEN_SACK_PERBLOCK) == 0)
 355                                 sack_adjust(skb, tcph, optoff+2,
 356                                             optoff+op[1], &nat->seq[!dir]);
 357                         optoff += op[1];
 358                 }
 359         }
 360         return 1;
 361 }
 362
 363 /* TCP sequence number adjustment.  Returns 1 on success, 0 on failure */
 364 int
 365 nf_nat_seq_adjust(struct sk_buff *skb,
 366                   struct nf_conn *ct,
 367                   enum ip_conntrack_info ctinfo,
 368                   unsigned int protoff)
 369 {
 370         struct tcphdr *tcph;
 371         int dir;
 372         __be32 newseq, newack;
 373         s16 seqoff, ackoff;
 374         struct nf_conn_nat *nat = nfct_nat(ct);
 375         struct nf_nat_seq *this_way, *other_way;
 376
 377         dir = CTINFO2DIR(ctinfo);
 378
 379         this_way = &nat->seq[dir];
 380         other_way = &nat->seq[!dir];
 381
 382         if (!skb_make_writable(skb, protoff + sizeof(*tcph)))
 383                 return 0;
 384
 385         tcph = (void *)skb->data + protoff;
 386         if (after(ntohl(tcph->seq), this_way->correction_pos))
 387                 seqoff = this_way->offset_after;
 388         else
 389                 seqoff = this_way->offset_before;
 390
 391         if (after(ntohl(tcph->ack_seq) - other_way->offset_before,
 392                   other_way->correction_pos))
 393                 ackoff = other_way->offset_after;
 394         else
 395                 ackoff = other_way->offset_before;
 396
 397         newseq = htonl(ntohl(tcph->seq) + seqoff);
 398         newack = htonl(ntohl(tcph->ack_seq) - ackoff);
 399
 400         inet_proto_csum_replace4(&tcph->check, skb, tcph->seq, newseq, 0);
 401         inet_proto_csum_replace4(&tcph->check, skb, tcph->ack_seq, newack, 0);
 402
 403         pr_debug("Adjusting sequence number from %u->%u, ack from %u->%u\n",
 404                  ntohl(tcph->seq), ntohl(newseq), ntohl(tcph->ack_seq),
 405                  ntohl(newack));
 406
 407         tcph->seq = newseq;
 408         tcph->ack_seq = newack;
 409
 410         return nf_nat_sack_adjust(skb, protoff, tcph, ct, ctinfo);
 411 }
 412
 413 /* Setup NAT on this expected conntrack so it follows master. */
 414 /* If we fail to get a free NAT slot, we'll get dropped on confirm */
 415 void nf_nat_follow_master(struct nf_conn *ct,
 416                           struct nf_conntrack_expect *exp)
 417 {
 418         struct nf_nat_range range;
 419
 420         /* This must be a fresh one. */
 421         BUG_ON(ct->status & IPS_NAT_DONE_MASK);
 422
 423         /* Change src to where master sends to */
 424         range.flags = NF_NAT_RANGE_MAP_IPS;
 425         range.min_addr = range.max_addr
 426                 = ct->master->tuplehash[!exp->dir].tuple.dst.u3;
 427         nf_nat_setup_info(ct, &range, NF_NAT_MANIP_SRC);
 428
 429         /* For DST manip, map port here to where it's expected. */
 430         range.flags = (NF_NAT_RANGE_MAP_IPS | NF_NAT_RANGE_PROTO_SPECIFIED);
 431         range.min_proto = range.max_proto = exp->saved_proto;
 432         range.min_addr = range.max_addr
 433                 = ct->master->tuplehash[!exp->dir].tuple.src.u3;
 434         nf_nat_setup_info(ct, &range, NF_NAT_MANIP_DST);
 435 }
 436 EXPORT_SYMBOL(nf_nat_follow_master);