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         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
  45         struct nf_conn_nat *nat = nfct_nat(ct);
  46         struct nf_nat_seq *this_way = &nat->seq[dir];
  47
  48         pr_debug("adjust_tcp_sequence: seq = %u, sizediff = %d\n",
  49                  seq, sizediff);
  50
  51         pr_debug("adjust_tcp_sequence: Seq_offset before: ");
  52         DUMP_OFFSET(this_way);
  53
  54         spin_lock_bh(&nf_nat_seqofs_lock);
  55
  56         /* SYN adjust. If it's uninitialized, or this is after last
  57          * correction, record it: we don't handle more than one
  58          * adjustment in the window, but do deal with common case of a
  59          * retransmit */
  60         if (this_way->offset_before == this_way->offset_after ||
  61             before(this_way->correction_pos, seq)) {
  62                 this_way->correction_pos = seq;
  63                 this_way->offset_before = this_way->offset_after;
  64                 this_way->offset_after += sizediff;
  65         }
  66         spin_unlock_bh(&nf_nat_seqofs_lock);
  67
  68         pr_debug("adjust_tcp_sequence: Seq_offset after: ");
  69         DUMP_OFFSET(this_way);
  70 }
  71
  72 /* Get the offset value, for conntrack */
  73 s16 nf_nat_get_offset(const struct nf_conn *ct,
  74                       enum ip_conntrack_dir dir,
  75                       u32 seq)
  76 {
  77         struct nf_conn_nat *nat = nfct_nat(ct);
  78         struct nf_nat_seq *this_way;
  79         s16 offset;
  80
  81         if (!nat)
  82                 return 0;
  83
  84         this_way = &nat->seq[dir];
  85         spin_lock_bh(&nf_nat_seqofs_lock);
  86         offset = after(seq, this_way->correction_pos)
  87                  ? this_way->offset_after : this_way->offset_before;
  88         spin_unlock_bh(&nf_nat_seqofs_lock);
  89
  90         return offset;
  91 }
  92 EXPORT_SYMBOL_GPL(nf_nat_get_offset);
  93
  94 /* Frobs data inside this packet, which is linear. */
  95 static void mangle_contents(struct sk_buff *skb,
  96                             unsigned int dataoff,
  97                             unsigned int match_offset,
  98                             unsigned int match_len,
  99                             const char *rep_buffer,
 100                             unsigned int rep_len)
 101 {
 102         unsigned char *data;
 103
 104         BUG_ON(skb_is_nonlinear(skb));
 105         data = skb_network_header(skb) + dataoff;
 106
 107         /* move post-replacement */
 108         memmove(data + match_offset + rep_len,
 109                 data + match_offset + match_len,
 110                 skb->tail - (skb->network_header + dataoff +
 111                              match_offset + match_len));
 112
 113         /* insert data from buffer */
 114         memcpy(data + match_offset, rep_buffer, rep_len);
 115
 116         /* update skb info */
 117         if (rep_len > match_len) {
 118                 pr_debug("nf_nat_mangle_packet: Extending packet by "
 119                          "%u from %u bytes\n", rep_len - match_len, skb->len);
 120                 skb_put(skb, rep_len - match_len);
 121         } else {
 122                 pr_debug("nf_nat_mangle_packet: Shrinking packet from "
 123                          "%u from %u bytes\n", match_len - rep_len, skb->len);
 124                 __skb_trim(skb, skb->len + rep_len - match_len);
 125         }
 126
 127         /* fix IP hdr checksum information */
 128         ip_hdr(skb)->tot_len = htons(skb->len);
 129         ip_send_check(ip_hdr(skb));
 130 }
 131
 132 /* Unusual, but possible case. */
 133 static int enlarge_skb(struct sk_buff *skb, unsigned int extra)
 134 {
 135         if (skb->len + extra > 65535)
 136                 return 0;
 137
 138         if (pskb_expand_head(skb, 0, extra - skb_tailroom(skb), GFP_ATOMIC))
 139                 return 0;
 140
 141         return 1;
 142 }
 143
 144 void nf_nat_set_seq_adjust(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
 145                            __be32 seq, s16 off)
 146 {
 147         if (!off)
 148                 return;
 149         set_bit(IPS_SEQ_ADJUST_BIT, &ct->status);
 150         adjust_tcp_sequence(ntohl(seq), off, ct, ctinfo);
 151         nf_conntrack_event_cache(IPCT_NATSEQADJ, ct);
 152 }
 153 EXPORT_SYMBOL_GPL(nf_nat_set_seq_adjust);
 154
 155 /* Generic function for mangling variable-length address changes inside
 156  * NATed TCP connections (like the PORT XXX,XXX,XXX,XXX,XXX,XXX
 157  * command in FTP).
 158  *
 159  * Takes care about all the nasty sequence number changes, checksumming,
 160  * skb enlargement, ...
 161  *
 162  * */
 163 int __nf_nat_mangle_tcp_packet(struct sk_buff *skb,
 164                                struct nf_conn *ct,
 165                                enum ip_conntrack_info ctinfo,
 166                                unsigned int match_offset,
 167                                unsigned int match_len,
 168                                const char *rep_buffer,
 169                                unsigned int rep_len, bool adjust)
 170 {
 171         struct rtable *rt = skb_rtable(skb);
 172         struct iphdr *iph;
 173         struct tcphdr *tcph;
 174         int oldlen, datalen;
 175
 176         if (!skb_make_writable(skb, skb->len))
 177                 return 0;
 178
 179         if (rep_len > match_len &&
 180             rep_len - match_len > skb_tailroom(skb) &&
 181             !enlarge_skb(skb, rep_len - match_len))
 182                 return 0;
 183
 184         SKB_LINEAR_ASSERT(skb);
 185
 186         iph = ip_hdr(skb);
 187         tcph = (void *)iph + iph->ihl*4;
 188
 189         oldlen = skb->len - iph->ihl*4;
 190         mangle_contents(skb, iph->ihl*4 + tcph->doff*4,
 191                         match_offset, match_len, rep_buffer, rep_len);
 192
 193         datalen = skb->len - iph->ihl*4;
 194         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 195                 if (!(rt->rt_flags & RTCF_LOCAL) &&
 196                     skb->dev->features & NETIF_F_V4_CSUM) {
 197                         skb->ip_summed = CHECKSUM_PARTIAL;
 198                         skb->csum_start = skb_headroom(skb) +
 199                                           skb_network_offset(skb) +
 200                                           iph->ihl * 4;
 201                         skb->csum_offset = offsetof(struct tcphdr, check);
 202                         tcph->check = ~tcp_v4_check(datalen,
 203                                                     iph->saddr, iph->daddr, 0);
 204                 } else {
 205                         tcph->check = 0;
 206                         tcph->check = tcp_v4_check(datalen,
 207                                                    iph->saddr, iph->daddr,
 208                                                    csum_partial(tcph,
 209                                                                 datalen, 0));
 210                 }
 211         } else
 212                 inet_proto_csum_replace2(&tcph->check, skb,
 213                                          htons(oldlen), htons(datalen), 1);
 214
 215         if (adjust && rep_len != match_len)
 216                 nf_nat_set_seq_adjust(ct, ctinfo, tcph->seq,
 217                                       (int)rep_len - (int)match_len);
 218
 219         return 1;
 220 }
 221 EXPORT_SYMBOL(__nf_nat_mangle_tcp_packet);
 222
 223 /* Generic function for mangling variable-length address changes inside
 224  * NATed UDP connections (like the CONNECT DATA XXXXX MESG XXXXX INDEX XXXXX
 225  * command in the Amanda protocol)
 226  *
 227  * Takes care about all the nasty sequence number changes, checksumming,
 228  * skb enlargement, ...
 229  *
 230  * XXX - This function could be merged with nf_nat_mangle_tcp_packet which
 231  *       should be fairly easy to do.
 232  */
 233 int
 234 nf_nat_mangle_udp_packet(struct sk_buff *skb,
 235                          struct nf_conn *ct,
 236                          enum ip_conntrack_info ctinfo,
 237                          unsigned int match_offset,
 238                          unsigned int match_len,
 239                          const char *rep_buffer,
 240                          unsigned int rep_len)
 241 {
 242         struct rtable *rt = skb_rtable(skb);
 243         struct iphdr *iph;
 244         struct udphdr *udph;
 245         int datalen, oldlen;
 246
 247         /* UDP helpers might accidentally mangle the wrong packet */
 248         iph = ip_hdr(skb);
 249         if (skb->len < iph->ihl*4 + sizeof(*udph) +
 250                                match_offset + match_len)
 251                 return 0;
 252
 253         if (!skb_make_writable(skb, skb->len))
 254                 return 0;
 255
 256         if (rep_len > match_len &&
 257             rep_len - match_len > skb_tailroom(skb) &&
 258             !enlarge_skb(skb, rep_len - match_len))
 259                 return 0;
 260
 261         iph = ip_hdr(skb);
 262         udph = (void *)iph + iph->ihl*4;
 263
 264         oldlen = skb->len - iph->ihl*4;
 265         mangle_contents(skb, iph->ihl*4 + sizeof(*udph),
 266                         match_offset, match_len, rep_buffer, rep_len);
 267
 268         /* update the length of the UDP packet */
 269         datalen = skb->len - iph->ihl*4;
 270         udph->len = htons(datalen);
 271
 272         /* fix udp checksum if udp checksum was previously calculated */
 273         if (!udph->check && skb->ip_summed != CHECKSUM_PARTIAL)
 274                 return 1;
 275
 276         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 277                 if (!(rt->rt_flags & RTCF_LOCAL) &&
 278                     skb->dev->features & NETIF_F_V4_CSUM) {
 279                         skb->ip_summed = CHECKSUM_PARTIAL;
 280                         skb->csum_start = skb_headroom(skb) +
 281                                           skb_network_offset(skb) +
 282                                           iph->ihl * 4;
 283                         skb->csum_offset = offsetof(struct udphdr, check);
 284                         udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
 285                                                          datalen, IPPROTO_UDP,
 286                                                          0);
 287                 } else {
 288                         udph->check = 0;
 289                         udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 290                                                         datalen, IPPROTO_UDP,
 291                                                         csum_partial(udph,
 292                                                                      datalen, 0));
 293                         if (!udph->check)
 294                                 udph->check = CSUM_MANGLED_0;
 295                 }
 296         } else
 297                 inet_proto_csum_replace2(&udph->check, skb,
 298                                          htons(oldlen), htons(datalen), 1);
 299
 300         return 1;
 301 }
 302 EXPORT_SYMBOL(nf_nat_mangle_udp_packet);
 303
 304 /* Adjust one found SACK option including checksum correction */
 305 static void
 306 sack_adjust(struct sk_buff *skb,
 307             struct tcphdr *tcph,
 308             unsigned int sackoff,
 309             unsigned int sackend,
 310             struct nf_nat_seq *natseq)
 311 {
 312         while (sackoff < sackend) {
 313                 struct tcp_sack_block_wire *sack;
 314                 __be32 new_start_seq, new_end_seq;
 315
 316                 sack = (void *)skb->data + sackoff;
 317                 if (after(ntohl(sack->start_seq) - natseq->offset_before,
 318                           natseq->correction_pos))
 319                         new_start_seq = htonl(ntohl(sack->start_seq)
 320                                         - natseq->offset_after);
 321                 else
 322                         new_start_seq = htonl(ntohl(sack->start_seq)
 323                                         - natseq->offset_before);
 324
 325                 if (after(ntohl(sack->end_seq) - natseq->offset_before,
 326                           natseq->correction_pos))
 327                         new_end_seq = htonl(ntohl(sack->end_seq)
 328                                       - natseq->offset_after);
 329                 else
 330                         new_end_seq = htonl(ntohl(sack->end_seq)
 331                                       - natseq->offset_before);
 332
 333                 pr_debug("sack_adjust: start_seq: %d->%d, end_seq: %d->%d\n",
 334                          ntohl(sack->start_seq), new_start_seq,
 335                          ntohl(sack->end_seq), new_end_seq);
 336
 337                 inet_proto_csum_replace4(&tcph->check, skb,
 338                                          sack->start_seq, new_start_seq, 0);
 339                 inet_proto_csum_replace4(&tcph->check, skb,
 340                                          sack->end_seq, new_end_seq, 0);
 341                 sack->start_seq = new_start_seq;
 342                 sack->end_seq = new_end_seq;
 343                 sackoff += sizeof(*sack);
 344         }
 345 }
 346
 347 /* TCP SACK sequence number adjustment */
 348 static inline unsigned int
 349 nf_nat_sack_adjust(struct sk_buff *skb,
 350                    struct tcphdr *tcph,
 351                    struct nf_conn *ct,
 352                    enum ip_conntrack_info ctinfo)
 353 {
 354         unsigned int dir, optoff, optend;
 355         struct nf_conn_nat *nat = nfct_nat(ct);
 356
 357         optoff = ip_hdrlen(skb) + sizeof(struct tcphdr);
 358         optend = ip_hdrlen(skb) + tcph->doff * 4;
 359
 360         if (!skb_make_writable(skb, optend))
 361                 return 0;
 362
 363         dir = CTINFO2DIR(ctinfo);
 364
 365         while (optoff < optend) {
 366                 /* Usually: option, length. */
 367                 unsigned char *op = skb->data + optoff;
 368
 369                 switch (op[0]) {
 370                 case TCPOPT_EOL:
 371                         return 1;
 372                 case TCPOPT_NOP:
 373                         optoff++;
 374                         continue;
 375                 default:
 376                         /* no partial options */
 377                         if (optoff + 1 == optend ||
 378                             optoff + op[1] > optend ||
 379                             op[1] < 2)
 380                                 return 0;
 381                         if (op[0] == TCPOPT_SACK &&
 382                             op[1] >= 2+TCPOLEN_SACK_PERBLOCK &&
 383                             ((op[1] - 2) % TCPOLEN_SACK_PERBLOCK) == 0)
 384                                 sack_adjust(skb, tcph, optoff+2,
 385                                             optoff+op[1], &nat->seq[!dir]);
 386                         optoff += op[1];
 387                 }
 388         }
 389         return 1;
 390 }
 391
 392 /* TCP sequence number adjustment.  Returns 1 on success, 0 on failure */
 393 int
 394 nf_nat_seq_adjust(struct sk_buff *skb,
 395                   struct nf_conn *ct,
 396                   enum ip_conntrack_info ctinfo)
 397 {
 398         struct tcphdr *tcph;
 399         int dir;
 400         __be32 newseq, newack;
 401         s16 seqoff, ackoff;
 402         struct nf_conn_nat *nat = nfct_nat(ct);
 403         struct nf_nat_seq *this_way, *other_way;
 404
 405         dir = CTINFO2DIR(ctinfo);
 406
 407         this_way = &nat->seq[dir];
 408         other_way = &nat->seq[!dir];
 409
 410         if (!skb_make_writable(skb, ip_hdrlen(skb) + sizeof(*tcph)))
 411                 return 0;
 412
 413         tcph = (void *)skb->data + ip_hdrlen(skb);
 414         if (after(ntohl(tcph->seq), this_way->correction_pos))
 415                 seqoff = this_way->offset_after;
 416         else
 417                 seqoff = this_way->offset_before;
 418
 419         if (after(ntohl(tcph->ack_seq) - other_way->offset_before,
 420                   other_way->correction_pos))
 421                 ackoff = other_way->offset_after;
 422         else
 423                 ackoff = other_way->offset_before;
 424
 425         newseq = htonl(ntohl(tcph->seq) + seqoff);
 426         newack = htonl(ntohl(tcph->ack_seq) - ackoff);
 427
 428         inet_proto_csum_replace4(&tcph->check, skb, tcph->seq, newseq, 0);
 429         inet_proto_csum_replace4(&tcph->check, skb, tcph->ack_seq, newack, 0);
 430
 431         pr_debug("Adjusting sequence number from %u->%u, ack from %u->%u\n",
 432                  ntohl(tcph->seq), ntohl(newseq), ntohl(tcph->ack_seq),
 433                  ntohl(newack));
 434
 435         tcph->seq = newseq;
 436         tcph->ack_seq = newack;
 437
 438         return nf_nat_sack_adjust(skb, tcph, ct, ctinfo);
 439 }
 440
 441 /* Setup NAT on this expected conntrack so it follows master. */
 442 /* If we fail to get a free NAT slot, we'll get dropped on confirm */
 443 void nf_nat_follow_master(struct nf_conn *ct,
 444                           struct nf_conntrack_expect *exp)
 445 {
 446         struct nf_nat_range range;
 447
 448         /* This must be a fresh one. */
 449         BUG_ON(ct->status & IPS_NAT_DONE_MASK);
 450
 451         /* Change src to where master sends to */
 452         range.flags = IP_NAT_RANGE_MAP_IPS;
 453         range.min_ip = range.max_ip
 454                 = ct->master->tuplehash[!exp->dir].tuple.dst.u3.ip;
 455         nf_nat_setup_info(ct, &range, IP_NAT_MANIP_SRC);
 456
 457         /* For DST manip, map port here to where it's expected. */
 458         range.flags = (IP_NAT_RANGE_MAP_IPS | IP_NAT_RANGE_PROTO_SPECIFIED);
 459         range.min = range.max = exp->saved_proto;
 460         range.min_ip = range.max_ip
 461                 = ct->master->tuplehash[!exp->dir].tuple.src.u3.ip;
 462         nf_nat_setup_info(ct, &range, IP_NAT_MANIP_DST);
 463 }
 464 EXPORT_SYMBOL(nf_nat_follow_master);