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/gfp.h>
  12 #include <linux/kmod.h>
  13 #include <linux/types.h>
  14 #include <linux/timer.h>
  15 #include <linux/skbuff.h>
  16 #include <linux/tcp.h>
  17 #include <linux/udp.h>
  18 #include <net/checksum.h>
  19 #include <net/tcp.h>
  20 #include <net/route.h>
  21
  22 #include <linux/netfilter_ipv4.h>
  23 #include <net/netfilter/nf_conntrack.h>
  24 #include <net/netfilter/nf_conntrack_helper.h>
  25 #include <net/netfilter/nf_conntrack_ecache.h>
  26 #include <net/netfilter/nf_conntrack_expect.h>
  27 #include <net/netfilter/nf_nat.h>
  28 #include <net/netfilter/nf_nat_protocol.h>
  29 #include <net/netfilter/nf_nat_core.h>
  30 #include <net/netfilter/nf_nat_helper.h>
  31
  32 #define DUMP_OFFSET(x) \
  33         pr_debug("offset_before=%d, offset_after=%d, correction_pos=%u\n", \
  34                  x->offset_before, x->offset_after, x->correction_pos);
  35
  36 static DEFINE_SPINLOCK(nf_nat_seqofs_lock);
  37
  38 /* Setup TCP sequence correction given this change at this sequence */
  39 static inline void
  40 adjust_tcp_sequence(u32 seq,
  41                     int sizediff,
  42                     struct nf_conn *ct,
  43                     enum ip_conntrack_info ctinfo)
  44 {
  45         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
  46         struct nf_conn_nat *nat = nfct_nat(ct);
  47         struct nf_nat_seq *this_way = &nat->seq[dir];
  48
  49         pr_debug("adjust_tcp_sequence: seq = %u, sizediff = %d\n",
  50                  seq, sizediff);
  51
  52         pr_debug("adjust_tcp_sequence: Seq_offset before: ");
  53         DUMP_OFFSET(this_way);
  54
  55         spin_lock_bh(&nf_nat_seqofs_lock);
  56
  57         /* SYN adjust. If it's uninitialized, or this is after last
  58          * correction, record it: we don't handle more than one
  59          * adjustment in the window, but do deal with common case of a
  60          * retransmit */
  61         if (this_way->offset_before == this_way->offset_after ||
  62             before(this_way->correction_pos, seq)) {
  63                 this_way->correction_pos = seq;
  64                 this_way->offset_before = this_way->offset_after;
  65                 this_way->offset_after += sizediff;
  66         }
  67         spin_unlock_bh(&nf_nat_seqofs_lock);
  68
  69         pr_debug("adjust_tcp_sequence: Seq_offset after: ");
  70         DUMP_OFFSET(this_way);
  71 }
  72
  73 /* Get the offset value, for conntrack */
  74 s16 nf_nat_get_offset(const struct nf_conn *ct,
  75                       enum ip_conntrack_dir dir,
  76                       u32 seq)
  77 {
  78         struct nf_conn_nat *nat = nfct_nat(ct);
  79         struct nf_nat_seq *this_way;
  80         s16 offset;
  81
  82         if (!nat)
  83                 return 0;
  84
  85         this_way = &nat->seq[dir];
  86         spin_lock_bh(&nf_nat_seqofs_lock);
  87         offset = after(seq, this_way->correction_pos)
  88                  ? this_way->offset_after : this_way->offset_before;
  89         spin_unlock_bh(&nf_nat_seqofs_lock);
  90
  91         return offset;
  92 }
  93 EXPORT_SYMBOL_GPL(nf_nat_get_offset);
  94
  95 /* Frobs data inside this packet, which is linear. */
  96 static void mangle_contents(struct sk_buff *skb,
  97                             unsigned int dataoff,
  98                             unsigned int match_offset,
  99                             unsigned int match_len,
 100                             const char *rep_buffer,
 101                             unsigned int rep_len)
 102 {
 103         unsigned char *data;
 104
 105         BUG_ON(skb_is_nonlinear(skb));
 106         data = skb_network_header(skb) + dataoff;
 107
 108         /* move post-replacement */
 109         memmove(data + match_offset + rep_len,
 110                 data + match_offset + match_len,
 111                 skb->tail - (skb->network_header + dataoff +
 112                              match_offset + match_len));
 113
 114         /* insert data from buffer */
 115         memcpy(data + match_offset, rep_buffer, rep_len);
 116
 117         /* update skb info */
 118         if (rep_len > match_len) {
 119                 pr_debug("nf_nat_mangle_packet: Extending packet by "
 120                          "%u from %u bytes\n", rep_len - match_len, skb->len);
 121                 skb_put(skb, rep_len - match_len);
 122         } else {
 123                 pr_debug("nf_nat_mangle_packet: Shrinking packet from "
 124                          "%u from %u bytes\n", match_len - rep_len, skb->len);
 125                 __skb_trim(skb, skb->len + rep_len - match_len);
 126         }
 127
 128         /* fix IP hdr checksum information */
 129         ip_hdr(skb)->tot_len = htons(skb->len);
 130         ip_send_check(ip_hdr(skb));
 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 /* Generic function for mangling variable-length address changes inside
 157  * NATed TCP connections (like the PORT XXX,XXX,XXX,XXX,XXX,XXX
 158  * command in FTP).
 159  *
 160  * Takes care about all the nasty sequence number changes, checksumming,
 161  * skb enlargement, ...
 162  *
 163  * */
 164 int __nf_nat_mangle_tcp_packet(struct sk_buff *skb,
 165                                struct nf_conn *ct,
 166                                enum ip_conntrack_info ctinfo,
 167                                unsigned int match_offset,
 168                                unsigned int match_len,
 169                                const char *rep_buffer,
 170                                unsigned int rep_len, bool adjust)
 171 {
 172         struct rtable *rt = skb_rtable(skb);
 173         struct iphdr *iph;
 174         struct tcphdr *tcph;
 175         int oldlen, datalen;
 176
 177         if (!skb_make_writable(skb, skb->len))
 178                 return 0;
 179
 180         if (rep_len > match_len &&
 181             rep_len - match_len > skb_tailroom(skb) &&
 182             !enlarge_skb(skb, rep_len - match_len))
 183                 return 0;
 184
 185         SKB_LINEAR_ASSERT(skb);
 186
 187         iph = ip_hdr(skb);
 188         tcph = (void *)iph + iph->ihl*4;
 189
 190         oldlen = skb->len - iph->ihl*4;
 191         mangle_contents(skb, iph->ihl*4 + tcph->doff*4,
 192                         match_offset, match_len, rep_buffer, rep_len);
 193
 194         datalen = skb->len - iph->ihl*4;
 195         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 196                 if (!(rt->rt_flags & RTCF_LOCAL) &&
 197                     skb->dev->features & NETIF_F_V4_CSUM) {
 198                         skb->ip_summed = CHECKSUM_PARTIAL;
 199                         skb->csum_start = skb_headroom(skb) +
 200                                           skb_network_offset(skb) +
 201                                           iph->ihl * 4;
 202                         skb->csum_offset = offsetof(struct tcphdr, check);
 203                         tcph->check = ~tcp_v4_check(datalen,
 204                                                     iph->saddr, iph->daddr, 0);
 205                 } else {
 206                         tcph->check = 0;
 207                         tcph->check = tcp_v4_check(datalen,
 208                                                    iph->saddr, iph->daddr,
 209                                                    csum_partial(tcph,
 210                                                                 datalen, 0));
 211                 }
 212         } else
 213                 inet_proto_csum_replace2(&tcph->check, skb,
 214                                          htons(oldlen), htons(datalen), 1);
 215
 216         if (adjust && rep_len != match_len)
 217                 nf_nat_set_seq_adjust(ct, ctinfo, tcph->seq,
 218                                       (int)rep_len - (int)match_len);
 219
 220         return 1;
 221 }
 222 EXPORT_SYMBOL(__nf_nat_mangle_tcp_packet);
 223
 224 /* Generic function for mangling variable-length address changes inside
 225  * NATed UDP connections (like the CONNECT DATA XXXXX MESG XXXXX INDEX XXXXX
 226  * command in the Amanda protocol)
 227  *
 228  * Takes care about all the nasty sequence number changes, checksumming,
 229  * skb enlargement, ...
 230  *
 231  * XXX - This function could be merged with nf_nat_mangle_tcp_packet which
 232  *       should be fairly easy to do.
 233  */
 234 int
 235 nf_nat_mangle_udp_packet(struct sk_buff *skb,
 236                          struct nf_conn *ct,
 237                          enum ip_conntrack_info ctinfo,
 238                          unsigned int match_offset,
 239                          unsigned int match_len,
 240                          const char *rep_buffer,
 241                          unsigned int rep_len)
 242 {
 243         struct rtable *rt = skb_rtable(skb);
 244         struct iphdr *iph;
 245         struct udphdr *udph;
 246         int datalen, oldlen;
 247
 248         /* UDP helpers might accidentally mangle the wrong packet */
 249         iph = ip_hdr(skb);
 250         if (skb->len < iph->ihl*4 + sizeof(*udph) +
 251                                match_offset + match_len)
 252                 return 0;
 253
 254         if (!skb_make_writable(skb, skb->len))
 255                 return 0;
 256
 257         if (rep_len > match_len &&
 258             rep_len - match_len > skb_tailroom(skb) &&
 259             !enlarge_skb(skb, rep_len - match_len))
 260                 return 0;
 261
 262         iph = ip_hdr(skb);
 263         udph = (void *)iph + iph->ihl*4;
 264
 265         oldlen = skb->len - iph->ihl*4;
 266         mangle_contents(skb, iph->ihl*4 + sizeof(*udph),
 267                         match_offset, match_len, rep_buffer, rep_len);
 268
 269         /* update the length of the UDP packet */
 270         datalen = skb->len - iph->ihl*4;
 271         udph->len = htons(datalen);
 272
 273         /* fix udp checksum if udp checksum was previously calculated */
 274         if (!udph->check && skb->ip_summed != CHECKSUM_PARTIAL)
 275                 return 1;
 276
 277         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 278                 if (!(rt->rt_flags & RTCF_LOCAL) &&
 279                     skb->dev->features & NETIF_F_V4_CSUM) {
 280                         skb->ip_summed = CHECKSUM_PARTIAL;
 281                         skb->csum_start = skb_headroom(skb) +
 282                                           skb_network_offset(skb) +
 283                                           iph->ihl * 4;
 284                         skb->csum_offset = offsetof(struct udphdr, check);
 285                         udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
 286                                                          datalen, IPPROTO_UDP,
 287                                                          0);
 288                 } else {
 289                         udph->check = 0;
 290                         udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 291                                                         datalen, IPPROTO_UDP,
 292                                                         csum_partial(udph,
 293                                                                      datalen, 0));
 294                         if (!udph->check)
 295                                 udph->check = CSUM_MANGLED_0;
 296                 }
 297         } else
 298                 inet_proto_csum_replace2(&udph->check, skb,
 299                                          htons(oldlen), htons(datalen), 1);
 300
 301         return 1;
 302 }
 303 EXPORT_SYMBOL(nf_nat_mangle_udp_packet);
 304
 305 /* Adjust one found SACK option including checksum correction */
 306 static void
 307 sack_adjust(struct sk_buff *skb,
 308             struct tcphdr *tcph,
 309             unsigned int sackoff,
 310             unsigned int sackend,
 311             struct nf_nat_seq *natseq)
 312 {
 313         while (sackoff < sackend) {
 314                 struct tcp_sack_block_wire *sack;
 315                 __be32 new_start_seq, new_end_seq;
 316
 317                 sack = (void *)skb->data + sackoff;
 318                 if (after(ntohl(sack->start_seq) - natseq->offset_before,
 319                           natseq->correction_pos))
 320                         new_start_seq = htonl(ntohl(sack->start_seq)
 321                                         - natseq->offset_after);
 322                 else
 323                         new_start_seq = htonl(ntohl(sack->start_seq)
 324                                         - natseq->offset_before);
 325
 326                 if (after(ntohl(sack->end_seq) - natseq->offset_before,
 327                           natseq->correction_pos))
 328                         new_end_seq = htonl(ntohl(sack->end_seq)
 329                                       - natseq->offset_after);
 330                 else
 331                         new_end_seq = htonl(ntohl(sack->end_seq)
 332                                       - natseq->offset_before);
 333
 334                 pr_debug("sack_adjust: start_seq: %d->%d, end_seq: %d->%d\n",
 335                          ntohl(sack->start_seq), new_start_seq,
 336                          ntohl(sack->end_seq), new_end_seq);
 337
 338                 inet_proto_csum_replace4(&tcph->check, skb,
 339                                          sack->start_seq, new_start_seq, 0);
 340                 inet_proto_csum_replace4(&tcph->check, skb,
 341                                          sack->end_seq, new_end_seq, 0);
 342                 sack->start_seq = new_start_seq;
 343                 sack->end_seq = new_end_seq;
 344                 sackoff += sizeof(*sack);
 345         }
 346 }
 347
 348 /* TCP SACK sequence number adjustment */
 349 static inline unsigned int
 350 nf_nat_sack_adjust(struct sk_buff *skb,
 351                    struct tcphdr *tcph,
 352                    struct nf_conn *ct,
 353                    enum ip_conntrack_info ctinfo)
 354 {
 355         unsigned int dir, optoff, optend;
 356         struct nf_conn_nat *nat = nfct_nat(ct);
 357
 358         optoff = ip_hdrlen(skb) + sizeof(struct tcphdr);
 359         optend = ip_hdrlen(skb) + tcph->doff * 4;
 360
 361         if (!skb_make_writable(skb, optend))
 362                 return 0;
 363
 364         dir = CTINFO2DIR(ctinfo);
 365
 366         while (optoff < optend) {
 367                 /* Usually: option, length. */
 368                 unsigned char *op = skb->data + optoff;
 369
 370                 switch (op[0]) {
 371                 case TCPOPT_EOL:
 372                         return 1;
 373                 case TCPOPT_NOP:
 374                         optoff++;
 375                         continue;
 376                 default:
 377                         /* no partial options */
 378                         if (optoff + 1 == optend ||
 379                             optoff + op[1] > optend ||
 380                             op[1] < 2)
 381                                 return 0;
 382                         if (op[0] == TCPOPT_SACK &&
 383                             op[1] >= 2+TCPOLEN_SACK_PERBLOCK &&
 384                             ((op[1] - 2) % TCPOLEN_SACK_PERBLOCK) == 0)
 385                                 sack_adjust(skb, tcph, optoff+2,
 386                                             optoff+op[1], &nat->seq[!dir]);
 387                         optoff += op[1];
 388                 }
 389         }
 390         return 1;
 391 }
 392
 393 /* TCP sequence number adjustment.  Returns 1 on success, 0 on failure */
 394 int
 395 nf_nat_seq_adjust(struct sk_buff *skb,
 396                   struct nf_conn *ct,
 397                   enum ip_conntrack_info ctinfo)
 398 {
 399         struct tcphdr *tcph;
 400         int dir;
 401         __be32 newseq, newack;
 402         s16 seqoff, ackoff;
 403         struct nf_conn_nat *nat = nfct_nat(ct);
 404         struct nf_nat_seq *this_way, *other_way;
 405
 406         dir = CTINFO2DIR(ctinfo);
 407
 408         this_way = &nat->seq[dir];
 409         other_way = &nat->seq[!dir];
 410
 411         if (!skb_make_writable(skb, ip_hdrlen(skb) + sizeof(*tcph)))
 412                 return 0;
 413
 414         tcph = (void *)skb->data + ip_hdrlen(skb);
 415         if (after(ntohl(tcph->seq), this_way->correction_pos))
 416                 seqoff = this_way->offset_after;
 417         else
 418                 seqoff = this_way->offset_before;
 419
 420         if (after(ntohl(tcph->ack_seq) - other_way->offset_before,
 421                   other_way->correction_pos))
 422                 ackoff = other_way->offset_after;
 423         else
 424                 ackoff = other_way->offset_before;
 425
 426         newseq = htonl(ntohl(tcph->seq) + seqoff);
 427         newack = htonl(ntohl(tcph->ack_seq) - ackoff);
 428
 429         inet_proto_csum_replace4(&tcph->check, skb, tcph->seq, newseq, 0);
 430         inet_proto_csum_replace4(&tcph->check, skb, tcph->ack_seq, newack, 0);
 431
 432         pr_debug("Adjusting sequence number from %u->%u, ack from %u->%u\n",
 433                  ntohl(tcph->seq), ntohl(newseq), ntohl(tcph->ack_seq),
 434                  ntohl(newack));
 435
 436         tcph->seq = newseq;
 437         tcph->ack_seq = newack;
 438
 439         return nf_nat_sack_adjust(skb, tcph, ct, ctinfo);
 440 }
 441
 442 /* Setup NAT on this expected conntrack so it follows master. */
 443 /* If we fail to get a free NAT slot, we'll get dropped on confirm */
 444 void nf_nat_follow_master(struct nf_conn *ct,
 445                           struct nf_conntrack_expect *exp)
 446 {
 447         struct nf_nat_range range;
 448
 449         /* This must be a fresh one. */
 450         BUG_ON(ct->status & IPS_NAT_DONE_MASK);
 451
 452         /* Change src to where master sends to */
 453         range.flags = IP_NAT_RANGE_MAP_IPS;
 454         range.min_ip = range.max_ip
 455                 = ct->master->tuplehash[!exp->dir].tuple.dst.u3.ip;
 456         nf_nat_setup_info(ct, &range, IP_NAT_MANIP_SRC);
 457
 458         /* For DST manip, map port here to where it's expected. */
 459         range.flags = (IP_NAT_RANGE_MAP_IPS | IP_NAT_RANGE_PROTO_SPECIFIED);
 460         range.min = range.max = exp->saved_proto;
 461         range.min_ip = range.max_ip
 462                 = ct->master->tuplehash[!exp->dir].tuple.src.u3.ip;
 463         nf_nat_setup_info(ct, &range, IP_NAT_MANIP_DST);
 464 }
 465 EXPORT_SYMBOL(nf_nat_follow_master);