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 static void nf_nat_csum(struct sk_buff *skb, const struct iphdr *iph, void *data,
 157                         int datalen, __sum16 *check, int oldlen)
 158 {
 159         struct rtable *rt = skb_rtable(skb);
 160
 161         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 162                 if (!(rt->rt_flags & RTCF_LOCAL) &&
 163                     (!skb->dev || skb->dev->features & NETIF_F_V4_CSUM)) {
 164                         skb->ip_summed = CHECKSUM_PARTIAL;
 165                         skb->csum_start = skb_headroom(skb) +
 166                                           skb_network_offset(skb) +
 167                                           iph->ihl * 4;
 168                         skb->csum_offset = (void *)check - data;
 169                         *check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
 170                                                     datalen, iph->protocol, 0);
 171                 } else {
 172                         *check = 0;
 173                         *check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 174                                                    datalen, iph->protocol,
 175                                                    csum_partial(data, datalen,
 176                                                                 0));
 177                         if (iph->protocol == IPPROTO_UDP && !*check)
 178                                 *check = CSUM_MANGLED_0;
 179                 }
 180         } else
 181                 inet_proto_csum_replace2(check, skb,
 182                                          htons(oldlen), htons(datalen), 1);
 183 }
 184
 185 /* Generic function for mangling variable-length address changes inside
 186  * NATed TCP connections (like the PORT XXX,XXX,XXX,XXX,XXX,XXX
 187  * command in FTP).
 188  *
 189  * Takes care about all the nasty sequence number changes, checksumming,
 190  * skb enlargement, ...
 191  *
 192  * */
 193 int __nf_nat_mangle_tcp_packet(struct sk_buff *skb,
 194                                struct nf_conn *ct,
 195                                enum ip_conntrack_info ctinfo,
 196                                unsigned int match_offset,
 197                                unsigned int match_len,
 198                                const char *rep_buffer,
 199                                unsigned int rep_len, bool adjust)
 200 {
 201         struct iphdr *iph;
 202         struct tcphdr *tcph;
 203         int oldlen, datalen;
 204
 205         if (!skb_make_writable(skb, skb->len))
 206                 return 0;
 207
 208         if (rep_len > match_len &&
 209             rep_len - match_len > skb_tailroom(skb) &&
 210             !enlarge_skb(skb, rep_len - match_len))
 211                 return 0;
 212
 213         SKB_LINEAR_ASSERT(skb);
 214
 215         iph = ip_hdr(skb);
 216         tcph = (void *)iph + iph->ihl*4;
 217
 218         oldlen = skb->len - iph->ihl*4;
 219         mangle_contents(skb, iph->ihl*4 + tcph->doff*4,
 220                         match_offset, match_len, rep_buffer, rep_len);
 221
 222         datalen = skb->len - iph->ihl*4;
 223         nf_nat_csum(skb, iph, tcph, datalen, &tcph->check, oldlen);
 224
 225         if (adjust && rep_len != match_len)
 226                 nf_nat_set_seq_adjust(ct, ctinfo, tcph->seq,
 227                                       (int)rep_len - (int)match_len);
 228
 229         return 1;
 230 }
 231 EXPORT_SYMBOL(__nf_nat_mangle_tcp_packet);
 232
 233 /* Generic function for mangling variable-length address changes inside
 234  * NATed UDP connections (like the CONNECT DATA XXXXX MESG XXXXX INDEX XXXXX
 235  * command in the Amanda protocol)
 236  *
 237  * Takes care about all the nasty sequence number changes, checksumming,
 238  * skb enlargement, ...
 239  *
 240  * XXX - This function could be merged with nf_nat_mangle_tcp_packet which
 241  *       should be fairly easy to do.
 242  */
 243 int
 244 nf_nat_mangle_udp_packet(struct sk_buff *skb,
 245                          struct nf_conn *ct,
 246                          enum ip_conntrack_info ctinfo,
 247                          unsigned int match_offset,
 248                          unsigned int match_len,
 249                          const char *rep_buffer,
 250                          unsigned int rep_len)
 251 {
 252         struct iphdr *iph;
 253         struct udphdr *udph;
 254         int datalen, oldlen;
 255
 256         if (!skb_make_writable(skb, skb->len))
 257                 return 0;
 258
 259         if (rep_len > match_len &&
 260             rep_len - match_len > skb_tailroom(skb) &&
 261             !enlarge_skb(skb, rep_len - match_len))
 262                 return 0;
 263
 264         iph = ip_hdr(skb);
 265         udph = (void *)iph + iph->ihl*4;
 266
 267         oldlen = skb->len - iph->ihl*4;
 268         mangle_contents(skb, iph->ihl*4 + sizeof(*udph),
 269                         match_offset, match_len, rep_buffer, rep_len);
 270
 271         /* update the length of the UDP packet */
 272         datalen = skb->len - iph->ihl*4;
 273         udph->len = htons(datalen);
 274
 275         /* fix udp checksum if udp checksum was previously calculated */
 276         if (!udph->check && skb->ip_summed != CHECKSUM_PARTIAL)
 277                 return 1;
 278
 279         nf_nat_csum(skb, iph, udph, datalen, &udph->check, oldlen);
 280
 281         return 1;
 282 }
 283 EXPORT_SYMBOL(nf_nat_mangle_udp_packet);
 284
 285 /* Adjust one found SACK option including checksum correction */
 286 static void
 287 sack_adjust(struct sk_buff *skb,
 288             struct tcphdr *tcph,
 289             unsigned int sackoff,
 290             unsigned int sackend,
 291             struct nf_nat_seq *natseq)
 292 {
 293         while (sackoff < sackend) {
 294                 struct tcp_sack_block_wire *sack;
 295                 __be32 new_start_seq, new_end_seq;
 296
 297                 sack = (void *)skb->data + sackoff;
 298                 if (after(ntohl(sack->start_seq) - natseq->offset_before,
 299                           natseq->correction_pos))
 300                         new_start_seq = htonl(ntohl(sack->start_seq)
 301                                         - natseq->offset_after);
 302                 else
 303                         new_start_seq = htonl(ntohl(sack->start_seq)
 304                                         - natseq->offset_before);
 305
 306                 if (after(ntohl(sack->end_seq) - natseq->offset_before,
 307                           natseq->correction_pos))
 308                         new_end_seq = htonl(ntohl(sack->end_seq)
 309                                       - natseq->offset_after);
 310                 else
 311                         new_end_seq = htonl(ntohl(sack->end_seq)
 312                                       - natseq->offset_before);
 313
 314                 pr_debug("sack_adjust: start_seq: %d->%d, end_seq: %d->%d\n",
 315                          ntohl(sack->start_seq), new_start_seq,
 316                          ntohl(sack->end_seq), new_end_seq);
 317
 318                 inet_proto_csum_replace4(&tcph->check, skb,
 319                                          sack->start_seq, new_start_seq, 0);
 320                 inet_proto_csum_replace4(&tcph->check, skb,
 321                                          sack->end_seq, new_end_seq, 0);
 322                 sack->start_seq = new_start_seq;
 323                 sack->end_seq = new_end_seq;
 324                 sackoff += sizeof(*sack);
 325         }
 326 }
 327
 328 /* TCP SACK sequence number adjustment */
 329 static inline unsigned int
 330 nf_nat_sack_adjust(struct sk_buff *skb,
 331                    struct tcphdr *tcph,
 332                    struct nf_conn *ct,
 333                    enum ip_conntrack_info ctinfo)
 334 {
 335         unsigned int dir, optoff, optend;
 336         struct nf_conn_nat *nat = nfct_nat(ct);
 337
 338         optoff = ip_hdrlen(skb) + sizeof(struct tcphdr);
 339         optend = ip_hdrlen(skb) + tcph->doff * 4;
 340
 341         if (!skb_make_writable(skb, optend))
 342                 return 0;
 343
 344         dir = CTINFO2DIR(ctinfo);
 345
 346         while (optoff < optend) {
 347                 /* Usually: option, length. */
 348                 unsigned char *op = skb->data + optoff;
 349
 350                 switch (op[0]) {
 351                 case TCPOPT_EOL:
 352                         return 1;
 353                 case TCPOPT_NOP:
 354                         optoff++;
 355                         continue;
 356                 default:
 357                         /* no partial options */
 358                         if (optoff + 1 == optend ||
 359                             optoff + op[1] > optend ||
 360                             op[1] < 2)
 361                                 return 0;
 362                         if (op[0] == TCPOPT_SACK &&
 363                             op[1] >= 2+TCPOLEN_SACK_PERBLOCK &&
 364                             ((op[1] - 2) % TCPOLEN_SACK_PERBLOCK) == 0)
 365                                 sack_adjust(skb, tcph, optoff+2,
 366                                             optoff+op[1], &nat->seq[!dir]);
 367                         optoff += op[1];
 368                 }
 369         }
 370         return 1;
 371 }
 372
 373 /* TCP sequence number adjustment.  Returns 1 on success, 0 on failure */
 374 int
 375 nf_nat_seq_adjust(struct sk_buff *skb,
 376                   struct nf_conn *ct,
 377                   enum ip_conntrack_info ctinfo)
 378 {
 379         struct tcphdr *tcph;
 380         int dir;
 381         __be32 newseq, newack;
 382         s16 seqoff, ackoff;
 383         struct nf_conn_nat *nat = nfct_nat(ct);
 384         struct nf_nat_seq *this_way, *other_way;
 385
 386         dir = CTINFO2DIR(ctinfo);
 387
 388         this_way = &nat->seq[dir];
 389         other_way = &nat->seq[!dir];
 390
 391         if (!skb_make_writable(skb, ip_hdrlen(skb) + sizeof(*tcph)))
 392                 return 0;
 393
 394         tcph = (void *)skb->data + ip_hdrlen(skb);
 395         if (after(ntohl(tcph->seq), this_way->correction_pos))
 396                 seqoff = this_way->offset_after;
 397         else
 398                 seqoff = this_way->offset_before;
 399
 400         if (after(ntohl(tcph->ack_seq) - other_way->offset_before,
 401                   other_way->correction_pos))
 402                 ackoff = other_way->offset_after;
 403         else
 404                 ackoff = other_way->offset_before;
 405
 406         newseq = htonl(ntohl(tcph->seq) + seqoff);
 407         newack = htonl(ntohl(tcph->ack_seq) - ackoff);
 408
 409         inet_proto_csum_replace4(&tcph->check, skb, tcph->seq, newseq, 0);
 410         inet_proto_csum_replace4(&tcph->check, skb, tcph->ack_seq, newack, 0);
 411
 412         pr_debug("Adjusting sequence number from %u->%u, ack from %u->%u\n",
 413                  ntohl(tcph->seq), ntohl(newseq), ntohl(tcph->ack_seq),
 414                  ntohl(newack));
 415
 416         tcph->seq = newseq;
 417         tcph->ack_seq = newack;
 418
 419         return nf_nat_sack_adjust(skb, tcph, ct, ctinfo);
 420 }
 421
 422 /* Setup NAT on this expected conntrack so it follows master. */
 423 /* If we fail to get a free NAT slot, we'll get dropped on confirm */
 424 void nf_nat_follow_master(struct nf_conn *ct,
 425                           struct nf_conntrack_expect *exp)
 426 {
 427         struct nf_nat_ipv4_range range;
 428
 429         /* This must be a fresh one. */
 430         BUG_ON(ct->status & IPS_NAT_DONE_MASK);
 431
 432         /* Change src to where master sends to */
 433         range.flags = NF_NAT_RANGE_MAP_IPS;
 434         range.min_ip = range.max_ip
 435                 = ct->master->tuplehash[!exp->dir].tuple.dst.u3.ip;
 436         nf_nat_setup_info(ct, &range, NF_NAT_MANIP_SRC);
 437
 438         /* For DST manip, map port here to where it's expected. */
 439         range.flags = (NF_NAT_RANGE_MAP_IPS | NF_NAT_RANGE_PROTO_SPECIFIED);
 440         range.min = range.max = exp->saved_proto;
 441         range.min_ip = range.max_ip
 442                 = ct->master->tuplehash[!exp->dir].tuple.src.u3.ip;
 443         nf_nat_setup_info(ct, &range, NF_NAT_MANIP_DST);
 444 }
 445 EXPORT_SYMBOL(nf_nat_follow_master);