net/ipv4/netfilter/ip_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) 2003-2004 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  *      14 Jan 2002 Harald Welte <laforge@gnumonks.org>:
  11  *              - add support for SACK adjustment
  12  *      14 Mar 2002 Harald Welte <laforge@gnumonks.org>:
  13  *              - merge SACK support into newnat API
  14  *      16 Aug 2002 Brian J. Murrell <netfilter@interlinx.bc.ca>:
  15  *              - make ip_nat_resize_packet more generic (TCP and UDP)
  16  *              - add ip_nat_mangle_udp_packet
  17  */
  18 #include <linux/config.h>
  19 #include <linux/module.h>
  20 #include <linux/kmod.h>
  21 #include <linux/types.h>
  22 #include <linux/timer.h>
  23 #include <linux/skbuff.h>
  24 #include <linux/netfilter_ipv4.h>
  25 #include <net/checksum.h>
  26 #include <net/icmp.h>
  27 #include <net/ip.h>
  28 #include <net/tcp.h>
  29 #include <net/udp.h>
  30
  31 #define ASSERT_READ_LOCK(x) MUST_BE_READ_LOCKED(&ip_nat_lock)
  32 #define ASSERT_WRITE_LOCK(x) MUST_BE_WRITE_LOCKED(&ip_nat_lock)
  33
  34 #include <linux/netfilter_ipv4/ip_conntrack.h>
  35 #include <linux/netfilter_ipv4/ip_conntrack_helper.h>
  36 #include <linux/netfilter_ipv4/ip_nat.h>
  37 #include <linux/netfilter_ipv4/ip_nat_protocol.h>
  38 #include <linux/netfilter_ipv4/ip_nat_core.h>
  39 #include <linux/netfilter_ipv4/ip_nat_helper.h>
  40 #include <linux/netfilter_ipv4/listhelp.h>
  41
  42 #if 0
  43 #define DEBUGP printk
  44 #define DUMP_OFFSET(x)  printk("offset_before=%d, offset_after=%d, correction_pos=%u\n", x->offset_before, x->offset_after, x->correction_pos);
  45 #else
  46 #define DEBUGP(format, args...)
  47 #define DUMP_OFFSET(x)
  48 #endif
  49
  50 static LIST_HEAD(helpers);
  51 DECLARE_LOCK(ip_nat_seqofs_lock);
  52
  53 /* Setup TCP sequence correction given this change at this sequence */
  54 static inline void
  55 adjust_tcp_sequence(u32 seq,
  56                     int sizediff,
  57                     struct ip_conntrack *ct,
  58                     enum ip_conntrack_info ctinfo)
  59 {
  60         int dir;
  61         struct ip_nat_seq *this_way, *other_way;
  62
  63         DEBUGP("ip_nat_resize_packet: old_size = %u, new_size = %u\n",
  64                 (*skb)->len, new_size);
  65
  66         dir = CTINFO2DIR(ctinfo);
  67
  68         this_way = &ct->nat.info.seq[dir];
  69         other_way = &ct->nat.info.seq[!dir];
  70
  71         DEBUGP("ip_nat_resize_packet: Seq_offset before: ");
  72         DUMP_OFFSET(this_way);
  73
  74         LOCK_BH(&ip_nat_seqofs_lock);
  75
  76         /* SYN adjust. If it's uninitialized, or this is after last
  77          * correction, record it: we don't handle more than one
  78          * adjustment in the window, but do deal with common case of a
  79          * retransmit */
  80         if (this_way->offset_before == this_way->offset_after
  81             || before(this_way->correction_pos, seq)) {
  82                     this_way->correction_pos = seq;
  83                     this_way->offset_before = this_way->offset_after;
  84                     this_way->offset_after += sizediff;
  85         }
  86         UNLOCK_BH(&ip_nat_seqofs_lock);
  87
  88         DEBUGP("ip_nat_resize_packet: Seq_offset after: ");
  89         DUMP_OFFSET(this_way);
  90 }
  91
  92 /* Frobs data inside this packet, which is linear. */
  93 static void mangle_contents(struct sk_buff *skb,
  94                             unsigned int dataoff,
  95                             unsigned int match_offset,
  96                             unsigned int match_len,
  97                             const char *rep_buffer,
  98                             unsigned int rep_len)
  99 {
 100         unsigned char *data;
 101
 102         BUG_ON(skb_is_nonlinear(skb));
 103         data = (unsigned char *)skb->nh.iph + dataoff;
 104
 105         /* move post-replacement */
 106         memmove(data + match_offset + rep_len,
 107                 data + match_offset + match_len,
 108                 skb->tail - (data + 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                 DEBUGP("ip_nat_mangle_packet: Extending packet by "
 116                         "%u from %u bytes\n", rep_len - match_len,
 117                        skb->len);
 118                 skb_put(skb, rep_len - match_len);
 119         } else {
 120                 DEBUGP("ip_nat_mangle_packet: Shrinking packet from "
 121                         "%u from %u bytes\n", match_len - rep_len,
 122                        skb->len);
 123                 __skb_trim(skb, skb->len + rep_len - match_len);
 124         }
 125
 126         /* fix IP hdr checksum information */
 127         skb->nh.iph->tot_len = htons(skb->len);
 128         ip_send_check(skb->nh.iph);
 129 }
 130
 131 /* Unusual, but possible case. */
 132 static int enlarge_skb(struct sk_buff **pskb, unsigned int extra)
 133 {
 134         struct sk_buff *nskb;
 135
 136         if ((*pskb)->len + extra > 65535)
 137                 return 0;
 138
 139         nskb = skb_copy_expand(*pskb, skb_headroom(*pskb), extra, GFP_ATOMIC);
 140         if (!nskb)
 141                 return 0;
 142
 143         /* Transfer socket to new skb. */
 144         if ((*pskb)->sk)
 145                 skb_set_owner_w(nskb, (*pskb)->sk);
 146 #ifdef CONFIG_NETFILTER_DEBUG
 147         nskb->nf_debug = (*pskb)->nf_debug;
 148 #endif
 149         kfree_skb(*pskb);
 150         *pskb = nskb;
 151         return 1;
 152 }
 153
 154 /* Generic function for mangling variable-length address changes inside
 155  * NATed TCP connections (like the PORT XXX,XXX,XXX,XXX,XXX,XXX
 156  * command in FTP).
 157  *
 158  * Takes care about all the nasty sequence number changes, checksumming,
 159  * skb enlargement, ...
 160  *
 161  * */
 162 int
 163 ip_nat_mangle_tcp_packet(struct sk_buff **pskb,
 164                          struct ip_conntrack *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)
 170 {
 171         struct iphdr *iph;
 172         struct tcphdr *tcph;
 173         int datalen;
 174
 175         if (!skb_ip_make_writable(pskb, (*pskb)->len))
 176                 return 0;
 177
 178         if (rep_len > match_len
 179             && rep_len - match_len > skb_tailroom(*pskb)
 180             && !enlarge_skb(pskb, rep_len - match_len))
 181                 return 0;
 182
 183         SKB_LINEAR_ASSERT(*pskb);
 184
 185         iph = (*pskb)->nh.iph;
 186         tcph = (void *)iph + iph->ihl*4;
 187
 188         mangle_contents(*pskb, iph->ihl*4 + tcph->doff*4,
 189                         match_offset, match_len, rep_buffer, rep_len);
 190
 191         datalen = (*pskb)->len - iph->ihl*4;
 192         tcph->check = 0;
 193         tcph->check = tcp_v4_check(tcph, datalen, iph->saddr, iph->daddr,
 194                                    csum_partial((char *)tcph, datalen, 0));
 195
 196         adjust_tcp_sequence(ntohl(tcph->seq),
 197                             (int)rep_len - (int)match_len,
 198                             ct, ctinfo);
 199         return 1;
 200 }
 201
 202 /* Generic function for mangling variable-length address changes inside
 203  * NATed UDP connections (like the CONNECT DATA XXXXX MESG XXXXX INDEX XXXXX
 204  * command in the Amanda protocol)
 205  *
 206  * Takes care about all the nasty sequence number changes, checksumming,
 207  * skb enlargement, ...
 208  *
 209  * XXX - This function could be merged with ip_nat_mangle_tcp_packet which
 210  *       should be fairly easy to do.
 211  */
 212 int
 213 ip_nat_mangle_udp_packet(struct sk_buff **pskb,
 214                          struct ip_conntrack *ct,
 215                          enum ip_conntrack_info ctinfo,
 216                          unsigned int match_offset,
 217                          unsigned int match_len,
 218                          const char *rep_buffer,
 219                          unsigned int rep_len)
 220 {
 221         struct iphdr *iph;
 222         struct udphdr *udph;
 223
 224         /* UDP helpers might accidentally mangle the wrong packet */
 225         iph = (*pskb)->nh.iph;
 226         if ((*pskb)->len < iph->ihl*4 + sizeof(*udph) +
 227                                match_offset + match_len)
 228                 return 0;
 229
 230         if (!skb_ip_make_writable(pskb, (*pskb)->len))
 231                 return 0;
 232
 233         if (rep_len > match_len
 234             && rep_len - match_len > skb_tailroom(*pskb)
 235             && !enlarge_skb(pskb, rep_len - match_len))
 236                 return 0;
 237
 238         iph = (*pskb)->nh.iph;
 239         udph = (void *)iph + iph->ihl*4;
 240         mangle_contents(*pskb, iph->ihl*4 + sizeof(*udph),
 241                         match_offset, match_len, rep_buffer, rep_len);
 242
 243         /* update the length of the UDP packet */
 244         udph->len = htons((*pskb)->len - iph->ihl*4);
 245
 246         /* fix udp checksum if udp checksum was previously calculated */
 247         if (udph->check) {
 248                 int datalen = (*pskb)->len - iph->ihl * 4;
 249                 udph->check = 0;
 250                 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 251                                                 datalen, IPPROTO_UDP,
 252                                                 csum_partial((char *)udph,
 253                                                              datalen, 0));
 254         }
 255
 256         return 1;
 257 }
 258
 259 /* Adjust one found SACK option including checksum correction */
 260 static void
 261 sack_adjust(struct sk_buff *skb,
 262             struct tcphdr *tcph,
 263             unsigned int sackoff,
 264             unsigned int sackend,
 265             struct ip_nat_seq *natseq)
 266 {
 267         while (sackoff < sackend) {
 268                 struct tcp_sack_block *sack;
 269                 u_int32_t new_start_seq, new_end_seq;
 270
 271                 sack = (void *)skb->data + sackoff;
 272                 if (after(ntohl(sack->start_seq) - natseq->offset_before,
 273                           natseq->correction_pos))
 274                         new_start_seq = ntohl(sack->start_seq)
 275                                         - natseq->offset_after;
 276                 else
 277                         new_start_seq = ntohl(sack->start_seq)
 278                                         - natseq->offset_before;
 279                 new_start_seq = htonl(new_start_seq);
 280
 281                 if (after(ntohl(sack->end_seq) - natseq->offset_before,
 282                           natseq->correction_pos))
 283                         new_end_seq = ntohl(sack->end_seq)
 284                                       - natseq->offset_after;
 285                 else
 286                         new_end_seq = ntohl(sack->end_seq)
 287                                       - natseq->offset_before;
 288                 new_end_seq = htonl(new_end_seq);
 289
 290                 DEBUGP("sack_adjust: start_seq: %d->%d, end_seq: %d->%d\n",
 291                         ntohl(sack->start_seq), new_start_seq,
 292                         ntohl(sack->end_seq), new_end_seq);
 293
 294                 tcph->check =
 295                         ip_nat_cheat_check(~sack->start_seq, new_start_seq,
 296                                            ip_nat_cheat_check(~sack->end_seq,
 297                                                               new_end_seq,
 298                                                               tcph->check));
 299                 sack->start_seq = new_start_seq;
 300                 sack->end_seq = new_end_seq;
 301                 sackoff += sizeof(*sack);
 302         }
 303 }
 304
 305 /* TCP SACK sequence number adjustment */
 306 static inline unsigned int
 307 ip_nat_sack_adjust(struct sk_buff **pskb,
 308                    struct tcphdr *tcph,
 309                    struct ip_conntrack *ct,
 310                    enum ip_conntrack_info ctinfo)
 311 {
 312         unsigned int dir, optoff, optend;
 313
 314         optoff = (*pskb)->nh.iph->ihl*4 + sizeof(struct tcphdr);
 315         optend = (*pskb)->nh.iph->ihl*4 + tcph->doff*4;
 316
 317         if (!skb_ip_make_writable(pskb, optend))
 318                 return 0;
 319
 320         dir = CTINFO2DIR(ctinfo);
 321
 322         while (optoff < optend) {
 323                 /* Usually: option, length. */
 324                 unsigned char *op = (*pskb)->data + optoff;
 325
 326                 switch (op[0]) {
 327                 case TCPOPT_EOL:
 328                         return 1;
 329                 case TCPOPT_NOP:
 330                         optoff++;
 331                         continue;
 332                 default:
 333                         /* no partial options */
 334                         if (optoff + 1 == optend
 335                             || optoff + op[1] > optend
 336                             || op[1] < 2)
 337                                 return 0;
 338                         if (op[0] == TCPOPT_SACK
 339                             && op[1] >= 2+TCPOLEN_SACK_PERBLOCK
 340                             && ((op[1] - 2) % TCPOLEN_SACK_PERBLOCK) == 0)
 341                                 sack_adjust(*pskb, tcph, optoff+2,
 342                                             optoff+op[1],
 343                                             &ct->nat.info.seq[!dir]);
 344                         optoff += op[1];
 345                 }
 346         }
 347         return 1;
 348 }
 349
 350 /* TCP sequence number adjustment.  Returns 1 on success, 0 on failure */
 351 int
 352 ip_nat_seq_adjust(struct sk_buff **pskb,
 353                   struct ip_conntrack *ct,
 354                   enum ip_conntrack_info ctinfo)
 355 {
 356         struct tcphdr *tcph;
 357         int dir, newseq, newack;
 358         struct ip_nat_seq *this_way, *other_way;
 359
 360         dir = CTINFO2DIR(ctinfo);
 361
 362         this_way = &ct->nat.info.seq[dir];
 363         other_way = &ct->nat.info.seq[!dir];
 364
 365         /* No adjustments to make?  Very common case. */
 366         if (!this_way->offset_before && !this_way->offset_after
 367             && !other_way->offset_before && !other_way->offset_after)
 368                 return 1;
 369
 370         if (!skb_ip_make_writable(pskb, (*pskb)->nh.iph->ihl*4+sizeof(*tcph)))
 371                 return 0;
 372
 373         tcph = (void *)(*pskb)->data + (*pskb)->nh.iph->ihl*4;
 374         if (after(ntohl(tcph->seq), this_way->correction_pos))
 375                 newseq = ntohl(tcph->seq) + this_way->offset_after;
 376         else
 377                 newseq = ntohl(tcph->seq) + this_way->offset_before;
 378         newseq = htonl(newseq);
 379
 380         if (after(ntohl(tcph->ack_seq) - other_way->offset_before,
 381                   other_way->correction_pos))
 382                 newack = ntohl(tcph->ack_seq) - other_way->offset_after;
 383         else
 384                 newack = ntohl(tcph->ack_seq) - other_way->offset_before;
 385         newack = htonl(newack);
 386
 387         tcph->check = ip_nat_cheat_check(~tcph->seq, newseq,
 388                                          ip_nat_cheat_check(~tcph->ack_seq,
 389                                                             newack,
 390                                                             tcph->check));
 391
 392         DEBUGP("Adjusting sequence number from %u->%u, ack from %u->%u\n",
 393                 ntohl(tcph->seq), ntohl(newseq), ntohl(tcph->ack_seq),
 394                 ntohl(newack));
 395
 396         tcph->seq = newseq;
 397         tcph->ack_seq = newack;
 398
 399         if (!ip_nat_sack_adjust(pskb, tcph, ct, ctinfo))
 400                 return 0;
 401
 402         ip_conntrack_tcp_update(*pskb, ct, dir);
 403
 404         return 1;
 405 }
 406
 407 static inline int
 408 helper_cmp(const struct ip_nat_helper *helper,
 409            const struct ip_conntrack_tuple *tuple)
 410 {
 411         return ip_ct_tuple_mask_cmp(tuple, &helper->tuple, &helper->mask);
 412 }
 413
 414 int ip_nat_helper_register(struct ip_nat_helper *me)
 415 {
 416         int ret = 0;
 417
 418         WRITE_LOCK(&ip_nat_lock);
 419         if (LIST_FIND(&helpers, helper_cmp, struct ip_nat_helper *,&me->tuple))
 420                 ret = -EBUSY;
 421         else
 422                 list_prepend(&helpers, me);
 423         WRITE_UNLOCK(&ip_nat_lock);
 424
 425         return ret;
 426 }
 427
 428 struct ip_nat_helper *
 429 __ip_nat_find_helper(const struct ip_conntrack_tuple *tuple)
 430 {
 431         return LIST_FIND(&helpers, helper_cmp, struct ip_nat_helper *, tuple);
 432 }
 433
 434 struct ip_nat_helper *
 435 ip_nat_find_helper(const struct ip_conntrack_tuple *tuple)
 436 {
 437         struct ip_nat_helper *h;
 438
 439         READ_LOCK(&ip_nat_lock);
 440         h = __ip_nat_find_helper(tuple);
 441         READ_UNLOCK(&ip_nat_lock);
 442
 443         return h;
 444 }
 445
 446 static int
 447 kill_helper(const struct ip_conntrack *i, void *helper)
 448 {
 449         int ret;
 450
 451         READ_LOCK(&ip_nat_lock);
 452         ret = (i->nat.info.helper == helper);
 453         READ_UNLOCK(&ip_nat_lock);
 454
 455         return ret;
 456 }
 457
 458 void ip_nat_helper_unregister(struct ip_nat_helper *me)
 459 {
 460         WRITE_LOCK(&ip_nat_lock);
 461         /* Autoloading conntrack helper might have failed */
 462         if (LIST_FIND(&helpers, helper_cmp, struct ip_nat_helper *,&me->tuple)) {
 463                 LIST_DELETE(&helpers, me);
 464         }
 465         WRITE_UNLOCK(&ip_nat_lock);
 466
 467         /* Someone could be still looking at the helper in a bh. */
 468         synchronize_net();
 469
 470         /* Find anything using it, and umm, kill them.  We can't turn
 471            them into normal connections: if we've adjusted SYNs, then
 472            they'll ackstorm.  So we just drop it.  We used to just
 473            bump module count when a connection existed, but that
 474            forces admins to gen fake RSTs or bounce box, either of
 475            which is just a long-winded way of making things
 476            worse. --RR */
 477         ip_ct_selective_cleanup(kill_helper, me);
 478 }