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 DECLARE_LOCK(ip_nat_seqofs_lock);
  51
  52 /* Setup TCP sequence correction given this change at this sequence */
  53 static inline void
  54 adjust_tcp_sequence(u32 seq,
  55                     int sizediff,
  56                     struct ip_conntrack *ct,
  57                     enum ip_conntrack_info ctinfo)
  58 {
  59         int dir;
  60         struct ip_nat_seq *this_way, *other_way;
  61
  62         DEBUGP("ip_nat_resize_packet: old_size = %u, new_size = %u\n",
  63                 (*skb)->len, new_size);
  64
  65         dir = CTINFO2DIR(ctinfo);
  66
  67         this_way = &ct->nat.info.seq[dir];
  68         other_way = &ct->nat.info.seq[!dir];
  69
  70         DEBUGP("ip_nat_resize_packet: Seq_offset before: ");
  71         DUMP_OFFSET(this_way);
  72
  73         LOCK_BH(&ip_nat_seqofs_lock);
  74
  75         /* SYN adjust. If it's uninitialized, of this is after last
  76          * correction, record it: we don't handle more than one
  77          * adjustment in the window, but do deal with common case of a
  78          * retransmit */
  79         if (this_way->offset_before == this_way->offset_after
  80             || before(this_way->correction_pos, seq)) {
  81                     this_way->correction_pos = seq;
  82                     this_way->offset_before = this_way->offset_after;
  83                     this_way->offset_after += sizediff;
  84         }
  85         UNLOCK_BH(&ip_nat_seqofs_lock);
  86
  87         DEBUGP("ip_nat_resize_packet: Seq_offset after: ");
  88         DUMP_OFFSET(this_way);
  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 = (unsigned char *)skb->nh.iph + dataoff;
 103
 104         /* move post-replacement */
 105         memmove(data + match_offset + rep_len,
 106                 data + match_offset + match_len,
 107                 skb->tail - (data + match_offset + match_len));
 108
 109         /* insert data from buffer */
 110         memcpy(data + match_offset, rep_buffer, rep_len);
 111
 112         /* update skb info */
 113         if (rep_len > match_len) {
 114                 DEBUGP("ip_nat_mangle_packet: Extending packet by "
 115                         "%u from %u bytes\n", rep_len - match_len,
 116                        skb->len);
 117                 skb_put(skb, rep_len - match_len);
 118         } else {
 119                 DEBUGP("ip_nat_mangle_packet: Shrinking packet from "
 120                         "%u from %u bytes\n", match_len - rep_len,
 121                        skb->len);
 122                 __skb_trim(skb, skb->len + rep_len - match_len);
 123         }
 124
 125         /* fix IP hdr checksum information */
 126         skb->nh.iph->tot_len = htons(skb->len);
 127         ip_send_check(skb->nh.iph);
 128 }
 129
 130 /* Unusual, but possible case. */
 131 static int enlarge_skb(struct sk_buff **pskb, unsigned int extra)
 132 {
 133         struct sk_buff *nskb;
 134
 135         if ((*pskb)->len + extra > 65535)
 136                 return 0;
 137
 138         nskb = skb_copy_expand(*pskb, skb_headroom(*pskb), extra, GFP_ATOMIC);
 139         if (!nskb)
 140                 return 0;
 141
 142         /* Transfer socket to new skb. */
 143         if ((*pskb)->sk)
 144                 skb_set_owner_w(nskb, (*pskb)->sk);
 145 #ifdef CONFIG_NETFILTER_DEBUG
 146         nskb->nf_debug = (*pskb)->nf_debug;
 147 #endif
 148         kfree_skb(*pskb);
 149         *pskb = nskb;
 150         return 1;
 151 }
 152
 153 /* Generic function for mangling variable-length address changes inside
 154  * NATed TCP connections (like the PORT XXX,XXX,XXX,XXX,XXX,XXX
 155  * command in FTP).
 156  *
 157  * Takes care about all the nasty sequence number changes, checksumming,
 158  * skb enlargement, ...
 159  *
 160  * */
 161 int
 162 ip_nat_mangle_tcp_packet(struct sk_buff **pskb,
 163                          struct ip_conntrack *ct,
 164                          enum ip_conntrack_info ctinfo,
 165                          unsigned int match_offset,
 166                          unsigned int match_len,
 167                          const char *rep_buffer,
 168                          unsigned int rep_len)
 169 {
 170         struct iphdr *iph;
 171         struct tcphdr *tcph;
 172         int datalen;
 173
 174         if (!skb_ip_make_writable(pskb, (*pskb)->len))
 175                 return 0;
 176
 177         if (rep_len > match_len
 178             && rep_len - match_len > skb_tailroom(*pskb)
 179             && !enlarge_skb(pskb, rep_len - match_len))
 180                 return 0;
 181
 182         SKB_LINEAR_ASSERT(*pskb);
 183
 184         iph = (*pskb)->nh.iph;
 185         tcph = (void *)iph + iph->ihl*4;
 186
 187         mangle_contents(*pskb, iph->ihl*4 + tcph->doff*4,
 188                         match_offset, match_len, rep_buffer, rep_len);
 189
 190         datalen = (*pskb)->len - iph->ihl*4;
 191         tcph->check = 0;
 192         tcph->check = tcp_v4_check(tcph, datalen, iph->saddr, iph->daddr,
 193                                    csum_partial((char *)tcph, datalen, 0));
 194
 195         adjust_tcp_sequence(ntohl(tcph->seq),
 196                             (int)rep_len - (int)match_len,
 197                             ct, ctinfo);
 198         return 1;
 199 }
 200
 201 /* Generic function for mangling variable-length address changes inside
 202  * NATed UDP connections (like the CONNECT DATA XXXXX MESG XXXXX INDEX XXXXX
 203  * command in the Amanda protocol)
 204  *
 205  * Takes care about all the nasty sequence number changes, checksumming,
 206  * skb enlargement, ...
 207  *
 208  * XXX - This function could be merged with ip_nat_mangle_tcp_packet which
 209  *       should be fairly easy to do.
 210  */
 211 int
 212 ip_nat_mangle_udp_packet(struct sk_buff **pskb,
 213                          struct ip_conntrack *ct,
 214                          enum ip_conntrack_info ctinfo,
 215                          unsigned int match_offset,
 216                          unsigned int match_len,
 217                          const char *rep_buffer,
 218                          unsigned int rep_len)
 219 {
 220         struct iphdr *iph;
 221         struct udphdr *udph;
 222
 223         /* UDP helpers might accidentally mangle the wrong packet */
 224         iph = (*pskb)->nh.iph;
 225         if ((*pskb)->len < iph->ihl*4 + sizeof(*udph) +
 226                                match_offset + match_len)
 227                 return 0;
 228
 229         if (!skb_ip_make_writable(pskb, (*pskb)->len))
 230                 return 0;
 231
 232         if (rep_len > match_len
 233             && rep_len - match_len > skb_tailroom(*pskb)
 234             && !enlarge_skb(pskb, rep_len - match_len))
 235                 return 0;
 236
 237         iph = (*pskb)->nh.iph;
 238         udph = (void *)iph + iph->ihl*4;
 239         mangle_contents(*pskb, iph->ihl*4 + sizeof(*udph),
 240                         match_offset, match_len, rep_buffer, rep_len);
 241
 242         /* update the length of the UDP packet */
 243         udph->len = htons((*pskb)->len - iph->ihl*4);
 244
 245         /* fix udp checksum if udp checksum was previously calculated */
 246         if (udph->check) {
 247                 int datalen = (*pskb)->len - iph->ihl * 4;
 248                 udph->check = 0;
 249                 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 250                                                 datalen, IPPROTO_UDP,
 251                                                 csum_partial((char *)udph,
 252                                                              datalen, 0));
 253         }
 254
 255         return 1;
 256 }
 257
 258 /* Adjust one found SACK option including checksum correction */
 259 static void
 260 sack_adjust(struct sk_buff *skb,
 261             struct tcphdr *tcph,
 262             unsigned int sackoff,
 263             unsigned int sackend,
 264             struct ip_nat_seq *natseq)
 265 {
 266         while (sackoff < sackend) {
 267                 struct tcp_sack_block *sack;
 268                 u_int32_t new_start_seq, new_end_seq;
 269
 270                 sack = (void *)skb->data + sackoff;
 271                 if (after(ntohl(sack->start_seq) - natseq->offset_before,
 272                           natseq->correction_pos))
 273                         new_start_seq = ntohl(sack->start_seq)
 274                                         - natseq->offset_after;
 275                 else
 276                         new_start_seq = ntohl(sack->start_seq)
 277                                         - natseq->offset_before;
 278                 new_start_seq = htonl(new_start_seq);
 279
 280                 if (after(ntohl(sack->end_seq) - natseq->offset_before,
 281                           natseq->correction_pos))
 282                         new_end_seq = ntohl(sack->end_seq)
 283                                       - natseq->offset_after;
 284                 else
 285                         new_end_seq = ntohl(sack->end_seq)
 286                                       - natseq->offset_before;
 287                 new_end_seq = htonl(new_end_seq);
 288
 289                 DEBUGP("sack_adjust: start_seq: %d->%d, end_seq: %d->%d\n",
 290                         ntohl(sack->start_seq), new_start_seq,
 291                         ntohl(sack->end_seq), new_end_seq);
 292
 293                 tcph->check =
 294                         ip_nat_cheat_check(~sack->start_seq, new_start_seq,
 295                                            ip_nat_cheat_check(~sack->end_seq,
 296                                                               new_end_seq,
 297                                                               tcph->check));
 298                 sack->start_seq = new_start_seq;
 299                 sack->end_seq = new_end_seq;
 300                 sackoff += sizeof(*sack);
 301         }
 302 }
 303
 304 /* TCP SACK sequence number adjustment */
 305 static inline unsigned int
 306 ip_nat_sack_adjust(struct sk_buff **pskb,
 307                    struct tcphdr *tcph,
 308                    struct ip_conntrack *ct,
 309                    enum ip_conntrack_info ctinfo)
 310 {
 311         unsigned int dir, optoff, optend;
 312
 313         optoff = (*pskb)->nh.iph->ihl*4 + sizeof(struct tcphdr);
 314         optend = (*pskb)->nh.iph->ihl*4 + tcph->doff*4;
 315
 316         if (!skb_ip_make_writable(pskb, optend))
 317                 return 0;
 318
 319         dir = CTINFO2DIR(ctinfo);
 320
 321         while (optoff < optend) {
 322                 /* Usually: option, length. */
 323                 unsigned char *op = (*pskb)->data + optoff;
 324
 325                 switch (op[0]) {
 326                 case TCPOPT_EOL:
 327                         return 1;
 328                 case TCPOPT_NOP:
 329                         optoff++;
 330                         continue;
 331                 default:
 332                         /* no partial options */
 333                         if (optoff + 1 == optend
 334                             || optoff + op[1] > optend
 335                             || op[1] < 2)
 336                                 return 0;
 337                         if (op[0] == TCPOPT_SACK
 338                             && op[1] >= 2+TCPOLEN_SACK_PERBLOCK
 339                             && ((op[1] - 2) % TCPOLEN_SACK_PERBLOCK) == 0)
 340                                 sack_adjust(*pskb, tcph, optoff+2,
 341                                             optoff+op[1],
 342                                             &ct->nat.info.seq[!dir]);
 343                         optoff += op[1];
 344                 }
 345         }
 346         return 1;
 347 }
 348
 349 /* TCP sequence number adjustment.  Returns true or false.  */
 350 int
 351 ip_nat_seq_adjust(struct sk_buff **pskb,
 352                   struct ip_conntrack *ct,
 353                   enum ip_conntrack_info ctinfo)
 354 {
 355         struct tcphdr *tcph;
 356         int dir, newseq, newack;
 357         struct ip_nat_seq *this_way, *other_way;
 358
 359         dir = CTINFO2DIR(ctinfo);
 360
 361         this_way = &ct->nat.info.seq[dir];
 362         other_way = &ct->nat.info.seq[!dir];
 363
 364         /* No adjustments to make?  Very common case. */
 365         if (!this_way->offset_before && !this_way->offset_after
 366             && !other_way->offset_before && !other_way->offset_after)
 367                 return 1;
 368
 369         if (!skb_ip_make_writable(pskb, (*pskb)->nh.iph->ihl*4+sizeof(*tcph)))
 370                 return 0;
 371
 372         tcph = (void *)(*pskb)->data + (*pskb)->nh.iph->ihl*4;
 373         if (after(ntohl(tcph->seq), this_way->correction_pos))
 374                 newseq = ntohl(tcph->seq) + this_way->offset_after;
 375         else
 376                 newseq = ntohl(tcph->seq) + this_way->offset_before;
 377         newseq = htonl(newseq);
 378
 379         if (after(ntohl(tcph->ack_seq) - other_way->offset_before,
 380                   other_way->correction_pos))
 381                 newack = ntohl(tcph->ack_seq) - other_way->offset_after;
 382         else
 383                 newack = ntohl(tcph->ack_seq) - other_way->offset_before;
 384         newack = htonl(newack);
 385
 386         tcph->check = ip_nat_cheat_check(~tcph->seq, newseq,
 387                                          ip_nat_cheat_check(~tcph->ack_seq,
 388                                                             newack,
 389                                                             tcph->check));
 390
 391         DEBUGP("Adjusting sequence number from %u->%u, ack from %u->%u\n",
 392                 ntohl(tcph->seq), ntohl(newseq), ntohl(tcph->ack_seq),
 393                 ntohl(newack));
 394
 395         tcph->seq = newseq;
 396         tcph->ack_seq = newack;
 397
 398         return ip_nat_sack_adjust(pskb, tcph, ct, ctinfo);
 399 }
 400
 401 static inline int
 402 helper_cmp(const struct ip_nat_helper *helper,
 403            const struct ip_conntrack_tuple *tuple)
 404 {
 405         return ip_ct_tuple_mask_cmp(tuple, &helper->tuple, &helper->mask);
 406 }
 407
 408 int ip_nat_helper_register(struct ip_nat_helper *me)
 409 {
 410         int ret = 0;
 411
 412         WRITE_LOCK(&ip_nat_lock);
 413         if (LIST_FIND(&helpers, helper_cmp, struct ip_nat_helper *,&me->tuple))
 414                 ret = -EBUSY;
 415         else
 416                 list_prepend(&helpers, me);
 417         WRITE_UNLOCK(&ip_nat_lock);
 418
 419         return ret;
 420 }
 421
 422 static int
 423 kill_helper(const struct ip_conntrack *i, void *helper)
 424 {
 425         int ret;
 426
 427         READ_LOCK(&ip_nat_lock);
 428         ret = (i->nat.info.helper == helper);
 429         READ_UNLOCK(&ip_nat_lock);
 430
 431         return ret;
 432 }
 433
 434 void ip_nat_helper_unregister(struct ip_nat_helper *me)
 435 {
 436         WRITE_LOCK(&ip_nat_lock);
 437         /* Autoloading conntrack helper might have failed */
 438         if (LIST_FIND(&helpers, helper_cmp, struct ip_nat_helper *,&me->tuple)) {
 439                 LIST_DELETE(&helpers, me);
 440         }
 441         WRITE_UNLOCK(&ip_nat_lock);
 442
 443         /* Someone could be still looking at the helper in a bh. */
 444         synchronize_net();
 445
 446         /* Find anything using it, and umm, kill them.  We can't turn
 447            them into normal connections: if we've adjusted SYNs, then
 448            they'll ackstorm.  So we just drop it.  We used to just
 449            bump module count when a connection existed, but that
 450            forces admins to gen fake RSTs or bounce box, either of
 451            which is just a long-winded way of making things
 452            worse. --RR */
 453         ip_ct_selective_cleanup(kill_helper, me);
 454 }