Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 31 | #define ASSERT_READ_LOCK(x) |
| 32 | #define ASSERT_WRITE_LOCK(x) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 50 | static DEFINE_SPINLOCK(ip_nat_seqofs_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 73 | spin_lock_bh(&ip_nat_seqofs_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 74 | |
| 75 | /* SYN adjust. If it's uninitialized, or 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 | } |
Patrick McHardy | e45b1be | 2005-06-21 14:01:30 -0700 | [diff] [blame] | 85 | spin_unlock_bh(&ip_nat_seqofs_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 145 | kfree_skb(*pskb); |
| 146 | *pskb = nskb; |
| 147 | return 1; |
| 148 | } |
| 149 | |
| 150 | /* Generic function for mangling variable-length address changes inside |
| 151 | * NATed TCP connections (like the PORT XXX,XXX,XXX,XXX,XXX,XXX |
| 152 | * command in FTP). |
| 153 | * |
| 154 | * Takes care about all the nasty sequence number changes, checksumming, |
| 155 | * skb enlargement, ... |
| 156 | * |
| 157 | * */ |
| 158 | int |
| 159 | ip_nat_mangle_tcp_packet(struct sk_buff **pskb, |
| 160 | struct ip_conntrack *ct, |
| 161 | enum ip_conntrack_info ctinfo, |
| 162 | unsigned int match_offset, |
| 163 | unsigned int match_len, |
| 164 | const char *rep_buffer, |
| 165 | unsigned int rep_len) |
| 166 | { |
| 167 | struct iphdr *iph; |
| 168 | struct tcphdr *tcph; |
| 169 | int datalen; |
| 170 | |
Harald Welte | 089af26 | 2005-08-09 19:37:23 -0700 | [diff] [blame^] | 171 | if (!skb_make_writable(pskb, (*pskb)->len)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 172 | return 0; |
| 173 | |
| 174 | if (rep_len > match_len |
| 175 | && rep_len - match_len > skb_tailroom(*pskb) |
| 176 | && !enlarge_skb(pskb, rep_len - match_len)) |
| 177 | return 0; |
| 178 | |
| 179 | SKB_LINEAR_ASSERT(*pskb); |
| 180 | |
| 181 | iph = (*pskb)->nh.iph; |
| 182 | tcph = (void *)iph + iph->ihl*4; |
| 183 | |
| 184 | mangle_contents(*pskb, iph->ihl*4 + tcph->doff*4, |
| 185 | match_offset, match_len, rep_buffer, rep_len); |
| 186 | |
| 187 | datalen = (*pskb)->len - iph->ihl*4; |
| 188 | tcph->check = 0; |
| 189 | tcph->check = tcp_v4_check(tcph, datalen, iph->saddr, iph->daddr, |
| 190 | csum_partial((char *)tcph, datalen, 0)); |
| 191 | |
| 192 | if (rep_len != match_len) { |
| 193 | set_bit(IPS_SEQ_ADJUST_BIT, &ct->status); |
| 194 | adjust_tcp_sequence(ntohl(tcph->seq), |
| 195 | (int)rep_len - (int)match_len, |
| 196 | ct, ctinfo); |
| 197 | /* Tell TCP window tracking about seq change */ |
| 198 | ip_conntrack_tcp_update(*pskb, ct, CTINFO2DIR(ctinfo)); |
| 199 | } |
| 200 | return 1; |
| 201 | } |
| 202 | |
| 203 | /* Generic function for mangling variable-length address changes inside |
| 204 | * NATed UDP connections (like the CONNECT DATA XXXXX MESG XXXXX INDEX XXXXX |
| 205 | * command in the Amanda protocol) |
| 206 | * |
| 207 | * Takes care about all the nasty sequence number changes, checksumming, |
| 208 | * skb enlargement, ... |
| 209 | * |
| 210 | * XXX - This function could be merged with ip_nat_mangle_tcp_packet which |
| 211 | * should be fairly easy to do. |
| 212 | */ |
| 213 | int |
| 214 | ip_nat_mangle_udp_packet(struct sk_buff **pskb, |
| 215 | struct ip_conntrack *ct, |
| 216 | enum ip_conntrack_info ctinfo, |
| 217 | unsigned int match_offset, |
| 218 | unsigned int match_len, |
| 219 | const char *rep_buffer, |
| 220 | unsigned int rep_len) |
| 221 | { |
| 222 | struct iphdr *iph; |
| 223 | struct udphdr *udph; |
| 224 | |
| 225 | /* UDP helpers might accidentally mangle the wrong packet */ |
| 226 | iph = (*pskb)->nh.iph; |
| 227 | if ((*pskb)->len < iph->ihl*4 + sizeof(*udph) + |
| 228 | match_offset + match_len) |
| 229 | return 0; |
| 230 | |
Harald Welte | 089af26 | 2005-08-09 19:37:23 -0700 | [diff] [blame^] | 231 | if (!skb_make_writable(pskb, (*pskb)->len)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 232 | return 0; |
| 233 | |
| 234 | if (rep_len > match_len |
| 235 | && rep_len - match_len > skb_tailroom(*pskb) |
| 236 | && !enlarge_skb(pskb, rep_len - match_len)) |
| 237 | return 0; |
| 238 | |
| 239 | iph = (*pskb)->nh.iph; |
| 240 | udph = (void *)iph + iph->ihl*4; |
| 241 | mangle_contents(*pskb, iph->ihl*4 + sizeof(*udph), |
| 242 | match_offset, match_len, rep_buffer, rep_len); |
| 243 | |
| 244 | /* update the length of the UDP packet */ |
| 245 | udph->len = htons((*pskb)->len - iph->ihl*4); |
| 246 | |
| 247 | /* fix udp checksum if udp checksum was previously calculated */ |
| 248 | if (udph->check) { |
| 249 | int datalen = (*pskb)->len - iph->ihl * 4; |
| 250 | udph->check = 0; |
| 251 | udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr, |
| 252 | datalen, IPPROTO_UDP, |
| 253 | csum_partial((char *)udph, |
| 254 | datalen, 0)); |
| 255 | } |
| 256 | |
| 257 | return 1; |
| 258 | } |
| 259 | |
| 260 | /* Adjust one found SACK option including checksum correction */ |
| 261 | static void |
| 262 | sack_adjust(struct sk_buff *skb, |
| 263 | struct tcphdr *tcph, |
| 264 | unsigned int sackoff, |
| 265 | unsigned int sackend, |
| 266 | struct ip_nat_seq *natseq) |
| 267 | { |
| 268 | while (sackoff < sackend) { |
| 269 | struct tcp_sack_block *sack; |
| 270 | u_int32_t new_start_seq, new_end_seq; |
| 271 | |
| 272 | sack = (void *)skb->data + sackoff; |
| 273 | if (after(ntohl(sack->start_seq) - natseq->offset_before, |
| 274 | natseq->correction_pos)) |
| 275 | new_start_seq = ntohl(sack->start_seq) |
| 276 | - natseq->offset_after; |
| 277 | else |
| 278 | new_start_seq = ntohl(sack->start_seq) |
| 279 | - natseq->offset_before; |
| 280 | new_start_seq = htonl(new_start_seq); |
| 281 | |
| 282 | if (after(ntohl(sack->end_seq) - natseq->offset_before, |
| 283 | natseq->correction_pos)) |
| 284 | new_end_seq = ntohl(sack->end_seq) |
| 285 | - natseq->offset_after; |
| 286 | else |
| 287 | new_end_seq = ntohl(sack->end_seq) |
| 288 | - natseq->offset_before; |
| 289 | new_end_seq = htonl(new_end_seq); |
| 290 | |
| 291 | DEBUGP("sack_adjust: start_seq: %d->%d, end_seq: %d->%d\n", |
| 292 | ntohl(sack->start_seq), new_start_seq, |
| 293 | ntohl(sack->end_seq), new_end_seq); |
| 294 | |
| 295 | tcph->check = |
| 296 | ip_nat_cheat_check(~sack->start_seq, new_start_seq, |
| 297 | ip_nat_cheat_check(~sack->end_seq, |
| 298 | new_end_seq, |
| 299 | tcph->check)); |
| 300 | sack->start_seq = new_start_seq; |
| 301 | sack->end_seq = new_end_seq; |
| 302 | sackoff += sizeof(*sack); |
| 303 | } |
| 304 | } |
| 305 | |
| 306 | /* TCP SACK sequence number adjustment */ |
| 307 | static inline unsigned int |
| 308 | ip_nat_sack_adjust(struct sk_buff **pskb, |
| 309 | struct tcphdr *tcph, |
| 310 | struct ip_conntrack *ct, |
| 311 | enum ip_conntrack_info ctinfo) |
| 312 | { |
| 313 | unsigned int dir, optoff, optend; |
| 314 | |
| 315 | optoff = (*pskb)->nh.iph->ihl*4 + sizeof(struct tcphdr); |
| 316 | optend = (*pskb)->nh.iph->ihl*4 + tcph->doff*4; |
| 317 | |
Harald Welte | 089af26 | 2005-08-09 19:37:23 -0700 | [diff] [blame^] | 318 | if (!skb_make_writable(pskb, optend)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 319 | return 0; |
| 320 | |
| 321 | dir = CTINFO2DIR(ctinfo); |
| 322 | |
| 323 | while (optoff < optend) { |
| 324 | /* Usually: option, length. */ |
| 325 | unsigned char *op = (*pskb)->data + optoff; |
| 326 | |
| 327 | switch (op[0]) { |
| 328 | case TCPOPT_EOL: |
| 329 | return 1; |
| 330 | case TCPOPT_NOP: |
| 331 | optoff++; |
| 332 | continue; |
| 333 | default: |
| 334 | /* no partial options */ |
| 335 | if (optoff + 1 == optend |
| 336 | || optoff + op[1] > optend |
| 337 | || op[1] < 2) |
| 338 | return 0; |
| 339 | if (op[0] == TCPOPT_SACK |
| 340 | && op[1] >= 2+TCPOLEN_SACK_PERBLOCK |
| 341 | && ((op[1] - 2) % TCPOLEN_SACK_PERBLOCK) == 0) |
| 342 | sack_adjust(*pskb, tcph, optoff+2, |
| 343 | optoff+op[1], |
| 344 | &ct->nat.info.seq[!dir]); |
| 345 | optoff += op[1]; |
| 346 | } |
| 347 | } |
| 348 | return 1; |
| 349 | } |
| 350 | |
| 351 | /* TCP sequence number adjustment. Returns 1 on success, 0 on failure */ |
| 352 | int |
| 353 | ip_nat_seq_adjust(struct sk_buff **pskb, |
| 354 | struct ip_conntrack *ct, |
| 355 | enum ip_conntrack_info ctinfo) |
| 356 | { |
| 357 | struct tcphdr *tcph; |
| 358 | int dir, newseq, newack; |
| 359 | struct ip_nat_seq *this_way, *other_way; |
| 360 | |
| 361 | dir = CTINFO2DIR(ctinfo); |
| 362 | |
| 363 | this_way = &ct->nat.info.seq[dir]; |
| 364 | other_way = &ct->nat.info.seq[!dir]; |
| 365 | |
Harald Welte | 089af26 | 2005-08-09 19:37:23 -0700 | [diff] [blame^] | 366 | if (!skb_make_writable(pskb, (*pskb)->nh.iph->ihl*4+sizeof(*tcph))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 367 | return 0; |
| 368 | |
| 369 | tcph = (void *)(*pskb)->data + (*pskb)->nh.iph->ihl*4; |
| 370 | if (after(ntohl(tcph->seq), this_way->correction_pos)) |
| 371 | newseq = ntohl(tcph->seq) + this_way->offset_after; |
| 372 | else |
| 373 | newseq = ntohl(tcph->seq) + this_way->offset_before; |
| 374 | newseq = htonl(newseq); |
| 375 | |
| 376 | if (after(ntohl(tcph->ack_seq) - other_way->offset_before, |
| 377 | other_way->correction_pos)) |
| 378 | newack = ntohl(tcph->ack_seq) - other_way->offset_after; |
| 379 | else |
| 380 | newack = ntohl(tcph->ack_seq) - other_way->offset_before; |
| 381 | newack = htonl(newack); |
| 382 | |
| 383 | tcph->check = ip_nat_cheat_check(~tcph->seq, newseq, |
| 384 | ip_nat_cheat_check(~tcph->ack_seq, |
| 385 | newack, |
| 386 | tcph->check)); |
| 387 | |
| 388 | DEBUGP("Adjusting sequence number from %u->%u, ack from %u->%u\n", |
| 389 | ntohl(tcph->seq), ntohl(newseq), ntohl(tcph->ack_seq), |
| 390 | ntohl(newack)); |
| 391 | |
| 392 | tcph->seq = newseq; |
| 393 | tcph->ack_seq = newack; |
| 394 | |
| 395 | if (!ip_nat_sack_adjust(pskb, tcph, ct, ctinfo)) |
| 396 | return 0; |
| 397 | |
| 398 | ip_conntrack_tcp_update(*pskb, ct, dir); |
| 399 | |
| 400 | return 1; |
| 401 | } |
| 402 | |
| 403 | /* Setup NAT on this expected conntrack so it follows master. */ |
| 404 | /* If we fail to get a free NAT slot, we'll get dropped on confirm */ |
| 405 | void ip_nat_follow_master(struct ip_conntrack *ct, |
| 406 | struct ip_conntrack_expect *exp) |
| 407 | { |
| 408 | struct ip_nat_range range; |
| 409 | |
| 410 | /* This must be a fresh one. */ |
| 411 | BUG_ON(ct->status & IPS_NAT_DONE_MASK); |
| 412 | |
| 413 | /* Change src to where master sends to */ |
| 414 | range.flags = IP_NAT_RANGE_MAP_IPS; |
| 415 | range.min_ip = range.max_ip |
| 416 | = ct->master->tuplehash[!exp->dir].tuple.dst.ip; |
| 417 | /* hook doesn't matter, but it has to do source manip */ |
| 418 | ip_nat_setup_info(ct, &range, NF_IP_POST_ROUTING); |
| 419 | |
| 420 | /* For DST manip, map port here to where it's expected. */ |
| 421 | range.flags = (IP_NAT_RANGE_MAP_IPS | IP_NAT_RANGE_PROTO_SPECIFIED); |
| 422 | range.min = range.max = exp->saved_proto; |
| 423 | range.min_ip = range.max_ip |
| 424 | = ct->master->tuplehash[!exp->dir].tuple.src.ip; |
| 425 | /* hook doesn't matter, but it has to do destination manip */ |
| 426 | ip_nat_setup_info(ct, &range, NF_IP_PRE_ROUTING); |
| 427 | } |