1637b96d8c0110c155efd1fe7492570cfdb34650
[linux-2.6.git] / 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 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, 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         }
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         if (rep_len != match_len) {
196                 set_bit(IPS_SEQ_ADJUST_BIT, &ct->status);
197                 adjust_tcp_sequence(ntohl(tcph->seq),
198                                     (int)rep_len - (int)match_len,
199                                     ct, ctinfo);
200                 /* Tell TCP window tracking about seq change */
201                 ip_conntrack_tcp_update(*pskb, ct, CTINFO2DIR(ctinfo));
202         }
203         return 1;
204 }
205                         
206 /* Generic function for mangling variable-length address changes inside
207  * NATed UDP connections (like the CONNECT DATA XXXXX MESG XXXXX INDEX XXXXX
208  * command in the Amanda protocol)
209  *
210  * Takes care about all the nasty sequence number changes, checksumming,
211  * skb enlargement, ...
212  *
213  * XXX - This function could be merged with ip_nat_mangle_tcp_packet which
214  *       should be fairly easy to do.
215  */
216 int 
217 ip_nat_mangle_udp_packet(struct sk_buff **pskb,
218                          struct ip_conntrack *ct,
219                          enum ip_conntrack_info ctinfo,
220                          unsigned int match_offset,
221                          unsigned int match_len,
222                          const char *rep_buffer,
223                          unsigned int rep_len)
224 {
225         struct iphdr *iph;
226         struct udphdr *udph;
227
228         /* UDP helpers might accidentally mangle the wrong packet */
229         iph = (*pskb)->nh.iph;
230         if ((*pskb)->len < iph->ihl*4 + sizeof(*udph) + 
231                                match_offset + match_len)
232                 return 0;
233
234         if (!skb_ip_make_writable(pskb, (*pskb)->len))
235                 return 0;
236
237         if (rep_len > match_len
238             && rep_len - match_len > skb_tailroom(*pskb)
239             && !enlarge_skb(pskb, rep_len - match_len))
240                 return 0;
241
242         iph = (*pskb)->nh.iph;
243         udph = (void *)iph + iph->ihl*4;
244         mangle_contents(*pskb, iph->ihl*4 + sizeof(*udph),
245                         match_offset, match_len, rep_buffer, rep_len);
246
247         /* update the length of the UDP packet */
248         udph->len = htons((*pskb)->len - iph->ihl*4);
249
250         /* fix udp checksum if udp checksum was previously calculated */
251         if (udph->check) {
252                 int datalen = (*pskb)->len - iph->ihl * 4;
253                 udph->check = 0;
254                 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
255                                                 datalen, IPPROTO_UDP,
256                                                 csum_partial((char *)udph,
257                                                              datalen, 0));
258         }
259
260         return 1;
261 }
262
263 /* Adjust one found SACK option including checksum correction */
264 static void
265 sack_adjust(struct sk_buff *skb,
266             struct tcphdr *tcph, 
267             unsigned int sackoff,
268             unsigned int sackend,
269             struct ip_nat_seq *natseq)
270 {
271         while (sackoff < sackend) {
272                 struct tcp_sack_block *sack;
273                 u_int32_t new_start_seq, new_end_seq;
274
275                 sack = (void *)skb->data + sackoff;
276                 if (after(ntohl(sack->start_seq) - natseq->offset_before,
277                           natseq->correction_pos))
278                         new_start_seq = ntohl(sack->start_seq) 
279                                         - natseq->offset_after;
280                 else
281                         new_start_seq = ntohl(sack->start_seq) 
282                                         - natseq->offset_before;
283                 new_start_seq = htonl(new_start_seq);
284
285                 if (after(ntohl(sack->end_seq) - natseq->offset_before,
286                           natseq->correction_pos))
287                         new_end_seq = ntohl(sack->end_seq)
288                                       - natseq->offset_after;
289                 else
290                         new_end_seq = ntohl(sack->end_seq)
291                                       - natseq->offset_before;
292                 new_end_seq = htonl(new_end_seq);
293
294                 DEBUGP("sack_adjust: start_seq: %d->%d, end_seq: %d->%d\n",
295                         ntohl(sack->start_seq), new_start_seq,
296                         ntohl(sack->end_seq), new_end_seq);
297
298                 tcph->check = 
299                         ip_nat_cheat_check(~sack->start_seq, new_start_seq,
300                                            ip_nat_cheat_check(~sack->end_seq, 
301                                                               new_end_seq,
302                                                               tcph->check));
303                 sack->start_seq = new_start_seq;
304                 sack->end_seq = new_end_seq;
305                 sackoff += sizeof(*sack);
306         }
307 }
308
309 /* TCP SACK sequence number adjustment */
310 static inline unsigned int
311 ip_nat_sack_adjust(struct sk_buff **pskb,
312                    struct tcphdr *tcph,
313                    struct ip_conntrack *ct,
314                    enum ip_conntrack_info ctinfo)
315 {
316         unsigned int dir, optoff, optend;
317
318         optoff = (*pskb)->nh.iph->ihl*4 + sizeof(struct tcphdr);
319         optend = (*pskb)->nh.iph->ihl*4 + tcph->doff*4;
320
321         if (!skb_ip_make_writable(pskb, optend))
322                 return 0;
323
324         dir = CTINFO2DIR(ctinfo);
325
326         while (optoff < optend) {
327                 /* Usually: option, length. */
328                 unsigned char *op = (*pskb)->data + optoff;
329
330                 switch (op[0]) {
331                 case TCPOPT_EOL:
332                         return 1;
333                 case TCPOPT_NOP:
334                         optoff++;
335                         continue;
336                 default:
337                         /* no partial options */
338                         if (optoff + 1 == optend
339                             || optoff + op[1] > optend
340                             || op[1] < 2)
341                                 return 0;
342                         if (op[0] == TCPOPT_SACK
343                             && op[1] >= 2+TCPOLEN_SACK_PERBLOCK
344                             && ((op[1] - 2) % TCPOLEN_SACK_PERBLOCK) == 0)
345                                 sack_adjust(*pskb, tcph, optoff+2,
346                                             optoff+op[1],
347                                             &ct->nat.info.seq[!dir]);
348                         optoff += op[1];
349                 }
350         }
351         return 1;
352 }
353
354 /* TCP sequence number adjustment.  Returns 1 on success, 0 on failure */
355 int
356 ip_nat_seq_adjust(struct sk_buff **pskb, 
357                   struct ip_conntrack *ct, 
358                   enum ip_conntrack_info ctinfo)
359 {
360         struct tcphdr *tcph;
361         int dir, newseq, newack;
362         struct ip_nat_seq *this_way, *other_way;        
363
364         dir = CTINFO2DIR(ctinfo);
365
366         this_way = &ct->nat.info.seq[dir];
367         other_way = &ct->nat.info.seq[!dir];
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         if (!ip_nat_sack_adjust(pskb, tcph, ct, ctinfo))
399                 return 0;
400
401         ip_conntrack_tcp_update(*pskb, ct, dir);
402
403         return 1;
404 }
405
406 /* Setup NAT on this expected conntrack so it follows master. */
407 /* If we fail to get a free NAT slot, we'll get dropped on confirm */
408 void ip_nat_follow_master(struct ip_conntrack *ct,
409                           struct ip_conntrack_expect *exp)
410 {
411         struct ip_nat_range range;
412
413         /* This must be a fresh one. */
414         BUG_ON(ct->status & IPS_NAT_DONE_MASK);
415
416         /* Change src to where master sends to */
417         range.flags = IP_NAT_RANGE_MAP_IPS;
418         range.min_ip = range.max_ip
419                 = ct->master->tuplehash[!exp->dir].tuple.dst.ip;
420         /* hook doesn't matter, but it has to do source manip */
421         ip_nat_setup_info(ct, &range, NF_IP_POST_ROUTING);
422
423         /* For DST manip, map port here to where it's expected. */
424         range.flags = (IP_NAT_RANGE_MAP_IPS | IP_NAT_RANGE_PROTO_SPECIFIED);
425         range.min = range.max = exp->saved_proto;
426         range.min_ip = range.max_ip
427                 = ct->master->tuplehash[!exp->dir].tuple.src.ip;
428         /* hook doesn't matter, but it has to do destination manip */
429         ip_nat_setup_info(ct, &range, NF_IP_PRE_ROUTING);
430 }