dc778cfef58b43b37ac523f0257ad593c6b09dae
[linux-3.10.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/module.h>
19 #include <linux/kmod.h>
20 #include <linux/types.h>
21 #include <linux/timer.h>
22 #include <linux/skbuff.h>
23 #include <linux/netfilter_ipv4.h>
24 #include <net/checksum.h>
25 #include <net/icmp.h>
26 #include <net/ip.h>
27 #include <net/tcp.h>
28 #include <net/udp.h>
29
30 #include <linux/netfilter_ipv4/ip_conntrack.h>
31 #include <linux/netfilter_ipv4/ip_conntrack_helper.h>
32 #include <linux/netfilter_ipv4/ip_nat.h>
33 #include <linux/netfilter_ipv4/ip_nat_protocol.h>
34 #include <linux/netfilter_ipv4/ip_nat_core.h>
35 #include <linux/netfilter_ipv4/ip_nat_helper.h>
36
37 #if 0
38 #define DEBUGP printk
39 #define DUMP_OFFSET(x)  printk("offset_before=%d, offset_after=%d, correction_pos=%u\n", x->offset_before, x->offset_after, x->correction_pos);
40 #else
41 #define DEBUGP(format, args...)
42 #define DUMP_OFFSET(x)
43 #endif
44
45 static DEFINE_SPINLOCK(ip_nat_seqofs_lock);
46
47 /* Setup TCP sequence correction given this change at this sequence */
48 static inline void
49 adjust_tcp_sequence(u32 seq,
50                     int sizediff,
51                     struct ip_conntrack *ct,
52                     enum ip_conntrack_info ctinfo)
53 {
54         int dir;
55         struct ip_nat_seq *this_way, *other_way;
56
57         DEBUGP("ip_nat_resize_packet: old_size = %u, new_size = %u\n",
58                 (*skb)->len, new_size);
59
60         dir = CTINFO2DIR(ctinfo);
61
62         this_way = &ct->nat.info.seq[dir];
63         other_way = &ct->nat.info.seq[!dir];
64
65         DEBUGP("ip_nat_resize_packet: Seq_offset before: ");
66         DUMP_OFFSET(this_way);
67
68         spin_lock_bh(&ip_nat_seqofs_lock);
69
70         /* SYN adjust. If it's uninitialized, or this is after last
71          * correction, record it: we don't handle more than one
72          * adjustment in the window, but do deal with common case of a
73          * retransmit */
74         if (this_way->offset_before == this_way->offset_after
75             || before(this_way->correction_pos, seq)) {
76                     this_way->correction_pos = seq;
77                     this_way->offset_before = this_way->offset_after;
78                     this_way->offset_after += sizediff;
79         }
80         spin_unlock_bh(&ip_nat_seqofs_lock);
81
82         DEBUGP("ip_nat_resize_packet: Seq_offset after: ");
83         DUMP_OFFSET(this_way);
84 }
85
86 /* Frobs data inside this packet, which is linear. */
87 static void mangle_contents(struct sk_buff *skb,
88                             unsigned int dataoff,
89                             unsigned int match_offset,
90                             unsigned int match_len,
91                             const char *rep_buffer,
92                             unsigned int rep_len)
93 {
94         unsigned char *data;
95
96         BUG_ON(skb_is_nonlinear(skb));
97         data = (unsigned char *)skb->nh.iph + dataoff;
98
99         /* move post-replacement */
100         memmove(data + match_offset + rep_len,
101                 data + match_offset + match_len,
102                 skb->tail - (data + match_offset + match_len));
103
104         /* insert data from buffer */
105         memcpy(data + match_offset, rep_buffer, rep_len);
106
107         /* update skb info */
108         if (rep_len > match_len) {
109                 DEBUGP("ip_nat_mangle_packet: Extending packet by "
110                         "%u from %u bytes\n", rep_len - match_len,
111                        skb->len);
112                 skb_put(skb, rep_len - match_len);
113         } else {
114                 DEBUGP("ip_nat_mangle_packet: Shrinking packet from "
115                         "%u from %u bytes\n", match_len - rep_len,
116                        skb->len);
117                 __skb_trim(skb, skb->len + rep_len - match_len);
118         }
119
120         /* fix IP hdr checksum information */
121         skb->nh.iph->tot_len = htons(skb->len);
122         ip_send_check(skb->nh.iph);
123 }
124
125 /* Unusual, but possible case. */
126 static int enlarge_skb(struct sk_buff **pskb, unsigned int extra)
127 {
128         struct sk_buff *nskb;
129
130         if ((*pskb)->len + extra > 65535)
131                 return 0;
132
133         nskb = skb_copy_expand(*pskb, skb_headroom(*pskb), extra, GFP_ATOMIC);
134         if (!nskb)
135                 return 0;
136
137         /* Transfer socket to new skb. */
138         if ((*pskb)->sk)
139                 skb_set_owner_w(nskb, (*pskb)->sk);
140         kfree_skb(*pskb);
141         *pskb = nskb;
142         return 1;
143 }
144
145 /* Generic function for mangling variable-length address changes inside
146  * NATed TCP connections (like the PORT XXX,XXX,XXX,XXX,XXX,XXX
147  * command in FTP).
148  *
149  * Takes care about all the nasty sequence number changes, checksumming,
150  * skb enlargement, ...
151  *
152  * */
153 int
154 ip_nat_mangle_tcp_packet(struct sk_buff **pskb,
155                          struct ip_conntrack *ct,
156                          enum ip_conntrack_info ctinfo,
157                          unsigned int match_offset,
158                          unsigned int match_len,
159                          const char *rep_buffer,
160                          unsigned int rep_len)
161 {
162         struct iphdr *iph;
163         struct tcphdr *tcph;
164         int oldlen, datalen;
165
166         if (!skb_make_writable(pskb, (*pskb)->len))
167                 return 0;
168
169         if (rep_len > match_len
170             && rep_len - match_len > skb_tailroom(*pskb)
171             && !enlarge_skb(pskb, rep_len - match_len))
172                 return 0;
173
174         SKB_LINEAR_ASSERT(*pskb);
175
176         iph = (*pskb)->nh.iph;
177         tcph = (void *)iph + iph->ihl*4;
178
179         oldlen = (*pskb)->len - iph->ihl*4;
180         mangle_contents(*pskb, iph->ihl*4 + tcph->doff*4,
181                         match_offset, match_len, rep_buffer, rep_len);
182
183         datalen = (*pskb)->len - iph->ihl*4;
184         if ((*pskb)->ip_summed != CHECKSUM_PARTIAL) {
185                 tcph->check = 0;
186                 tcph->check = tcp_v4_check(datalen,
187                                            iph->saddr, iph->daddr,
188                                            csum_partial((char *)tcph,
189                                                         datalen, 0));
190         } else
191                 nf_proto_csum_replace2(&tcph->check, *pskb,
192                                         htons(oldlen), htons(datalen), 1);
193
194         if (rep_len != match_len) {
195                 set_bit(IPS_SEQ_ADJUST_BIT, &ct->status);
196                 adjust_tcp_sequence(ntohl(tcph->seq),
197                                     (int)rep_len - (int)match_len,
198                                     ct, ctinfo);
199                 /* Tell TCP window tracking about seq change */
200                 ip_conntrack_tcp_update(*pskb, ct, CTINFO2DIR(ctinfo));
201         }
202         return 1;
203 }
204 EXPORT_SYMBOL(ip_nat_mangle_tcp_packet);
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         int datalen, oldlen;
228
229         /* UDP helpers might accidentally mangle the wrong packet */
230         iph = (*pskb)->nh.iph;
231         if ((*pskb)->len < iph->ihl*4 + sizeof(*udph) +
232                                match_offset + match_len)
233                 return 0;
234
235         if (!skb_make_writable(pskb, (*pskb)->len))
236                 return 0;
237
238         if (rep_len > match_len
239             && rep_len - match_len > skb_tailroom(*pskb)
240             && !enlarge_skb(pskb, rep_len - match_len))
241                 return 0;
242
243         iph = (*pskb)->nh.iph;
244         udph = (void *)iph + iph->ihl*4;
245
246         oldlen = (*pskb)->len - iph->ihl*4;
247         mangle_contents(*pskb, iph->ihl*4 + sizeof(*udph),
248                         match_offset, match_len, rep_buffer, rep_len);
249
250         /* update the length of the UDP packet */
251         datalen = (*pskb)->len - iph->ihl*4;
252         udph->len = htons(datalen);
253
254         /* fix udp checksum if udp checksum was previously calculated */
255         if (!udph->check && (*pskb)->ip_summed != CHECKSUM_PARTIAL)
256                 return 1;
257
258         if ((*pskb)->ip_summed != CHECKSUM_PARTIAL) {
259                 udph->check = 0;
260                 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
261                                                 datalen, IPPROTO_UDP,
262                                                 csum_partial((char *)udph,
263                                                              datalen, 0));
264                 if (!udph->check)
265                         udph->check = CSUM_MANGLED_0;
266         } else
267                 nf_proto_csum_replace2(&udph->check, *pskb,
268                                         htons(oldlen), htons(datalen), 1);
269         return 1;
270 }
271 EXPORT_SYMBOL(ip_nat_mangle_udp_packet);
272
273 /* Adjust one found SACK option including checksum correction */
274 static void
275 sack_adjust(struct sk_buff *skb,
276             struct tcphdr *tcph,
277             unsigned int sackoff,
278             unsigned int sackend,
279             struct ip_nat_seq *natseq)
280 {
281         while (sackoff < sackend) {
282                 struct tcp_sack_block_wire *sack;
283                 __be32 new_start_seq, new_end_seq;
284
285                 sack = (void *)skb->data + sackoff;
286                 if (after(ntohl(sack->start_seq) - natseq->offset_before,
287                           natseq->correction_pos))
288                         new_start_seq = htonl(ntohl(sack->start_seq)
289                                         - natseq->offset_after);
290                 else
291                         new_start_seq = htonl(ntohl(sack->start_seq)
292                                         - natseq->offset_before);
293
294                 if (after(ntohl(sack->end_seq) - natseq->offset_before,
295                           natseq->correction_pos))
296                         new_end_seq = htonl(ntohl(sack->end_seq)
297                                       - natseq->offset_after);
298                 else
299                         new_end_seq = htonl(ntohl(sack->end_seq)
300                                       - natseq->offset_before);
301
302                 DEBUGP("sack_adjust: start_seq: %d->%d, end_seq: %d->%d\n",
303                         ntohl(sack->start_seq), new_start_seq,
304                         ntohl(sack->end_seq), new_end_seq);
305
306                 nf_proto_csum_replace4(&tcph->check, skb,
307                                         sack->start_seq, new_start_seq, 0);
308                 nf_proto_csum_replace4(&tcph->check, skb,
309                                         sack->end_seq, new_end_seq, 0);
310                 sack->start_seq = new_start_seq;
311                 sack->end_seq = new_end_seq;
312                 sackoff += sizeof(*sack);
313         }
314 }
315
316 /* TCP SACK sequence number adjustment */
317 static inline unsigned int
318 ip_nat_sack_adjust(struct sk_buff **pskb,
319                    struct tcphdr *tcph,
320                    struct ip_conntrack *ct,
321                    enum ip_conntrack_info ctinfo)
322 {
323         unsigned int dir, optoff, optend;
324
325         optoff = (*pskb)->nh.iph->ihl*4 + sizeof(struct tcphdr);
326         optend = (*pskb)->nh.iph->ihl*4 + tcph->doff*4;
327
328         if (!skb_make_writable(pskb, optend))
329                 return 0;
330
331         dir = CTINFO2DIR(ctinfo);
332
333         while (optoff < optend) {
334                 /* Usually: option, length. */
335                 unsigned char *op = (*pskb)->data + optoff;
336
337                 switch (op[0]) {
338                 case TCPOPT_EOL:
339                         return 1;
340                 case TCPOPT_NOP:
341                         optoff++;
342                         continue;
343                 default:
344                         /* no partial options */
345                         if (optoff + 1 == optend
346                             || optoff + op[1] > optend
347                             || op[1] < 2)
348                                 return 0;
349                         if (op[0] == TCPOPT_SACK
350                             && op[1] >= 2+TCPOLEN_SACK_PERBLOCK
351                             && ((op[1] - 2) % TCPOLEN_SACK_PERBLOCK) == 0)
352                                 sack_adjust(*pskb, tcph, optoff+2,
353                                             optoff+op[1],
354                                             &ct->nat.info.seq[!dir]);
355                         optoff += op[1];
356                 }
357         }
358         return 1;
359 }
360
361 /* TCP sequence number adjustment.  Returns 1 on success, 0 on failure */
362 int
363 ip_nat_seq_adjust(struct sk_buff **pskb,
364                   struct ip_conntrack *ct,
365                   enum ip_conntrack_info ctinfo)
366 {
367         struct tcphdr *tcph;
368         int dir;
369         __be32 newseq, newack;
370         struct ip_nat_seq *this_way, *other_way;
371
372         dir = CTINFO2DIR(ctinfo);
373
374         this_way = &ct->nat.info.seq[dir];
375         other_way = &ct->nat.info.seq[!dir];
376
377         if (!skb_make_writable(pskb, (*pskb)->nh.iph->ihl*4+sizeof(*tcph)))
378                 return 0;
379
380         tcph = (void *)(*pskb)->data + (*pskb)->nh.iph->ihl*4;
381         if (after(ntohl(tcph->seq), this_way->correction_pos))
382                 newseq = htonl(ntohl(tcph->seq) + this_way->offset_after);
383         else
384                 newseq = htonl(ntohl(tcph->seq) + this_way->offset_before);
385
386         if (after(ntohl(tcph->ack_seq) - other_way->offset_before,
387                   other_way->correction_pos))
388                 newack = htonl(ntohl(tcph->ack_seq) - other_way->offset_after);
389         else
390                 newack = htonl(ntohl(tcph->ack_seq) - other_way->offset_before);
391
392         nf_proto_csum_replace4(&tcph->check, *pskb, tcph->seq, newseq, 0);
393         nf_proto_csum_replace4(&tcph->check, *pskb, tcph->ack_seq, newack, 0);
394
395         DEBUGP("Adjusting sequence number from %u->%u, ack from %u->%u\n",
396                 ntohl(tcph->seq), ntohl(newseq), ntohl(tcph->ack_seq),
397                 ntohl(newack));
398
399         tcph->seq = newseq;
400         tcph->ack_seq = newack;
401
402         if (!ip_nat_sack_adjust(pskb, tcph, ct, ctinfo))
403                 return 0;
404
405         ip_conntrack_tcp_update(*pskb, ct, dir);
406
407         return 1;
408 }
409 EXPORT_SYMBOL(ip_nat_seq_adjust);
410
411 /* Setup NAT on this expected conntrack so it follows master. */
412 /* If we fail to get a free NAT slot, we'll get dropped on confirm */
413 void ip_nat_follow_master(struct ip_conntrack *ct,
414                           struct ip_conntrack_expect *exp)
415 {
416         struct ip_nat_range range;
417
418         /* This must be a fresh one. */
419         BUG_ON(ct->status & IPS_NAT_DONE_MASK);
420
421         /* Change src to where master sends to */
422         range.flags = IP_NAT_RANGE_MAP_IPS;
423         range.min_ip = range.max_ip
424                 = ct->master->tuplehash[!exp->dir].tuple.dst.ip;
425         /* hook doesn't matter, but it has to do source manip */
426         ip_nat_setup_info(ct, &range, NF_IP_POST_ROUTING);
427
428         /* For DST manip, map port here to where it's expected. */
429         range.flags = (IP_NAT_RANGE_MAP_IPS | IP_NAT_RANGE_PROTO_SPECIFIED);
430         range.min = range.max = exp->saved_proto;
431         range.min_ip = range.max_ip
432                 = ct->master->tuplehash[!exp->dir].tuple.src.ip;
433         /* hook doesn't matter, but it has to do destination manip */
434         ip_nat_setup_info(ct, &range, NF_IP_PRE_ROUTING);
435 }
436 EXPORT_SYMBOL(ip_nat_follow_master);