Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/kaber/nf-next-2.6
[linux-2.6.git] / net / ipv4 / netfilter / nf_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) 2003-2006 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 #include <linux/module.h>
11 #include <linux/kmod.h>
12 #include <linux/types.h>
13 #include <linux/timer.h>
14 #include <linux/skbuff.h>
15 #include <linux/tcp.h>
16 #include <linux/udp.h>
17 #include <net/checksum.h>
18 #include <net/tcp.h>
19 #include <net/route.h>
20
21 #include <linux/netfilter_ipv4.h>
22 #include <net/netfilter/nf_conntrack.h>
23 #include <net/netfilter/nf_conntrack_helper.h>
24 #include <net/netfilter/nf_conntrack_ecache.h>
25 #include <net/netfilter/nf_conntrack_expect.h>
26 #include <net/netfilter/nf_nat.h>
27 #include <net/netfilter/nf_nat_protocol.h>
28 #include <net/netfilter/nf_nat_core.h>
29 #include <net/netfilter/nf_nat_helper.h>
30
31 #define DUMP_OFFSET(x) \
32         pr_debug("offset_before=%d, offset_after=%d, correction_pos=%u\n", \
33                  x->offset_before, x->offset_after, x->correction_pos);
34
35 static DEFINE_SPINLOCK(nf_nat_seqofs_lock);
36
37 /* Setup TCP sequence correction given this change at this sequence */
38 static inline void
39 adjust_tcp_sequence(u32 seq,
40                     int sizediff,
41                     struct nf_conn *ct,
42                     enum ip_conntrack_info ctinfo)
43 {
44         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
45         struct nf_conn_nat *nat = nfct_nat(ct);
46         struct nf_nat_seq *this_way = &nat->seq[dir];
47
48         pr_debug("adjust_tcp_sequence: seq = %u, sizediff = %d\n",
49                  seq, sizediff);
50
51         pr_debug("adjust_tcp_sequence: Seq_offset before: ");
52         DUMP_OFFSET(this_way);
53
54         spin_lock_bh(&nf_nat_seqofs_lock);
55
56         /* SYN adjust. If it's uninitialized, or this is after last
57          * correction, record it: we don't handle more than one
58          * adjustment in the window, but do deal with common case of a
59          * retransmit */
60         if (this_way->offset_before == this_way->offset_after ||
61             before(this_way->correction_pos, seq)) {
62                 this_way->correction_pos = seq;
63                 this_way->offset_before = this_way->offset_after;
64                 this_way->offset_after += sizediff;
65         }
66         spin_unlock_bh(&nf_nat_seqofs_lock);
67
68         pr_debug("adjust_tcp_sequence: Seq_offset after: ");
69         DUMP_OFFSET(this_way);
70 }
71
72 /* Get the offset value, for conntrack */
73 s16 nf_nat_get_offset(const struct nf_conn *ct,
74                       enum ip_conntrack_dir dir,
75                       u32 seq)
76 {
77         struct nf_conn_nat *nat = nfct_nat(ct);
78         struct nf_nat_seq *this_way;
79         s16 offset;
80
81         if (!nat)
82                 return 0;
83
84         this_way = &nat->seq[dir];
85         spin_lock_bh(&nf_nat_seqofs_lock);
86         offset = after(seq, this_way->correction_pos)
87                  ? this_way->offset_after : this_way->offset_before;
88         spin_unlock_bh(&nf_nat_seqofs_lock);
89
90         return offset;
91 }
92 EXPORT_SYMBOL_GPL(nf_nat_get_offset);
93
94 /* Frobs data inside this packet, which is linear. */
95 static void mangle_contents(struct sk_buff *skb,
96                             unsigned int dataoff,
97                             unsigned int match_offset,
98                             unsigned int match_len,
99                             const char *rep_buffer,
100                             unsigned int rep_len)
101 {
102         unsigned char *data;
103
104         BUG_ON(skb_is_nonlinear(skb));
105         data = skb_network_header(skb) + dataoff;
106
107         /* move post-replacement */
108         memmove(data + match_offset + rep_len,
109                 data + match_offset + match_len,
110                 skb->tail - (skb->network_header + dataoff +
111                              match_offset + match_len));
112
113         /* insert data from buffer */
114         memcpy(data + match_offset, rep_buffer, rep_len);
115
116         /* update skb info */
117         if (rep_len > match_len) {
118                 pr_debug("nf_nat_mangle_packet: Extending packet by "
119                          "%u from %u bytes\n", rep_len - match_len, skb->len);
120                 skb_put(skb, rep_len - match_len);
121         } else {
122                 pr_debug("nf_nat_mangle_packet: Shrinking packet from "
123                          "%u from %u bytes\n", match_len - rep_len, skb->len);
124                 __skb_trim(skb, skb->len + rep_len - match_len);
125         }
126
127         /* fix IP hdr checksum information */
128         ip_hdr(skb)->tot_len = htons(skb->len);
129         ip_send_check(ip_hdr(skb));
130 }
131
132 /* Unusual, but possible case. */
133 static int enlarge_skb(struct sk_buff *skb, unsigned int extra)
134 {
135         if (skb->len + extra > 65535)
136                 return 0;
137
138         if (pskb_expand_head(skb, 0, extra - skb_tailroom(skb), GFP_ATOMIC))
139                 return 0;
140
141         return 1;
142 }
143
144 /* Generic function for mangling variable-length address changes inside
145  * NATed TCP connections (like the PORT XXX,XXX,XXX,XXX,XXX,XXX
146  * command in FTP).
147  *
148  * Takes care about all the nasty sequence number changes, checksumming,
149  * skb enlargement, ...
150  *
151  * */
152 int
153 nf_nat_mangle_tcp_packet(struct sk_buff *skb,
154                          struct nf_conn *ct,
155                          enum ip_conntrack_info ctinfo,
156                          unsigned int match_offset,
157                          unsigned int match_len,
158                          const char *rep_buffer,
159                          unsigned int rep_len)
160 {
161         struct rtable *rt = skb_rtable(skb);
162         struct iphdr *iph;
163         struct tcphdr *tcph;
164         int oldlen, datalen;
165
166         if (!skb_make_writable(skb, skb->len))
167                 return 0;
168
169         if (rep_len > match_len &&
170             rep_len - match_len > skb_tailroom(skb) &&
171             !enlarge_skb(skb, rep_len - match_len))
172                 return 0;
173
174         SKB_LINEAR_ASSERT(skb);
175
176         iph = ip_hdr(skb);
177         tcph = (void *)iph + iph->ihl*4;
178
179         oldlen = skb->len - iph->ihl*4;
180         mangle_contents(skb, iph->ihl*4 + tcph->doff*4,
181                         match_offset, match_len, rep_buffer, rep_len);
182
183         datalen = skb->len - iph->ihl*4;
184         if (skb->ip_summed != CHECKSUM_PARTIAL) {
185                 if (!(rt->rt_flags & RTCF_LOCAL) &&
186                     skb->dev->features & NETIF_F_V4_CSUM) {
187                         skb->ip_summed = CHECKSUM_PARTIAL;
188                         skb->csum_start = skb_headroom(skb) +
189                                           skb_network_offset(skb) +
190                                           iph->ihl * 4;
191                         skb->csum_offset = offsetof(struct tcphdr, check);
192                         tcph->check = ~tcp_v4_check(datalen,
193                                                     iph->saddr, iph->daddr, 0);
194                 } else {
195                         tcph->check = 0;
196                         tcph->check = tcp_v4_check(datalen,
197                                                    iph->saddr, iph->daddr,
198                                                    csum_partial(tcph,
199                                                                 datalen, 0));
200                 }
201         } else
202                 inet_proto_csum_replace2(&tcph->check, skb,
203                                          htons(oldlen), htons(datalen), 1);
204
205         if (rep_len != match_len) {
206                 set_bit(IPS_SEQ_ADJUST_BIT, &ct->status);
207                 adjust_tcp_sequence(ntohl(tcph->seq),
208                                     (int)rep_len - (int)match_len,
209                                     ct, ctinfo);
210                 nf_conntrack_event_cache(IPCT_NATSEQADJ, ct);
211         }
212         return 1;
213 }
214 EXPORT_SYMBOL(nf_nat_mangle_tcp_packet);
215
216 /* Generic function for mangling variable-length address changes inside
217  * NATed UDP connections (like the CONNECT DATA XXXXX MESG XXXXX INDEX XXXXX
218  * command in the Amanda protocol)
219  *
220  * Takes care about all the nasty sequence number changes, checksumming,
221  * skb enlargement, ...
222  *
223  * XXX - This function could be merged with nf_nat_mangle_tcp_packet which
224  *       should be fairly easy to do.
225  */
226 int
227 nf_nat_mangle_udp_packet(struct sk_buff *skb,
228                          struct nf_conn *ct,
229                          enum ip_conntrack_info ctinfo,
230                          unsigned int match_offset,
231                          unsigned int match_len,
232                          const char *rep_buffer,
233                          unsigned int rep_len)
234 {
235         struct rtable *rt = skb_rtable(skb);
236         struct iphdr *iph;
237         struct udphdr *udph;
238         int datalen, oldlen;
239
240         /* UDP helpers might accidentally mangle the wrong packet */
241         iph = ip_hdr(skb);
242         if (skb->len < iph->ihl*4 + sizeof(*udph) +
243                                match_offset + match_len)
244                 return 0;
245
246         if (!skb_make_writable(skb, skb->len))
247                 return 0;
248
249         if (rep_len > match_len &&
250             rep_len - match_len > skb_tailroom(skb) &&
251             !enlarge_skb(skb, rep_len - match_len))
252                 return 0;
253
254         iph = ip_hdr(skb);
255         udph = (void *)iph + iph->ihl*4;
256
257         oldlen = skb->len - iph->ihl*4;
258         mangle_contents(skb, iph->ihl*4 + sizeof(*udph),
259                         match_offset, match_len, rep_buffer, rep_len);
260
261         /* update the length of the UDP packet */
262         datalen = skb->len - iph->ihl*4;
263         udph->len = htons(datalen);
264
265         /* fix udp checksum if udp checksum was previously calculated */
266         if (!udph->check && skb->ip_summed != CHECKSUM_PARTIAL)
267                 return 1;
268
269         if (skb->ip_summed != CHECKSUM_PARTIAL) {
270                 if (!(rt->rt_flags & RTCF_LOCAL) &&
271                     skb->dev->features & NETIF_F_V4_CSUM) {
272                         skb->ip_summed = CHECKSUM_PARTIAL;
273                         skb->csum_start = skb_headroom(skb) +
274                                           skb_network_offset(skb) +
275                                           iph->ihl * 4;
276                         skb->csum_offset = offsetof(struct udphdr, check);
277                         udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
278                                                          datalen, IPPROTO_UDP,
279                                                          0);
280                 } else {
281                         udph->check = 0;
282                         udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
283                                                         datalen, IPPROTO_UDP,
284                                                         csum_partial(udph,
285                                                                      datalen, 0));
286                         if (!udph->check)
287                                 udph->check = CSUM_MANGLED_0;
288                 }
289         } else
290                 inet_proto_csum_replace2(&udph->check, skb,
291                                          htons(oldlen), htons(datalen), 1);
292
293         return 1;
294 }
295 EXPORT_SYMBOL(nf_nat_mangle_udp_packet);
296
297 /* Adjust one found SACK option including checksum correction */
298 static void
299 sack_adjust(struct sk_buff *skb,
300             struct tcphdr *tcph,
301             unsigned int sackoff,
302             unsigned int sackend,
303             struct nf_nat_seq *natseq)
304 {
305         while (sackoff < sackend) {
306                 struct tcp_sack_block_wire *sack;
307                 __be32 new_start_seq, new_end_seq;
308
309                 sack = (void *)skb->data + sackoff;
310                 if (after(ntohl(sack->start_seq) - natseq->offset_before,
311                           natseq->correction_pos))
312                         new_start_seq = htonl(ntohl(sack->start_seq)
313                                         - natseq->offset_after);
314                 else
315                         new_start_seq = htonl(ntohl(sack->start_seq)
316                                         - natseq->offset_before);
317
318                 if (after(ntohl(sack->end_seq) - natseq->offset_before,
319                           natseq->correction_pos))
320                         new_end_seq = htonl(ntohl(sack->end_seq)
321                                       - natseq->offset_after);
322                 else
323                         new_end_seq = htonl(ntohl(sack->end_seq)
324                                       - natseq->offset_before);
325
326                 pr_debug("sack_adjust: start_seq: %d->%d, end_seq: %d->%d\n",
327                          ntohl(sack->start_seq), new_start_seq,
328                          ntohl(sack->end_seq), new_end_seq);
329
330                 inet_proto_csum_replace4(&tcph->check, skb,
331                                          sack->start_seq, new_start_seq, 0);
332                 inet_proto_csum_replace4(&tcph->check, skb,
333                                          sack->end_seq, new_end_seq, 0);
334                 sack->start_seq = new_start_seq;
335                 sack->end_seq = new_end_seq;
336                 sackoff += sizeof(*sack);
337         }
338 }
339
340 /* TCP SACK sequence number adjustment */
341 static inline unsigned int
342 nf_nat_sack_adjust(struct sk_buff *skb,
343                    struct tcphdr *tcph,
344                    struct nf_conn *ct,
345                    enum ip_conntrack_info ctinfo)
346 {
347         unsigned int dir, optoff, optend;
348         struct nf_conn_nat *nat = nfct_nat(ct);
349
350         optoff = ip_hdrlen(skb) + sizeof(struct tcphdr);
351         optend = ip_hdrlen(skb) + tcph->doff * 4;
352
353         if (!skb_make_writable(skb, optend))
354                 return 0;
355
356         dir = CTINFO2DIR(ctinfo);
357
358         while (optoff < optend) {
359                 /* Usually: option, length. */
360                 unsigned char *op = skb->data + optoff;
361
362                 switch (op[0]) {
363                 case TCPOPT_EOL:
364                         return 1;
365                 case TCPOPT_NOP:
366                         optoff++;
367                         continue;
368                 default:
369                         /* no partial options */
370                         if (optoff + 1 == optend ||
371                             optoff + op[1] > optend ||
372                             op[1] < 2)
373                                 return 0;
374                         if (op[0] == TCPOPT_SACK &&
375                             op[1] >= 2+TCPOLEN_SACK_PERBLOCK &&
376                             ((op[1] - 2) % TCPOLEN_SACK_PERBLOCK) == 0)
377                                 sack_adjust(skb, tcph, optoff+2,
378                                             optoff+op[1], &nat->seq[!dir]);
379                         optoff += op[1];
380                 }
381         }
382         return 1;
383 }
384
385 /* TCP sequence number adjustment.  Returns 1 on success, 0 on failure */
386 int
387 nf_nat_seq_adjust(struct sk_buff *skb,
388                   struct nf_conn *ct,
389                   enum ip_conntrack_info ctinfo)
390 {
391         struct tcphdr *tcph;
392         int dir;
393         __be32 newseq, newack;
394         s16 seqoff, ackoff;
395         struct nf_conn_nat *nat = nfct_nat(ct);
396         struct nf_nat_seq *this_way, *other_way;
397
398         dir = CTINFO2DIR(ctinfo);
399
400         this_way = &nat->seq[dir];
401         other_way = &nat->seq[!dir];
402
403         if (!skb_make_writable(skb, ip_hdrlen(skb) + sizeof(*tcph)))
404                 return 0;
405
406         tcph = (void *)skb->data + ip_hdrlen(skb);
407         if (after(ntohl(tcph->seq), this_way->correction_pos))
408                 seqoff = this_way->offset_after;
409         else
410                 seqoff = this_way->offset_before;
411
412         if (after(ntohl(tcph->ack_seq) - other_way->offset_before,
413                   other_way->correction_pos))
414                 ackoff = other_way->offset_after;
415         else
416                 ackoff = other_way->offset_before;
417
418         newseq = htonl(ntohl(tcph->seq) + seqoff);
419         newack = htonl(ntohl(tcph->ack_seq) - ackoff);
420
421         inet_proto_csum_replace4(&tcph->check, skb, tcph->seq, newseq, 0);
422         inet_proto_csum_replace4(&tcph->check, skb, tcph->ack_seq, newack, 0);
423
424         pr_debug("Adjusting sequence number from %u->%u, ack from %u->%u\n",
425                  ntohl(tcph->seq), ntohl(newseq), ntohl(tcph->ack_seq),
426                  ntohl(newack));
427
428         tcph->seq = newseq;
429         tcph->ack_seq = newack;
430
431         return nf_nat_sack_adjust(skb, tcph, ct, ctinfo);
432 }
433
434 /* Setup NAT on this expected conntrack so it follows master. */
435 /* If we fail to get a free NAT slot, we'll get dropped on confirm */
436 void nf_nat_follow_master(struct nf_conn *ct,
437                           struct nf_conntrack_expect *exp)
438 {
439         struct nf_nat_range range;
440
441         /* This must be a fresh one. */
442         BUG_ON(ct->status & IPS_NAT_DONE_MASK);
443
444         /* Change src to where master sends to */
445         range.flags = IP_NAT_RANGE_MAP_IPS;
446         range.min_ip = range.max_ip
447                 = ct->master->tuplehash[!exp->dir].tuple.dst.u3.ip;
448         nf_nat_setup_info(ct, &range, IP_NAT_MANIP_SRC);
449
450         /* For DST manip, map port here to where it's expected. */
451         range.flags = (IP_NAT_RANGE_MAP_IPS | IP_NAT_RANGE_PROTO_SPECIFIED);
452         range.min = range.max = exp->saved_proto;
453         range.min_ip = range.max_ip
454                 = ct->master->tuplehash[!exp->dir].tuple.src.u3.ip;
455         nf_nat_setup_info(ct, &range, IP_NAT_MANIP_DST);
456 }
457 EXPORT_SYMBOL(nf_nat_follow_master);