[NETFILTER]: ctnetlink: use netlink attribute helpers
[linux-2.6.git] / net / ipv4 / netfilter / nf_nat_core.c
1 /* NAT for netfilter; shared with compatibility layer. */
2
3 /* (C) 1999-2001 Paul `Rusty' Russell
4  * (C) 2002-2006 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
11 #include <linux/module.h>
12 #include <linux/types.h>
13 #include <linux/timer.h>
14 #include <linux/skbuff.h>
15 #include <net/checksum.h>
16 #include <net/icmp.h>
17 #include <net/ip.h>
18 #include <net/tcp.h>  /* For tcp_prot in getorigdst */
19 #include <linux/icmp.h>
20 #include <linux/udp.h>
21 #include <linux/jhash.h>
22
23 #include <linux/netfilter_ipv4.h>
24 #include <net/netfilter/nf_conntrack.h>
25 #include <net/netfilter/nf_conntrack_core.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 #include <net/netfilter/nf_conntrack_helper.h>
31 #include <net/netfilter/nf_conntrack_l3proto.h>
32 #include <net/netfilter/nf_conntrack_l4proto.h>
33
34 static DEFINE_RWLOCK(nf_nat_lock);
35
36 static struct nf_conntrack_l3proto *l3proto = NULL;
37
38 /* Calculated at init based on memory size */
39 static unsigned int nf_nat_htable_size;
40 static int nf_nat_vmalloced;
41
42 static struct hlist_head *bysource;
43
44 #define MAX_IP_NAT_PROTO 256
45 static struct nf_nat_protocol *nf_nat_protos[MAX_IP_NAT_PROTO];
46
47 static inline struct nf_nat_protocol *
48 __nf_nat_proto_find(u_int8_t protonum)
49 {
50         return rcu_dereference(nf_nat_protos[protonum]);
51 }
52
53 struct nf_nat_protocol *
54 nf_nat_proto_find_get(u_int8_t protonum)
55 {
56         struct nf_nat_protocol *p;
57
58         rcu_read_lock();
59         p = __nf_nat_proto_find(protonum);
60         if (!try_module_get(p->me))
61                 p = &nf_nat_unknown_protocol;
62         rcu_read_unlock();
63
64         return p;
65 }
66 EXPORT_SYMBOL_GPL(nf_nat_proto_find_get);
67
68 void
69 nf_nat_proto_put(struct nf_nat_protocol *p)
70 {
71         module_put(p->me);
72 }
73 EXPORT_SYMBOL_GPL(nf_nat_proto_put);
74
75 /* We keep an extra hash for each conntrack, for fast searching. */
76 static inline unsigned int
77 hash_by_src(const struct nf_conntrack_tuple *tuple)
78 {
79         /* Original src, to ensure we map it consistently if poss. */
80         return jhash_3words((__force u32)tuple->src.u3.ip,
81                             (__force u32)tuple->src.u.all,
82                             tuple->dst.protonum, 0) % nf_nat_htable_size;
83 }
84
85 /* Is this tuple already taken? (not by us) */
86 int
87 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
88                   const struct nf_conn *ignored_conntrack)
89 {
90         /* Conntrack tracking doesn't keep track of outgoing tuples; only
91            incoming ones.  NAT means they don't have a fixed mapping,
92            so we invert the tuple and look for the incoming reply.
93
94            We could keep a separate hash if this proves too slow. */
95         struct nf_conntrack_tuple reply;
96
97         nf_ct_invert_tuplepr(&reply, tuple);
98         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
99 }
100 EXPORT_SYMBOL(nf_nat_used_tuple);
101
102 /* If we source map this tuple so reply looks like reply_tuple, will
103  * that meet the constraints of range. */
104 static int
105 in_range(const struct nf_conntrack_tuple *tuple,
106          const struct nf_nat_range *range)
107 {
108         struct nf_nat_protocol *proto;
109         int ret = 0;
110
111         /* If we are supposed to map IPs, then we must be in the
112            range specified, otherwise let this drag us onto a new src IP. */
113         if (range->flags & IP_NAT_RANGE_MAP_IPS) {
114                 if (ntohl(tuple->src.u3.ip) < ntohl(range->min_ip) ||
115                     ntohl(tuple->src.u3.ip) > ntohl(range->max_ip))
116                         return 0;
117         }
118
119         rcu_read_lock();
120         proto = __nf_nat_proto_find(tuple->dst.protonum);
121         if (!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
122             proto->in_range(tuple, IP_NAT_MANIP_SRC,
123                             &range->min, &range->max))
124                 ret = 1;
125         rcu_read_unlock();
126
127         return ret;
128 }
129
130 static inline int
131 same_src(const struct nf_conn *ct,
132          const struct nf_conntrack_tuple *tuple)
133 {
134         const struct nf_conntrack_tuple *t;
135
136         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
137         return (t->dst.protonum == tuple->dst.protonum &&
138                 t->src.u3.ip == tuple->src.u3.ip &&
139                 t->src.u.all == tuple->src.u.all);
140 }
141
142 /* Only called for SRC manip */
143 static int
144 find_appropriate_src(const struct nf_conntrack_tuple *tuple,
145                      struct nf_conntrack_tuple *result,
146                      const struct nf_nat_range *range)
147 {
148         unsigned int h = hash_by_src(tuple);
149         struct nf_conn_nat *nat;
150         struct nf_conn *ct;
151         struct hlist_node *n;
152
153         read_lock_bh(&nf_nat_lock);
154         hlist_for_each_entry(nat, n, &bysource[h], bysource) {
155                 ct = nat->ct;
156                 if (same_src(ct, tuple)) {
157                         /* Copy source part from reply tuple. */
158                         nf_ct_invert_tuplepr(result,
159                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
160                         result->dst = tuple->dst;
161
162                         if (in_range(result, range)) {
163                                 read_unlock_bh(&nf_nat_lock);
164                                 return 1;
165                         }
166                 }
167         }
168         read_unlock_bh(&nf_nat_lock);
169         return 0;
170 }
171
172 /* For [FUTURE] fragmentation handling, we want the least-used
173    src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
174    if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
175    1-65535, we don't do pro-rata allocation based on ports; we choose
176    the ip with the lowest src-ip/dst-ip/proto usage.
177 */
178 static void
179 find_best_ips_proto(struct nf_conntrack_tuple *tuple,
180                     const struct nf_nat_range *range,
181                     const struct nf_conn *ct,
182                     enum nf_nat_manip_type maniptype)
183 {
184         __be32 *var_ipp;
185         /* Host order */
186         u_int32_t minip, maxip, j;
187
188         /* No IP mapping?  Do nothing. */
189         if (!(range->flags & IP_NAT_RANGE_MAP_IPS))
190                 return;
191
192         if (maniptype == IP_NAT_MANIP_SRC)
193                 var_ipp = &tuple->src.u3.ip;
194         else
195                 var_ipp = &tuple->dst.u3.ip;
196
197         /* Fast path: only one choice. */
198         if (range->min_ip == range->max_ip) {
199                 *var_ipp = range->min_ip;
200                 return;
201         }
202
203         /* Hashing source and destination IPs gives a fairly even
204          * spread in practice (if there are a small number of IPs
205          * involved, there usually aren't that many connections
206          * anyway).  The consistency means that servers see the same
207          * client coming from the same IP (some Internet Banking sites
208          * like this), even across reboots. */
209         minip = ntohl(range->min_ip);
210         maxip = ntohl(range->max_ip);
211         j = jhash_2words((__force u32)tuple->src.u3.ip,
212                          (__force u32)tuple->dst.u3.ip, 0);
213         *var_ipp = htonl(minip + j % (maxip - minip + 1));
214 }
215
216 /* Manipulate the tuple into the range given.  For NF_INET_POST_ROUTING,
217  * we change the source to map into the range.  For NF_INET_PRE_ROUTING
218  * and NF_INET_LOCAL_OUT, we change the destination to map into the
219  * range.  It might not be possible to get a unique tuple, but we try.
220  * At worst (or if we race), we will end up with a final duplicate in
221  * __ip_conntrack_confirm and drop the packet. */
222 static void
223 get_unique_tuple(struct nf_conntrack_tuple *tuple,
224                  const struct nf_conntrack_tuple *orig_tuple,
225                  const struct nf_nat_range *range,
226                  struct nf_conn *ct,
227                  enum nf_nat_manip_type maniptype)
228 {
229         struct nf_nat_protocol *proto;
230
231         /* 1) If this srcip/proto/src-proto-part is currently mapped,
232            and that same mapping gives a unique tuple within the given
233            range, use that.
234
235            This is only required for source (ie. NAT/masq) mappings.
236            So far, we don't do local source mappings, so multiple
237            manips not an issue.  */
238         if (maniptype == IP_NAT_MANIP_SRC) {
239                 if (find_appropriate_src(orig_tuple, tuple, range)) {
240                         pr_debug("get_unique_tuple: Found current src map\n");
241                         if (!(range->flags & IP_NAT_RANGE_PROTO_RANDOM))
242                                 if (!nf_nat_used_tuple(tuple, ct))
243                                         return;
244                 }
245         }
246
247         /* 2) Select the least-used IP/proto combination in the given
248            range. */
249         *tuple = *orig_tuple;
250         find_best_ips_proto(tuple, range, ct, maniptype);
251
252         /* 3) The per-protocol part of the manip is made to map into
253            the range to make a unique tuple. */
254
255         rcu_read_lock();
256         proto = __nf_nat_proto_find(orig_tuple->dst.protonum);
257
258         /* Change protocol info to have some randomization */
259         if (range->flags & IP_NAT_RANGE_PROTO_RANDOM) {
260                 proto->unique_tuple(tuple, range, maniptype, ct);
261                 goto out;
262         }
263
264         /* Only bother mapping if it's not already in range and unique */
265         if ((!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
266              proto->in_range(tuple, maniptype, &range->min, &range->max)) &&
267             !nf_nat_used_tuple(tuple, ct))
268                 goto out;
269
270         /* Last change: get protocol to try to obtain unique tuple. */
271         proto->unique_tuple(tuple, range, maniptype, ct);
272 out:
273         rcu_read_unlock();
274 }
275
276 unsigned int
277 nf_nat_setup_info(struct nf_conn *ct,
278                   const struct nf_nat_range *range,
279                   unsigned int hooknum)
280 {
281         struct nf_conntrack_tuple curr_tuple, new_tuple;
282         struct nf_conn_nat *nat;
283         int have_to_hash = !(ct->status & IPS_NAT_DONE_MASK);
284         enum nf_nat_manip_type maniptype = HOOK2MANIP(hooknum);
285
286         /* nat helper or nfctnetlink also setup binding */
287         nat = nfct_nat(ct);
288         if (!nat) {
289                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
290                 if (nat == NULL) {
291                         pr_debug("failed to add NAT extension\n");
292                         return NF_ACCEPT;
293                 }
294         }
295
296         NF_CT_ASSERT(hooknum == NF_INET_PRE_ROUTING ||
297                      hooknum == NF_INET_POST_ROUTING ||
298                      hooknum == NF_INET_LOCAL_IN ||
299                      hooknum == NF_INET_LOCAL_OUT);
300         BUG_ON(nf_nat_initialized(ct, maniptype));
301
302         /* What we've got will look like inverse of reply. Normally
303            this is what is in the conntrack, except for prior
304            manipulations (future optimization: if num_manips == 0,
305            orig_tp =
306            conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple) */
307         nf_ct_invert_tuplepr(&curr_tuple,
308                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
309
310         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
311
312         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
313                 struct nf_conntrack_tuple reply;
314
315                 /* Alter conntrack table so will recognize replies. */
316                 nf_ct_invert_tuplepr(&reply, &new_tuple);
317                 nf_conntrack_alter_reply(ct, &reply);
318
319                 /* Non-atomic: we own this at the moment. */
320                 if (maniptype == IP_NAT_MANIP_SRC)
321                         ct->status |= IPS_SRC_NAT;
322                 else
323                         ct->status |= IPS_DST_NAT;
324         }
325
326         /* Place in source hash if this is the first time. */
327         if (have_to_hash) {
328                 unsigned int srchash;
329
330                 srchash = hash_by_src(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
331                 write_lock_bh(&nf_nat_lock);
332                 /* nf_conntrack_alter_reply might re-allocate exntension aera */
333                 nat = nfct_nat(ct);
334                 nat->ct = ct;
335                 hlist_add_head(&nat->bysource, &bysource[srchash]);
336                 write_unlock_bh(&nf_nat_lock);
337         }
338
339         /* It's done. */
340         if (maniptype == IP_NAT_MANIP_DST)
341                 set_bit(IPS_DST_NAT_DONE_BIT, &ct->status);
342         else
343                 set_bit(IPS_SRC_NAT_DONE_BIT, &ct->status);
344
345         return NF_ACCEPT;
346 }
347 EXPORT_SYMBOL(nf_nat_setup_info);
348
349 /* Returns true if succeeded. */
350 static int
351 manip_pkt(u_int16_t proto,
352           struct sk_buff *skb,
353           unsigned int iphdroff,
354           const struct nf_conntrack_tuple *target,
355           enum nf_nat_manip_type maniptype)
356 {
357         struct iphdr *iph;
358         struct nf_nat_protocol *p;
359
360         if (!skb_make_writable(skb, iphdroff + sizeof(*iph)))
361                 return 0;
362
363         iph = (void *)skb->data + iphdroff;
364
365         /* Manipulate protcol part. */
366
367         /* rcu_read_lock()ed by nf_hook_slow */
368         p = __nf_nat_proto_find(proto);
369         if (!p->manip_pkt(skb, iphdroff, target, maniptype))
370                 return 0;
371
372         iph = (void *)skb->data + iphdroff;
373
374         if (maniptype == IP_NAT_MANIP_SRC) {
375                 csum_replace4(&iph->check, iph->saddr, target->src.u3.ip);
376                 iph->saddr = target->src.u3.ip;
377         } else {
378                 csum_replace4(&iph->check, iph->daddr, target->dst.u3.ip);
379                 iph->daddr = target->dst.u3.ip;
380         }
381         return 1;
382 }
383
384 /* Do packet manipulations according to nf_nat_setup_info. */
385 unsigned int nf_nat_packet(struct nf_conn *ct,
386                            enum ip_conntrack_info ctinfo,
387                            unsigned int hooknum,
388                            struct sk_buff *skb)
389 {
390         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
391         unsigned long statusbit;
392         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
393
394         if (mtype == IP_NAT_MANIP_SRC)
395                 statusbit = IPS_SRC_NAT;
396         else
397                 statusbit = IPS_DST_NAT;
398
399         /* Invert if this is reply dir. */
400         if (dir == IP_CT_DIR_REPLY)
401                 statusbit ^= IPS_NAT_MASK;
402
403         /* Non-atomic: these bits don't change. */
404         if (ct->status & statusbit) {
405                 struct nf_conntrack_tuple target;
406
407                 /* We are aiming to look like inverse of other direction. */
408                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
409
410                 if (!manip_pkt(target.dst.protonum, skb, 0, &target, mtype))
411                         return NF_DROP;
412         }
413         return NF_ACCEPT;
414 }
415 EXPORT_SYMBOL_GPL(nf_nat_packet);
416
417 /* Dir is direction ICMP is coming from (opposite to packet it contains) */
418 int nf_nat_icmp_reply_translation(struct nf_conn *ct,
419                                   enum ip_conntrack_info ctinfo,
420                                   unsigned int hooknum,
421                                   struct sk_buff *skb)
422 {
423         struct {
424                 struct icmphdr icmp;
425                 struct iphdr ip;
426         } *inside;
427         struct nf_conntrack_l4proto *l4proto;
428         struct nf_conntrack_tuple inner, target;
429         int hdrlen = ip_hdrlen(skb);
430         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
431         unsigned long statusbit;
432         enum nf_nat_manip_type manip = HOOK2MANIP(hooknum);
433
434         if (!skb_make_writable(skb, hdrlen + sizeof(*inside)))
435                 return 0;
436
437         inside = (void *)skb->data + ip_hdrlen(skb);
438
439         /* We're actually going to mangle it beyond trivial checksum
440            adjustment, so make sure the current checksum is correct. */
441         if (nf_ip_checksum(skb, hooknum, hdrlen, 0))
442                 return 0;
443
444         /* Must be RELATED */
445         NF_CT_ASSERT(skb->nfctinfo == IP_CT_RELATED ||
446                      skb->nfctinfo == IP_CT_RELATED+IP_CT_IS_REPLY);
447
448         /* Redirects on non-null nats must be dropped, else they'll
449            start talking to each other without our translation, and be
450            confused... --RR */
451         if (inside->icmp.type == ICMP_REDIRECT) {
452                 /* If NAT isn't finished, assume it and drop. */
453                 if ((ct->status & IPS_NAT_DONE_MASK) != IPS_NAT_DONE_MASK)
454                         return 0;
455
456                 if (ct->status & IPS_NAT_MASK)
457                         return 0;
458         }
459
460         pr_debug("icmp_reply_translation: translating error %p manip %u "
461                  "dir %s\n", skb, manip,
462                  dir == IP_CT_DIR_ORIGINAL ? "ORIG" : "REPLY");
463
464         /* rcu_read_lock()ed by nf_hook_slow */
465         l4proto = __nf_ct_l4proto_find(PF_INET, inside->ip.protocol);
466
467         if (!nf_ct_get_tuple(skb,
468                              ip_hdrlen(skb) + sizeof(struct icmphdr),
469                              (ip_hdrlen(skb) +
470                               sizeof(struct icmphdr) + inside->ip.ihl * 4),
471                              (u_int16_t)AF_INET,
472                              inside->ip.protocol,
473                              &inner, l3proto, l4proto))
474                 return 0;
475
476         /* Change inner back to look like incoming packet.  We do the
477            opposite manip on this hook to normal, because it might not
478            pass all hooks (locally-generated ICMP).  Consider incoming
479            packet: PREROUTING (DST manip), routing produces ICMP, goes
480            through POSTROUTING (which must correct the DST manip). */
481         if (!manip_pkt(inside->ip.protocol, skb,
482                        ip_hdrlen(skb) + sizeof(inside->icmp),
483                        &ct->tuplehash[!dir].tuple,
484                        !manip))
485                 return 0;
486
487         if (skb->ip_summed != CHECKSUM_PARTIAL) {
488                 /* Reloading "inside" here since manip_pkt inner. */
489                 inside = (void *)skb->data + ip_hdrlen(skb);
490                 inside->icmp.checksum = 0;
491                 inside->icmp.checksum =
492                         csum_fold(skb_checksum(skb, hdrlen,
493                                                skb->len - hdrlen, 0));
494         }
495
496         /* Change outer to look the reply to an incoming packet
497          * (proto 0 means don't invert per-proto part). */
498         if (manip == IP_NAT_MANIP_SRC)
499                 statusbit = IPS_SRC_NAT;
500         else
501                 statusbit = IPS_DST_NAT;
502
503         /* Invert if this is reply dir. */
504         if (dir == IP_CT_DIR_REPLY)
505                 statusbit ^= IPS_NAT_MASK;
506
507         if (ct->status & statusbit) {
508                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
509                 if (!manip_pkt(0, skb, 0, &target, manip))
510                         return 0;
511         }
512
513         return 1;
514 }
515 EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation);
516
517 /* Protocol registration. */
518 int nf_nat_protocol_register(struct nf_nat_protocol *proto)
519 {
520         int ret = 0;
521
522         write_lock_bh(&nf_nat_lock);
523         if (nf_nat_protos[proto->protonum] != &nf_nat_unknown_protocol) {
524                 ret = -EBUSY;
525                 goto out;
526         }
527         rcu_assign_pointer(nf_nat_protos[proto->protonum], proto);
528  out:
529         write_unlock_bh(&nf_nat_lock);
530         return ret;
531 }
532 EXPORT_SYMBOL(nf_nat_protocol_register);
533
534 /* Noone stores the protocol anywhere; simply delete it. */
535 void nf_nat_protocol_unregister(struct nf_nat_protocol *proto)
536 {
537         write_lock_bh(&nf_nat_lock);
538         rcu_assign_pointer(nf_nat_protos[proto->protonum],
539                            &nf_nat_unknown_protocol);
540         write_unlock_bh(&nf_nat_lock);
541         synchronize_rcu();
542 }
543 EXPORT_SYMBOL(nf_nat_protocol_unregister);
544
545 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
546 int
547 nf_nat_port_range_to_nlattr(struct sk_buff *skb,
548                             const struct nf_nat_range *range)
549 {
550         NLA_PUT_BE16(skb, CTA_PROTONAT_PORT_MIN, range->min.tcp.port);
551         NLA_PUT_BE16(skb, CTA_PROTONAT_PORT_MAX, range->max.tcp.port);
552
553         return 0;
554
555 nla_put_failure:
556         return -1;
557 }
558 EXPORT_SYMBOL_GPL(nf_nat_port_nlattr_to_range);
559
560 int
561 nf_nat_port_nlattr_to_range(struct nlattr *tb[], struct nf_nat_range *range)
562 {
563         int ret = 0;
564
565         /* we have to return whether we actually parsed something or not */
566
567         if (tb[CTA_PROTONAT_PORT_MIN]) {
568                 ret = 1;
569                 range->min.tcp.port = nla_get_be16(tb[CTA_PROTONAT_PORT_MIN]);
570         }
571
572         if (!tb[CTA_PROTONAT_PORT_MAX]) {
573                 if (ret)
574                         range->max.tcp.port = range->min.tcp.port;
575         } else {
576                 ret = 1;
577                 range->max.tcp.port = nla_get_be16(tb[CTA_PROTONAT_PORT_MAX]);
578         }
579
580         return ret;
581 }
582 EXPORT_SYMBOL_GPL(nf_nat_port_range_to_nlattr);
583 #endif
584
585 /* Noone using conntrack by the time this called. */
586 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
587 {
588         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
589
590         if (nat == NULL || nat->ct == NULL)
591                 return;
592
593         NF_CT_ASSERT(nat->ct->status & IPS_NAT_DONE_MASK);
594
595         write_lock_bh(&nf_nat_lock);
596         hlist_del(&nat->bysource);
597         nat->ct = NULL;
598         write_unlock_bh(&nf_nat_lock);
599 }
600
601 static void nf_nat_move_storage(struct nf_conn *conntrack, void *old)
602 {
603         struct nf_conn_nat *new_nat = nf_ct_ext_find(conntrack, NF_CT_EXT_NAT);
604         struct nf_conn_nat *old_nat = (struct nf_conn_nat *)old;
605         struct nf_conn *ct = old_nat->ct;
606
607         if (!ct || !(ct->status & IPS_NAT_DONE_MASK))
608                 return;
609
610         write_lock_bh(&nf_nat_lock);
611         hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
612         new_nat->ct = ct;
613         write_unlock_bh(&nf_nat_lock);
614 }
615
616 static struct nf_ct_ext_type nat_extend __read_mostly = {
617         .len            = sizeof(struct nf_conn_nat),
618         .align          = __alignof__(struct nf_conn_nat),
619         .destroy        = nf_nat_cleanup_conntrack,
620         .move           = nf_nat_move_storage,
621         .id             = NF_CT_EXT_NAT,
622         .flags          = NF_CT_EXT_F_PREALLOC,
623 };
624
625 static int __init nf_nat_init(void)
626 {
627         size_t i;
628         int ret;
629
630         ret = nf_ct_extend_register(&nat_extend);
631         if (ret < 0) {
632                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
633                 return ret;
634         }
635
636         /* Leave them the same for the moment. */
637         nf_nat_htable_size = nf_conntrack_htable_size;
638
639         bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size,
640                                          &nf_nat_vmalloced);
641         if (!bysource) {
642                 ret = -ENOMEM;
643                 goto cleanup_extend;
644         }
645
646         /* Sew in builtin protocols. */
647         write_lock_bh(&nf_nat_lock);
648         for (i = 0; i < MAX_IP_NAT_PROTO; i++)
649                 rcu_assign_pointer(nf_nat_protos[i], &nf_nat_unknown_protocol);
650         rcu_assign_pointer(nf_nat_protos[IPPROTO_TCP], &nf_nat_protocol_tcp);
651         rcu_assign_pointer(nf_nat_protos[IPPROTO_UDP], &nf_nat_protocol_udp);
652         rcu_assign_pointer(nf_nat_protos[IPPROTO_ICMP], &nf_nat_protocol_icmp);
653         write_unlock_bh(&nf_nat_lock);
654
655         for (i = 0; i < nf_nat_htable_size; i++) {
656                 INIT_HLIST_HEAD(&bysource[i]);
657         }
658
659         /* Initialize fake conntrack so that NAT will skip it */
660         nf_conntrack_untracked.status |= IPS_NAT_DONE_MASK;
661
662         l3proto = nf_ct_l3proto_find_get((u_int16_t)AF_INET);
663         return 0;
664
665  cleanup_extend:
666         nf_ct_extend_unregister(&nat_extend);
667         return ret;
668 }
669
670 /* Clear NAT section of all conntracks, in case we're loaded again. */
671 static int clean_nat(struct nf_conn *i, void *data)
672 {
673         struct nf_conn_nat *nat = nfct_nat(i);
674
675         if (!nat)
676                 return 0;
677         memset(nat, 0, sizeof(*nat));
678         i->status &= ~(IPS_NAT_MASK | IPS_NAT_DONE_MASK | IPS_SEQ_ADJUST);
679         return 0;
680 }
681
682 static void __exit nf_nat_cleanup(void)
683 {
684         nf_ct_iterate_cleanup(&clean_nat, NULL);
685         synchronize_rcu();
686         nf_ct_free_hashtable(bysource, nf_nat_vmalloced, nf_nat_htable_size);
687         nf_ct_l3proto_put(l3proto);
688         nf_ct_extend_unregister(&nat_extend);
689 }
690
691 MODULE_LICENSE("GPL");
692
693 module_init(nf_nat_init);
694 module_exit(nf_nat_cleanup);