ipvs: convert lblc scheduler to rcu
[linux-3.10.git] / net / netfilter / nf_nat_core.c
1 /*
2  * (C) 1999-2001 Paul `Rusty' Russell
3  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
4  * (C) 2011 Patrick McHardy <kaber@trash.net>
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 <linux/gfp.h>
16 #include <net/xfrm.h>
17 #include <linux/jhash.h>
18 #include <linux/rtnetlink.h>
19
20 #include <net/netfilter/nf_conntrack.h>
21 #include <net/netfilter/nf_conntrack_core.h>
22 #include <net/netfilter/nf_nat.h>
23 #include <net/netfilter/nf_nat_l3proto.h>
24 #include <net/netfilter/nf_nat_l4proto.h>
25 #include <net/netfilter/nf_nat_core.h>
26 #include <net/netfilter/nf_nat_helper.h>
27 #include <net/netfilter/nf_conntrack_helper.h>
28 #include <net/netfilter/nf_conntrack_l3proto.h>
29 #include <net/netfilter/nf_conntrack_zones.h>
30 #include <linux/netfilter/nf_nat.h>
31
32 static DEFINE_SPINLOCK(nf_nat_lock);
33
34 static DEFINE_MUTEX(nf_nat_proto_mutex);
35 static const struct nf_nat_l3proto __rcu *nf_nat_l3protos[NFPROTO_NUMPROTO]
36                                                 __read_mostly;
37 static const struct nf_nat_l4proto __rcu **nf_nat_l4protos[NFPROTO_NUMPROTO]
38                                                 __read_mostly;
39
40
41 inline const struct nf_nat_l3proto *
42 __nf_nat_l3proto_find(u8 family)
43 {
44         return rcu_dereference(nf_nat_l3protos[family]);
45 }
46
47 inline const struct nf_nat_l4proto *
48 __nf_nat_l4proto_find(u8 family, u8 protonum)
49 {
50         return rcu_dereference(nf_nat_l4protos[family][protonum]);
51 }
52 EXPORT_SYMBOL_GPL(__nf_nat_l4proto_find);
53
54 #ifdef CONFIG_XFRM
55 static void __nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl)
56 {
57         const struct nf_nat_l3proto *l3proto;
58         const struct nf_conn *ct;
59         enum ip_conntrack_info ctinfo;
60         enum ip_conntrack_dir dir;
61         unsigned  long statusbit;
62         u8 family;
63
64         ct = nf_ct_get(skb, &ctinfo);
65         if (ct == NULL)
66                 return;
67
68         family = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.l3num;
69         rcu_read_lock();
70         l3proto = __nf_nat_l3proto_find(family);
71         if (l3proto == NULL)
72                 goto out;
73
74         dir = CTINFO2DIR(ctinfo);
75         if (dir == IP_CT_DIR_ORIGINAL)
76                 statusbit = IPS_DST_NAT;
77         else
78                 statusbit = IPS_SRC_NAT;
79
80         l3proto->decode_session(skb, ct, dir, statusbit, fl);
81 out:
82         rcu_read_unlock();
83 }
84
85 int nf_xfrm_me_harder(struct sk_buff *skb, unsigned int family)
86 {
87         struct flowi fl;
88         unsigned int hh_len;
89         struct dst_entry *dst;
90
91         if (xfrm_decode_session(skb, &fl, family) < 0)
92                 return -1;
93
94         dst = skb_dst(skb);
95         if (dst->xfrm)
96                 dst = ((struct xfrm_dst *)dst)->route;
97         dst_hold(dst);
98
99         dst = xfrm_lookup(dev_net(dst->dev), dst, &fl, skb->sk, 0);
100         if (IS_ERR(dst))
101                 return -1;
102
103         skb_dst_drop(skb);
104         skb_dst_set(skb, dst);
105
106         /* Change in oif may mean change in hh_len. */
107         hh_len = skb_dst(skb)->dev->hard_header_len;
108         if (skb_headroom(skb) < hh_len &&
109             pskb_expand_head(skb, hh_len - skb_headroom(skb), 0, GFP_ATOMIC))
110                 return -1;
111         return 0;
112 }
113 EXPORT_SYMBOL(nf_xfrm_me_harder);
114 #endif /* CONFIG_XFRM */
115
116 /* We keep an extra hash for each conntrack, for fast searching. */
117 static inline unsigned int
118 hash_by_src(const struct net *net, u16 zone,
119             const struct nf_conntrack_tuple *tuple)
120 {
121         unsigned int hash;
122
123         /* Original src, to ensure we map it consistently if poss. */
124         hash = jhash2((u32 *)&tuple->src, sizeof(tuple->src) / sizeof(u32),
125                       tuple->dst.protonum ^ zone ^ nf_conntrack_hash_rnd);
126         return ((u64)hash * net->ct.nat_htable_size) >> 32;
127 }
128
129 /* Is this tuple already taken? (not by us) */
130 int
131 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
132                   const struct nf_conn *ignored_conntrack)
133 {
134         /* Conntrack tracking doesn't keep track of outgoing tuples; only
135          * incoming ones.  NAT means they don't have a fixed mapping,
136          * so we invert the tuple and look for the incoming reply.
137          *
138          * We could keep a separate hash if this proves too slow.
139          */
140         struct nf_conntrack_tuple reply;
141
142         nf_ct_invert_tuplepr(&reply, tuple);
143         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
144 }
145 EXPORT_SYMBOL(nf_nat_used_tuple);
146
147 /* If we source map this tuple so reply looks like reply_tuple, will
148  * that meet the constraints of range.
149  */
150 static int in_range(const struct nf_nat_l3proto *l3proto,
151                     const struct nf_nat_l4proto *l4proto,
152                     const struct nf_conntrack_tuple *tuple,
153                     const struct nf_nat_range *range)
154 {
155         /* If we are supposed to map IPs, then we must be in the
156          * range specified, otherwise let this drag us onto a new src IP.
157          */
158         if (range->flags & NF_NAT_RANGE_MAP_IPS &&
159             !l3proto->in_range(tuple, range))
160                 return 0;
161
162         if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) ||
163             l4proto->in_range(tuple, NF_NAT_MANIP_SRC,
164                               &range->min_proto, &range->max_proto))
165                 return 1;
166
167         return 0;
168 }
169
170 static inline int
171 same_src(const struct nf_conn *ct,
172          const struct nf_conntrack_tuple *tuple)
173 {
174         const struct nf_conntrack_tuple *t;
175
176         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
177         return (t->dst.protonum == tuple->dst.protonum &&
178                 nf_inet_addr_cmp(&t->src.u3, &tuple->src.u3) &&
179                 t->src.u.all == tuple->src.u.all);
180 }
181
182 /* Only called for SRC manip */
183 static int
184 find_appropriate_src(struct net *net, u16 zone,
185                      const struct nf_nat_l3proto *l3proto,
186                      const struct nf_nat_l4proto *l4proto,
187                      const struct nf_conntrack_tuple *tuple,
188                      struct nf_conntrack_tuple *result,
189                      const struct nf_nat_range *range)
190 {
191         unsigned int h = hash_by_src(net, zone, tuple);
192         const struct nf_conn_nat *nat;
193         const struct nf_conn *ct;
194
195         hlist_for_each_entry_rcu(nat, &net->ct.nat_bysource[h], bysource) {
196                 ct = nat->ct;
197                 if (same_src(ct, tuple) && nf_ct_zone(ct) == zone) {
198                         /* Copy source part from reply tuple. */
199                         nf_ct_invert_tuplepr(result,
200                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
201                         result->dst = tuple->dst;
202
203                         if (in_range(l3proto, l4proto, result, range))
204                                 return 1;
205                 }
206         }
207         return 0;
208 }
209
210 /* For [FUTURE] fragmentation handling, we want the least-used
211  * src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
212  * if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
213  * 1-65535, we don't do pro-rata allocation based on ports; we choose
214  * the ip with the lowest src-ip/dst-ip/proto usage.
215  */
216 static void
217 find_best_ips_proto(u16 zone, struct nf_conntrack_tuple *tuple,
218                     const struct nf_nat_range *range,
219                     const struct nf_conn *ct,
220                     enum nf_nat_manip_type maniptype)
221 {
222         union nf_inet_addr *var_ipp;
223         unsigned int i, max;
224         /* Host order */
225         u32 minip, maxip, j, dist;
226         bool full_range;
227
228         /* No IP mapping?  Do nothing. */
229         if (!(range->flags & NF_NAT_RANGE_MAP_IPS))
230                 return;
231
232         if (maniptype == NF_NAT_MANIP_SRC)
233                 var_ipp = &tuple->src.u3;
234         else
235                 var_ipp = &tuple->dst.u3;
236
237         /* Fast path: only one choice. */
238         if (nf_inet_addr_cmp(&range->min_addr, &range->max_addr)) {
239                 *var_ipp = range->min_addr;
240                 return;
241         }
242
243         if (nf_ct_l3num(ct) == NFPROTO_IPV4)
244                 max = sizeof(var_ipp->ip) / sizeof(u32) - 1;
245         else
246                 max = sizeof(var_ipp->ip6) / sizeof(u32) - 1;
247
248         /* Hashing source and destination IPs gives a fairly even
249          * spread in practice (if there are a small number of IPs
250          * involved, there usually aren't that many connections
251          * anyway).  The consistency means that servers see the same
252          * client coming from the same IP (some Internet Banking sites
253          * like this), even across reboots.
254          */
255         j = jhash2((u32 *)&tuple->src.u3, sizeof(tuple->src.u3) / sizeof(u32),
256                    range->flags & NF_NAT_RANGE_PERSISTENT ?
257                         0 : (__force u32)tuple->dst.u3.all[max] ^ zone);
258
259         full_range = false;
260         for (i = 0; i <= max; i++) {
261                 /* If first bytes of the address are at the maximum, use the
262                  * distance. Otherwise use the full range.
263                  */
264                 if (!full_range) {
265                         minip = ntohl((__force __be32)range->min_addr.all[i]);
266                         maxip = ntohl((__force __be32)range->max_addr.all[i]);
267                         dist  = maxip - minip + 1;
268                 } else {
269                         minip = 0;
270                         dist  = ~0;
271                 }
272
273                 var_ipp->all[i] = (__force __u32)
274                         htonl(minip + (((u64)j * dist) >> 32));
275                 if (var_ipp->all[i] != range->max_addr.all[i])
276                         full_range = true;
277
278                 if (!(range->flags & NF_NAT_RANGE_PERSISTENT))
279                         j ^= (__force u32)tuple->dst.u3.all[i];
280         }
281 }
282
283 /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING,
284  * we change the source to map into the range. For NF_INET_PRE_ROUTING
285  * and NF_INET_LOCAL_OUT, we change the destination to map into the
286  * range. It might not be possible to get a unique tuple, but we try.
287  * At worst (or if we race), we will end up with a final duplicate in
288  * __ip_conntrack_confirm and drop the packet. */
289 static void
290 get_unique_tuple(struct nf_conntrack_tuple *tuple,
291                  const struct nf_conntrack_tuple *orig_tuple,
292                  const struct nf_nat_range *range,
293                  struct nf_conn *ct,
294                  enum nf_nat_manip_type maniptype)
295 {
296         const struct nf_nat_l3proto *l3proto;
297         const struct nf_nat_l4proto *l4proto;
298         struct net *net = nf_ct_net(ct);
299         u16 zone = nf_ct_zone(ct);
300
301         rcu_read_lock();
302         l3proto = __nf_nat_l3proto_find(orig_tuple->src.l3num);
303         l4proto = __nf_nat_l4proto_find(orig_tuple->src.l3num,
304                                         orig_tuple->dst.protonum);
305
306         /* 1) If this srcip/proto/src-proto-part is currently mapped,
307          * and that same mapping gives a unique tuple within the given
308          * range, use that.
309          *
310          * This is only required for source (ie. NAT/masq) mappings.
311          * So far, we don't do local source mappings, so multiple
312          * manips not an issue.
313          */
314         if (maniptype == NF_NAT_MANIP_SRC &&
315             !(range->flags & NF_NAT_RANGE_PROTO_RANDOM)) {
316                 /* try the original tuple first */
317                 if (in_range(l3proto, l4proto, orig_tuple, range)) {
318                         if (!nf_nat_used_tuple(orig_tuple, ct)) {
319                                 *tuple = *orig_tuple;
320                                 goto out;
321                         }
322                 } else if (find_appropriate_src(net, zone, l3proto, l4proto,
323                                                 orig_tuple, tuple, range)) {
324                         pr_debug("get_unique_tuple: Found current src map\n");
325                         if (!nf_nat_used_tuple(tuple, ct))
326                                 goto out;
327                 }
328         }
329
330         /* 2) Select the least-used IP/proto combination in the given range */
331         *tuple = *orig_tuple;
332         find_best_ips_proto(zone, tuple, range, ct, maniptype);
333
334         /* 3) The per-protocol part of the manip is made to map into
335          * the range to make a unique tuple.
336          */
337
338         /* Only bother mapping if it's not already in range and unique */
339         if (!(range->flags & NF_NAT_RANGE_PROTO_RANDOM)) {
340                 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
341                         if (l4proto->in_range(tuple, maniptype,
342                                               &range->min_proto,
343                                               &range->max_proto) &&
344                             (range->min_proto.all == range->max_proto.all ||
345                              !nf_nat_used_tuple(tuple, ct)))
346                                 goto out;
347                 } else if (!nf_nat_used_tuple(tuple, ct)) {
348                         goto out;
349                 }
350         }
351
352         /* Last change: get protocol to try to obtain unique tuple. */
353         l4proto->unique_tuple(l3proto, tuple, range, maniptype, ct);
354 out:
355         rcu_read_unlock();
356 }
357
358 unsigned int
359 nf_nat_setup_info(struct nf_conn *ct,
360                   const struct nf_nat_range *range,
361                   enum nf_nat_manip_type maniptype)
362 {
363         struct net *net = nf_ct_net(ct);
364         struct nf_conntrack_tuple curr_tuple, new_tuple;
365         struct nf_conn_nat *nat;
366
367         /* nat helper or nfctnetlink also setup binding */
368         nat = nfct_nat(ct);
369         if (!nat) {
370                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
371                 if (nat == NULL) {
372                         pr_debug("failed to add NAT extension\n");
373                         return NF_ACCEPT;
374                 }
375         }
376
377         NF_CT_ASSERT(maniptype == NF_NAT_MANIP_SRC ||
378                      maniptype == NF_NAT_MANIP_DST);
379         BUG_ON(nf_nat_initialized(ct, maniptype));
380
381         /* What we've got will look like inverse of reply. Normally
382          * this is what is in the conntrack, except for prior
383          * manipulations (future optimization: if num_manips == 0,
384          * orig_tp = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple)
385          */
386         nf_ct_invert_tuplepr(&curr_tuple,
387                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
388
389         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
390
391         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
392                 struct nf_conntrack_tuple reply;
393
394                 /* Alter conntrack table so will recognize replies. */
395                 nf_ct_invert_tuplepr(&reply, &new_tuple);
396                 nf_conntrack_alter_reply(ct, &reply);
397
398                 /* Non-atomic: we own this at the moment. */
399                 if (maniptype == NF_NAT_MANIP_SRC)
400                         ct->status |= IPS_SRC_NAT;
401                 else
402                         ct->status |= IPS_DST_NAT;
403         }
404
405         if (maniptype == NF_NAT_MANIP_SRC) {
406                 unsigned int srchash;
407
408                 srchash = hash_by_src(net, nf_ct_zone(ct),
409                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
410                 spin_lock_bh(&nf_nat_lock);
411                 /* nf_conntrack_alter_reply might re-allocate extension aera */
412                 nat = nfct_nat(ct);
413                 nat->ct = ct;
414                 hlist_add_head_rcu(&nat->bysource,
415                                    &net->ct.nat_bysource[srchash]);
416                 spin_unlock_bh(&nf_nat_lock);
417         }
418
419         /* It's done. */
420         if (maniptype == NF_NAT_MANIP_DST)
421                 ct->status |= IPS_DST_NAT_DONE;
422         else
423                 ct->status |= IPS_SRC_NAT_DONE;
424
425         return NF_ACCEPT;
426 }
427 EXPORT_SYMBOL(nf_nat_setup_info);
428
429 /* Do packet manipulations according to nf_nat_setup_info. */
430 unsigned int nf_nat_packet(struct nf_conn *ct,
431                            enum ip_conntrack_info ctinfo,
432                            unsigned int hooknum,
433                            struct sk_buff *skb)
434 {
435         const struct nf_nat_l3proto *l3proto;
436         const struct nf_nat_l4proto *l4proto;
437         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
438         unsigned long statusbit;
439         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
440
441         if (mtype == NF_NAT_MANIP_SRC)
442                 statusbit = IPS_SRC_NAT;
443         else
444                 statusbit = IPS_DST_NAT;
445
446         /* Invert if this is reply dir. */
447         if (dir == IP_CT_DIR_REPLY)
448                 statusbit ^= IPS_NAT_MASK;
449
450         /* Non-atomic: these bits don't change. */
451         if (ct->status & statusbit) {
452                 struct nf_conntrack_tuple target;
453
454                 /* We are aiming to look like inverse of other direction. */
455                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
456
457                 l3proto = __nf_nat_l3proto_find(target.src.l3num);
458                 l4proto = __nf_nat_l4proto_find(target.src.l3num,
459                                                 target.dst.protonum);
460                 if (!l3proto->manip_pkt(skb, 0, l4proto, &target, mtype))
461                         return NF_DROP;
462         }
463         return NF_ACCEPT;
464 }
465 EXPORT_SYMBOL_GPL(nf_nat_packet);
466
467 struct nf_nat_proto_clean {
468         u8      l3proto;
469         u8      l4proto;
470         bool    hash;
471 };
472
473 /* Clear NAT section of all conntracks, in case we're loaded again. */
474 static int nf_nat_proto_clean(struct nf_conn *i, void *data)
475 {
476         const struct nf_nat_proto_clean *clean = data;
477         struct nf_conn_nat *nat = nfct_nat(i);
478
479         if (!nat)
480                 return 0;
481         if (!(i->status & IPS_SRC_NAT_DONE))
482                 return 0;
483         if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) ||
484             (clean->l4proto && nf_ct_protonum(i) != clean->l4proto))
485                 return 0;
486
487         if (clean->hash) {
488                 spin_lock_bh(&nf_nat_lock);
489                 hlist_del_rcu(&nat->bysource);
490                 spin_unlock_bh(&nf_nat_lock);
491         } else {
492                 memset(nat, 0, sizeof(*nat));
493                 i->status &= ~(IPS_NAT_MASK | IPS_NAT_DONE_MASK |
494                                IPS_SEQ_ADJUST);
495         }
496         return 0;
497 }
498
499 static void nf_nat_l4proto_clean(u8 l3proto, u8 l4proto)
500 {
501         struct nf_nat_proto_clean clean = {
502                 .l3proto = l3proto,
503                 .l4proto = l4proto,
504         };
505         struct net *net;
506
507         rtnl_lock();
508         /* Step 1 - remove from bysource hash */
509         clean.hash = true;
510         for_each_net(net)
511                 nf_ct_iterate_cleanup(net, nf_nat_proto_clean, &clean);
512         synchronize_rcu();
513
514         /* Step 2 - clean NAT section */
515         clean.hash = false;
516         for_each_net(net)
517                 nf_ct_iterate_cleanup(net, nf_nat_proto_clean, &clean);
518         rtnl_unlock();
519 }
520
521 static void nf_nat_l3proto_clean(u8 l3proto)
522 {
523         struct nf_nat_proto_clean clean = {
524                 .l3proto = l3proto,
525         };
526         struct net *net;
527
528         rtnl_lock();
529         /* Step 1 - remove from bysource hash */
530         clean.hash = true;
531         for_each_net(net)
532                 nf_ct_iterate_cleanup(net, nf_nat_proto_clean, &clean);
533         synchronize_rcu();
534
535         /* Step 2 - clean NAT section */
536         clean.hash = false;
537         for_each_net(net)
538                 nf_ct_iterate_cleanup(net, nf_nat_proto_clean, &clean);
539         rtnl_unlock();
540 }
541
542 /* Protocol registration. */
543 int nf_nat_l4proto_register(u8 l3proto, const struct nf_nat_l4proto *l4proto)
544 {
545         const struct nf_nat_l4proto **l4protos;
546         unsigned int i;
547         int ret = 0;
548
549         mutex_lock(&nf_nat_proto_mutex);
550         if (nf_nat_l4protos[l3proto] == NULL) {
551                 l4protos = kmalloc(IPPROTO_MAX * sizeof(struct nf_nat_l4proto *),
552                                    GFP_KERNEL);
553                 if (l4protos == NULL) {
554                         ret = -ENOMEM;
555                         goto out;
556                 }
557
558                 for (i = 0; i < IPPROTO_MAX; i++)
559                         RCU_INIT_POINTER(l4protos[i], &nf_nat_l4proto_unknown);
560
561                 /* Before making proto_array visible to lockless readers,
562                  * we must make sure its content is committed to memory.
563                  */
564                 smp_wmb();
565
566                 nf_nat_l4protos[l3proto] = l4protos;
567         }
568
569         if (rcu_dereference_protected(
570                         nf_nat_l4protos[l3proto][l4proto->l4proto],
571                         lockdep_is_held(&nf_nat_proto_mutex)
572                         ) != &nf_nat_l4proto_unknown) {
573                 ret = -EBUSY;
574                 goto out;
575         }
576         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto], l4proto);
577  out:
578         mutex_unlock(&nf_nat_proto_mutex);
579         return ret;
580 }
581 EXPORT_SYMBOL_GPL(nf_nat_l4proto_register);
582
583 /* No one stores the protocol anywhere; simply delete it. */
584 void nf_nat_l4proto_unregister(u8 l3proto, const struct nf_nat_l4proto *l4proto)
585 {
586         mutex_lock(&nf_nat_proto_mutex);
587         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto],
588                          &nf_nat_l4proto_unknown);
589         mutex_unlock(&nf_nat_proto_mutex);
590         synchronize_rcu();
591
592         nf_nat_l4proto_clean(l3proto, l4proto->l4proto);
593 }
594 EXPORT_SYMBOL_GPL(nf_nat_l4proto_unregister);
595
596 int nf_nat_l3proto_register(const struct nf_nat_l3proto *l3proto)
597 {
598         int err;
599
600         err = nf_ct_l3proto_try_module_get(l3proto->l3proto);
601         if (err < 0)
602                 return err;
603
604         mutex_lock(&nf_nat_proto_mutex);
605         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_TCP],
606                          &nf_nat_l4proto_tcp);
607         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDP],
608                          &nf_nat_l4proto_udp);
609         mutex_unlock(&nf_nat_proto_mutex);
610
611         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], l3proto);
612         return 0;
613 }
614 EXPORT_SYMBOL_GPL(nf_nat_l3proto_register);
615
616 void nf_nat_l3proto_unregister(const struct nf_nat_l3proto *l3proto)
617 {
618         mutex_lock(&nf_nat_proto_mutex);
619         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], NULL);
620         mutex_unlock(&nf_nat_proto_mutex);
621         synchronize_rcu();
622
623         nf_nat_l3proto_clean(l3proto->l3proto);
624         nf_ct_l3proto_module_put(l3proto->l3proto);
625 }
626 EXPORT_SYMBOL_GPL(nf_nat_l3proto_unregister);
627
628 /* No one using conntrack by the time this called. */
629 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
630 {
631         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
632
633         if (nat == NULL || nat->ct == NULL)
634                 return;
635
636         NF_CT_ASSERT(nat->ct->status & IPS_SRC_NAT_DONE);
637
638         spin_lock_bh(&nf_nat_lock);
639         hlist_del_rcu(&nat->bysource);
640         spin_unlock_bh(&nf_nat_lock);
641 }
642
643 static void nf_nat_move_storage(void *new, void *old)
644 {
645         struct nf_conn_nat *new_nat = new;
646         struct nf_conn_nat *old_nat = old;
647         struct nf_conn *ct = old_nat->ct;
648
649         if (!ct || !(ct->status & IPS_SRC_NAT_DONE))
650                 return;
651
652         spin_lock_bh(&nf_nat_lock);
653         hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
654         spin_unlock_bh(&nf_nat_lock);
655 }
656
657 static struct nf_ct_ext_type nat_extend __read_mostly = {
658         .len            = sizeof(struct nf_conn_nat),
659         .align          = __alignof__(struct nf_conn_nat),
660         .destroy        = nf_nat_cleanup_conntrack,
661         .move           = nf_nat_move_storage,
662         .id             = NF_CT_EXT_NAT,
663         .flags          = NF_CT_EXT_F_PREALLOC,
664 };
665
666 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
667
668 #include <linux/netfilter/nfnetlink.h>
669 #include <linux/netfilter/nfnetlink_conntrack.h>
670
671 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
672         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
673         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
674 };
675
676 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
677                                      const struct nf_conn *ct,
678                                      struct nf_nat_range *range)
679 {
680         struct nlattr *tb[CTA_PROTONAT_MAX+1];
681         const struct nf_nat_l4proto *l4proto;
682         int err;
683
684         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, protonat_nla_policy);
685         if (err < 0)
686                 return err;
687
688         l4proto = __nf_nat_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
689         if (l4proto->nlattr_to_range)
690                 err = l4proto->nlattr_to_range(tb, range);
691
692         return err;
693 }
694
695 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
696         [CTA_NAT_V4_MINIP]      = { .type = NLA_U32 },
697         [CTA_NAT_V4_MAXIP]      = { .type = NLA_U32 },
698         [CTA_NAT_V6_MINIP]      = { .len = sizeof(struct in6_addr) },
699         [CTA_NAT_V6_MAXIP]      = { .len = sizeof(struct in6_addr) },
700         [CTA_NAT_PROTO]         = { .type = NLA_NESTED },
701 };
702
703 static int
704 nfnetlink_parse_nat(const struct nlattr *nat,
705                     const struct nf_conn *ct, struct nf_nat_range *range)
706 {
707         const struct nf_nat_l3proto *l3proto;
708         struct nlattr *tb[CTA_NAT_MAX+1];
709         int err;
710
711         memset(range, 0, sizeof(*range));
712
713         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy);
714         if (err < 0)
715                 return err;
716
717         rcu_read_lock();
718         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
719         if (l3proto == NULL) {
720                 err = -EAGAIN;
721                 goto out;
722         }
723         err = l3proto->nlattr_to_range(tb, range);
724         if (err < 0)
725                 goto out;
726
727         if (!tb[CTA_NAT_PROTO])
728                 goto out;
729
730         err = nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
731 out:
732         rcu_read_unlock();
733         return err;
734 }
735
736 static int
737 nfnetlink_parse_nat_setup(struct nf_conn *ct,
738                           enum nf_nat_manip_type manip,
739                           const struct nlattr *attr)
740 {
741         struct nf_nat_range range;
742         int err;
743
744         err = nfnetlink_parse_nat(attr, ct, &range);
745         if (err < 0)
746                 return err;
747         if (nf_nat_initialized(ct, manip))
748                 return -EEXIST;
749
750         return nf_nat_setup_info(ct, &range, manip);
751 }
752 #else
753 static int
754 nfnetlink_parse_nat_setup(struct nf_conn *ct,
755                           enum nf_nat_manip_type manip,
756                           const struct nlattr *attr)
757 {
758         return -EOPNOTSUPP;
759 }
760 #endif
761
762 static int __net_init nf_nat_net_init(struct net *net)
763 {
764         /* Leave them the same for the moment. */
765         net->ct.nat_htable_size = net->ct.htable_size;
766         net->ct.nat_bysource = nf_ct_alloc_hashtable(&net->ct.nat_htable_size, 0);
767         if (!net->ct.nat_bysource)
768                 return -ENOMEM;
769         return 0;
770 }
771
772 static void __net_exit nf_nat_net_exit(struct net *net)
773 {
774         struct nf_nat_proto_clean clean = {};
775
776         nf_ct_iterate_cleanup(net, &nf_nat_proto_clean, &clean);
777         synchronize_rcu();
778         nf_ct_free_hashtable(net->ct.nat_bysource, net->ct.nat_htable_size);
779 }
780
781 static struct pernet_operations nf_nat_net_ops = {
782         .init = nf_nat_net_init,
783         .exit = nf_nat_net_exit,
784 };
785
786 static struct nf_ct_helper_expectfn follow_master_nat = {
787         .name           = "nat-follow-master",
788         .expectfn       = nf_nat_follow_master,
789 };
790
791 static struct nfq_ct_nat_hook nfq_ct_nat = {
792         .seq_adjust     = nf_nat_tcp_seq_adjust,
793 };
794
795 static int __init nf_nat_init(void)
796 {
797         int ret;
798
799         ret = nf_ct_extend_register(&nat_extend);
800         if (ret < 0) {
801                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
802                 return ret;
803         }
804
805         ret = register_pernet_subsys(&nf_nat_net_ops);
806         if (ret < 0)
807                 goto cleanup_extend;
808
809         nf_ct_helper_expectfn_register(&follow_master_nat);
810
811         /* Initialize fake conntrack so that NAT will skip it */
812         nf_ct_untracked_status_or(IPS_NAT_DONE_MASK);
813
814         BUG_ON(nf_nat_seq_adjust_hook != NULL);
815         RCU_INIT_POINTER(nf_nat_seq_adjust_hook, nf_nat_seq_adjust);
816         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
817         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook,
818                            nfnetlink_parse_nat_setup);
819         BUG_ON(nf_ct_nat_offset != NULL);
820         RCU_INIT_POINTER(nf_ct_nat_offset, nf_nat_get_offset);
821         RCU_INIT_POINTER(nfq_ct_nat_hook, &nfq_ct_nat);
822 #ifdef CONFIG_XFRM
823         BUG_ON(nf_nat_decode_session_hook != NULL);
824         RCU_INIT_POINTER(nf_nat_decode_session_hook, __nf_nat_decode_session);
825 #endif
826         return 0;
827
828  cleanup_extend:
829         nf_ct_extend_unregister(&nat_extend);
830         return ret;
831 }
832
833 static void __exit nf_nat_cleanup(void)
834 {
835         unsigned int i;
836
837         unregister_pernet_subsys(&nf_nat_net_ops);
838         nf_ct_extend_unregister(&nat_extend);
839         nf_ct_helper_expectfn_unregister(&follow_master_nat);
840         RCU_INIT_POINTER(nf_nat_seq_adjust_hook, NULL);
841         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook, NULL);
842         RCU_INIT_POINTER(nf_ct_nat_offset, NULL);
843         RCU_INIT_POINTER(nfq_ct_nat_hook, NULL);
844 #ifdef CONFIG_XFRM
845         RCU_INIT_POINTER(nf_nat_decode_session_hook, NULL);
846 #endif
847         for (i = 0; i < NFPROTO_NUMPROTO; i++)
848                 kfree(nf_nat_l4protos[i]);
849         synchronize_net();
850 }
851
852 MODULE_LICENSE("GPL");
853
854 module_init(nf_nat_init);
855 module_exit(nf_nat_cleanup);