37e2b88313f26d0867cbd30c3f523043900a9fcf
[linux-2.6.git] / net / netfilter / nf_conntrack_core.c
1 /* Connection state tracking for netfilter.  This is separated from,
2    but required by, the NAT layer; it can also be used by an iptables
3    extension. */
4
5 /* (C) 1999-2001 Paul `Rusty' Russell
6  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
7  * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License version 2 as
11  * published by the Free Software Foundation.
12  */
13
14 #include <linux/types.h>
15 #include <linux/netfilter.h>
16 #include <linux/module.h>
17 #include <linux/sched.h>
18 #include <linux/skbuff.h>
19 #include <linux/proc_fs.h>
20 #include <linux/vmalloc.h>
21 #include <linux/stddef.h>
22 #include <linux/slab.h>
23 #include <linux/random.h>
24 #include <linux/jhash.h>
25 #include <linux/err.h>
26 #include <linux/percpu.h>
27 #include <linux/moduleparam.h>
28 #include <linux/notifier.h>
29 #include <linux/kernel.h>
30 #include <linux/netdevice.h>
31 #include <linux/socket.h>
32 #include <linux/mm.h>
33 #include <linux/rculist_nulls.h>
34
35 #include <net/netfilter/nf_conntrack.h>
36 #include <net/netfilter/nf_conntrack_l3proto.h>
37 #include <net/netfilter/nf_conntrack_l4proto.h>
38 #include <net/netfilter/nf_conntrack_expect.h>
39 #include <net/netfilter/nf_conntrack_helper.h>
40 #include <net/netfilter/nf_conntrack_core.h>
41 #include <net/netfilter/nf_conntrack_extend.h>
42 #include <net/netfilter/nf_conntrack_acct.h>
43 #include <net/netfilter/nf_conntrack_ecache.h>
44 #include <net/netfilter/nf_nat.h>
45 #include <net/netfilter/nf_nat_core.h>
46
47 #define NF_CONNTRACK_VERSION    "0.5.0"
48
49 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
50                                       enum nf_nat_manip_type manip,
51                                       const struct nlattr *attr) __read_mostly;
52 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
53
54 DEFINE_SPINLOCK(nf_conntrack_lock);
55 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
56
57 unsigned int nf_conntrack_htable_size __read_mostly;
58 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
59
60 unsigned int nf_conntrack_max __read_mostly;
61 EXPORT_SYMBOL_GPL(nf_conntrack_max);
62
63 struct nf_conn nf_conntrack_untracked __read_mostly;
64 EXPORT_SYMBOL_GPL(nf_conntrack_untracked);
65
66 static struct kmem_cache *nf_conntrack_cachep __read_mostly;
67
68 static int nf_conntrack_hash_rnd_initted;
69 static unsigned int nf_conntrack_hash_rnd;
70
71 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
72                                   unsigned int size, unsigned int rnd)
73 {
74         unsigned int n;
75         u_int32_t h;
76
77         /* The direction must be ignored, so we hash everything up to the
78          * destination ports (which is a multiple of 4) and treat the last
79          * three bytes manually.
80          */
81         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
82         h = jhash2((u32 *)tuple, n,
83                    rnd ^ (((__force __u16)tuple->dst.u.all << 16) |
84                           tuple->dst.protonum));
85
86         return ((u64)h * size) >> 32;
87 }
88
89 static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
90 {
91         return __hash_conntrack(tuple, nf_conntrack_htable_size,
92                                 nf_conntrack_hash_rnd);
93 }
94
95 bool
96 nf_ct_get_tuple(const struct sk_buff *skb,
97                 unsigned int nhoff,
98                 unsigned int dataoff,
99                 u_int16_t l3num,
100                 u_int8_t protonum,
101                 struct nf_conntrack_tuple *tuple,
102                 const struct nf_conntrack_l3proto *l3proto,
103                 const struct nf_conntrack_l4proto *l4proto)
104 {
105         memset(tuple, 0, sizeof(*tuple));
106
107         tuple->src.l3num = l3num;
108         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
109                 return false;
110
111         tuple->dst.protonum = protonum;
112         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
113
114         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
115 }
116 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
117
118 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
119                        u_int16_t l3num, struct nf_conntrack_tuple *tuple)
120 {
121         struct nf_conntrack_l3proto *l3proto;
122         struct nf_conntrack_l4proto *l4proto;
123         unsigned int protoff;
124         u_int8_t protonum;
125         int ret;
126
127         rcu_read_lock();
128
129         l3proto = __nf_ct_l3proto_find(l3num);
130         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
131         if (ret != NF_ACCEPT) {
132                 rcu_read_unlock();
133                 return false;
134         }
135
136         l4proto = __nf_ct_l4proto_find(l3num, protonum);
137
138         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
139                               l3proto, l4proto);
140
141         rcu_read_unlock();
142         return ret;
143 }
144 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
145
146 bool
147 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
148                    const struct nf_conntrack_tuple *orig,
149                    const struct nf_conntrack_l3proto *l3proto,
150                    const struct nf_conntrack_l4proto *l4proto)
151 {
152         memset(inverse, 0, sizeof(*inverse));
153
154         inverse->src.l3num = orig->src.l3num;
155         if (l3proto->invert_tuple(inverse, orig) == 0)
156                 return false;
157
158         inverse->dst.dir = !orig->dst.dir;
159
160         inverse->dst.protonum = orig->dst.protonum;
161         return l4proto->invert_tuple(inverse, orig);
162 }
163 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
164
165 static void
166 clean_from_lists(struct nf_conn *ct)
167 {
168         pr_debug("clean_from_lists(%p)\n", ct);
169         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
170         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
171
172         /* Destroy all pending expectations */
173         nf_ct_remove_expectations(ct);
174 }
175
176 static void
177 destroy_conntrack(struct nf_conntrack *nfct)
178 {
179         struct nf_conn *ct = (struct nf_conn *)nfct;
180         struct net *net = nf_ct_net(ct);
181         struct nf_conntrack_l4proto *l4proto;
182
183         pr_debug("destroy_conntrack(%p)\n", ct);
184         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
185         NF_CT_ASSERT(!timer_pending(&ct->timeout));
186
187         /* To make sure we don't get any weird locking issues here:
188          * destroy_conntrack() MUST NOT be called with a write lock
189          * to nf_conntrack_lock!!! -HW */
190         rcu_read_lock();
191         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
192         if (l4proto && l4proto->destroy)
193                 l4proto->destroy(ct);
194
195         rcu_read_unlock();
196
197         spin_lock_bh(&nf_conntrack_lock);
198         /* Expectations will have been removed in clean_from_lists,
199          * except TFTP can create an expectation on the first packet,
200          * before connection is in the list, so we need to clean here,
201          * too. */
202         nf_ct_remove_expectations(ct);
203
204         /* We overload first tuple to link into unconfirmed list. */
205         if (!nf_ct_is_confirmed(ct)) {
206                 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
207                 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
208         }
209
210         NF_CT_STAT_INC(net, delete);
211         spin_unlock_bh(&nf_conntrack_lock);
212
213         if (ct->master)
214                 nf_ct_put(ct->master);
215
216         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
217         nf_conntrack_free(ct);
218 }
219
220 void nf_ct_delete_from_lists(struct nf_conn *ct)
221 {
222         struct net *net = nf_ct_net(ct);
223
224         nf_ct_helper_destroy(ct);
225         spin_lock_bh(&nf_conntrack_lock);
226         /* Inside lock so preempt is disabled on module removal path.
227          * Otherwise we can get spurious warnings. */
228         NF_CT_STAT_INC(net, delete_list);
229         clean_from_lists(ct);
230         spin_unlock_bh(&nf_conntrack_lock);
231 }
232 EXPORT_SYMBOL_GPL(nf_ct_delete_from_lists);
233
234 static void death_by_event(unsigned long ul_conntrack)
235 {
236         struct nf_conn *ct = (void *)ul_conntrack;
237         struct net *net = nf_ct_net(ct);
238
239         if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) {
240                 /* bad luck, let's retry again */
241                 ct->timeout.expires = jiffies +
242                         (random32() % net->ct.sysctl_events_retry_timeout);
243                 add_timer(&ct->timeout);
244                 return;
245         }
246         /* we've got the event delivered, now it's dying */
247         set_bit(IPS_DYING_BIT, &ct->status);
248         spin_lock(&nf_conntrack_lock);
249         hlist_nulls_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
250         spin_unlock(&nf_conntrack_lock);
251         nf_ct_put(ct);
252 }
253
254 void nf_ct_insert_dying_list(struct nf_conn *ct)
255 {
256         struct net *net = nf_ct_net(ct);
257
258         /* add this conntrack to the dying list */
259         spin_lock_bh(&nf_conntrack_lock);
260         hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
261                              &net->ct.dying);
262         spin_unlock_bh(&nf_conntrack_lock);
263         /* set a new timer to retry event delivery */
264         setup_timer(&ct->timeout, death_by_event, (unsigned long)ct);
265         ct->timeout.expires = jiffies +
266                 (random32() % net->ct.sysctl_events_retry_timeout);
267         add_timer(&ct->timeout);
268 }
269 EXPORT_SYMBOL_GPL(nf_ct_insert_dying_list);
270
271 static void death_by_timeout(unsigned long ul_conntrack)
272 {
273         struct nf_conn *ct = (void *)ul_conntrack;
274
275         if (!test_bit(IPS_DYING_BIT, &ct->status) &&
276             unlikely(nf_conntrack_event(IPCT_DESTROY, ct) < 0)) {
277                 /* destroy event was not delivered */
278                 nf_ct_delete_from_lists(ct);
279                 nf_ct_insert_dying_list(ct);
280                 return;
281         }
282         set_bit(IPS_DYING_BIT, &ct->status);
283         nf_ct_delete_from_lists(ct);
284         nf_ct_put(ct);
285 }
286
287 /*
288  * Warning :
289  * - Caller must take a reference on returned object
290  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
291  * OR
292  * - Caller must lock nf_conntrack_lock before calling this function
293  */
294 struct nf_conntrack_tuple_hash *
295 __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple)
296 {
297         struct nf_conntrack_tuple_hash *h;
298         struct hlist_nulls_node *n;
299         unsigned int hash = hash_conntrack(tuple);
300
301         /* Disable BHs the entire time since we normally need to disable them
302          * at least once for the stats anyway.
303          */
304         local_bh_disable();
305 begin:
306         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
307                 if (nf_ct_tuple_equal(tuple, &h->tuple)) {
308                         NF_CT_STAT_INC(net, found);
309                         local_bh_enable();
310                         return h;
311                 }
312                 NF_CT_STAT_INC(net, searched);
313         }
314         /*
315          * if the nulls value we got at the end of this lookup is
316          * not the expected one, we must restart lookup.
317          * We probably met an item that was moved to another chain.
318          */
319         if (get_nulls_value(n) != hash)
320                 goto begin;
321         local_bh_enable();
322
323         return NULL;
324 }
325 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
326
327 /* Find a connection corresponding to a tuple. */
328 struct nf_conntrack_tuple_hash *
329 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple)
330 {
331         struct nf_conntrack_tuple_hash *h;
332         struct nf_conn *ct;
333
334         rcu_read_lock();
335 begin:
336         h = __nf_conntrack_find(net, tuple);
337         if (h) {
338                 ct = nf_ct_tuplehash_to_ctrack(h);
339                 if (unlikely(nf_ct_is_dying(ct) ||
340                              !atomic_inc_not_zero(&ct->ct_general.use)))
341                         h = NULL;
342                 else {
343                         if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple))) {
344                                 nf_ct_put(ct);
345                                 goto begin;
346                         }
347                 }
348         }
349         rcu_read_unlock();
350
351         return h;
352 }
353 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
354
355 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
356                                        unsigned int hash,
357                                        unsigned int repl_hash)
358 {
359         struct net *net = nf_ct_net(ct);
360
361         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
362                            &net->ct.hash[hash]);
363         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
364                            &net->ct.hash[repl_hash]);
365 }
366
367 void nf_conntrack_hash_insert(struct nf_conn *ct)
368 {
369         unsigned int hash, repl_hash;
370
371         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
372         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
373
374         __nf_conntrack_hash_insert(ct, hash, repl_hash);
375 }
376 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
377
378 /* Confirm a connection given skb; places it in hash table */
379 int
380 __nf_conntrack_confirm(struct sk_buff *skb)
381 {
382         unsigned int hash, repl_hash;
383         struct nf_conntrack_tuple_hash *h;
384         struct nf_conn *ct;
385         struct nf_conn_help *help;
386         struct hlist_nulls_node *n;
387         enum ip_conntrack_info ctinfo;
388         struct net *net;
389
390         ct = nf_ct_get(skb, &ctinfo);
391         net = nf_ct_net(ct);
392
393         /* ipt_REJECT uses nf_conntrack_attach to attach related
394            ICMP/TCP RST packets in other direction.  Actual packet
395            which created connection will be IP_CT_NEW or for an
396            expected connection, IP_CT_RELATED. */
397         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
398                 return NF_ACCEPT;
399
400         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
401         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
402
403         /* We're not in hash table, and we refuse to set up related
404            connections for unconfirmed conns.  But packet copies and
405            REJECT will give spurious warnings here. */
406         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
407
408         /* No external references means noone else could have
409            confirmed us. */
410         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
411         pr_debug("Confirming conntrack %p\n", ct);
412
413         spin_lock_bh(&nf_conntrack_lock);
414
415         /* See if there's one in the list already, including reverse:
416            NAT could have grabbed it without realizing, since we're
417            not in the hash.  If there is, we lost race. */
418         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
419                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
420                                       &h->tuple))
421                         goto out;
422         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
423                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
424                                       &h->tuple))
425                         goto out;
426
427         /* Remove from unconfirmed list */
428         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
429
430         /* Timer relative to confirmation time, not original
431            setting time, otherwise we'd get timer wrap in
432            weird delay cases. */
433         ct->timeout.expires += jiffies;
434         add_timer(&ct->timeout);
435         atomic_inc(&ct->ct_general.use);
436         set_bit(IPS_CONFIRMED_BIT, &ct->status);
437
438         /* Since the lookup is lockless, hash insertion must be done after
439          * starting the timer and setting the CONFIRMED bit. The RCU barriers
440          * guarantee that no other CPU can find the conntrack before the above
441          * stores are visible.
442          */
443         __nf_conntrack_hash_insert(ct, hash, repl_hash);
444         NF_CT_STAT_INC(net, insert);
445         spin_unlock_bh(&nf_conntrack_lock);
446
447         help = nfct_help(ct);
448         if (help && help->helper)
449                 nf_conntrack_event_cache(IPCT_HELPER, ct);
450
451         nf_conntrack_event_cache(master_ct(ct) ?
452                                  IPCT_RELATED : IPCT_NEW, ct);
453         return NF_ACCEPT;
454
455 out:
456         NF_CT_STAT_INC(net, insert_failed);
457         spin_unlock_bh(&nf_conntrack_lock);
458         return NF_DROP;
459 }
460 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
461
462 /* Returns true if a connection correspondings to the tuple (required
463    for NAT). */
464 int
465 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
466                          const struct nf_conn *ignored_conntrack)
467 {
468         struct net *net = nf_ct_net(ignored_conntrack);
469         struct nf_conntrack_tuple_hash *h;
470         struct hlist_nulls_node *n;
471         unsigned int hash = hash_conntrack(tuple);
472
473         /* Disable BHs the entire time since we need to disable them at
474          * least once for the stats anyway.
475          */
476         rcu_read_lock_bh();
477         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
478                 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
479                     nf_ct_tuple_equal(tuple, &h->tuple)) {
480                         NF_CT_STAT_INC(net, found);
481                         rcu_read_unlock_bh();
482                         return 1;
483                 }
484                 NF_CT_STAT_INC(net, searched);
485         }
486         rcu_read_unlock_bh();
487
488         return 0;
489 }
490 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
491
492 #define NF_CT_EVICTION_RANGE    8
493
494 /* There's a small race here where we may free a just-assured
495    connection.  Too bad: we're in trouble anyway. */
496 static noinline int early_drop(struct net *net, unsigned int hash)
497 {
498         /* Use oldest entry, which is roughly LRU */
499         struct nf_conntrack_tuple_hash *h;
500         struct nf_conn *ct = NULL, *tmp;
501         struct hlist_nulls_node *n;
502         unsigned int i, cnt = 0;
503         int dropped = 0;
504
505         rcu_read_lock();
506         for (i = 0; i < nf_conntrack_htable_size; i++) {
507                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
508                                          hnnode) {
509                         tmp = nf_ct_tuplehash_to_ctrack(h);
510                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
511                                 ct = tmp;
512                         cnt++;
513                 }
514
515                 if (ct != NULL) {
516                         if (likely(!nf_ct_is_dying(ct) &&
517                                    atomic_inc_not_zero(&ct->ct_general.use)))
518                                 break;
519                         else
520                                 ct = NULL;
521                 }
522
523                 if (cnt >= NF_CT_EVICTION_RANGE)
524                         break;
525
526                 hash = (hash + 1) % nf_conntrack_htable_size;
527         }
528         rcu_read_unlock();
529
530         if (!ct)
531                 return dropped;
532
533         if (del_timer(&ct->timeout)) {
534                 death_by_timeout((unsigned long)ct);
535                 dropped = 1;
536                 NF_CT_STAT_INC_ATOMIC(net, early_drop);
537         }
538         nf_ct_put(ct);
539         return dropped;
540 }
541
542 struct nf_conn *nf_conntrack_alloc(struct net *net,
543                                    const struct nf_conntrack_tuple *orig,
544                                    const struct nf_conntrack_tuple *repl,
545                                    gfp_t gfp)
546 {
547         struct nf_conn *ct;
548
549         if (unlikely(!nf_conntrack_hash_rnd_initted)) {
550                 get_random_bytes(&nf_conntrack_hash_rnd,
551                                 sizeof(nf_conntrack_hash_rnd));
552                 nf_conntrack_hash_rnd_initted = 1;
553         }
554
555         /* We don't want any race condition at early drop stage */
556         atomic_inc(&net->ct.count);
557
558         if (nf_conntrack_max &&
559             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
560                 unsigned int hash = hash_conntrack(orig);
561                 if (!early_drop(net, hash)) {
562                         atomic_dec(&net->ct.count);
563                         if (net_ratelimit())
564                                 printk(KERN_WARNING
565                                        "nf_conntrack: table full, dropping"
566                                        " packet.\n");
567                         return ERR_PTR(-ENOMEM);
568                 }
569         }
570
571         /*
572          * Do not use kmem_cache_zalloc(), as this cache uses
573          * SLAB_DESTROY_BY_RCU.
574          */
575         ct = kmem_cache_alloc(nf_conntrack_cachep, gfp);
576         if (ct == NULL) {
577                 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
578                 atomic_dec(&net->ct.count);
579                 return ERR_PTR(-ENOMEM);
580         }
581         /*
582          * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next
583          * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged.
584          */
585         memset(&ct->tuplehash[IP_CT_DIR_MAX], 0,
586                sizeof(*ct) - offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX]));
587         spin_lock_init(&ct->lock);
588         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
589         ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
590         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
591         ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev = NULL;
592         /* Don't set timer yet: wait for confirmation */
593         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
594 #ifdef CONFIG_NET_NS
595         ct->ct_net = net;
596 #endif
597
598         /*
599          * changes to lookup keys must be done before setting refcnt to 1
600          */
601         smp_wmb();
602         atomic_set(&ct->ct_general.use, 1);
603         return ct;
604 }
605 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
606
607 void nf_conntrack_free(struct nf_conn *ct)
608 {
609         struct net *net = nf_ct_net(ct);
610
611         nf_ct_ext_destroy(ct);
612         atomic_dec(&net->ct.count);
613         nf_ct_ext_free(ct);
614         kmem_cache_free(nf_conntrack_cachep, ct);
615 }
616 EXPORT_SYMBOL_GPL(nf_conntrack_free);
617
618 /* Allocate a new conntrack: we return -ENOMEM if classification
619    failed due to stress.  Otherwise it really is unclassifiable. */
620 static struct nf_conntrack_tuple_hash *
621 init_conntrack(struct net *net,
622                const struct nf_conntrack_tuple *tuple,
623                struct nf_conntrack_l3proto *l3proto,
624                struct nf_conntrack_l4proto *l4proto,
625                struct sk_buff *skb,
626                unsigned int dataoff)
627 {
628         struct nf_conn *ct;
629         struct nf_conn_help *help;
630         struct nf_conntrack_tuple repl_tuple;
631         struct nf_conntrack_expect *exp;
632
633         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
634                 pr_debug("Can't invert tuple.\n");
635                 return NULL;
636         }
637
638         ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC);
639         if (IS_ERR(ct)) {
640                 pr_debug("Can't allocate conntrack.\n");
641                 return (struct nf_conntrack_tuple_hash *)ct;
642         }
643
644         if (!l4proto->new(ct, skb, dataoff)) {
645                 nf_conntrack_free(ct);
646                 pr_debug("init conntrack: can't track with proto module\n");
647                 return NULL;
648         }
649
650         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
651         nf_ct_ecache_ext_add(ct, GFP_ATOMIC);
652
653         spin_lock_bh(&nf_conntrack_lock);
654         exp = nf_ct_find_expectation(net, tuple);
655         if (exp) {
656                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
657                          ct, exp);
658                 /* Welcome, Mr. Bond.  We've been expecting you... */
659                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
660                 ct->master = exp->master;
661                 if (exp->helper) {
662                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
663                         if (help)
664                                 rcu_assign_pointer(help->helper, exp->helper);
665                 }
666
667 #ifdef CONFIG_NF_CONNTRACK_MARK
668                 ct->mark = exp->master->mark;
669 #endif
670 #ifdef CONFIG_NF_CONNTRACK_SECMARK
671                 ct->secmark = exp->master->secmark;
672 #endif
673                 nf_conntrack_get(&ct->master->ct_general);
674                 NF_CT_STAT_INC(net, expect_new);
675         } else {
676                 __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
677                 NF_CT_STAT_INC(net, new);
678         }
679
680         /* Overload tuple linked list to put us in unconfirmed list. */
681         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
682                        &net->ct.unconfirmed);
683
684         spin_unlock_bh(&nf_conntrack_lock);
685
686         if (exp) {
687                 if (exp->expectfn)
688                         exp->expectfn(ct, exp);
689                 nf_ct_expect_put(exp);
690         }
691
692         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
693 }
694
695 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
696 static inline struct nf_conn *
697 resolve_normal_ct(struct net *net,
698                   struct sk_buff *skb,
699                   unsigned int dataoff,
700                   u_int16_t l3num,
701                   u_int8_t protonum,
702                   struct nf_conntrack_l3proto *l3proto,
703                   struct nf_conntrack_l4proto *l4proto,
704                   int *set_reply,
705                   enum ip_conntrack_info *ctinfo)
706 {
707         struct nf_conntrack_tuple tuple;
708         struct nf_conntrack_tuple_hash *h;
709         struct nf_conn *ct;
710
711         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
712                              dataoff, l3num, protonum, &tuple, l3proto,
713                              l4proto)) {
714                 pr_debug("resolve_normal_ct: Can't get tuple\n");
715                 return NULL;
716         }
717
718         /* look for tuple match */
719         h = nf_conntrack_find_get(net, &tuple);
720         if (!h) {
721                 h = init_conntrack(net, &tuple, l3proto, l4proto, skb, dataoff);
722                 if (!h)
723                         return NULL;
724                 if (IS_ERR(h))
725                         return (void *)h;
726         }
727         ct = nf_ct_tuplehash_to_ctrack(h);
728
729         /* It exists; we have (non-exclusive) reference. */
730         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
731                 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
732                 /* Please set reply bit if this packet OK */
733                 *set_reply = 1;
734         } else {
735                 /* Once we've had two way comms, always ESTABLISHED. */
736                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
737                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
738                         *ctinfo = IP_CT_ESTABLISHED;
739                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
740                         pr_debug("nf_conntrack_in: related packet for %p\n",
741                                  ct);
742                         *ctinfo = IP_CT_RELATED;
743                 } else {
744                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
745                         *ctinfo = IP_CT_NEW;
746                 }
747                 *set_reply = 0;
748         }
749         skb->nfct = &ct->ct_general;
750         skb->nfctinfo = *ctinfo;
751         return ct;
752 }
753
754 unsigned int
755 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
756                 struct sk_buff *skb)
757 {
758         struct nf_conn *ct;
759         enum ip_conntrack_info ctinfo;
760         struct nf_conntrack_l3proto *l3proto;
761         struct nf_conntrack_l4proto *l4proto;
762         unsigned int dataoff;
763         u_int8_t protonum;
764         int set_reply = 0;
765         int ret;
766
767         /* Previously seen (loopback or untracked)?  Ignore. */
768         if (skb->nfct) {
769                 NF_CT_STAT_INC_ATOMIC(net, ignore);
770                 return NF_ACCEPT;
771         }
772
773         /* rcu_read_lock()ed by nf_hook_slow */
774         l3proto = __nf_ct_l3proto_find(pf);
775         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
776                                    &dataoff, &protonum);
777         if (ret <= 0) {
778                 pr_debug("not prepared to track yet or error occured\n");
779                 NF_CT_STAT_INC_ATOMIC(net, error);
780                 NF_CT_STAT_INC_ATOMIC(net, invalid);
781                 return -ret;
782         }
783
784         l4proto = __nf_ct_l4proto_find(pf, protonum);
785
786         /* It may be an special packet, error, unclean...
787          * inverse of the return code tells to the netfilter
788          * core what to do with the packet. */
789         if (l4proto->error != NULL) {
790                 ret = l4proto->error(net, skb, dataoff, &ctinfo, pf, hooknum);
791                 if (ret <= 0) {
792                         NF_CT_STAT_INC_ATOMIC(net, error);
793                         NF_CT_STAT_INC_ATOMIC(net, invalid);
794                         return -ret;
795                 }
796         }
797
798         ct = resolve_normal_ct(net, skb, dataoff, pf, protonum,
799                                l3proto, l4proto, &set_reply, &ctinfo);
800         if (!ct) {
801                 /* Not valid part of a connection */
802                 NF_CT_STAT_INC_ATOMIC(net, invalid);
803                 return NF_ACCEPT;
804         }
805
806         if (IS_ERR(ct)) {
807                 /* Too stressed to deal. */
808                 NF_CT_STAT_INC_ATOMIC(net, drop);
809                 return NF_DROP;
810         }
811
812         NF_CT_ASSERT(skb->nfct);
813
814         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
815         if (ret <= 0) {
816                 /* Invalid: inverse of the return code tells
817                  * the netfilter core what to do */
818                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
819                 nf_conntrack_put(skb->nfct);
820                 skb->nfct = NULL;
821                 NF_CT_STAT_INC_ATOMIC(net, invalid);
822                 if (ret == -NF_DROP)
823                         NF_CT_STAT_INC_ATOMIC(net, drop);
824                 return -ret;
825         }
826
827         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
828                 nf_conntrack_event_cache(IPCT_STATUS, ct);
829
830         return ret;
831 }
832 EXPORT_SYMBOL_GPL(nf_conntrack_in);
833
834 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
835                           const struct nf_conntrack_tuple *orig)
836 {
837         bool ret;
838
839         rcu_read_lock();
840         ret = nf_ct_invert_tuple(inverse, orig,
841                                  __nf_ct_l3proto_find(orig->src.l3num),
842                                  __nf_ct_l4proto_find(orig->src.l3num,
843                                                       orig->dst.protonum));
844         rcu_read_unlock();
845         return ret;
846 }
847 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
848
849 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
850    implicitly racy: see __nf_conntrack_confirm */
851 void nf_conntrack_alter_reply(struct nf_conn *ct,
852                               const struct nf_conntrack_tuple *newreply)
853 {
854         struct nf_conn_help *help = nfct_help(ct);
855
856         /* Should be unconfirmed, so not in hash table yet */
857         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
858
859         pr_debug("Altering reply tuple of %p to ", ct);
860         nf_ct_dump_tuple(newreply);
861
862         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
863         if (ct->master || (help && !hlist_empty(&help->expectations)))
864                 return;
865
866         rcu_read_lock();
867         __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
868         rcu_read_unlock();
869 }
870 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
871
872 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
873 void __nf_ct_refresh_acct(struct nf_conn *ct,
874                           enum ip_conntrack_info ctinfo,
875                           const struct sk_buff *skb,
876                           unsigned long extra_jiffies,
877                           int do_acct)
878 {
879         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
880         NF_CT_ASSERT(skb);
881
882         /* Only update if this is not a fixed timeout */
883         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
884                 goto acct;
885
886         /* If not in hash table, timer will not be active yet */
887         if (!nf_ct_is_confirmed(ct)) {
888                 ct->timeout.expires = extra_jiffies;
889         } else {
890                 unsigned long newtime = jiffies + extra_jiffies;
891
892                 /* Only update the timeout if the new timeout is at least
893                    HZ jiffies from the old timeout. Need del_timer for race
894                    avoidance (may already be dying). */
895                 if (newtime - ct->timeout.expires >= HZ)
896                         mod_timer_pending(&ct->timeout, newtime);
897         }
898
899 acct:
900         if (do_acct) {
901                 struct nf_conn_counter *acct;
902
903                 acct = nf_conn_acct_find(ct);
904                 if (acct) {
905                         spin_lock_bh(&ct->lock);
906                         acct[CTINFO2DIR(ctinfo)].packets++;
907                         acct[CTINFO2DIR(ctinfo)].bytes +=
908                                 skb->len - skb_network_offset(skb);
909                         spin_unlock_bh(&ct->lock);
910                 }
911         }
912 }
913 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
914
915 bool __nf_ct_kill_acct(struct nf_conn *ct,
916                        enum ip_conntrack_info ctinfo,
917                        const struct sk_buff *skb,
918                        int do_acct)
919 {
920         if (do_acct) {
921                 struct nf_conn_counter *acct;
922
923                 acct = nf_conn_acct_find(ct);
924                 if (acct) {
925                         spin_lock_bh(&ct->lock);
926                         acct[CTINFO2DIR(ctinfo)].packets++;
927                         acct[CTINFO2DIR(ctinfo)].bytes +=
928                                 skb->len - skb_network_offset(skb);
929                         spin_unlock_bh(&ct->lock);
930                 }
931         }
932
933         if (del_timer(&ct->timeout)) {
934                 ct->timeout.function((unsigned long)ct);
935                 return true;
936         }
937         return false;
938 }
939 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
940
941 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
942
943 #include <linux/netfilter/nfnetlink.h>
944 #include <linux/netfilter/nfnetlink_conntrack.h>
945 #include <linux/mutex.h>
946
947 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
948  * in ip_conntrack_core, since we don't want the protocols to autoload
949  * or depend on ctnetlink */
950 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
951                                const struct nf_conntrack_tuple *tuple)
952 {
953         NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
954         NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
955         return 0;
956
957 nla_put_failure:
958         return -1;
959 }
960 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
961
962 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
963         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
964         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
965 };
966 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
967
968 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
969                                struct nf_conntrack_tuple *t)
970 {
971         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
972                 return -EINVAL;
973
974         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
975         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
976
977         return 0;
978 }
979 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
980
981 int nf_ct_port_nlattr_tuple_size(void)
982 {
983         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
984 }
985 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
986 #endif
987
988 /* Used by ipt_REJECT and ip6t_REJECT. */
989 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
990 {
991         struct nf_conn *ct;
992         enum ip_conntrack_info ctinfo;
993
994         /* This ICMP is in reverse direction to the packet which caused it */
995         ct = nf_ct_get(skb, &ctinfo);
996         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
997                 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
998         else
999                 ctinfo = IP_CT_RELATED;
1000
1001         /* Attach to new skbuff, and increment count */
1002         nskb->nfct = &ct->ct_general;
1003         nskb->nfctinfo = ctinfo;
1004         nf_conntrack_get(nskb->nfct);
1005 }
1006
1007 /* Bring out ya dead! */
1008 static struct nf_conn *
1009 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1010                 void *data, unsigned int *bucket)
1011 {
1012         struct nf_conntrack_tuple_hash *h;
1013         struct nf_conn *ct;
1014         struct hlist_nulls_node *n;
1015
1016         spin_lock_bh(&nf_conntrack_lock);
1017         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
1018                 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
1019                         ct = nf_ct_tuplehash_to_ctrack(h);
1020                         if (iter(ct, data))
1021                                 goto found;
1022                 }
1023         }
1024         hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
1025                 ct = nf_ct_tuplehash_to_ctrack(h);
1026                 if (iter(ct, data))
1027                         set_bit(IPS_DYING_BIT, &ct->status);
1028         }
1029         spin_unlock_bh(&nf_conntrack_lock);
1030         return NULL;
1031 found:
1032         atomic_inc(&ct->ct_general.use);
1033         spin_unlock_bh(&nf_conntrack_lock);
1034         return ct;
1035 }
1036
1037 void nf_ct_iterate_cleanup(struct net *net,
1038                            int (*iter)(struct nf_conn *i, void *data),
1039                            void *data)
1040 {
1041         struct nf_conn *ct;
1042         unsigned int bucket = 0;
1043
1044         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1045                 /* Time to push up daises... */
1046                 if (del_timer(&ct->timeout))
1047                         death_by_timeout((unsigned long)ct);
1048                 /* ... else the timer will get him soon. */
1049
1050                 nf_ct_put(ct);
1051         }
1052 }
1053 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1054
1055 struct __nf_ct_flush_report {
1056         u32 pid;
1057         int report;
1058 };
1059
1060 static int kill_report(struct nf_conn *i, void *data)
1061 {
1062         struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
1063
1064         /* If we fail to deliver the event, death_by_timeout() will retry */
1065         if (nf_conntrack_event_report(IPCT_DESTROY, i,
1066                                       fr->pid, fr->report) < 0)
1067                 return 1;
1068
1069         /* Avoid the delivery of the destroy event in death_by_timeout(). */
1070         set_bit(IPS_DYING_BIT, &i->status);
1071         return 1;
1072 }
1073
1074 static int kill_all(struct nf_conn *i, void *data)
1075 {
1076         return 1;
1077 }
1078
1079 void nf_ct_free_hashtable(void *hash, int vmalloced, unsigned int size)
1080 {
1081         if (vmalloced)
1082                 vfree(hash);
1083         else
1084                 free_pages((unsigned long)hash,
1085                            get_order(sizeof(struct hlist_head) * size));
1086 }
1087 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1088
1089 void nf_conntrack_flush_report(struct net *net, u32 pid, int report)
1090 {
1091         struct __nf_ct_flush_report fr = {
1092                 .pid    = pid,
1093                 .report = report,
1094         };
1095         nf_ct_iterate_cleanup(net, kill_report, &fr);
1096 }
1097 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1098
1099 static void nf_ct_release_dying_list(struct net *net)
1100 {
1101         struct nf_conntrack_tuple_hash *h;
1102         struct nf_conn *ct;
1103         struct hlist_nulls_node *n;
1104
1105         spin_lock_bh(&nf_conntrack_lock);
1106         hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) {
1107                 ct = nf_ct_tuplehash_to_ctrack(h);
1108                 /* never fails to remove them, no listeners at this point */
1109                 nf_ct_kill(ct);
1110         }
1111         spin_unlock_bh(&nf_conntrack_lock);
1112 }
1113
1114 static void nf_conntrack_cleanup_init_net(void)
1115 {
1116         /* wait until all references to nf_conntrack_untracked are dropped */
1117         while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1118                 schedule();
1119
1120         nf_conntrack_helper_fini();
1121         nf_conntrack_proto_fini();
1122         kmem_cache_destroy(nf_conntrack_cachep);
1123 }
1124
1125 static void nf_conntrack_cleanup_net(struct net *net)
1126 {
1127  i_see_dead_people:
1128         nf_ct_iterate_cleanup(net, kill_all, NULL);
1129         nf_ct_release_dying_list(net);
1130         if (atomic_read(&net->ct.count) != 0) {
1131                 schedule();
1132                 goto i_see_dead_people;
1133         }
1134
1135         nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1136                              nf_conntrack_htable_size);
1137         nf_conntrack_ecache_fini(net);
1138         nf_conntrack_acct_fini(net);
1139         nf_conntrack_expect_fini(net);
1140         free_percpu(net->ct.stat);
1141 }
1142
1143 /* Mishearing the voices in his head, our hero wonders how he's
1144    supposed to kill the mall. */
1145 void nf_conntrack_cleanup(struct net *net)
1146 {
1147         if (net_eq(net, &init_net))
1148                 rcu_assign_pointer(ip_ct_attach, NULL);
1149
1150         /* This makes sure all current packets have passed through
1151            netfilter framework.  Roll on, two-stage module
1152            delete... */
1153         synchronize_net();
1154
1155         nf_conntrack_cleanup_net(net);
1156
1157         if (net_eq(net, &init_net)) {
1158                 rcu_assign_pointer(nf_ct_destroy, NULL);
1159                 nf_conntrack_cleanup_init_net();
1160         }
1161 }
1162
1163 void *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced, int nulls)
1164 {
1165         struct hlist_nulls_head *hash;
1166         unsigned int nr_slots, i;
1167         size_t sz;
1168
1169         *vmalloced = 0;
1170
1171         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1172         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1173         sz = nr_slots * sizeof(struct hlist_nulls_head);
1174         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1175                                         get_order(sz));
1176         if (!hash) {
1177                 *vmalloced = 1;
1178                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1179                 hash = __vmalloc(sz, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
1180         }
1181
1182         if (hash && nulls)
1183                 for (i = 0; i < nr_slots; i++)
1184                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1185
1186         return hash;
1187 }
1188 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1189
1190 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1191 {
1192         int i, bucket, vmalloced, old_vmalloced;
1193         unsigned int hashsize, old_size;
1194         int rnd;
1195         struct hlist_nulls_head *hash, *old_hash;
1196         struct nf_conntrack_tuple_hash *h;
1197
1198         /* On boot, we can set this without any fancy locking. */
1199         if (!nf_conntrack_htable_size)
1200                 return param_set_uint(val, kp);
1201
1202         hashsize = simple_strtoul(val, NULL, 0);
1203         if (!hashsize)
1204                 return -EINVAL;
1205
1206         hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced, 1);
1207         if (!hash)
1208                 return -ENOMEM;
1209
1210         /* We have to rehahs for the new table anyway, so we also can
1211          * use a newrandom seed */
1212         get_random_bytes(&rnd, sizeof(rnd));
1213
1214         /* Lookups in the old hash might happen in parallel, which means we
1215          * might get false negatives during connection lookup. New connections
1216          * created because of a false negative won't make it into the hash
1217          * though since that required taking the lock.
1218          */
1219         spin_lock_bh(&nf_conntrack_lock);
1220         for (i = 0; i < nf_conntrack_htable_size; i++) {
1221                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1222                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1223                                         struct nf_conntrack_tuple_hash, hnnode);
1224                         hlist_nulls_del_rcu(&h->hnnode);
1225                         bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1226                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1227                 }
1228         }
1229         old_size = nf_conntrack_htable_size;
1230         old_vmalloced = init_net.ct.hash_vmalloc;
1231         old_hash = init_net.ct.hash;
1232
1233         nf_conntrack_htable_size = hashsize;
1234         init_net.ct.hash_vmalloc = vmalloced;
1235         init_net.ct.hash = hash;
1236         nf_conntrack_hash_rnd = rnd;
1237         spin_unlock_bh(&nf_conntrack_lock);
1238
1239         nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1240         return 0;
1241 }
1242 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1243
1244 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1245                   &nf_conntrack_htable_size, 0600);
1246
1247 static int nf_conntrack_init_init_net(void)
1248 {
1249         int max_factor = 8;
1250         int ret;
1251
1252         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1253          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1254         if (!nf_conntrack_htable_size) {
1255                 nf_conntrack_htable_size
1256                         = (((totalram_pages << PAGE_SHIFT) / 16384)
1257                            / sizeof(struct hlist_head));
1258                 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1259                         nf_conntrack_htable_size = 16384;
1260                 if (nf_conntrack_htable_size < 32)
1261                         nf_conntrack_htable_size = 32;
1262
1263                 /* Use a max. factor of four by default to get the same max as
1264                  * with the old struct list_heads. When a table size is given
1265                  * we use the old value of 8 to avoid reducing the max.
1266                  * entries. */
1267                 max_factor = 4;
1268         }
1269         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1270
1271         printk("nf_conntrack version %s (%u buckets, %d max)\n",
1272                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1273                nf_conntrack_max);
1274
1275         nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1276                                                 sizeof(struct nf_conn),
1277                                                 0, SLAB_DESTROY_BY_RCU, NULL);
1278         if (!nf_conntrack_cachep) {
1279                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1280                 ret = -ENOMEM;
1281                 goto err_cache;
1282         }
1283
1284         ret = nf_conntrack_proto_init();
1285         if (ret < 0)
1286                 goto err_proto;
1287
1288         ret = nf_conntrack_helper_init();
1289         if (ret < 0)
1290                 goto err_helper;
1291
1292         /* Set up fake conntrack: to never be deleted, not in any hashes */
1293 #ifdef CONFIG_NET_NS
1294         nf_conntrack_untracked.ct_net = &init_net;
1295 #endif
1296         atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1297         /*  - and look it like as a confirmed connection */
1298         set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1299
1300         return 0;
1301
1302 err_helper:
1303         nf_conntrack_proto_fini();
1304 err_proto:
1305         kmem_cache_destroy(nf_conntrack_cachep);
1306 err_cache:
1307         return ret;
1308 }
1309
1310 /*
1311  * We need to use special "null" values, not used in hash table
1312  */
1313 #define UNCONFIRMED_NULLS_VAL   ((1<<30)+0)
1314 #define DYING_NULLS_VAL         ((1<<30)+1)
1315
1316 static int nf_conntrack_init_net(struct net *net)
1317 {
1318         int ret;
1319
1320         atomic_set(&net->ct.count, 0);
1321         INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
1322         INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL);
1323         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1324         if (!net->ct.stat) {
1325                 ret = -ENOMEM;
1326                 goto err_stat;
1327         }
1328         net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1329                                              &net->ct.hash_vmalloc, 1);
1330         if (!net->ct.hash) {
1331                 ret = -ENOMEM;
1332                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1333                 goto err_hash;
1334         }
1335         ret = nf_conntrack_expect_init(net);
1336         if (ret < 0)
1337                 goto err_expect;
1338         ret = nf_conntrack_acct_init(net);
1339         if (ret < 0)
1340                 goto err_acct;
1341         ret = nf_conntrack_ecache_init(net);
1342         if (ret < 0)
1343                 goto err_ecache;
1344
1345         return 0;
1346
1347 err_ecache:
1348         nf_conntrack_acct_fini(net);
1349 err_acct:
1350         nf_conntrack_expect_fini(net);
1351 err_expect:
1352         nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1353                              nf_conntrack_htable_size);
1354 err_hash:
1355         free_percpu(net->ct.stat);
1356 err_stat:
1357         return ret;
1358 }
1359
1360 s16 (*nf_ct_nat_offset)(const struct nf_conn *ct,
1361                         enum ip_conntrack_dir dir,
1362                         u32 seq);
1363 EXPORT_SYMBOL_GPL(nf_ct_nat_offset);
1364
1365 int nf_conntrack_init(struct net *net)
1366 {
1367         int ret;
1368
1369         if (net_eq(net, &init_net)) {
1370                 ret = nf_conntrack_init_init_net();
1371                 if (ret < 0)
1372                         goto out_init_net;
1373         }
1374         ret = nf_conntrack_init_net(net);
1375         if (ret < 0)
1376                 goto out_net;
1377
1378         if (net_eq(net, &init_net)) {
1379                 /* For use by REJECT target */
1380                 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1381                 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1382
1383                 /* Howto get NAT offsets */
1384                 rcu_assign_pointer(nf_ct_nat_offset, NULL);
1385         }
1386         return 0;
1387
1388 out_net:
1389         if (net_eq(net, &init_net))
1390                 nf_conntrack_cleanup_init_net();
1391 out_init_net:
1392         return ret;
1393 }