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