8cce814f6bee1f16d4a8d7d501b2ebff135ce0c0
[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
33 #include <net/netfilter/nf_conntrack.h>
34 #include <net/netfilter/nf_conntrack_l3proto.h>
35 #include <net/netfilter/nf_conntrack_l4proto.h>
36 #include <net/netfilter/nf_conntrack_expect.h>
37 #include <net/netfilter/nf_conntrack_helper.h>
38 #include <net/netfilter/nf_conntrack_core.h>
39 #include <net/netfilter/nf_conntrack_extend.h>
40
41 #define NF_CONNTRACK_VERSION    "0.5.0"
42
43 DEFINE_RWLOCK(nf_conntrack_lock);
44 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
45
46 /* nf_conntrack_standalone needs this */
47 atomic_t nf_conntrack_count = ATOMIC_INIT(0);
48 EXPORT_SYMBOL_GPL(nf_conntrack_count);
49
50 unsigned int nf_conntrack_htable_size __read_mostly;
51 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
52
53 int nf_conntrack_max __read_mostly;
54 EXPORT_SYMBOL_GPL(nf_conntrack_max);
55
56 struct hlist_head *nf_conntrack_hash __read_mostly;
57 EXPORT_SYMBOL_GPL(nf_conntrack_hash);
58
59 struct nf_conn nf_conntrack_untracked __read_mostly;
60 EXPORT_SYMBOL_GPL(nf_conntrack_untracked);
61
62 unsigned int nf_ct_log_invalid __read_mostly;
63 HLIST_HEAD(unconfirmed);
64 static int nf_conntrack_vmalloc __read_mostly;
65 static struct kmem_cache *nf_conntrack_cachep __read_mostly;
66 static unsigned int nf_conntrack_next_id;
67
68 DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat);
69 EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat);
70
71 static int nf_conntrack_hash_rnd_initted;
72 static unsigned int nf_conntrack_hash_rnd;
73
74 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
75                                   unsigned int size, unsigned int rnd)
76 {
77         unsigned int a, b;
78
79         a = jhash2(tuple->src.u3.all, ARRAY_SIZE(tuple->src.u3.all),
80                    (tuple->src.l3num << 16) | tuple->dst.protonum);
81         b = jhash2(tuple->dst.u3.all, ARRAY_SIZE(tuple->dst.u3.all),
82                    (tuple->src.u.all << 16) | tuple->dst.u.all);
83
84         return jhash_2words(a, b, rnd) % size;
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 int
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         NF_CT_TUPLE_U_BLANK(tuple);
104
105         tuple->src.l3num = l3num;
106         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
107                 return 0;
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 int nf_ct_get_tuplepr(const struct sk_buff *skb,
117                       unsigned int nhoff,
118                       u_int16_t l3num,
119                       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 0;
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 int
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         NF_CT_TUPLE_U_BLANK(inverse);
153
154         inverse->src.l3num = orig->src.l3num;
155         if (l3proto->invert_tuple(inverse, orig) == 0)
156                 return 0;
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_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
170         hlist_del(&ct->tuplehash[IP_CT_DIR_REPLY].hnode);
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 nf_conntrack_l4proto *l4proto;
181
182         pr_debug("destroy_conntrack(%p)\n", ct);
183         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
184         NF_CT_ASSERT(!timer_pending(&ct->timeout));
185
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(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num,
194                                        ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum);
195         if (l4proto && l4proto->destroy)
196                 l4proto->destroy(ct);
197
198         nf_ct_ext_destroy(ct);
199
200         rcu_read_unlock();
201
202         write_lock_bh(&nf_conntrack_lock);
203         /* Expectations will have been removed in clean_from_lists,
204          * except TFTP can create an expectation on the first packet,
205          * before connection is in the list, so we need to clean here,
206          * too. */
207         nf_ct_remove_expectations(ct);
208
209         /* We overload first tuple to link into unconfirmed list. */
210         if (!nf_ct_is_confirmed(ct)) {
211                 BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode));
212                 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
213         }
214
215         NF_CT_STAT_INC(delete);
216         write_unlock_bh(&nf_conntrack_lock);
217
218         if (ct->master)
219                 nf_ct_put(ct->master);
220
221         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
222         nf_conntrack_free(ct);
223 }
224
225 static void death_by_timeout(unsigned long ul_conntrack)
226 {
227         struct nf_conn *ct = (void *)ul_conntrack;
228         struct nf_conn_help *help = nfct_help(ct);
229         struct nf_conntrack_helper *helper;
230
231         if (help) {
232                 rcu_read_lock();
233                 helper = rcu_dereference(help->helper);
234                 if (helper && helper->destroy)
235                         helper->destroy(ct);
236                 rcu_read_unlock();
237         }
238
239         write_lock_bh(&nf_conntrack_lock);
240         /* Inside lock so preempt is disabled on module removal path.
241          * Otherwise we can get spurious warnings. */
242         NF_CT_STAT_INC(delete_list);
243         clean_from_lists(ct);
244         write_unlock_bh(&nf_conntrack_lock);
245         nf_ct_put(ct);
246 }
247
248 struct nf_conntrack_tuple_hash *
249 __nf_conntrack_find(const struct nf_conntrack_tuple *tuple,
250                     const struct nf_conn *ignored_conntrack)
251 {
252         struct nf_conntrack_tuple_hash *h;
253         struct hlist_node *n;
254         unsigned int hash = hash_conntrack(tuple);
255
256         hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode) {
257                 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
258                     nf_ct_tuple_equal(tuple, &h->tuple)) {
259                         NF_CT_STAT_INC(found);
260                         return h;
261                 }
262                 NF_CT_STAT_INC(searched);
263         }
264
265         return NULL;
266 }
267 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
268
269 /* Find a connection corresponding to a tuple. */
270 struct nf_conntrack_tuple_hash *
271 nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple)
272 {
273         struct nf_conntrack_tuple_hash *h;
274
275         read_lock_bh(&nf_conntrack_lock);
276         h = __nf_conntrack_find(tuple, NULL);
277         if (h)
278                 atomic_inc(&nf_ct_tuplehash_to_ctrack(h)->ct_general.use);
279         read_unlock_bh(&nf_conntrack_lock);
280
281         return h;
282 }
283 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
284
285 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
286                                        unsigned int hash,
287                                        unsigned int repl_hash)
288 {
289         ct->id = ++nf_conntrack_next_id;
290         hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
291                        &nf_conntrack_hash[hash]);
292         hlist_add_head(&ct->tuplehash[IP_CT_DIR_REPLY].hnode,
293                        &nf_conntrack_hash[repl_hash]);
294 }
295
296 void nf_conntrack_hash_insert(struct nf_conn *ct)
297 {
298         unsigned int hash, repl_hash;
299
300         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
301         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
302
303         write_lock_bh(&nf_conntrack_lock);
304         __nf_conntrack_hash_insert(ct, hash, repl_hash);
305         write_unlock_bh(&nf_conntrack_lock);
306 }
307 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
308
309 /* Confirm a connection given skb; places it in hash table */
310 int
311 __nf_conntrack_confirm(struct sk_buff **pskb)
312 {
313         unsigned int hash, repl_hash;
314         struct nf_conntrack_tuple_hash *h;
315         struct nf_conn *ct;
316         struct nf_conn_help *help;
317         struct hlist_node *n;
318         enum ip_conntrack_info ctinfo;
319
320         ct = nf_ct_get(*pskb, &ctinfo);
321
322         /* ipt_REJECT uses nf_conntrack_attach to attach related
323            ICMP/TCP RST packets in other direction.  Actual packet
324            which created connection will be IP_CT_NEW or for an
325            expected connection, IP_CT_RELATED. */
326         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
327                 return NF_ACCEPT;
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         /* We're not in hash table, and we refuse to set up related
333            connections for unconfirmed conns.  But packet copies and
334            REJECT will give spurious warnings here. */
335         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
336
337         /* No external references means noone else could have
338            confirmed us. */
339         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
340         pr_debug("Confirming conntrack %p\n", ct);
341
342         write_lock_bh(&nf_conntrack_lock);
343
344         /* See if there's one in the list already, including reverse:
345            NAT could have grabbed it without realizing, since we're
346            not in the hash.  If there is, we lost race. */
347         hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode)
348                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
349                                       &h->tuple))
350                         goto out;
351         hlist_for_each_entry(h, n, &nf_conntrack_hash[repl_hash], hnode)
352                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
353                                       &h->tuple))
354                         goto out;
355
356         /* Remove from unconfirmed list */
357         hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
358
359         __nf_conntrack_hash_insert(ct, hash, repl_hash);
360         /* Timer relative to confirmation time, not original
361            setting time, otherwise we'd get timer wrap in
362            weird delay cases. */
363         ct->timeout.expires += jiffies;
364         add_timer(&ct->timeout);
365         atomic_inc(&ct->ct_general.use);
366         set_bit(IPS_CONFIRMED_BIT, &ct->status);
367         NF_CT_STAT_INC(insert);
368         write_unlock_bh(&nf_conntrack_lock);
369         help = nfct_help(ct);
370         if (help && help->helper)
371                 nf_conntrack_event_cache(IPCT_HELPER, *pskb);
372 #ifdef CONFIG_NF_NAT_NEEDED
373         if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
374             test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
375                 nf_conntrack_event_cache(IPCT_NATINFO, *pskb);
376 #endif
377         nf_conntrack_event_cache(master_ct(ct) ?
378                                  IPCT_RELATED : IPCT_NEW, *pskb);
379         return NF_ACCEPT;
380
381 out:
382         NF_CT_STAT_INC(insert_failed);
383         write_unlock_bh(&nf_conntrack_lock);
384         return NF_DROP;
385 }
386 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
387
388 /* Returns true if a connection correspondings to the tuple (required
389    for NAT). */
390 int
391 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
392                          const struct nf_conn *ignored_conntrack)
393 {
394         struct nf_conntrack_tuple_hash *h;
395
396         read_lock_bh(&nf_conntrack_lock);
397         h = __nf_conntrack_find(tuple, ignored_conntrack);
398         read_unlock_bh(&nf_conntrack_lock);
399
400         return h != NULL;
401 }
402 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
403
404 #define NF_CT_EVICTION_RANGE    8
405
406 /* There's a small race here where we may free a just-assured
407    connection.  Too bad: we're in trouble anyway. */
408 static int early_drop(unsigned int hash)
409 {
410         /* Use oldest entry, which is roughly LRU */
411         struct nf_conntrack_tuple_hash *h;
412         struct nf_conn *ct = NULL, *tmp;
413         struct hlist_node *n;
414         unsigned int i, cnt = 0;
415         int dropped = 0;
416
417         read_lock_bh(&nf_conntrack_lock);
418         for (i = 0; i < nf_conntrack_htable_size; i++) {
419                 hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode) {
420                         tmp = nf_ct_tuplehash_to_ctrack(h);
421                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
422                                 ct = tmp;
423                         cnt++;
424                 }
425                 if (ct || cnt >= NF_CT_EVICTION_RANGE)
426                         break;
427                 hash = (hash + 1) % nf_conntrack_htable_size;
428         }
429         if (ct)
430                 atomic_inc(&ct->ct_general.use);
431         read_unlock_bh(&nf_conntrack_lock);
432
433         if (!ct)
434                 return dropped;
435
436         if (del_timer(&ct->timeout)) {
437                 death_by_timeout((unsigned long)ct);
438                 dropped = 1;
439                 NF_CT_STAT_INC_ATOMIC(early_drop);
440         }
441         nf_ct_put(ct);
442         return dropped;
443 }
444
445 struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
446                                    const struct nf_conntrack_tuple *repl)
447 {
448         struct nf_conn *conntrack = NULL;
449
450         if (unlikely(!nf_conntrack_hash_rnd_initted)) {
451                 get_random_bytes(&nf_conntrack_hash_rnd, 4);
452                 nf_conntrack_hash_rnd_initted = 1;
453         }
454
455         /* We don't want any race condition at early drop stage */
456         atomic_inc(&nf_conntrack_count);
457
458         if (nf_conntrack_max
459             && atomic_read(&nf_conntrack_count) > nf_conntrack_max) {
460                 unsigned int hash = hash_conntrack(orig);
461                 if (!early_drop(hash)) {
462                         atomic_dec(&nf_conntrack_count);
463                         if (net_ratelimit())
464                                 printk(KERN_WARNING
465                                        "nf_conntrack: table full, dropping"
466                                        " packet.\n");
467                         return ERR_PTR(-ENOMEM);
468                 }
469         }
470
471         conntrack = kmem_cache_zalloc(nf_conntrack_cachep, GFP_ATOMIC);
472         if (conntrack == NULL) {
473                 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
474                 atomic_dec(&nf_conntrack_count);
475                 return ERR_PTR(-ENOMEM);
476         }
477
478         atomic_set(&conntrack->ct_general.use, 1);
479         conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
480         conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
481         /* Don't set timer yet: wait for confirmation */
482         setup_timer(&conntrack->timeout, death_by_timeout,
483                     (unsigned long)conntrack);
484
485         return conntrack;
486 }
487 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
488
489 void nf_conntrack_free(struct nf_conn *conntrack)
490 {
491         nf_ct_ext_free(conntrack);
492         kmem_cache_free(nf_conntrack_cachep, conntrack);
493         atomic_dec(&nf_conntrack_count);
494 }
495 EXPORT_SYMBOL_GPL(nf_conntrack_free);
496
497 /* Allocate a new conntrack: we return -ENOMEM if classification
498    failed due to stress.  Otherwise it really is unclassifiable. */
499 static struct nf_conntrack_tuple_hash *
500 init_conntrack(const struct nf_conntrack_tuple *tuple,
501                struct nf_conntrack_l3proto *l3proto,
502                struct nf_conntrack_l4proto *l4proto,
503                struct sk_buff *skb,
504                unsigned int dataoff)
505 {
506         struct nf_conn *conntrack;
507         struct nf_conn_help *help;
508         struct nf_conntrack_tuple repl_tuple;
509         struct nf_conntrack_expect *exp;
510
511         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
512                 pr_debug("Can't invert tuple.\n");
513                 return NULL;
514         }
515
516         conntrack = nf_conntrack_alloc(tuple, &repl_tuple);
517         if (conntrack == NULL || IS_ERR(conntrack)) {
518                 pr_debug("Can't allocate conntrack.\n");
519                 return (struct nf_conntrack_tuple_hash *)conntrack;
520         }
521
522         if (!l4proto->new(conntrack, skb, dataoff)) {
523                 nf_conntrack_free(conntrack);
524                 pr_debug("init conntrack: can't track with proto module\n");
525                 return NULL;
526         }
527
528         write_lock_bh(&nf_conntrack_lock);
529         exp = nf_ct_find_expectation(tuple);
530         if (exp) {
531                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
532                          conntrack, exp);
533                 /* Welcome, Mr. Bond.  We've been expecting you... */
534                 __set_bit(IPS_EXPECTED_BIT, &conntrack->status);
535                 conntrack->master = exp->master;
536                 if (exp->helper) {
537                         help = nf_ct_helper_ext_add(conntrack, GFP_ATOMIC);
538                         if (help)
539                                 rcu_assign_pointer(help->helper, exp->helper);
540                 }
541
542 #ifdef CONFIG_NF_CONNTRACK_MARK
543                 conntrack->mark = exp->master->mark;
544 #endif
545 #ifdef CONFIG_NF_CONNTRACK_SECMARK
546                 conntrack->secmark = exp->master->secmark;
547 #endif
548                 nf_conntrack_get(&conntrack->master->ct_general);
549                 NF_CT_STAT_INC(expect_new);
550         } else {
551                 struct nf_conntrack_helper *helper;
552
553                 helper = __nf_ct_helper_find(&repl_tuple);
554                 if (helper) {
555                         help = nf_ct_helper_ext_add(conntrack, GFP_ATOMIC);
556                         if (help)
557                                 rcu_assign_pointer(help->helper, helper);
558                 }
559                 NF_CT_STAT_INC(new);
560         }
561
562         /* Overload tuple linked list to put us in unconfirmed list. */
563         hlist_add_head(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
564                        &unconfirmed);
565
566         write_unlock_bh(&nf_conntrack_lock);
567
568         if (exp) {
569                 if (exp->expectfn)
570                         exp->expectfn(conntrack, exp);
571                 nf_ct_expect_put(exp);
572         }
573
574         return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL];
575 }
576
577 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
578 static inline struct nf_conn *
579 resolve_normal_ct(struct sk_buff *skb,
580                   unsigned int dataoff,
581                   u_int16_t l3num,
582                   u_int8_t protonum,
583                   struct nf_conntrack_l3proto *l3proto,
584                   struct nf_conntrack_l4proto *l4proto,
585                   int *set_reply,
586                   enum ip_conntrack_info *ctinfo)
587 {
588         struct nf_conntrack_tuple tuple;
589         struct nf_conntrack_tuple_hash *h;
590         struct nf_conn *ct;
591
592         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
593                              dataoff, l3num, protonum, &tuple, l3proto,
594                              l4proto)) {
595                 pr_debug("resolve_normal_ct: Can't get tuple\n");
596                 return NULL;
597         }
598
599         /* look for tuple match */
600         h = nf_conntrack_find_get(&tuple);
601         if (!h) {
602                 h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff);
603                 if (!h)
604                         return NULL;
605                 if (IS_ERR(h))
606                         return (void *)h;
607         }
608         ct = nf_ct_tuplehash_to_ctrack(h);
609
610         /* It exists; we have (non-exclusive) reference. */
611         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
612                 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
613                 /* Please set reply bit if this packet OK */
614                 *set_reply = 1;
615         } else {
616                 /* Once we've had two way comms, always ESTABLISHED. */
617                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
618                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
619                         *ctinfo = IP_CT_ESTABLISHED;
620                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
621                         pr_debug("nf_conntrack_in: related packet for %p\n",
622                                  ct);
623                         *ctinfo = IP_CT_RELATED;
624                 } else {
625                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
626                         *ctinfo = IP_CT_NEW;
627                 }
628                 *set_reply = 0;
629         }
630         skb->nfct = &ct->ct_general;
631         skb->nfctinfo = *ctinfo;
632         return ct;
633 }
634
635 unsigned int
636 nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff **pskb)
637 {
638         struct nf_conn *ct;
639         enum ip_conntrack_info ctinfo;
640         struct nf_conntrack_l3proto *l3proto;
641         struct nf_conntrack_l4proto *l4proto;
642         unsigned int dataoff;
643         u_int8_t protonum;
644         int set_reply = 0;
645         int ret;
646
647         /* Previously seen (loopback or untracked)?  Ignore. */
648         if ((*pskb)->nfct) {
649                 NF_CT_STAT_INC_ATOMIC(ignore);
650                 return NF_ACCEPT;
651         }
652
653         /* rcu_read_lock()ed by nf_hook_slow */
654         l3proto = __nf_ct_l3proto_find((u_int16_t)pf);
655         ret = l3proto->get_l4proto(*pskb, skb_network_offset(*pskb),
656                                    &dataoff, &protonum);
657         if (ret <= 0) {
658                 pr_debug("not prepared to track yet or error occured\n");
659                 NF_CT_STAT_INC_ATOMIC(error);
660                 NF_CT_STAT_INC_ATOMIC(invalid);
661                 return -ret;
662         }
663
664         l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum);
665
666         /* It may be an special packet, error, unclean...
667          * inverse of the return code tells to the netfilter
668          * core what to do with the packet. */
669         if (l4proto->error != NULL &&
670             (ret = l4proto->error(*pskb, dataoff, &ctinfo, pf, hooknum)) <= 0) {
671                 NF_CT_STAT_INC_ATOMIC(error);
672                 NF_CT_STAT_INC_ATOMIC(invalid);
673                 return -ret;
674         }
675
676         ct = resolve_normal_ct(*pskb, dataoff, pf, protonum, l3proto, l4proto,
677                                &set_reply, &ctinfo);
678         if (!ct) {
679                 /* Not valid part of a connection */
680                 NF_CT_STAT_INC_ATOMIC(invalid);
681                 return NF_ACCEPT;
682         }
683
684         if (IS_ERR(ct)) {
685                 /* Too stressed to deal. */
686                 NF_CT_STAT_INC_ATOMIC(drop);
687                 return NF_DROP;
688         }
689
690         NF_CT_ASSERT((*pskb)->nfct);
691
692         ret = l4proto->packet(ct, *pskb, dataoff, ctinfo, pf, hooknum);
693         if (ret < 0) {
694                 /* Invalid: inverse of the return code tells
695                  * the netfilter core what to do */
696                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
697                 nf_conntrack_put((*pskb)->nfct);
698                 (*pskb)->nfct = NULL;
699                 NF_CT_STAT_INC_ATOMIC(invalid);
700                 return -ret;
701         }
702
703         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
704                 nf_conntrack_event_cache(IPCT_STATUS, *pskb);
705
706         return ret;
707 }
708 EXPORT_SYMBOL_GPL(nf_conntrack_in);
709
710 int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
711                          const struct nf_conntrack_tuple *orig)
712 {
713         int ret;
714
715         rcu_read_lock();
716         ret = nf_ct_invert_tuple(inverse, orig,
717                                  __nf_ct_l3proto_find(orig->src.l3num),
718                                  __nf_ct_l4proto_find(orig->src.l3num,
719                                                       orig->dst.protonum));
720         rcu_read_unlock();
721         return ret;
722 }
723 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
724
725 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
726    implicitly racy: see __nf_conntrack_confirm */
727 void nf_conntrack_alter_reply(struct nf_conn *ct,
728                               const struct nf_conntrack_tuple *newreply)
729 {
730         struct nf_conn_help *help = nfct_help(ct);
731         struct nf_conntrack_helper *helper;
732
733         write_lock_bh(&nf_conntrack_lock);
734         /* Should be unconfirmed, so not in hash table yet */
735         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
736
737         pr_debug("Altering reply tuple of %p to ", ct);
738         NF_CT_DUMP_TUPLE(newreply);
739
740         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
741         if (ct->master || (help && help->expecting != 0))
742                 goto out;
743
744         helper = __nf_ct_helper_find(newreply);
745         if (helper == NULL) {
746                 if (help)
747                         rcu_assign_pointer(help->helper, NULL);
748                 goto out;
749         }
750
751         if (help == NULL) {
752                 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
753                 if (help == NULL)
754                         goto out;
755         } else {
756                 memset(&help->help, 0, sizeof(help->help));
757         }
758
759         rcu_assign_pointer(help->helper, helper);
760 out:
761         write_unlock_bh(&nf_conntrack_lock);
762 }
763 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
764
765 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
766 void __nf_ct_refresh_acct(struct nf_conn *ct,
767                           enum ip_conntrack_info ctinfo,
768                           const struct sk_buff *skb,
769                           unsigned long extra_jiffies,
770                           int do_acct)
771 {
772         int event = 0;
773
774         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
775         NF_CT_ASSERT(skb);
776
777         write_lock_bh(&nf_conntrack_lock);
778
779         /* Only update if this is not a fixed timeout */
780         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) {
781                 write_unlock_bh(&nf_conntrack_lock);
782                 return;
783         }
784
785         /* If not in hash table, timer will not be active yet */
786         if (!nf_ct_is_confirmed(ct)) {
787                 ct->timeout.expires = extra_jiffies;
788                 event = IPCT_REFRESH;
789         } else {
790                 unsigned long newtime = jiffies + extra_jiffies;
791
792                 /* Only update the timeout if the new timeout is at least
793                    HZ jiffies from the old timeout. Need del_timer for race
794                    avoidance (may already be dying). */
795                 if (newtime - ct->timeout.expires >= HZ
796                     && del_timer(&ct->timeout)) {
797                         ct->timeout.expires = newtime;
798                         add_timer(&ct->timeout);
799                         event = IPCT_REFRESH;
800                 }
801         }
802
803 #ifdef CONFIG_NF_CT_ACCT
804         if (do_acct) {
805                 ct->counters[CTINFO2DIR(ctinfo)].packets++;
806                 ct->counters[CTINFO2DIR(ctinfo)].bytes +=
807                         skb->len - skb_network_offset(skb);
808
809                 if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000)
810                     || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000))
811                         event |= IPCT_COUNTER_FILLING;
812         }
813 #endif
814
815         write_unlock_bh(&nf_conntrack_lock);
816
817         /* must be unlocked when calling event cache */
818         if (event)
819                 nf_conntrack_event_cache(event, skb);
820 }
821 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
822
823 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
824
825 #include <linux/netfilter/nfnetlink.h>
826 #include <linux/netfilter/nfnetlink_conntrack.h>
827 #include <linux/mutex.h>
828
829
830 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
831  * in ip_conntrack_core, since we don't want the protocols to autoload
832  * or depend on ctnetlink */
833 int nf_ct_port_tuple_to_nfattr(struct sk_buff *skb,
834                                const struct nf_conntrack_tuple *tuple)
835 {
836         NFA_PUT(skb, CTA_PROTO_SRC_PORT, sizeof(u_int16_t),
837                 &tuple->src.u.tcp.port);
838         NFA_PUT(skb, CTA_PROTO_DST_PORT, sizeof(u_int16_t),
839                 &tuple->dst.u.tcp.port);
840         return 0;
841
842 nfattr_failure:
843         return -1;
844 }
845 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nfattr);
846
847 static const size_t cta_min_proto[CTA_PROTO_MAX] = {
848         [CTA_PROTO_SRC_PORT-1]  = sizeof(u_int16_t),
849         [CTA_PROTO_DST_PORT-1]  = sizeof(u_int16_t)
850 };
851
852 int nf_ct_port_nfattr_to_tuple(struct nfattr *tb[],
853                                struct nf_conntrack_tuple *t)
854 {
855         if (!tb[CTA_PROTO_SRC_PORT-1] || !tb[CTA_PROTO_DST_PORT-1])
856                 return -EINVAL;
857
858         if (nfattr_bad_size(tb, CTA_PROTO_MAX, cta_min_proto))
859                 return -EINVAL;
860
861         t->src.u.tcp.port = *(__be16 *)NFA_DATA(tb[CTA_PROTO_SRC_PORT-1]);
862         t->dst.u.tcp.port = *(__be16 *)NFA_DATA(tb[CTA_PROTO_DST_PORT-1]);
863
864         return 0;
865 }
866 EXPORT_SYMBOL_GPL(nf_ct_port_nfattr_to_tuple);
867 #endif
868
869 /* Used by ipt_REJECT and ip6t_REJECT. */
870 void __nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
871 {
872         struct nf_conn *ct;
873         enum ip_conntrack_info ctinfo;
874
875         /* This ICMP is in reverse direction to the packet which caused it */
876         ct = nf_ct_get(skb, &ctinfo);
877         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
878                 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
879         else
880                 ctinfo = IP_CT_RELATED;
881
882         /* Attach to new skbuff, and increment count */
883         nskb->nfct = &ct->ct_general;
884         nskb->nfctinfo = ctinfo;
885         nf_conntrack_get(nskb->nfct);
886 }
887 EXPORT_SYMBOL_GPL(__nf_conntrack_attach);
888
889 static inline int
890 do_iter(const struct nf_conntrack_tuple_hash *i,
891         int (*iter)(struct nf_conn *i, void *data),
892         void *data)
893 {
894         return iter(nf_ct_tuplehash_to_ctrack(i), data);
895 }
896
897 /* Bring out ya dead! */
898 static struct nf_conn *
899 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
900                 void *data, unsigned int *bucket)
901 {
902         struct nf_conntrack_tuple_hash *h;
903         struct nf_conn *ct;
904         struct hlist_node *n;
905
906         write_lock_bh(&nf_conntrack_lock);
907         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
908                 hlist_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnode) {
909                         ct = nf_ct_tuplehash_to_ctrack(h);
910                         if (iter(ct, data))
911                                 goto found;
912                 }
913         }
914         hlist_for_each_entry(h, n, &unconfirmed, hnode) {
915                 ct = nf_ct_tuplehash_to_ctrack(h);
916                 if (iter(ct, data))
917                         set_bit(IPS_DYING_BIT, &ct->status);
918         }
919         write_unlock_bh(&nf_conntrack_lock);
920         return NULL;
921 found:
922         atomic_inc(&ct->ct_general.use);
923         write_unlock_bh(&nf_conntrack_lock);
924         return ct;
925 }
926
927 void
928 nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data)
929 {
930         struct nf_conn *ct;
931         unsigned int bucket = 0;
932
933         while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
934                 /* Time to push up daises... */
935                 if (del_timer(&ct->timeout))
936                         death_by_timeout((unsigned long)ct);
937                 /* ... else the timer will get him soon. */
938
939                 nf_ct_put(ct);
940         }
941 }
942 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
943
944 static int kill_all(struct nf_conn *i, void *data)
945 {
946         return 1;
947 }
948
949 void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, int size)
950 {
951         if (vmalloced)
952                 vfree(hash);
953         else
954                 free_pages((unsigned long)hash,
955                            get_order(sizeof(struct hlist_head) * size));
956 }
957 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
958
959 void nf_conntrack_flush(void)
960 {
961         nf_ct_iterate_cleanup(kill_all, NULL);
962 }
963 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
964
965 /* Mishearing the voices in his head, our hero wonders how he's
966    supposed to kill the mall. */
967 void nf_conntrack_cleanup(void)
968 {
969         rcu_assign_pointer(ip_ct_attach, NULL);
970
971         /* This makes sure all current packets have passed through
972            netfilter framework.  Roll on, two-stage module
973            delete... */
974         synchronize_net();
975
976         nf_ct_event_cache_flush();
977  i_see_dead_people:
978         nf_conntrack_flush();
979         if (atomic_read(&nf_conntrack_count) != 0) {
980                 schedule();
981                 goto i_see_dead_people;
982         }
983         /* wait until all references to nf_conntrack_untracked are dropped */
984         while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
985                 schedule();
986
987         rcu_assign_pointer(nf_ct_destroy, NULL);
988
989         kmem_cache_destroy(nf_conntrack_cachep);
990         nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
991                              nf_conntrack_htable_size);
992
993         nf_conntrack_proto_fini();
994         nf_conntrack_helper_fini();
995         nf_conntrack_expect_fini();
996 }
997
998 struct hlist_head *nf_ct_alloc_hashtable(int *sizep, int *vmalloced)
999 {
1000         struct hlist_head *hash;
1001         unsigned int size, i;
1002
1003         *vmalloced = 0;
1004
1005         size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head));
1006         hash = (void*)__get_free_pages(GFP_KERNEL,
1007                                        get_order(sizeof(struct hlist_head)
1008                                                  * size));
1009         if (!hash) {
1010                 *vmalloced = 1;
1011                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1012                 hash = vmalloc(sizeof(struct hlist_head) * size);
1013         }
1014
1015         if (hash)
1016                 for (i = 0; i < size; i++)
1017                         INIT_HLIST_HEAD(&hash[i]);
1018
1019         return hash;
1020 }
1021 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1022
1023 int set_hashsize(const char *val, struct kernel_param *kp)
1024 {
1025         int i, bucket, hashsize, vmalloced;
1026         int old_vmalloced, old_size;
1027         int rnd;
1028         struct hlist_head *hash, *old_hash;
1029         struct nf_conntrack_tuple_hash *h;
1030
1031         /* On boot, we can set this without any fancy locking. */
1032         if (!nf_conntrack_htable_size)
1033                 return param_set_uint(val, kp);
1034
1035         hashsize = simple_strtol(val, NULL, 0);
1036         if (!hashsize)
1037                 return -EINVAL;
1038
1039         hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced);
1040         if (!hash)
1041                 return -ENOMEM;
1042
1043         /* We have to rehahs for the new table anyway, so we also can
1044          * use a newrandom seed */
1045         get_random_bytes(&rnd, 4);
1046
1047         write_lock_bh(&nf_conntrack_lock);
1048         for (i = 0; i < nf_conntrack_htable_size; i++) {
1049                 while (!hlist_empty(&nf_conntrack_hash[i])) {
1050                         h = hlist_entry(nf_conntrack_hash[i].first,
1051                                         struct nf_conntrack_tuple_hash, hnode);
1052                         hlist_del(&h->hnode);
1053                         bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1054                         hlist_add_head(&h->hnode, &hash[bucket]);
1055                 }
1056         }
1057         old_size = nf_conntrack_htable_size;
1058         old_vmalloced = nf_conntrack_vmalloc;
1059         old_hash = nf_conntrack_hash;
1060
1061         nf_conntrack_htable_size = hashsize;
1062         nf_conntrack_vmalloc = vmalloced;
1063         nf_conntrack_hash = hash;
1064         nf_conntrack_hash_rnd = rnd;
1065         write_unlock_bh(&nf_conntrack_lock);
1066
1067         nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1068         return 0;
1069 }
1070
1071 module_param_call(hashsize, set_hashsize, param_get_uint,
1072                   &nf_conntrack_htable_size, 0600);
1073
1074 int __init nf_conntrack_init(void)
1075 {
1076         int max_factor = 8;
1077         int ret;
1078
1079         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1080          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1081         if (!nf_conntrack_htable_size) {
1082                 nf_conntrack_htable_size
1083                         = (((num_physpages << PAGE_SHIFT) / 16384)
1084                            / sizeof(struct hlist_head));
1085                 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1086                         nf_conntrack_htable_size = 16384;
1087                 if (nf_conntrack_htable_size < 32)
1088                         nf_conntrack_htable_size = 32;
1089
1090                 /* Use a max. factor of four by default to get the same max as
1091                  * with the old struct list_heads. When a table size is given
1092                  * we use the old value of 8 to avoid reducing the max.
1093                  * entries. */
1094                 max_factor = 4;
1095         }
1096         nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1097                                                   &nf_conntrack_vmalloc);
1098         if (!nf_conntrack_hash) {
1099                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1100                 goto err_out;
1101         }
1102
1103         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1104
1105         printk("nf_conntrack version %s (%u buckets, %d max)\n",
1106                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1107                nf_conntrack_max);
1108
1109         nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1110                                                 sizeof(struct nf_conn),
1111                                                 0, 0, NULL, NULL);
1112         if (!nf_conntrack_cachep) {
1113                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1114                 goto err_free_hash;
1115         }
1116
1117         ret = nf_conntrack_proto_init();
1118         if (ret < 0)
1119                 goto err_free_conntrack_slab;
1120
1121         ret = nf_conntrack_expect_init();
1122         if (ret < 0)
1123                 goto out_fini_proto;
1124
1125         ret = nf_conntrack_helper_init();
1126         if (ret < 0)
1127                 goto out_fini_expect;
1128
1129         /* For use by REJECT target */
1130         rcu_assign_pointer(ip_ct_attach, __nf_conntrack_attach);
1131         rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1132
1133         /* Set up fake conntrack:
1134             - to never be deleted, not in any hashes */
1135         atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1136         /*  - and look it like as a confirmed connection */
1137         set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1138
1139         return ret;
1140
1141 out_fini_expect:
1142         nf_conntrack_expect_fini();
1143 out_fini_proto:
1144         nf_conntrack_proto_fini();
1145 err_free_conntrack_slab:
1146         kmem_cache_destroy(nf_conntrack_cachep);
1147 err_free_hash:
1148         nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
1149                              nf_conntrack_htable_size);
1150 err_out:
1151         return -ENOMEM;
1152 }