f27c99246a4cce80a46f65ce809ea01c982883d8
[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_SPINLOCK(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
67 DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat);
68 EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat);
69
70 static int nf_conntrack_hash_rnd_initted;
71 static unsigned int nf_conntrack_hash_rnd;
72
73 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
74                                   unsigned int size, unsigned int rnd)
75 {
76         unsigned int n;
77         u_int32_t h;
78
79         /* The direction must be ignored, so we hash everything up to the
80          * destination ports (which is a multiple of 4) and treat the last
81          * three bytes manually.
82          */
83         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
84         h = jhash2((u32 *)tuple, n,
85                    rnd ^ (((__force __u16)tuple->dst.u.all << 16) |
86                           tuple->dst.protonum));
87
88         return ((u64)h * size) >> 32;
89 }
90
91 static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
92 {
93         return __hash_conntrack(tuple, nf_conntrack_htable_size,
94                                 nf_conntrack_hash_rnd);
95 }
96
97 bool
98 nf_ct_get_tuple(const struct sk_buff *skb,
99                 unsigned int nhoff,
100                 unsigned int dataoff,
101                 u_int16_t l3num,
102                 u_int8_t protonum,
103                 struct nf_conntrack_tuple *tuple,
104                 const struct nf_conntrack_l3proto *l3proto,
105                 const struct nf_conntrack_l4proto *l4proto)
106 {
107         memset(tuple, 0, sizeof(*tuple));
108
109         tuple->src.l3num = l3num;
110         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
111                 return false;
112
113         tuple->dst.protonum = protonum;
114         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
115
116         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
117 }
118 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
119
120 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
121                        u_int16_t l3num, struct nf_conntrack_tuple *tuple)
122 {
123         struct nf_conntrack_l3proto *l3proto;
124         struct nf_conntrack_l4proto *l4proto;
125         unsigned int protoff;
126         u_int8_t protonum;
127         int ret;
128
129         rcu_read_lock();
130
131         l3proto = __nf_ct_l3proto_find(l3num);
132         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
133         if (ret != NF_ACCEPT) {
134                 rcu_read_unlock();
135                 return false;
136         }
137
138         l4proto = __nf_ct_l4proto_find(l3num, protonum);
139
140         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
141                               l3proto, l4proto);
142
143         rcu_read_unlock();
144         return ret;
145 }
146 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
147
148 bool
149 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
150                    const struct nf_conntrack_tuple *orig,
151                    const struct nf_conntrack_l3proto *l3proto,
152                    const struct nf_conntrack_l4proto *l4proto)
153 {
154         memset(inverse, 0, sizeof(*inverse));
155
156         inverse->src.l3num = orig->src.l3num;
157         if (l3proto->invert_tuple(inverse, orig) == 0)
158                 return false;
159
160         inverse->dst.dir = !orig->dst.dir;
161
162         inverse->dst.protonum = orig->dst.protonum;
163         return l4proto->invert_tuple(inverse, orig);
164 }
165 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
166
167 static void
168 clean_from_lists(struct nf_conn *ct)
169 {
170         pr_debug("clean_from_lists(%p)\n", ct);
171         hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
172         hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode);
173
174         /* Destroy all pending expectations */
175         nf_ct_remove_expectations(ct);
176 }
177
178 static void
179 destroy_conntrack(struct nf_conntrack *nfct)
180 {
181         struct nf_conn *ct = (struct nf_conn *)nfct;
182         struct nf_conntrack_l4proto *l4proto;
183
184         pr_debug("destroy_conntrack(%p)\n", ct);
185         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
186         NF_CT_ASSERT(!timer_pending(&ct->timeout));
187
188         nf_conntrack_event(IPCT_DESTROY, ct);
189         set_bit(IPS_DYING_BIT, &ct->status);
190
191         /* To make sure we don't get any weird locking issues here:
192          * destroy_conntrack() MUST NOT be called with a write lock
193          * to nf_conntrack_lock!!! -HW */
194         rcu_read_lock();
195         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
196         if (l4proto && l4proto->destroy)
197                 l4proto->destroy(ct);
198
199         rcu_read_unlock();
200
201         spin_lock_bh(&nf_conntrack_lock);
202         /* Expectations will have been removed in clean_from_lists,
203          * except TFTP can create an expectation on the first packet,
204          * before connection is in the list, so we need to clean here,
205          * too. */
206         nf_ct_remove_expectations(ct);
207
208         /* We overload first tuple to link into unconfirmed list. */
209         if (!nf_ct_is_confirmed(ct)) {
210                 BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode));
211                 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
212         }
213
214         NF_CT_STAT_INC(delete);
215         spin_unlock_bh(&nf_conntrack_lock);
216
217         if (ct->master)
218                 nf_ct_put(ct->master);
219
220         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
221         nf_conntrack_free(ct);
222 }
223
224 static void death_by_timeout(unsigned long ul_conntrack)
225 {
226         struct nf_conn *ct = (void *)ul_conntrack;
227         struct nf_conn_help *help = nfct_help(ct);
228         struct nf_conntrack_helper *helper;
229
230         if (help) {
231                 rcu_read_lock();
232                 helper = rcu_dereference(help->helper);
233                 if (helper && helper->destroy)
234                         helper->destroy(ct);
235                 rcu_read_unlock();
236         }
237
238         spin_lock_bh(&nf_conntrack_lock);
239         /* Inside lock so preempt is disabled on module removal path.
240          * Otherwise we can get spurious warnings. */
241         NF_CT_STAT_INC(delete_list);
242         clean_from_lists(ct);
243         spin_unlock_bh(&nf_conntrack_lock);
244         nf_ct_put(ct);
245 }
246
247 struct nf_conntrack_tuple_hash *
248 __nf_conntrack_find(const struct nf_conntrack_tuple *tuple)
249 {
250         struct nf_conntrack_tuple_hash *h;
251         struct hlist_node *n;
252         unsigned int hash = hash_conntrack(tuple);
253
254         /* Disable BHs the entire time since we normally need to disable them
255          * at least once for the stats anyway.
256          */
257         local_bh_disable();
258         hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], hnode) {
259                 if (nf_ct_tuple_equal(tuple, &h->tuple)) {
260                         NF_CT_STAT_INC(found);
261                         local_bh_enable();
262                         return h;
263                 }
264                 NF_CT_STAT_INC(searched);
265         }
266         local_bh_enable();
267
268         return NULL;
269 }
270 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
271
272 /* Find a connection corresponding to a tuple. */
273 struct nf_conntrack_tuple_hash *
274 nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple)
275 {
276         struct nf_conntrack_tuple_hash *h;
277         struct nf_conn *ct;
278
279         rcu_read_lock();
280         h = __nf_conntrack_find(tuple);
281         if (h) {
282                 ct = nf_ct_tuplehash_to_ctrack(h);
283                 if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
284                         h = NULL;
285         }
286         rcu_read_unlock();
287
288         return h;
289 }
290 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
291
292 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
293                                        unsigned int hash,
294                                        unsigned int repl_hash)
295 {
296         hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
297                            &nf_conntrack_hash[hash]);
298         hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode,
299                            &nf_conntrack_hash[repl_hash]);
300 }
301
302 void nf_conntrack_hash_insert(struct nf_conn *ct)
303 {
304         unsigned int hash, repl_hash;
305
306         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
307         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
308
309         spin_lock_bh(&nf_conntrack_lock);
310         __nf_conntrack_hash_insert(ct, hash, repl_hash);
311         spin_unlock_bh(&nf_conntrack_lock);
312 }
313 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
314
315 /* Confirm a connection given skb; places it in hash table */
316 int
317 __nf_conntrack_confirm(struct sk_buff *skb)
318 {
319         unsigned int hash, repl_hash;
320         struct nf_conntrack_tuple_hash *h;
321         struct nf_conn *ct;
322         struct nf_conn_help *help;
323         struct hlist_node *n;
324         enum ip_conntrack_info ctinfo;
325
326         ct = nf_ct_get(skb, &ctinfo);
327
328         /* ipt_REJECT uses nf_conntrack_attach to attach related
329            ICMP/TCP RST packets in other direction.  Actual packet
330            which created connection will be IP_CT_NEW or for an
331            expected connection, IP_CT_RELATED. */
332         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
333                 return NF_ACCEPT;
334
335         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
336         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
337
338         /* We're not in hash table, and we refuse to set up related
339            connections for unconfirmed conns.  But packet copies and
340            REJECT will give spurious warnings here. */
341         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
342
343         /* No external references means noone else could have
344            confirmed us. */
345         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
346         pr_debug("Confirming conntrack %p\n", ct);
347
348         spin_lock_bh(&nf_conntrack_lock);
349
350         /* See if there's one in the list already, including reverse:
351            NAT could have grabbed it without realizing, since we're
352            not in the hash.  If there is, we lost race. */
353         hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode)
354                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
355                                       &h->tuple))
356                         goto out;
357         hlist_for_each_entry(h, n, &nf_conntrack_hash[repl_hash], hnode)
358                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
359                                       &h->tuple))
360                         goto out;
361
362         /* Remove from unconfirmed list */
363         hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
364
365         __nf_conntrack_hash_insert(ct, hash, repl_hash);
366         /* Timer relative to confirmation time, not original
367            setting time, otherwise we'd get timer wrap in
368            weird delay cases. */
369         ct->timeout.expires += jiffies;
370         add_timer(&ct->timeout);
371         atomic_inc(&ct->ct_general.use);
372         set_bit(IPS_CONFIRMED_BIT, &ct->status);
373         NF_CT_STAT_INC(insert);
374         spin_unlock_bh(&nf_conntrack_lock);
375         help = nfct_help(ct);
376         if (help && help->helper)
377                 nf_conntrack_event_cache(IPCT_HELPER, skb);
378 #ifdef CONFIG_NF_NAT_NEEDED
379         if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
380             test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
381                 nf_conntrack_event_cache(IPCT_NATINFO, skb);
382 #endif
383         nf_conntrack_event_cache(master_ct(ct) ?
384                                  IPCT_RELATED : IPCT_NEW, skb);
385         return NF_ACCEPT;
386
387 out:
388         NF_CT_STAT_INC(insert_failed);
389         spin_unlock_bh(&nf_conntrack_lock);
390         return NF_DROP;
391 }
392 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
393
394 /* Returns true if a connection correspondings to the tuple (required
395    for NAT). */
396 int
397 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
398                          const struct nf_conn *ignored_conntrack)
399 {
400         struct nf_conntrack_tuple_hash *h;
401         struct hlist_node *n;
402         unsigned int hash = hash_conntrack(tuple);
403
404         /* Disable BHs the entire time since we need to disable them at
405          * least once for the stats anyway.
406          */
407         rcu_read_lock_bh();
408         hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], hnode) {
409                 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
410                     nf_ct_tuple_equal(tuple, &h->tuple)) {
411                         NF_CT_STAT_INC(found);
412                         rcu_read_unlock_bh();
413                         return 1;
414                 }
415                 NF_CT_STAT_INC(searched);
416         }
417         rcu_read_unlock_bh();
418
419         return 0;
420 }
421 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
422
423 #define NF_CT_EVICTION_RANGE    8
424
425 /* There's a small race here where we may free a just-assured
426    connection.  Too bad: we're in trouble anyway. */
427 static noinline int early_drop(unsigned int hash)
428 {
429         /* Use oldest entry, which is roughly LRU */
430         struct nf_conntrack_tuple_hash *h;
431         struct nf_conn *ct = NULL, *tmp;
432         struct hlist_node *n;
433         unsigned int i, cnt = 0;
434         int dropped = 0;
435
436         rcu_read_lock();
437         for (i = 0; i < nf_conntrack_htable_size; i++) {
438                 hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash],
439                                          hnode) {
440                         tmp = nf_ct_tuplehash_to_ctrack(h);
441                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
442                                 ct = tmp;
443                         cnt++;
444                 }
445
446                 if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
447                         ct = NULL;
448                 if (ct || cnt >= NF_CT_EVICTION_RANGE)
449                         break;
450                 hash = (hash + 1) % nf_conntrack_htable_size;
451         }
452         rcu_read_unlock();
453
454         if (!ct)
455                 return dropped;
456
457         if (del_timer(&ct->timeout)) {
458                 death_by_timeout((unsigned long)ct);
459                 dropped = 1;
460                 NF_CT_STAT_INC_ATOMIC(early_drop);
461         }
462         nf_ct_put(ct);
463         return dropped;
464 }
465
466 struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
467                                    const struct nf_conntrack_tuple *repl)
468 {
469         struct nf_conn *ct = NULL;
470
471         if (unlikely(!nf_conntrack_hash_rnd_initted)) {
472                 get_random_bytes(&nf_conntrack_hash_rnd, 4);
473                 nf_conntrack_hash_rnd_initted = 1;
474         }
475
476         /* We don't want any race condition at early drop stage */
477         atomic_inc(&nf_conntrack_count);
478
479         if (nf_conntrack_max &&
480             unlikely(atomic_read(&nf_conntrack_count) > nf_conntrack_max)) {
481                 unsigned int hash = hash_conntrack(orig);
482                 if (!early_drop(hash)) {
483                         atomic_dec(&nf_conntrack_count);
484                         if (net_ratelimit())
485                                 printk(KERN_WARNING
486                                        "nf_conntrack: table full, dropping"
487                                        " packet.\n");
488                         return ERR_PTR(-ENOMEM);
489                 }
490         }
491
492         ct = kmem_cache_zalloc(nf_conntrack_cachep, GFP_ATOMIC);
493         if (ct == NULL) {
494                 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
495                 atomic_dec(&nf_conntrack_count);
496                 return ERR_PTR(-ENOMEM);
497         }
498
499         atomic_set(&ct->ct_general.use, 1);
500         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
501         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
502         /* Don't set timer yet: wait for confirmation */
503         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
504         INIT_RCU_HEAD(&ct->rcu);
505
506         return ct;
507 }
508 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
509
510 static void nf_conntrack_free_rcu(struct rcu_head *head)
511 {
512         struct nf_conn *ct = container_of(head, struct nf_conn, rcu);
513
514         nf_ct_ext_free(ct);
515         kmem_cache_free(nf_conntrack_cachep, ct);
516         atomic_dec(&nf_conntrack_count);
517 }
518
519 void nf_conntrack_free(struct nf_conn *ct)
520 {
521         nf_ct_ext_destroy(ct);
522         call_rcu(&ct->rcu, nf_conntrack_free_rcu);
523 }
524 EXPORT_SYMBOL_GPL(nf_conntrack_free);
525
526 /* Allocate a new conntrack: we return -ENOMEM if classification
527    failed due to stress.  Otherwise it really is unclassifiable. */
528 static struct nf_conntrack_tuple_hash *
529 init_conntrack(const struct nf_conntrack_tuple *tuple,
530                struct nf_conntrack_l3proto *l3proto,
531                struct nf_conntrack_l4proto *l4proto,
532                struct sk_buff *skb,
533                unsigned int dataoff)
534 {
535         struct nf_conn *ct;
536         struct nf_conn_help *help;
537         struct nf_conntrack_tuple repl_tuple;
538         struct nf_conntrack_expect *exp;
539
540         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
541                 pr_debug("Can't invert tuple.\n");
542                 return NULL;
543         }
544
545         ct = nf_conntrack_alloc(tuple, &repl_tuple);
546         if (ct == NULL || IS_ERR(ct)) {
547                 pr_debug("Can't allocate conntrack.\n");
548                 return (struct nf_conntrack_tuple_hash *)ct;
549         }
550
551         if (!l4proto->new(ct, skb, dataoff)) {
552                 nf_conntrack_free(ct);
553                 pr_debug("init conntrack: can't track with proto module\n");
554                 return NULL;
555         }
556
557         spin_lock_bh(&nf_conntrack_lock);
558         exp = nf_ct_find_expectation(tuple);
559         if (exp) {
560                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
561                          ct, exp);
562                 /* Welcome, Mr. Bond.  We've been expecting you... */
563                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
564                 ct->master = exp->master;
565                 if (exp->helper) {
566                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
567                         if (help)
568                                 rcu_assign_pointer(help->helper, exp->helper);
569                 }
570
571 #ifdef CONFIG_NF_CONNTRACK_MARK
572                 ct->mark = exp->master->mark;
573 #endif
574 #ifdef CONFIG_NF_CONNTRACK_SECMARK
575                 ct->secmark = exp->master->secmark;
576 #endif
577                 nf_conntrack_get(&ct->master->ct_general);
578                 NF_CT_STAT_INC(expect_new);
579         } else {
580                 struct nf_conntrack_helper *helper;
581
582                 helper = __nf_ct_helper_find(&repl_tuple);
583                 if (helper) {
584                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
585                         if (help)
586                                 rcu_assign_pointer(help->helper, helper);
587                 }
588                 NF_CT_STAT_INC(new);
589         }
590
591         /* Overload tuple linked list to put us in unconfirmed list. */
592         hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, &unconfirmed);
593
594         spin_unlock_bh(&nf_conntrack_lock);
595
596         if (exp) {
597                 if (exp->expectfn)
598                         exp->expectfn(ct, exp);
599                 nf_ct_expect_put(exp);
600         }
601
602         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
603 }
604
605 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
606 static inline struct nf_conn *
607 resolve_normal_ct(struct sk_buff *skb,
608                   unsigned int dataoff,
609                   u_int16_t l3num,
610                   u_int8_t protonum,
611                   struct nf_conntrack_l3proto *l3proto,
612                   struct nf_conntrack_l4proto *l4proto,
613                   int *set_reply,
614                   enum ip_conntrack_info *ctinfo)
615 {
616         struct nf_conntrack_tuple tuple;
617         struct nf_conntrack_tuple_hash *h;
618         struct nf_conn *ct;
619
620         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
621                              dataoff, l3num, protonum, &tuple, l3proto,
622                              l4proto)) {
623                 pr_debug("resolve_normal_ct: Can't get tuple\n");
624                 return NULL;
625         }
626
627         /* look for tuple match */
628         h = nf_conntrack_find_get(&tuple);
629         if (!h) {
630                 h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff);
631                 if (!h)
632                         return NULL;
633                 if (IS_ERR(h))
634                         return (void *)h;
635         }
636         ct = nf_ct_tuplehash_to_ctrack(h);
637
638         /* It exists; we have (non-exclusive) reference. */
639         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
640                 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
641                 /* Please set reply bit if this packet OK */
642                 *set_reply = 1;
643         } else {
644                 /* Once we've had two way comms, always ESTABLISHED. */
645                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
646                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
647                         *ctinfo = IP_CT_ESTABLISHED;
648                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
649                         pr_debug("nf_conntrack_in: related packet for %p\n",
650                                  ct);
651                         *ctinfo = IP_CT_RELATED;
652                 } else {
653                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
654                         *ctinfo = IP_CT_NEW;
655                 }
656                 *set_reply = 0;
657         }
658         skb->nfct = &ct->ct_general;
659         skb->nfctinfo = *ctinfo;
660         return ct;
661 }
662
663 unsigned int
664 nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff *skb)
665 {
666         struct nf_conn *ct;
667         enum ip_conntrack_info ctinfo;
668         struct nf_conntrack_l3proto *l3proto;
669         struct nf_conntrack_l4proto *l4proto;
670         unsigned int dataoff;
671         u_int8_t protonum;
672         int set_reply = 0;
673         int ret;
674
675         /* Previously seen (loopback or untracked)?  Ignore. */
676         if (skb->nfct) {
677                 NF_CT_STAT_INC_ATOMIC(ignore);
678                 return NF_ACCEPT;
679         }
680
681         /* rcu_read_lock()ed by nf_hook_slow */
682         l3proto = __nf_ct_l3proto_find((u_int16_t)pf);
683         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
684                                    &dataoff, &protonum);
685         if (ret <= 0) {
686                 pr_debug("not prepared to track yet or error occured\n");
687                 NF_CT_STAT_INC_ATOMIC(error);
688                 NF_CT_STAT_INC_ATOMIC(invalid);
689                 return -ret;
690         }
691
692         l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum);
693
694         /* It may be an special packet, error, unclean...
695          * inverse of the return code tells to the netfilter
696          * core what to do with the packet. */
697         if (l4proto->error != NULL &&
698             (ret = l4proto->error(skb, dataoff, &ctinfo, pf, hooknum)) <= 0) {
699                 NF_CT_STAT_INC_ATOMIC(error);
700                 NF_CT_STAT_INC_ATOMIC(invalid);
701                 return -ret;
702         }
703
704         ct = resolve_normal_ct(skb, dataoff, pf, protonum, l3proto, l4proto,
705                                &set_reply, &ctinfo);
706         if (!ct) {
707                 /* Not valid part of a connection */
708                 NF_CT_STAT_INC_ATOMIC(invalid);
709                 return NF_ACCEPT;
710         }
711
712         if (IS_ERR(ct)) {
713                 /* Too stressed to deal. */
714                 NF_CT_STAT_INC_ATOMIC(drop);
715                 return NF_DROP;
716         }
717
718         NF_CT_ASSERT(skb->nfct);
719
720         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
721         if (ret < 0) {
722                 /* Invalid: inverse of the return code tells
723                  * the netfilter core what to do */
724                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
725                 nf_conntrack_put(skb->nfct);
726                 skb->nfct = NULL;
727                 NF_CT_STAT_INC_ATOMIC(invalid);
728                 return -ret;
729         }
730
731         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
732                 nf_conntrack_event_cache(IPCT_STATUS, skb);
733
734         return ret;
735 }
736 EXPORT_SYMBOL_GPL(nf_conntrack_in);
737
738 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
739                           const struct nf_conntrack_tuple *orig)
740 {
741         bool ret;
742
743         rcu_read_lock();
744         ret = nf_ct_invert_tuple(inverse, orig,
745                                  __nf_ct_l3proto_find(orig->src.l3num),
746                                  __nf_ct_l4proto_find(orig->src.l3num,
747                                                       orig->dst.protonum));
748         rcu_read_unlock();
749         return ret;
750 }
751 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
752
753 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
754    implicitly racy: see __nf_conntrack_confirm */
755 void nf_conntrack_alter_reply(struct nf_conn *ct,
756                               const struct nf_conntrack_tuple *newreply)
757 {
758         struct nf_conn_help *help = nfct_help(ct);
759         struct nf_conntrack_helper *helper;
760
761         /* Should be unconfirmed, so not in hash table yet */
762         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
763
764         pr_debug("Altering reply tuple of %p to ", ct);
765         nf_ct_dump_tuple(newreply);
766
767         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
768         if (ct->master || (help && !hlist_empty(&help->expectations)))
769                 return;
770
771         rcu_read_lock();
772         helper = __nf_ct_helper_find(newreply);
773         if (helper == NULL) {
774                 if (help)
775                         rcu_assign_pointer(help->helper, NULL);
776                 goto out;
777         }
778
779         if (help == NULL) {
780                 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
781                 if (help == NULL)
782                         goto out;
783         } else {
784                 memset(&help->help, 0, sizeof(help->help));
785         }
786
787         rcu_assign_pointer(help->helper, helper);
788 out:
789         rcu_read_unlock();
790 }
791 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
792
793 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
794 void __nf_ct_refresh_acct(struct nf_conn *ct,
795                           enum ip_conntrack_info ctinfo,
796                           const struct sk_buff *skb,
797                           unsigned long extra_jiffies,
798                           int do_acct)
799 {
800         int event = 0;
801
802         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
803         NF_CT_ASSERT(skb);
804
805         spin_lock_bh(&nf_conntrack_lock);
806
807         /* Only update if this is not a fixed timeout */
808         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
809                 goto acct;
810
811         /* If not in hash table, timer will not be active yet */
812         if (!nf_ct_is_confirmed(ct)) {
813                 ct->timeout.expires = extra_jiffies;
814                 event = IPCT_REFRESH;
815         } else {
816                 unsigned long newtime = jiffies + extra_jiffies;
817
818                 /* Only update the timeout if the new timeout is at least
819                    HZ jiffies from the old timeout. Need del_timer for race
820                    avoidance (may already be dying). */
821                 if (newtime - ct->timeout.expires >= HZ
822                     && del_timer(&ct->timeout)) {
823                         ct->timeout.expires = newtime;
824                         add_timer(&ct->timeout);
825                         event = IPCT_REFRESH;
826                 }
827         }
828
829 acct:
830 #ifdef CONFIG_NF_CT_ACCT
831         if (do_acct) {
832                 ct->counters[CTINFO2DIR(ctinfo)].packets++;
833                 ct->counters[CTINFO2DIR(ctinfo)].bytes +=
834                         skb->len - skb_network_offset(skb);
835
836                 if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000)
837                     || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000))
838                         event |= IPCT_COUNTER_FILLING;
839         }
840 #endif
841
842         spin_unlock_bh(&nf_conntrack_lock);
843
844         /* must be unlocked when calling event cache */
845         if (event)
846                 nf_conntrack_event_cache(event, skb);
847 }
848 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
849
850 void __nf_ct_kill_acct(struct nf_conn *ct,
851                 enum ip_conntrack_info ctinfo,
852                 const struct sk_buff *skb,
853                 int do_acct)
854 {
855 #ifdef CONFIG_NF_CT_ACCT
856         if (do_acct) {
857                 spin_lock_bh(&nf_conntrack_lock);
858                 ct->counters[CTINFO2DIR(ctinfo)].packets++;
859                 ct->counters[CTINFO2DIR(ctinfo)].bytes +=
860                         skb->len - skb_network_offset(skb);
861                 spin_unlock_bh(&nf_conntrack_lock);
862         }
863 #endif
864         if (del_timer(&ct->timeout))
865                 ct->timeout.function((unsigned long)ct);
866 }
867 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
868
869 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
870
871 #include <linux/netfilter/nfnetlink.h>
872 #include <linux/netfilter/nfnetlink_conntrack.h>
873 #include <linux/mutex.h>
874
875 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
876  * in ip_conntrack_core, since we don't want the protocols to autoload
877  * or depend on ctnetlink */
878 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
879                                const struct nf_conntrack_tuple *tuple)
880 {
881         NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
882         NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
883         return 0;
884
885 nla_put_failure:
886         return -1;
887 }
888 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
889
890 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
891         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
892         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
893 };
894 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
895
896 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
897                                struct nf_conntrack_tuple *t)
898 {
899         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
900                 return -EINVAL;
901
902         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
903         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
904
905         return 0;
906 }
907 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
908 #endif
909
910 /* Used by ipt_REJECT and ip6t_REJECT. */
911 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
912 {
913         struct nf_conn *ct;
914         enum ip_conntrack_info ctinfo;
915
916         /* This ICMP is in reverse direction to the packet which caused it */
917         ct = nf_ct_get(skb, &ctinfo);
918         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
919                 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
920         else
921                 ctinfo = IP_CT_RELATED;
922
923         /* Attach to new skbuff, and increment count */
924         nskb->nfct = &ct->ct_general;
925         nskb->nfctinfo = ctinfo;
926         nf_conntrack_get(nskb->nfct);
927 }
928
929 /* Bring out ya dead! */
930 static struct nf_conn *
931 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
932                 void *data, unsigned int *bucket)
933 {
934         struct nf_conntrack_tuple_hash *h;
935         struct nf_conn *ct;
936         struct hlist_node *n;
937
938         spin_lock_bh(&nf_conntrack_lock);
939         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
940                 hlist_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnode) {
941                         ct = nf_ct_tuplehash_to_ctrack(h);
942                         if (iter(ct, data))
943                                 goto found;
944                 }
945         }
946         hlist_for_each_entry(h, n, &unconfirmed, hnode) {
947                 ct = nf_ct_tuplehash_to_ctrack(h);
948                 if (iter(ct, data))
949                         set_bit(IPS_DYING_BIT, &ct->status);
950         }
951         spin_unlock_bh(&nf_conntrack_lock);
952         return NULL;
953 found:
954         atomic_inc(&ct->ct_general.use);
955         spin_unlock_bh(&nf_conntrack_lock);
956         return ct;
957 }
958
959 void
960 nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data)
961 {
962         struct nf_conn *ct;
963         unsigned int bucket = 0;
964
965         while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
966                 /* Time to push up daises... */
967                 if (del_timer(&ct->timeout))
968                         death_by_timeout((unsigned long)ct);
969                 /* ... else the timer will get him soon. */
970
971                 nf_ct_put(ct);
972         }
973 }
974 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
975
976 static int kill_all(struct nf_conn *i, void *data)
977 {
978         return 1;
979 }
980
981 void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size)
982 {
983         if (vmalloced)
984                 vfree(hash);
985         else
986                 free_pages((unsigned long)hash,
987                            get_order(sizeof(struct hlist_head) * size));
988 }
989 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
990
991 void nf_conntrack_flush(void)
992 {
993         nf_ct_iterate_cleanup(kill_all, NULL);
994 }
995 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
996
997 /* Mishearing the voices in his head, our hero wonders how he's
998    supposed to kill the mall. */
999 void nf_conntrack_cleanup(void)
1000 {
1001         rcu_assign_pointer(ip_ct_attach, NULL);
1002
1003         /* This makes sure all current packets have passed through
1004            netfilter framework.  Roll on, two-stage module
1005            delete... */
1006         synchronize_net();
1007
1008         nf_ct_event_cache_flush();
1009  i_see_dead_people:
1010         nf_conntrack_flush();
1011         if (atomic_read(&nf_conntrack_count) != 0) {
1012                 schedule();
1013                 goto i_see_dead_people;
1014         }
1015         /* wait until all references to nf_conntrack_untracked are dropped */
1016         while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1017                 schedule();
1018
1019         rcu_assign_pointer(nf_ct_destroy, NULL);
1020
1021         kmem_cache_destroy(nf_conntrack_cachep);
1022         nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
1023                              nf_conntrack_htable_size);
1024
1025         nf_conntrack_proto_fini();
1026         nf_conntrack_helper_fini();
1027         nf_conntrack_expect_fini();
1028 }
1029
1030 struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced)
1031 {
1032         struct hlist_head *hash;
1033         unsigned int size, i;
1034
1035         *vmalloced = 0;
1036
1037         size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head));
1038         hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN,
1039                                        get_order(sizeof(struct hlist_head)
1040                                                  * size));
1041         if (!hash) {
1042                 *vmalloced = 1;
1043                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1044                 hash = vmalloc(sizeof(struct hlist_head) * size);
1045         }
1046
1047         if (hash)
1048                 for (i = 0; i < size; i++)
1049                         INIT_HLIST_HEAD(&hash[i]);
1050
1051         return hash;
1052 }
1053 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1054
1055 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1056 {
1057         int i, bucket, vmalloced, old_vmalloced;
1058         unsigned int hashsize, old_size;
1059         int rnd;
1060         struct hlist_head *hash, *old_hash;
1061         struct nf_conntrack_tuple_hash *h;
1062
1063         /* On boot, we can set this without any fancy locking. */
1064         if (!nf_conntrack_htable_size)
1065                 return param_set_uint(val, kp);
1066
1067         hashsize = simple_strtoul(val, NULL, 0);
1068         if (!hashsize)
1069                 return -EINVAL;
1070
1071         hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced);
1072         if (!hash)
1073                 return -ENOMEM;
1074
1075         /* We have to rehahs for the new table anyway, so we also can
1076          * use a newrandom seed */
1077         get_random_bytes(&rnd, 4);
1078
1079         /* Lookups in the old hash might happen in parallel, which means we
1080          * might get false negatives during connection lookup. New connections
1081          * created because of a false negative won't make it into the hash
1082          * though since that required taking the lock.
1083          */
1084         spin_lock_bh(&nf_conntrack_lock);
1085         for (i = 0; i < nf_conntrack_htable_size; i++) {
1086                 while (!hlist_empty(&nf_conntrack_hash[i])) {
1087                         h = hlist_entry(nf_conntrack_hash[i].first,
1088                                         struct nf_conntrack_tuple_hash, hnode);
1089                         hlist_del_rcu(&h->hnode);
1090                         bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1091                         hlist_add_head(&h->hnode, &hash[bucket]);
1092                 }
1093         }
1094         old_size = nf_conntrack_htable_size;
1095         old_vmalloced = nf_conntrack_vmalloc;
1096         old_hash = nf_conntrack_hash;
1097
1098         nf_conntrack_htable_size = hashsize;
1099         nf_conntrack_vmalloc = vmalloced;
1100         nf_conntrack_hash = hash;
1101         nf_conntrack_hash_rnd = rnd;
1102         spin_unlock_bh(&nf_conntrack_lock);
1103
1104         nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1105         return 0;
1106 }
1107 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1108
1109 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1110                   &nf_conntrack_htable_size, 0600);
1111
1112 int __init nf_conntrack_init(void)
1113 {
1114         int max_factor = 8;
1115         int ret;
1116
1117         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1118          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1119         if (!nf_conntrack_htable_size) {
1120                 nf_conntrack_htable_size
1121                         = (((num_physpages << PAGE_SHIFT) / 16384)
1122                            / sizeof(struct hlist_head));
1123                 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1124                         nf_conntrack_htable_size = 16384;
1125                 if (nf_conntrack_htable_size < 32)
1126                         nf_conntrack_htable_size = 32;
1127
1128                 /* Use a max. factor of four by default to get the same max as
1129                  * with the old struct list_heads. When a table size is given
1130                  * we use the old value of 8 to avoid reducing the max.
1131                  * entries. */
1132                 max_factor = 4;
1133         }
1134         nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1135                                                   &nf_conntrack_vmalloc);
1136         if (!nf_conntrack_hash) {
1137                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1138                 goto err_out;
1139         }
1140
1141         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1142
1143         printk("nf_conntrack version %s (%u buckets, %d max)\n",
1144                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1145                nf_conntrack_max);
1146
1147         nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1148                                                 sizeof(struct nf_conn),
1149                                                 0, 0, NULL);
1150         if (!nf_conntrack_cachep) {
1151                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1152                 goto err_free_hash;
1153         }
1154
1155         ret = nf_conntrack_proto_init();
1156         if (ret < 0)
1157                 goto err_free_conntrack_slab;
1158
1159         ret = nf_conntrack_expect_init();
1160         if (ret < 0)
1161                 goto out_fini_proto;
1162
1163         ret = nf_conntrack_helper_init();
1164         if (ret < 0)
1165                 goto out_fini_expect;
1166
1167         /* For use by REJECT target */
1168         rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1169         rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1170
1171         /* Set up fake conntrack:
1172             - to never be deleted, not in any hashes */
1173         atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1174         /*  - and look it like as a confirmed connection */
1175         set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1176
1177         return ret;
1178
1179 out_fini_expect:
1180         nf_conntrack_expect_fini();
1181 out_fini_proto:
1182         nf_conntrack_proto_fini();
1183 err_free_conntrack_slab:
1184         kmem_cache_destroy(nf_conntrack_cachep);
1185 err_free_hash:
1186         nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
1187                              nf_conntrack_htable_size);
1188 err_out:
1189         return -ENOMEM;
1190 }