netfilter: nf_conntrack: add allocation flag to nf_conntrack_alloc
[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                                    gfp_t gfp)
469 {
470         struct nf_conn *ct = NULL;
471
472         if (unlikely(!nf_conntrack_hash_rnd_initted)) {
473                 get_random_bytes(&nf_conntrack_hash_rnd, 4);
474                 nf_conntrack_hash_rnd_initted = 1;
475         }
476
477         /* We don't want any race condition at early drop stage */
478         atomic_inc(&nf_conntrack_count);
479
480         if (nf_conntrack_max &&
481             unlikely(atomic_read(&nf_conntrack_count) > nf_conntrack_max)) {
482                 unsigned int hash = hash_conntrack(orig);
483                 if (!early_drop(hash)) {
484                         atomic_dec(&nf_conntrack_count);
485                         if (net_ratelimit())
486                                 printk(KERN_WARNING
487                                        "nf_conntrack: table full, dropping"
488                                        " packet.\n");
489                         return ERR_PTR(-ENOMEM);
490                 }
491         }
492
493         ct = kmem_cache_zalloc(nf_conntrack_cachep, gfp);
494         if (ct == NULL) {
495                 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
496                 atomic_dec(&nf_conntrack_count);
497                 return ERR_PTR(-ENOMEM);
498         }
499
500         atomic_set(&ct->ct_general.use, 1);
501         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
502         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
503         /* Don't set timer yet: wait for confirmation */
504         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
505         INIT_RCU_HEAD(&ct->rcu);
506
507         return ct;
508 }
509 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
510
511 static void nf_conntrack_free_rcu(struct rcu_head *head)
512 {
513         struct nf_conn *ct = container_of(head, struct nf_conn, rcu);
514
515         nf_ct_ext_free(ct);
516         kmem_cache_free(nf_conntrack_cachep, ct);
517         atomic_dec(&nf_conntrack_count);
518 }
519
520 void nf_conntrack_free(struct nf_conn *ct)
521 {
522         nf_ct_ext_destroy(ct);
523         call_rcu(&ct->rcu, nf_conntrack_free_rcu);
524 }
525 EXPORT_SYMBOL_GPL(nf_conntrack_free);
526
527 /* Allocate a new conntrack: we return -ENOMEM if classification
528    failed due to stress.  Otherwise it really is unclassifiable. */
529 static struct nf_conntrack_tuple_hash *
530 init_conntrack(const struct nf_conntrack_tuple *tuple,
531                struct nf_conntrack_l3proto *l3proto,
532                struct nf_conntrack_l4proto *l4proto,
533                struct sk_buff *skb,
534                unsigned int dataoff)
535 {
536         struct nf_conn *ct;
537         struct nf_conn_help *help;
538         struct nf_conntrack_tuple repl_tuple;
539         struct nf_conntrack_expect *exp;
540
541         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
542                 pr_debug("Can't invert tuple.\n");
543                 return NULL;
544         }
545
546         ct = nf_conntrack_alloc(tuple, &repl_tuple, GFP_ATOMIC);
547         if (ct == NULL || IS_ERR(ct)) {
548                 pr_debug("Can't allocate conntrack.\n");
549                 return (struct nf_conntrack_tuple_hash *)ct;
550         }
551
552         if (!l4proto->new(ct, skb, dataoff)) {
553                 nf_conntrack_free(ct);
554                 pr_debug("init conntrack: can't track with proto module\n");
555                 return NULL;
556         }
557
558         spin_lock_bh(&nf_conntrack_lock);
559         exp = nf_ct_find_expectation(tuple);
560         if (exp) {
561                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
562                          ct, exp);
563                 /* Welcome, Mr. Bond.  We've been expecting you... */
564                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
565                 ct->master = exp->master;
566                 if (exp->helper) {
567                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
568                         if (help)
569                                 rcu_assign_pointer(help->helper, exp->helper);
570                 }
571
572 #ifdef CONFIG_NF_CONNTRACK_MARK
573                 ct->mark = exp->master->mark;
574 #endif
575 #ifdef CONFIG_NF_CONNTRACK_SECMARK
576                 ct->secmark = exp->master->secmark;
577 #endif
578                 nf_conntrack_get(&ct->master->ct_general);
579                 NF_CT_STAT_INC(expect_new);
580         } else {
581                 struct nf_conntrack_helper *helper;
582
583                 helper = __nf_ct_helper_find(&repl_tuple);
584                 if (helper) {
585                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
586                         if (help)
587                                 rcu_assign_pointer(help->helper, helper);
588                 }
589                 NF_CT_STAT_INC(new);
590         }
591
592         /* Overload tuple linked list to put us in unconfirmed list. */
593         hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, &unconfirmed);
594
595         spin_unlock_bh(&nf_conntrack_lock);
596
597         if (exp) {
598                 if (exp->expectfn)
599                         exp->expectfn(ct, exp);
600                 nf_ct_expect_put(exp);
601         }
602
603         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
604 }
605
606 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
607 static inline struct nf_conn *
608 resolve_normal_ct(struct sk_buff *skb,
609                   unsigned int dataoff,
610                   u_int16_t l3num,
611                   u_int8_t protonum,
612                   struct nf_conntrack_l3proto *l3proto,
613                   struct nf_conntrack_l4proto *l4proto,
614                   int *set_reply,
615                   enum ip_conntrack_info *ctinfo)
616 {
617         struct nf_conntrack_tuple tuple;
618         struct nf_conntrack_tuple_hash *h;
619         struct nf_conn *ct;
620
621         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
622                              dataoff, l3num, protonum, &tuple, l3proto,
623                              l4proto)) {
624                 pr_debug("resolve_normal_ct: Can't get tuple\n");
625                 return NULL;
626         }
627
628         /* look for tuple match */
629         h = nf_conntrack_find_get(&tuple);
630         if (!h) {
631                 h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff);
632                 if (!h)
633                         return NULL;
634                 if (IS_ERR(h))
635                         return (void *)h;
636         }
637         ct = nf_ct_tuplehash_to_ctrack(h);
638
639         /* It exists; we have (non-exclusive) reference. */
640         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
641                 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
642                 /* Please set reply bit if this packet OK */
643                 *set_reply = 1;
644         } else {
645                 /* Once we've had two way comms, always ESTABLISHED. */
646                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
647                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
648                         *ctinfo = IP_CT_ESTABLISHED;
649                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
650                         pr_debug("nf_conntrack_in: related packet for %p\n",
651                                  ct);
652                         *ctinfo = IP_CT_RELATED;
653                 } else {
654                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
655                         *ctinfo = IP_CT_NEW;
656                 }
657                 *set_reply = 0;
658         }
659         skb->nfct = &ct->ct_general;
660         skb->nfctinfo = *ctinfo;
661         return ct;
662 }
663
664 unsigned int
665 nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff *skb)
666 {
667         struct nf_conn *ct;
668         enum ip_conntrack_info ctinfo;
669         struct nf_conntrack_l3proto *l3proto;
670         struct nf_conntrack_l4proto *l4proto;
671         unsigned int dataoff;
672         u_int8_t protonum;
673         int set_reply = 0;
674         int ret;
675
676         /* Previously seen (loopback or untracked)?  Ignore. */
677         if (skb->nfct) {
678                 NF_CT_STAT_INC_ATOMIC(ignore);
679                 return NF_ACCEPT;
680         }
681
682         /* rcu_read_lock()ed by nf_hook_slow */
683         l3proto = __nf_ct_l3proto_find((u_int16_t)pf);
684         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
685                                    &dataoff, &protonum);
686         if (ret <= 0) {
687                 pr_debug("not prepared to track yet or error occured\n");
688                 NF_CT_STAT_INC_ATOMIC(error);
689                 NF_CT_STAT_INC_ATOMIC(invalid);
690                 return -ret;
691         }
692
693         l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum);
694
695         /* It may be an special packet, error, unclean...
696          * inverse of the return code tells to the netfilter
697          * core what to do with the packet. */
698         if (l4proto->error != NULL &&
699             (ret = l4proto->error(skb, dataoff, &ctinfo, pf, hooknum)) <= 0) {
700                 NF_CT_STAT_INC_ATOMIC(error);
701                 NF_CT_STAT_INC_ATOMIC(invalid);
702                 return -ret;
703         }
704
705         ct = resolve_normal_ct(skb, dataoff, pf, protonum, l3proto, l4proto,
706                                &set_reply, &ctinfo);
707         if (!ct) {
708                 /* Not valid part of a connection */
709                 NF_CT_STAT_INC_ATOMIC(invalid);
710                 return NF_ACCEPT;
711         }
712
713         if (IS_ERR(ct)) {
714                 /* Too stressed to deal. */
715                 NF_CT_STAT_INC_ATOMIC(drop);
716                 return NF_DROP;
717         }
718
719         NF_CT_ASSERT(skb->nfct);
720
721         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
722         if (ret < 0) {
723                 /* Invalid: inverse of the return code tells
724                  * the netfilter core what to do */
725                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
726                 nf_conntrack_put(skb->nfct);
727                 skb->nfct = NULL;
728                 NF_CT_STAT_INC_ATOMIC(invalid);
729                 return -ret;
730         }
731
732         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
733                 nf_conntrack_event_cache(IPCT_STATUS, skb);
734
735         return ret;
736 }
737 EXPORT_SYMBOL_GPL(nf_conntrack_in);
738
739 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
740                           const struct nf_conntrack_tuple *orig)
741 {
742         bool ret;
743
744         rcu_read_lock();
745         ret = nf_ct_invert_tuple(inverse, orig,
746                                  __nf_ct_l3proto_find(orig->src.l3num),
747                                  __nf_ct_l4proto_find(orig->src.l3num,
748                                                       orig->dst.protonum));
749         rcu_read_unlock();
750         return ret;
751 }
752 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
753
754 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
755    implicitly racy: see __nf_conntrack_confirm */
756 void nf_conntrack_alter_reply(struct nf_conn *ct,
757                               const struct nf_conntrack_tuple *newreply)
758 {
759         struct nf_conn_help *help = nfct_help(ct);
760         struct nf_conntrack_helper *helper;
761
762         /* Should be unconfirmed, so not in hash table yet */
763         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
764
765         pr_debug("Altering reply tuple of %p to ", ct);
766         nf_ct_dump_tuple(newreply);
767
768         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
769         if (ct->master || (help && !hlist_empty(&help->expectations)))
770                 return;
771
772         rcu_read_lock();
773         helper = __nf_ct_helper_find(newreply);
774         if (helper == NULL) {
775                 if (help)
776                         rcu_assign_pointer(help->helper, NULL);
777                 goto out;
778         }
779
780         if (help == NULL) {
781                 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
782                 if (help == NULL)
783                         goto out;
784         } else {
785                 memset(&help->help, 0, sizeof(help->help));
786         }
787
788         rcu_assign_pointer(help->helper, helper);
789 out:
790         rcu_read_unlock();
791 }
792 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
793
794 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
795 void __nf_ct_refresh_acct(struct nf_conn *ct,
796                           enum ip_conntrack_info ctinfo,
797                           const struct sk_buff *skb,
798                           unsigned long extra_jiffies,
799                           int do_acct)
800 {
801         int event = 0;
802
803         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
804         NF_CT_ASSERT(skb);
805
806         spin_lock_bh(&nf_conntrack_lock);
807
808         /* Only update if this is not a fixed timeout */
809         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
810                 goto acct;
811
812         /* If not in hash table, timer will not be active yet */
813         if (!nf_ct_is_confirmed(ct)) {
814                 ct->timeout.expires = extra_jiffies;
815                 event = IPCT_REFRESH;
816         } else {
817                 unsigned long newtime = jiffies + extra_jiffies;
818
819                 /* Only update the timeout if the new timeout is at least
820                    HZ jiffies from the old timeout. Need del_timer for race
821                    avoidance (may already be dying). */
822                 if (newtime - ct->timeout.expires >= HZ
823                     && del_timer(&ct->timeout)) {
824                         ct->timeout.expires = newtime;
825                         add_timer(&ct->timeout);
826                         event = IPCT_REFRESH;
827                 }
828         }
829
830 acct:
831 #ifdef CONFIG_NF_CT_ACCT
832         if (do_acct) {
833                 ct->counters[CTINFO2DIR(ctinfo)].packets++;
834                 ct->counters[CTINFO2DIR(ctinfo)].bytes +=
835                         skb->len - skb_network_offset(skb);
836
837                 if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000)
838                     || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000))
839                         event |= IPCT_COUNTER_FILLING;
840         }
841 #endif
842
843         spin_unlock_bh(&nf_conntrack_lock);
844
845         /* must be unlocked when calling event cache */
846         if (event)
847                 nf_conntrack_event_cache(event, skb);
848 }
849 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
850
851 void __nf_ct_kill_acct(struct nf_conn *ct,
852                 enum ip_conntrack_info ctinfo,
853                 const struct sk_buff *skb,
854                 int do_acct)
855 {
856 #ifdef CONFIG_NF_CT_ACCT
857         if (do_acct) {
858                 spin_lock_bh(&nf_conntrack_lock);
859                 ct->counters[CTINFO2DIR(ctinfo)].packets++;
860                 ct->counters[CTINFO2DIR(ctinfo)].bytes +=
861                         skb->len - skb_network_offset(skb);
862                 spin_unlock_bh(&nf_conntrack_lock);
863         }
864 #endif
865         if (del_timer(&ct->timeout))
866                 ct->timeout.function((unsigned long)ct);
867 }
868 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
869
870 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
871
872 #include <linux/netfilter/nfnetlink.h>
873 #include <linux/netfilter/nfnetlink_conntrack.h>
874 #include <linux/mutex.h>
875
876 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
877  * in ip_conntrack_core, since we don't want the protocols to autoload
878  * or depend on ctnetlink */
879 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
880                                const struct nf_conntrack_tuple *tuple)
881 {
882         NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
883         NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
884         return 0;
885
886 nla_put_failure:
887         return -1;
888 }
889 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
890
891 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
892         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
893         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
894 };
895 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
896
897 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
898                                struct nf_conntrack_tuple *t)
899 {
900         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
901                 return -EINVAL;
902
903         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
904         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
905
906         return 0;
907 }
908 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
909 #endif
910
911 /* Used by ipt_REJECT and ip6t_REJECT. */
912 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
913 {
914         struct nf_conn *ct;
915         enum ip_conntrack_info ctinfo;
916
917         /* This ICMP is in reverse direction to the packet which caused it */
918         ct = nf_ct_get(skb, &ctinfo);
919         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
920                 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
921         else
922                 ctinfo = IP_CT_RELATED;
923
924         /* Attach to new skbuff, and increment count */
925         nskb->nfct = &ct->ct_general;
926         nskb->nfctinfo = ctinfo;
927         nf_conntrack_get(nskb->nfct);
928 }
929
930 /* Bring out ya dead! */
931 static struct nf_conn *
932 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
933                 void *data, unsigned int *bucket)
934 {
935         struct nf_conntrack_tuple_hash *h;
936         struct nf_conn *ct;
937         struct hlist_node *n;
938
939         spin_lock_bh(&nf_conntrack_lock);
940         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
941                 hlist_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnode) {
942                         ct = nf_ct_tuplehash_to_ctrack(h);
943                         if (iter(ct, data))
944                                 goto found;
945                 }
946         }
947         hlist_for_each_entry(h, n, &unconfirmed, hnode) {
948                 ct = nf_ct_tuplehash_to_ctrack(h);
949                 if (iter(ct, data))
950                         set_bit(IPS_DYING_BIT, &ct->status);
951         }
952         spin_unlock_bh(&nf_conntrack_lock);
953         return NULL;
954 found:
955         atomic_inc(&ct->ct_general.use);
956         spin_unlock_bh(&nf_conntrack_lock);
957         return ct;
958 }
959
960 void
961 nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data)
962 {
963         struct nf_conn *ct;
964         unsigned int bucket = 0;
965
966         while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
967                 /* Time to push up daises... */
968                 if (del_timer(&ct->timeout))
969                         death_by_timeout((unsigned long)ct);
970                 /* ... else the timer will get him soon. */
971
972                 nf_ct_put(ct);
973         }
974 }
975 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
976
977 static int kill_all(struct nf_conn *i, void *data)
978 {
979         return 1;
980 }
981
982 void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size)
983 {
984         if (vmalloced)
985                 vfree(hash);
986         else
987                 free_pages((unsigned long)hash,
988                            get_order(sizeof(struct hlist_head) * size));
989 }
990 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
991
992 void nf_conntrack_flush(void)
993 {
994         nf_ct_iterate_cleanup(kill_all, NULL);
995 }
996 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
997
998 /* Mishearing the voices in his head, our hero wonders how he's
999    supposed to kill the mall. */
1000 void nf_conntrack_cleanup(void)
1001 {
1002         rcu_assign_pointer(ip_ct_attach, NULL);
1003
1004         /* This makes sure all current packets have passed through
1005            netfilter framework.  Roll on, two-stage module
1006            delete... */
1007         synchronize_net();
1008
1009         nf_ct_event_cache_flush();
1010  i_see_dead_people:
1011         nf_conntrack_flush();
1012         if (atomic_read(&nf_conntrack_count) != 0) {
1013                 schedule();
1014                 goto i_see_dead_people;
1015         }
1016         /* wait until all references to nf_conntrack_untracked are dropped */
1017         while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1018                 schedule();
1019
1020         rcu_assign_pointer(nf_ct_destroy, NULL);
1021
1022         kmem_cache_destroy(nf_conntrack_cachep);
1023         nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
1024                              nf_conntrack_htable_size);
1025
1026         nf_conntrack_proto_fini();
1027         nf_conntrack_helper_fini();
1028         nf_conntrack_expect_fini();
1029 }
1030
1031 struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced)
1032 {
1033         struct hlist_head *hash;
1034         unsigned int size, i;
1035
1036         *vmalloced = 0;
1037
1038         size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head));
1039         hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN,
1040                                        get_order(sizeof(struct hlist_head)
1041                                                  * size));
1042         if (!hash) {
1043                 *vmalloced = 1;
1044                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1045                 hash = vmalloc(sizeof(struct hlist_head) * size);
1046         }
1047
1048         if (hash)
1049                 for (i = 0; i < size; i++)
1050                         INIT_HLIST_HEAD(&hash[i]);
1051
1052         return hash;
1053 }
1054 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1055
1056 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1057 {
1058         int i, bucket, vmalloced, old_vmalloced;
1059         unsigned int hashsize, old_size;
1060         int rnd;
1061         struct hlist_head *hash, *old_hash;
1062         struct nf_conntrack_tuple_hash *h;
1063
1064         /* On boot, we can set this without any fancy locking. */
1065         if (!nf_conntrack_htable_size)
1066                 return param_set_uint(val, kp);
1067
1068         hashsize = simple_strtoul(val, NULL, 0);
1069         if (!hashsize)
1070                 return -EINVAL;
1071
1072         hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced);
1073         if (!hash)
1074                 return -ENOMEM;
1075
1076         /* We have to rehahs for the new table anyway, so we also can
1077          * use a newrandom seed */
1078         get_random_bytes(&rnd, 4);
1079
1080         /* Lookups in the old hash might happen in parallel, which means we
1081          * might get false negatives during connection lookup. New connections
1082          * created because of a false negative won't make it into the hash
1083          * though since that required taking the lock.
1084          */
1085         spin_lock_bh(&nf_conntrack_lock);
1086         for (i = 0; i < nf_conntrack_htable_size; i++) {
1087                 while (!hlist_empty(&nf_conntrack_hash[i])) {
1088                         h = hlist_entry(nf_conntrack_hash[i].first,
1089                                         struct nf_conntrack_tuple_hash, hnode);
1090                         hlist_del_rcu(&h->hnode);
1091                         bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1092                         hlist_add_head(&h->hnode, &hash[bucket]);
1093                 }
1094         }
1095         old_size = nf_conntrack_htable_size;
1096         old_vmalloced = nf_conntrack_vmalloc;
1097         old_hash = nf_conntrack_hash;
1098
1099         nf_conntrack_htable_size = hashsize;
1100         nf_conntrack_vmalloc = vmalloced;
1101         nf_conntrack_hash = hash;
1102         nf_conntrack_hash_rnd = rnd;
1103         spin_unlock_bh(&nf_conntrack_lock);
1104
1105         nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1106         return 0;
1107 }
1108 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1109
1110 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1111                   &nf_conntrack_htable_size, 0600);
1112
1113 int __init nf_conntrack_init(void)
1114 {
1115         int max_factor = 8;
1116         int ret;
1117
1118         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1119          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1120         if (!nf_conntrack_htable_size) {
1121                 nf_conntrack_htable_size
1122                         = (((num_physpages << PAGE_SHIFT) / 16384)
1123                            / sizeof(struct hlist_head));
1124                 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1125                         nf_conntrack_htable_size = 16384;
1126                 if (nf_conntrack_htable_size < 32)
1127                         nf_conntrack_htable_size = 32;
1128
1129                 /* Use a max. factor of four by default to get the same max as
1130                  * with the old struct list_heads. When a table size is given
1131                  * we use the old value of 8 to avoid reducing the max.
1132                  * entries. */
1133                 max_factor = 4;
1134         }
1135         nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1136                                                   &nf_conntrack_vmalloc);
1137         if (!nf_conntrack_hash) {
1138                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1139                 goto err_out;
1140         }
1141
1142         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1143
1144         printk("nf_conntrack version %s (%u buckets, %d max)\n",
1145                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1146                nf_conntrack_max);
1147
1148         nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1149                                                 sizeof(struct nf_conn),
1150                                                 0, 0, NULL);
1151         if (!nf_conntrack_cachep) {
1152                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1153                 goto err_free_hash;
1154         }
1155
1156         ret = nf_conntrack_proto_init();
1157         if (ret < 0)
1158                 goto err_free_conntrack_slab;
1159
1160         ret = nf_conntrack_expect_init();
1161         if (ret < 0)
1162                 goto out_fini_proto;
1163
1164         ret = nf_conntrack_helper_init();
1165         if (ret < 0)
1166                 goto out_fini_expect;
1167
1168         /* For use by REJECT target */
1169         rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1170         rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1171
1172         /* Set up fake conntrack:
1173             - to never be deleted, not in any hashes */
1174         atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1175         /*  - and look it like as a confirmed connection */
1176         set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1177
1178         return ret;
1179
1180 out_fini_expect:
1181         nf_conntrack_expect_fini();
1182 out_fini_proto:
1183         nf_conntrack_proto_fini();
1184 err_free_conntrack_slab:
1185         kmem_cache_destroy(nf_conntrack_cachep);
1186 err_free_hash:
1187         nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
1188                              nf_conntrack_htable_size);
1189 err_out:
1190         return -ENOMEM;
1191 }