[NETFILTER]: ctnetlink: use netlink attribute helpers
[linux-2.6.git] / net / netfilter / nf_conntrack_core.c
1 /* Connection state tracking for netfilter.  This is separated from,
2    but required by, the NAT layer; it can also be used by an iptables
3    extension. */
4
5 /* (C) 1999-2001 Paul `Rusty' Russell
6  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
7  * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License version 2 as
11  * published by the Free Software Foundation.
12  */
13
14 #include <linux/types.h>
15 #include <linux/netfilter.h>
16 #include <linux/module.h>
17 #include <linux/skbuff.h>
18 #include <linux/proc_fs.h>
19 #include <linux/vmalloc.h>
20 #include <linux/stddef.h>
21 #include <linux/slab.h>
22 #include <linux/random.h>
23 #include <linux/jhash.h>
24 #include <linux/err.h>
25 #include <linux/percpu.h>
26 #include <linux/moduleparam.h>
27 #include <linux/notifier.h>
28 #include <linux/kernel.h>
29 #include <linux/netdevice.h>
30 #include <linux/socket.h>
31 #include <linux/mm.h>
32
33 #include <net/netfilter/nf_conntrack.h>
34 #include <net/netfilter/nf_conntrack_l3proto.h>
35 #include <net/netfilter/nf_conntrack_l4proto.h>
36 #include <net/netfilter/nf_conntrack_expect.h>
37 #include <net/netfilter/nf_conntrack_helper.h>
38 #include <net/netfilter/nf_conntrack_core.h>
39 #include <net/netfilter/nf_conntrack_extend.h>
40
41 #define NF_CONNTRACK_VERSION    "0.5.0"
42
43 DEFINE_RWLOCK(nf_conntrack_lock);
44 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
45
46 /* nf_conntrack_standalone needs this */
47 atomic_t nf_conntrack_count = ATOMIC_INIT(0);
48 EXPORT_SYMBOL_GPL(nf_conntrack_count);
49
50 unsigned int nf_conntrack_htable_size __read_mostly;
51 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
52
53 int nf_conntrack_max __read_mostly;
54 EXPORT_SYMBOL_GPL(nf_conntrack_max);
55
56 struct hlist_head *nf_conntrack_hash __read_mostly;
57 EXPORT_SYMBOL_GPL(nf_conntrack_hash);
58
59 struct nf_conn nf_conntrack_untracked __read_mostly;
60 EXPORT_SYMBOL_GPL(nf_conntrack_untracked);
61
62 unsigned int nf_ct_log_invalid __read_mostly;
63 HLIST_HEAD(unconfirmed);
64 static int nf_conntrack_vmalloc __read_mostly;
65 static struct kmem_cache *nf_conntrack_cachep __read_mostly;
66
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 a, b;
77
78         a = jhash2(tuple->src.u3.all, ARRAY_SIZE(tuple->src.u3.all),
79                    (tuple->src.l3num << 16) | tuple->dst.protonum);
80         b = jhash2(tuple->dst.u3.all, ARRAY_SIZE(tuple->dst.u3.all),
81                    ((__force __u16)tuple->src.u.all << 16) |
82                     (__force __u16)tuple->dst.u.all);
83
84         return jhash_2words(a, b, rnd) % size;
85 }
86
87 static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
88 {
89         return __hash_conntrack(tuple, nf_conntrack_htable_size,
90                                 nf_conntrack_hash_rnd);
91 }
92
93 int
94 nf_ct_get_tuple(const struct sk_buff *skb,
95                 unsigned int nhoff,
96                 unsigned int dataoff,
97                 u_int16_t l3num,
98                 u_int8_t protonum,
99                 struct nf_conntrack_tuple *tuple,
100                 const struct nf_conntrack_l3proto *l3proto,
101                 const struct nf_conntrack_l4proto *l4proto)
102 {
103         NF_CT_TUPLE_U_BLANK(tuple);
104
105         tuple->src.l3num = l3num;
106         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
107                 return 0;
108
109         tuple->dst.protonum = protonum;
110         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
111
112         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
113 }
114 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
115
116 int nf_ct_get_tuplepr(const struct sk_buff *skb,
117                       unsigned int nhoff,
118                       u_int16_t l3num,
119                       struct nf_conntrack_tuple *tuple)
120 {
121         struct nf_conntrack_l3proto *l3proto;
122         struct nf_conntrack_l4proto *l4proto;
123         unsigned int protoff;
124         u_int8_t protonum;
125         int ret;
126
127         rcu_read_lock();
128
129         l3proto = __nf_ct_l3proto_find(l3num);
130         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
131         if (ret != NF_ACCEPT) {
132                 rcu_read_unlock();
133                 return 0;
134         }
135
136         l4proto = __nf_ct_l4proto_find(l3num, protonum);
137
138         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
139                               l3proto, l4proto);
140
141         rcu_read_unlock();
142         return ret;
143 }
144 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
145
146 int
147 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
148                    const struct nf_conntrack_tuple *orig,
149                    const struct nf_conntrack_l3proto *l3proto,
150                    const struct nf_conntrack_l4proto *l4proto)
151 {
152         NF_CT_TUPLE_U_BLANK(inverse);
153
154         inverse->src.l3num = orig->src.l3num;
155         if (l3proto->invert_tuple(inverse, orig) == 0)
156                 return 0;
157
158         inverse->dst.dir = !orig->dst.dir;
159
160         inverse->dst.protonum = orig->dst.protonum;
161         return l4proto->invert_tuple(inverse, orig);
162 }
163 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
164
165 static void
166 clean_from_lists(struct nf_conn *ct)
167 {
168         pr_debug("clean_from_lists(%p)\n", ct);
169         hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
170         hlist_del(&ct->tuplehash[IP_CT_DIR_REPLY].hnode);
171
172         /* Destroy all pending expectations */
173         nf_ct_remove_expectations(ct);
174 }
175
176 static void
177 destroy_conntrack(struct nf_conntrack *nfct)
178 {
179         struct nf_conn *ct = (struct nf_conn *)nfct;
180         struct nf_conntrack_l4proto *l4proto;
181
182         pr_debug("destroy_conntrack(%p)\n", ct);
183         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
184         NF_CT_ASSERT(!timer_pending(&ct->timeout));
185
186         nf_conntrack_event(IPCT_DESTROY, ct);
187         set_bit(IPS_DYING_BIT, &ct->status);
188
189         /* To make sure we don't get any weird locking issues here:
190          * destroy_conntrack() MUST NOT be called with a write lock
191          * to nf_conntrack_lock!!! -HW */
192         rcu_read_lock();
193         l4proto = __nf_ct_l4proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num,
194                                        ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum);
195         if (l4proto && l4proto->destroy)
196                 l4proto->destroy(ct);
197
198         nf_ct_ext_destroy(ct);
199
200         rcu_read_unlock();
201
202         write_lock_bh(&nf_conntrack_lock);
203         /* Expectations will have been removed in clean_from_lists,
204          * except TFTP can create an expectation on the first packet,
205          * before connection is in the list, so we need to clean here,
206          * too. */
207         nf_ct_remove_expectations(ct);
208
209         /* We overload first tuple to link into unconfirmed list. */
210         if (!nf_ct_is_confirmed(ct)) {
211                 BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode));
212                 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
213         }
214
215         NF_CT_STAT_INC(delete);
216         write_unlock_bh(&nf_conntrack_lock);
217
218         if (ct->master)
219                 nf_ct_put(ct->master);
220
221         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
222         nf_conntrack_free(ct);
223 }
224
225 static void death_by_timeout(unsigned long ul_conntrack)
226 {
227         struct nf_conn *ct = (void *)ul_conntrack;
228         struct nf_conn_help *help = nfct_help(ct);
229         struct nf_conntrack_helper *helper;
230
231         if (help) {
232                 rcu_read_lock();
233                 helper = rcu_dereference(help->helper);
234                 if (helper && helper->destroy)
235                         helper->destroy(ct);
236                 rcu_read_unlock();
237         }
238
239         write_lock_bh(&nf_conntrack_lock);
240         /* Inside lock so preempt is disabled on module removal path.
241          * Otherwise we can get spurious warnings. */
242         NF_CT_STAT_INC(delete_list);
243         clean_from_lists(ct);
244         write_unlock_bh(&nf_conntrack_lock);
245         nf_ct_put(ct);
246 }
247
248 struct nf_conntrack_tuple_hash *
249 __nf_conntrack_find(const struct nf_conntrack_tuple *tuple,
250                     const struct nf_conn *ignored_conntrack)
251 {
252         struct nf_conntrack_tuple_hash *h;
253         struct hlist_node *n;
254         unsigned int hash = hash_conntrack(tuple);
255
256         hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode) {
257                 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
258                     nf_ct_tuple_equal(tuple, &h->tuple)) {
259                         NF_CT_STAT_INC(found);
260                         return h;
261                 }
262                 NF_CT_STAT_INC(searched);
263         }
264
265         return NULL;
266 }
267 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
268
269 /* Find a connection corresponding to a tuple. */
270 struct nf_conntrack_tuple_hash *
271 nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple)
272 {
273         struct nf_conntrack_tuple_hash *h;
274
275         read_lock_bh(&nf_conntrack_lock);
276         h = __nf_conntrack_find(tuple, NULL);
277         if (h)
278                 atomic_inc(&nf_ct_tuplehash_to_ctrack(h)->ct_general.use);
279         read_unlock_bh(&nf_conntrack_lock);
280
281         return h;
282 }
283 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
284
285 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
286                                        unsigned int hash,
287                                        unsigned int repl_hash)
288 {
289         hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
290                        &nf_conntrack_hash[hash]);
291         hlist_add_head(&ct->tuplehash[IP_CT_DIR_REPLY].hnode,
292                        &nf_conntrack_hash[repl_hash]);
293 }
294
295 void nf_conntrack_hash_insert(struct nf_conn *ct)
296 {
297         unsigned int hash, repl_hash;
298
299         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
300         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
301
302         write_lock_bh(&nf_conntrack_lock);
303         __nf_conntrack_hash_insert(ct, hash, repl_hash);
304         write_unlock_bh(&nf_conntrack_lock);
305 }
306 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
307
308 /* Confirm a connection given skb; places it in hash table */
309 int
310 __nf_conntrack_confirm(struct sk_buff *skb)
311 {
312         unsigned int hash, repl_hash;
313         struct nf_conntrack_tuple_hash *h;
314         struct nf_conn *ct;
315         struct nf_conn_help *help;
316         struct hlist_node *n;
317         enum ip_conntrack_info ctinfo;
318
319         ct = nf_ct_get(skb, &ctinfo);
320
321         /* ipt_REJECT uses nf_conntrack_attach to attach related
322            ICMP/TCP RST packets in other direction.  Actual packet
323            which created connection will be IP_CT_NEW or for an
324            expected connection, IP_CT_RELATED. */
325         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
326                 return NF_ACCEPT;
327
328         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
329         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
330
331         /* We're not in hash table, and we refuse to set up related
332            connections for unconfirmed conns.  But packet copies and
333            REJECT will give spurious warnings here. */
334         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
335
336         /* No external references means noone else could have
337            confirmed us. */
338         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
339         pr_debug("Confirming conntrack %p\n", ct);
340
341         write_lock_bh(&nf_conntrack_lock);
342
343         /* See if there's one in the list already, including reverse:
344            NAT could have grabbed it without realizing, since we're
345            not in the hash.  If there is, we lost race. */
346         hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode)
347                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
348                                       &h->tuple))
349                         goto out;
350         hlist_for_each_entry(h, n, &nf_conntrack_hash[repl_hash], hnode)
351                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
352                                       &h->tuple))
353                         goto out;
354
355         /* Remove from unconfirmed list */
356         hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
357
358         __nf_conntrack_hash_insert(ct, hash, repl_hash);
359         /* Timer relative to confirmation time, not original
360            setting time, otherwise we'd get timer wrap in
361            weird delay cases. */
362         ct->timeout.expires += jiffies;
363         add_timer(&ct->timeout);
364         atomic_inc(&ct->ct_general.use);
365         set_bit(IPS_CONFIRMED_BIT, &ct->status);
366         NF_CT_STAT_INC(insert);
367         write_unlock_bh(&nf_conntrack_lock);
368         help = nfct_help(ct);
369         if (help && help->helper)
370                 nf_conntrack_event_cache(IPCT_HELPER, skb);
371 #ifdef CONFIG_NF_NAT_NEEDED
372         if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
373             test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
374                 nf_conntrack_event_cache(IPCT_NATINFO, skb);
375 #endif
376         nf_conntrack_event_cache(master_ct(ct) ?
377                                  IPCT_RELATED : IPCT_NEW, skb);
378         return NF_ACCEPT;
379
380 out:
381         NF_CT_STAT_INC(insert_failed);
382         write_unlock_bh(&nf_conntrack_lock);
383         return NF_DROP;
384 }
385 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
386
387 /* Returns true if a connection correspondings to the tuple (required
388    for NAT). */
389 int
390 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
391                          const struct nf_conn *ignored_conntrack)
392 {
393         struct nf_conntrack_tuple_hash *h;
394
395         read_lock_bh(&nf_conntrack_lock);
396         h = __nf_conntrack_find(tuple, ignored_conntrack);
397         read_unlock_bh(&nf_conntrack_lock);
398
399         return h != NULL;
400 }
401 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
402
403 #define NF_CT_EVICTION_RANGE    8
404
405 /* There's a small race here where we may free a just-assured
406    connection.  Too bad: we're in trouble anyway. */
407 static int early_drop(unsigned int hash)
408 {
409         /* Use oldest entry, which is roughly LRU */
410         struct nf_conntrack_tuple_hash *h;
411         struct nf_conn *ct = NULL, *tmp;
412         struct hlist_node *n;
413         unsigned int i, cnt = 0;
414         int dropped = 0;
415
416         read_lock_bh(&nf_conntrack_lock);
417         for (i = 0; i < nf_conntrack_htable_size; i++) {
418                 hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode) {
419                         tmp = nf_ct_tuplehash_to_ctrack(h);
420                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
421                                 ct = tmp;
422                         cnt++;
423                 }
424                 if (ct || cnt >= NF_CT_EVICTION_RANGE)
425                         break;
426                 hash = (hash + 1) % nf_conntrack_htable_size;
427         }
428         if (ct)
429                 atomic_inc(&ct->ct_general.use);
430         read_unlock_bh(&nf_conntrack_lock);
431
432         if (!ct)
433                 return dropped;
434
435         if (del_timer(&ct->timeout)) {
436                 death_by_timeout((unsigned long)ct);
437                 dropped = 1;
438                 NF_CT_STAT_INC_ATOMIC(early_drop);
439         }
440         nf_ct_put(ct);
441         return dropped;
442 }
443
444 struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
445                                    const struct nf_conntrack_tuple *repl)
446 {
447         struct nf_conn *conntrack = NULL;
448
449         if (unlikely(!nf_conntrack_hash_rnd_initted)) {
450                 get_random_bytes(&nf_conntrack_hash_rnd, 4);
451                 nf_conntrack_hash_rnd_initted = 1;
452         }
453
454         /* We don't want any race condition at early drop stage */
455         atomic_inc(&nf_conntrack_count);
456
457         if (nf_conntrack_max
458             && atomic_read(&nf_conntrack_count) > nf_conntrack_max) {
459                 unsigned int hash = hash_conntrack(orig);
460                 if (!early_drop(hash)) {
461                         atomic_dec(&nf_conntrack_count);
462                         if (net_ratelimit())
463                                 printk(KERN_WARNING
464                                        "nf_conntrack: table full, dropping"
465                                        " packet.\n");
466                         return ERR_PTR(-ENOMEM);
467                 }
468         }
469
470         conntrack = kmem_cache_zalloc(nf_conntrack_cachep, GFP_ATOMIC);
471         if (conntrack == NULL) {
472                 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
473                 atomic_dec(&nf_conntrack_count);
474                 return ERR_PTR(-ENOMEM);
475         }
476
477         atomic_set(&conntrack->ct_general.use, 1);
478         conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
479         conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
480         /* Don't set timer yet: wait for confirmation */
481         setup_timer(&conntrack->timeout, death_by_timeout,
482                     (unsigned long)conntrack);
483
484         return conntrack;
485 }
486 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
487
488 void nf_conntrack_free(struct nf_conn *conntrack)
489 {
490         nf_ct_ext_free(conntrack);
491         kmem_cache_free(nf_conntrack_cachep, conntrack);
492         atomic_dec(&nf_conntrack_count);
493 }
494 EXPORT_SYMBOL_GPL(nf_conntrack_free);
495
496 /* Allocate a new conntrack: we return -ENOMEM if classification
497    failed due to stress.  Otherwise it really is unclassifiable. */
498 static struct nf_conntrack_tuple_hash *
499 init_conntrack(const struct nf_conntrack_tuple *tuple,
500                struct nf_conntrack_l3proto *l3proto,
501                struct nf_conntrack_l4proto *l4proto,
502                struct sk_buff *skb,
503                unsigned int dataoff)
504 {
505         struct nf_conn *conntrack;
506         struct nf_conn_help *help;
507         struct nf_conntrack_tuple repl_tuple;
508         struct nf_conntrack_expect *exp;
509
510         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
511                 pr_debug("Can't invert tuple.\n");
512                 return NULL;
513         }
514
515         conntrack = nf_conntrack_alloc(tuple, &repl_tuple);
516         if (conntrack == NULL || IS_ERR(conntrack)) {
517                 pr_debug("Can't allocate conntrack.\n");
518                 return (struct nf_conntrack_tuple_hash *)conntrack;
519         }
520
521         if (!l4proto->new(conntrack, skb, dataoff)) {
522                 nf_conntrack_free(conntrack);
523                 pr_debug("init conntrack: can't track with proto module\n");
524                 return NULL;
525         }
526
527         write_lock_bh(&nf_conntrack_lock);
528         exp = nf_ct_find_expectation(tuple);
529         if (exp) {
530                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
531                          conntrack, exp);
532                 /* Welcome, Mr. Bond.  We've been expecting you... */
533                 __set_bit(IPS_EXPECTED_BIT, &conntrack->status);
534                 conntrack->master = exp->master;
535                 if (exp->helper) {
536                         help = nf_ct_helper_ext_add(conntrack, GFP_ATOMIC);
537                         if (help)
538                                 rcu_assign_pointer(help->helper, exp->helper);
539                 }
540
541 #ifdef CONFIG_NF_CONNTRACK_MARK
542                 conntrack->mark = exp->master->mark;
543 #endif
544 #ifdef CONFIG_NF_CONNTRACK_SECMARK
545                 conntrack->secmark = exp->master->secmark;
546 #endif
547                 nf_conntrack_get(&conntrack->master->ct_general);
548                 NF_CT_STAT_INC(expect_new);
549         } else {
550                 struct nf_conntrack_helper *helper;
551
552                 helper = __nf_ct_helper_find(&repl_tuple);
553                 if (helper) {
554                         help = nf_ct_helper_ext_add(conntrack, GFP_ATOMIC);
555                         if (help)
556                                 rcu_assign_pointer(help->helper, helper);
557                 }
558                 NF_CT_STAT_INC(new);
559         }
560
561         /* Overload tuple linked list to put us in unconfirmed list. */
562         hlist_add_head(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
563                        &unconfirmed);
564
565         write_unlock_bh(&nf_conntrack_lock);
566
567         if (exp) {
568                 if (exp->expectfn)
569                         exp->expectfn(conntrack, exp);
570                 nf_ct_expect_put(exp);
571         }
572
573         return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL];
574 }
575
576 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
577 static inline struct nf_conn *
578 resolve_normal_ct(struct sk_buff *skb,
579                   unsigned int dataoff,
580                   u_int16_t l3num,
581                   u_int8_t protonum,
582                   struct nf_conntrack_l3proto *l3proto,
583                   struct nf_conntrack_l4proto *l4proto,
584                   int *set_reply,
585                   enum ip_conntrack_info *ctinfo)
586 {
587         struct nf_conntrack_tuple tuple;
588         struct nf_conntrack_tuple_hash *h;
589         struct nf_conn *ct;
590
591         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
592                              dataoff, l3num, protonum, &tuple, l3proto,
593                              l4proto)) {
594                 pr_debug("resolve_normal_ct: Can't get tuple\n");
595                 return NULL;
596         }
597
598         /* look for tuple match */
599         h = nf_conntrack_find_get(&tuple);
600         if (!h) {
601                 h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff);
602                 if (!h)
603                         return NULL;
604                 if (IS_ERR(h))
605                         return (void *)h;
606         }
607         ct = nf_ct_tuplehash_to_ctrack(h);
608
609         /* It exists; we have (non-exclusive) reference. */
610         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
611                 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
612                 /* Please set reply bit if this packet OK */
613                 *set_reply = 1;
614         } else {
615                 /* Once we've had two way comms, always ESTABLISHED. */
616                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
617                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
618                         *ctinfo = IP_CT_ESTABLISHED;
619                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
620                         pr_debug("nf_conntrack_in: related packet for %p\n",
621                                  ct);
622                         *ctinfo = IP_CT_RELATED;
623                 } else {
624                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
625                         *ctinfo = IP_CT_NEW;
626                 }
627                 *set_reply = 0;
628         }
629         skb->nfct = &ct->ct_general;
630         skb->nfctinfo = *ctinfo;
631         return ct;
632 }
633
634 unsigned int
635 nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff *skb)
636 {
637         struct nf_conn *ct;
638         enum ip_conntrack_info ctinfo;
639         struct nf_conntrack_l3proto *l3proto;
640         struct nf_conntrack_l4proto *l4proto;
641         unsigned int dataoff;
642         u_int8_t protonum;
643         int set_reply = 0;
644         int ret;
645
646         /* Previously seen (loopback or untracked)?  Ignore. */
647         if (skb->nfct) {
648                 NF_CT_STAT_INC_ATOMIC(ignore);
649                 return NF_ACCEPT;
650         }
651
652         /* rcu_read_lock()ed by nf_hook_slow */
653         l3proto = __nf_ct_l3proto_find((u_int16_t)pf);
654         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
655                                    &dataoff, &protonum);
656         if (ret <= 0) {
657                 pr_debug("not prepared to track yet or error occured\n");
658                 NF_CT_STAT_INC_ATOMIC(error);
659                 NF_CT_STAT_INC_ATOMIC(invalid);
660                 return -ret;
661         }
662
663         l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum);
664
665         /* It may be an special packet, error, unclean...
666          * inverse of the return code tells to the netfilter
667          * core what to do with the packet. */
668         if (l4proto->error != NULL &&
669             (ret = l4proto->error(skb, dataoff, &ctinfo, pf, hooknum)) <= 0) {
670                 NF_CT_STAT_INC_ATOMIC(error);
671                 NF_CT_STAT_INC_ATOMIC(invalid);
672                 return -ret;
673         }
674
675         ct = resolve_normal_ct(skb, dataoff, pf, protonum, l3proto, l4proto,
676                                &set_reply, &ctinfo);
677         if (!ct) {
678                 /* Not valid part of a connection */
679                 NF_CT_STAT_INC_ATOMIC(invalid);
680                 return NF_ACCEPT;
681         }
682
683         if (IS_ERR(ct)) {
684                 /* Too stressed to deal. */
685                 NF_CT_STAT_INC_ATOMIC(drop);
686                 return NF_DROP;
687         }
688
689         NF_CT_ASSERT(skb->nfct);
690
691         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
692         if (ret < 0) {
693                 /* Invalid: inverse of the return code tells
694                  * the netfilter core what to do */
695                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
696                 nf_conntrack_put(skb->nfct);
697                 skb->nfct = NULL;
698                 NF_CT_STAT_INC_ATOMIC(invalid);
699                 return -ret;
700         }
701
702         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
703                 nf_conntrack_event_cache(IPCT_STATUS, skb);
704
705         return ret;
706 }
707 EXPORT_SYMBOL_GPL(nf_conntrack_in);
708
709 int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
710                          const struct nf_conntrack_tuple *orig)
711 {
712         int ret;
713
714         rcu_read_lock();
715         ret = nf_ct_invert_tuple(inverse, orig,
716                                  __nf_ct_l3proto_find(orig->src.l3num),
717                                  __nf_ct_l4proto_find(orig->src.l3num,
718                                                       orig->dst.protonum));
719         rcu_read_unlock();
720         return ret;
721 }
722 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
723
724 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
725    implicitly racy: see __nf_conntrack_confirm */
726 void nf_conntrack_alter_reply(struct nf_conn *ct,
727                               const struct nf_conntrack_tuple *newreply)
728 {
729         struct nf_conn_help *help = nfct_help(ct);
730         struct nf_conntrack_helper *helper;
731
732         write_lock_bh(&nf_conntrack_lock);
733         /* Should be unconfirmed, so not in hash table yet */
734         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
735
736         pr_debug("Altering reply tuple of %p to ", ct);
737         NF_CT_DUMP_TUPLE(newreply);
738
739         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
740         if (ct->master || (help && help->expecting != 0))
741                 goto out;
742
743         helper = __nf_ct_helper_find(newreply);
744         if (helper == NULL) {
745                 if (help)
746                         rcu_assign_pointer(help->helper, NULL);
747                 goto out;
748         }
749
750         if (help == NULL) {
751                 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
752                 if (help == NULL)
753                         goto out;
754         } else {
755                 memset(&help->help, 0, sizeof(help->help));
756         }
757
758         rcu_assign_pointer(help->helper, helper);
759 out:
760         write_unlock_bh(&nf_conntrack_lock);
761 }
762 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
763
764 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
765 void __nf_ct_refresh_acct(struct nf_conn *ct,
766                           enum ip_conntrack_info ctinfo,
767                           const struct sk_buff *skb,
768                           unsigned long extra_jiffies,
769                           int do_acct)
770 {
771         int event = 0;
772
773         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
774         NF_CT_ASSERT(skb);
775
776         write_lock_bh(&nf_conntrack_lock);
777
778         /* Only update if this is not a fixed timeout */
779         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) {
780                 write_unlock_bh(&nf_conntrack_lock);
781                 return;
782         }
783
784         /* If not in hash table, timer will not be active yet */
785         if (!nf_ct_is_confirmed(ct)) {
786                 ct->timeout.expires = extra_jiffies;
787                 event = IPCT_REFRESH;
788         } else {
789                 unsigned long newtime = jiffies + extra_jiffies;
790
791                 /* Only update the timeout if the new timeout is at least
792                    HZ jiffies from the old timeout. Need del_timer for race
793                    avoidance (may already be dying). */
794                 if (newtime - ct->timeout.expires >= HZ
795                     && del_timer(&ct->timeout)) {
796                         ct->timeout.expires = newtime;
797                         add_timer(&ct->timeout);
798                         event = IPCT_REFRESH;
799                 }
800         }
801
802 #ifdef CONFIG_NF_CT_ACCT
803         if (do_acct) {
804                 ct->counters[CTINFO2DIR(ctinfo)].packets++;
805                 ct->counters[CTINFO2DIR(ctinfo)].bytes +=
806                         skb->len - skb_network_offset(skb);
807
808                 if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000)
809                     || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000))
810                         event |= IPCT_COUNTER_FILLING;
811         }
812 #endif
813
814         write_unlock_bh(&nf_conntrack_lock);
815
816         /* must be unlocked when calling event cache */
817         if (event)
818                 nf_conntrack_event_cache(event, skb);
819 }
820 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
821
822 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
823
824 #include <linux/netfilter/nfnetlink.h>
825 #include <linux/netfilter/nfnetlink_conntrack.h>
826 #include <linux/mutex.h>
827
828 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
829  * in ip_conntrack_core, since we don't want the protocols to autoload
830  * or depend on ctnetlink */
831 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
832                                const struct nf_conntrack_tuple *tuple)
833 {
834         NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
835         NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
836         return 0;
837
838 nla_put_failure:
839         return -1;
840 }
841 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
842
843 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
844         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
845         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
846 };
847 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
848
849 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
850                                struct nf_conntrack_tuple *t)
851 {
852         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
853                 return -EINVAL;
854
855         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
856         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
857
858         return 0;
859 }
860 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
861 #endif
862
863 /* Used by ipt_REJECT and ip6t_REJECT. */
864 void __nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
865 {
866         struct nf_conn *ct;
867         enum ip_conntrack_info ctinfo;
868
869         /* This ICMP is in reverse direction to the packet which caused it */
870         ct = nf_ct_get(skb, &ctinfo);
871         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
872                 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
873         else
874                 ctinfo = IP_CT_RELATED;
875
876         /* Attach to new skbuff, and increment count */
877         nskb->nfct = &ct->ct_general;
878         nskb->nfctinfo = ctinfo;
879         nf_conntrack_get(nskb->nfct);
880 }
881 EXPORT_SYMBOL_GPL(__nf_conntrack_attach);
882
883 static inline int
884 do_iter(const struct nf_conntrack_tuple_hash *i,
885         int (*iter)(struct nf_conn *i, void *data),
886         void *data)
887 {
888         return iter(nf_ct_tuplehash_to_ctrack(i), data);
889 }
890
891 /* Bring out ya dead! */
892 static struct nf_conn *
893 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
894                 void *data, unsigned int *bucket)
895 {
896         struct nf_conntrack_tuple_hash *h;
897         struct nf_conn *ct;
898         struct hlist_node *n;
899
900         write_lock_bh(&nf_conntrack_lock);
901         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
902                 hlist_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnode) {
903                         ct = nf_ct_tuplehash_to_ctrack(h);
904                         if (iter(ct, data))
905                                 goto found;
906                 }
907         }
908         hlist_for_each_entry(h, n, &unconfirmed, hnode) {
909                 ct = nf_ct_tuplehash_to_ctrack(h);
910                 if (iter(ct, data))
911                         set_bit(IPS_DYING_BIT, &ct->status);
912         }
913         write_unlock_bh(&nf_conntrack_lock);
914         return NULL;
915 found:
916         atomic_inc(&ct->ct_general.use);
917         write_unlock_bh(&nf_conntrack_lock);
918         return ct;
919 }
920
921 void
922 nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data)
923 {
924         struct nf_conn *ct;
925         unsigned int bucket = 0;
926
927         while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
928                 /* Time to push up daises... */
929                 if (del_timer(&ct->timeout))
930                         death_by_timeout((unsigned long)ct);
931                 /* ... else the timer will get him soon. */
932
933                 nf_ct_put(ct);
934         }
935 }
936 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
937
938 static int kill_all(struct nf_conn *i, void *data)
939 {
940         return 1;
941 }
942
943 void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, int size)
944 {
945         if (vmalloced)
946                 vfree(hash);
947         else
948                 free_pages((unsigned long)hash,
949                            get_order(sizeof(struct hlist_head) * size));
950 }
951 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
952
953 void nf_conntrack_flush(void)
954 {
955         nf_ct_iterate_cleanup(kill_all, NULL);
956 }
957 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
958
959 /* Mishearing the voices in his head, our hero wonders how he's
960    supposed to kill the mall. */
961 void nf_conntrack_cleanup(void)
962 {
963         rcu_assign_pointer(ip_ct_attach, NULL);
964
965         /* This makes sure all current packets have passed through
966            netfilter framework.  Roll on, two-stage module
967            delete... */
968         synchronize_net();
969
970         nf_ct_event_cache_flush();
971  i_see_dead_people:
972         nf_conntrack_flush();
973         if (atomic_read(&nf_conntrack_count) != 0) {
974                 schedule();
975                 goto i_see_dead_people;
976         }
977         /* wait until all references to nf_conntrack_untracked are dropped */
978         while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
979                 schedule();
980
981         rcu_assign_pointer(nf_ct_destroy, NULL);
982
983         kmem_cache_destroy(nf_conntrack_cachep);
984         nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
985                              nf_conntrack_htable_size);
986
987         nf_conntrack_proto_fini();
988         nf_conntrack_helper_fini();
989         nf_conntrack_expect_fini();
990 }
991
992 struct hlist_head *nf_ct_alloc_hashtable(int *sizep, int *vmalloced)
993 {
994         struct hlist_head *hash;
995         unsigned int size, i;
996
997         *vmalloced = 0;
998
999         size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head));
1000         hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN,
1001                                        get_order(sizeof(struct hlist_head)
1002                                                  * size));
1003         if (!hash) {
1004                 *vmalloced = 1;
1005                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1006                 hash = vmalloc(sizeof(struct hlist_head) * size);
1007         }
1008
1009         if (hash)
1010                 for (i = 0; i < size; i++)
1011                         INIT_HLIST_HEAD(&hash[i]);
1012
1013         return hash;
1014 }
1015 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1016
1017 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1018 {
1019         int i, bucket, hashsize, vmalloced;
1020         int old_vmalloced, old_size;
1021         int rnd;
1022         struct hlist_head *hash, *old_hash;
1023         struct nf_conntrack_tuple_hash *h;
1024
1025         /* On boot, we can set this without any fancy locking. */
1026         if (!nf_conntrack_htable_size)
1027                 return param_set_uint(val, kp);
1028
1029         hashsize = simple_strtol(val, NULL, 0);
1030         if (!hashsize)
1031                 return -EINVAL;
1032
1033         hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced);
1034         if (!hash)
1035                 return -ENOMEM;
1036
1037         /* We have to rehahs for the new table anyway, so we also can
1038          * use a newrandom seed */
1039         get_random_bytes(&rnd, 4);
1040
1041         write_lock_bh(&nf_conntrack_lock);
1042         for (i = 0; i < nf_conntrack_htable_size; i++) {
1043                 while (!hlist_empty(&nf_conntrack_hash[i])) {
1044                         h = hlist_entry(nf_conntrack_hash[i].first,
1045                                         struct nf_conntrack_tuple_hash, hnode);
1046                         hlist_del(&h->hnode);
1047                         bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1048                         hlist_add_head(&h->hnode, &hash[bucket]);
1049                 }
1050         }
1051         old_size = nf_conntrack_htable_size;
1052         old_vmalloced = nf_conntrack_vmalloc;
1053         old_hash = nf_conntrack_hash;
1054
1055         nf_conntrack_htable_size = hashsize;
1056         nf_conntrack_vmalloc = vmalloced;
1057         nf_conntrack_hash = hash;
1058         nf_conntrack_hash_rnd = rnd;
1059         write_unlock_bh(&nf_conntrack_lock);
1060
1061         nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1062         return 0;
1063 }
1064 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1065
1066 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1067                   &nf_conntrack_htable_size, 0600);
1068
1069 int __init nf_conntrack_init(void)
1070 {
1071         int max_factor = 8;
1072         int ret;
1073
1074         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1075          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1076         if (!nf_conntrack_htable_size) {
1077                 nf_conntrack_htable_size
1078                         = (((num_physpages << PAGE_SHIFT) / 16384)
1079                            / sizeof(struct hlist_head));
1080                 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1081                         nf_conntrack_htable_size = 16384;
1082                 if (nf_conntrack_htable_size < 32)
1083                         nf_conntrack_htable_size = 32;
1084
1085                 /* Use a max. factor of four by default to get the same max as
1086                  * with the old struct list_heads. When a table size is given
1087                  * we use the old value of 8 to avoid reducing the max.
1088                  * entries. */
1089                 max_factor = 4;
1090         }
1091         nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1092                                                   &nf_conntrack_vmalloc);
1093         if (!nf_conntrack_hash) {
1094                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1095                 goto err_out;
1096         }
1097
1098         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1099
1100         printk("nf_conntrack version %s (%u buckets, %d max)\n",
1101                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1102                nf_conntrack_max);
1103
1104         nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1105                                                 sizeof(struct nf_conn),
1106                                                 0, 0, NULL);
1107         if (!nf_conntrack_cachep) {
1108                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1109                 goto err_free_hash;
1110         }
1111
1112         ret = nf_conntrack_proto_init();
1113         if (ret < 0)
1114                 goto err_free_conntrack_slab;
1115
1116         ret = nf_conntrack_expect_init();
1117         if (ret < 0)
1118                 goto out_fini_proto;
1119
1120         ret = nf_conntrack_helper_init();
1121         if (ret < 0)
1122                 goto out_fini_expect;
1123
1124         /* For use by REJECT target */
1125         rcu_assign_pointer(ip_ct_attach, __nf_conntrack_attach);
1126         rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1127
1128         /* Set up fake conntrack:
1129             - to never be deleted, not in any hashes */
1130         atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1131         /*  - and look it like as a confirmed connection */
1132         set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1133
1134         return ret;
1135
1136 out_fini_expect:
1137         nf_conntrack_expect_fini();
1138 out_fini_proto:
1139         nf_conntrack_proto_fini();
1140 err_free_conntrack_slab:
1141         kmem_cache_destroy(nf_conntrack_cachep);
1142 err_free_hash:
1143         nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
1144                              nf_conntrack_htable_size);
1145 err_out:
1146         return -ENOMEM;
1147 }