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
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 * (C) 2005-2012 Patrick McHardy <kaber@trash.net>
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License version 2 as
12 * published by the Free Software Foundation.
15 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
17 #include <linux/types.h>
18 #include <linux/netfilter.h>
19 #include <linux/module.h>
20 #include <linux/sched.h>
21 #include <linux/skbuff.h>
22 #include <linux/proc_fs.h>
23 #include <linux/vmalloc.h>
24 #include <linux/stddef.h>
25 #include <linux/slab.h>
26 #include <linux/random.h>
27 #include <linux/jhash.h>
28 #include <linux/err.h>
29 #include <linux/percpu.h>
30 #include <linux/moduleparam.h>
31 #include <linux/notifier.h>
32 #include <linux/kernel.h>
33 #include <linux/netdevice.h>
34 #include <linux/socket.h>
36 #include <linux/nsproxy.h>
37 #include <linux/rculist_nulls.h>
39 #include <net/netfilter/nf_conntrack.h>
40 #include <net/netfilter/nf_conntrack_l3proto.h>
41 #include <net/netfilter/nf_conntrack_l4proto.h>
42 #include <net/netfilter/nf_conntrack_expect.h>
43 #include <net/netfilter/nf_conntrack_helper.h>
44 #include <net/netfilter/nf_conntrack_seqadj.h>
45 #include <net/netfilter/nf_conntrack_core.h>
46 #include <net/netfilter/nf_conntrack_extend.h>
47 #include <net/netfilter/nf_conntrack_acct.h>
48 #include <net/netfilter/nf_conntrack_ecache.h>
49 #include <net/netfilter/nf_conntrack_zones.h>
50 #include <net/netfilter/nf_conntrack_timestamp.h>
51 #include <net/netfilter/nf_conntrack_timeout.h>
52 #include <net/netfilter/nf_conntrack_labels.h>
53 #include <net/netfilter/nf_conntrack_synproxy.h>
54 #include <net/netfilter/nf_nat.h>
55 #include <net/netfilter/nf_nat_core.h>
56 #include <net/netfilter/nf_nat_helper.h>
57 #include <net/netns/hash.h>
59 #define NF_CONNTRACK_VERSION "0.5.0"
61 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
62 enum nf_nat_manip_type manip,
63 const struct nlattr *attr) __read_mostly;
64 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
66 __cacheline_aligned_in_smp spinlock_t nf_conntrack_locks[CONNTRACK_LOCKS];
67 EXPORT_SYMBOL_GPL(nf_conntrack_locks);
69 __cacheline_aligned_in_smp DEFINE_SPINLOCK(nf_conntrack_expect_lock);
70 EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock);
72 struct hlist_nulls_head *nf_conntrack_hash __read_mostly;
73 EXPORT_SYMBOL_GPL(nf_conntrack_hash);
75 static __read_mostly struct kmem_cache *nf_conntrack_cachep;
76 static __read_mostly spinlock_t nf_conntrack_locks_all_lock;
77 static __read_mostly seqcount_t nf_conntrack_generation;
78 static __read_mostly DEFINE_SPINLOCK(nf_conntrack_locks_all_lock);
79 static __read_mostly bool nf_conntrack_locks_all;
81 void nf_conntrack_lock(spinlock_t *lock) __acquires(lock)
84 while (unlikely(nf_conntrack_locks_all)) {
88 * Order the 'nf_conntrack_locks_all' load vs. the
89 * spin_unlock_wait() loads below, to ensure
90 * that 'nf_conntrack_locks_all_lock' is indeed held:
92 smp_rmb(); /* spin_lock(&nf_conntrack_locks_all_lock) */
93 spin_unlock_wait(&nf_conntrack_locks_all_lock);
97 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
99 static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2)
101 h1 %= CONNTRACK_LOCKS;
102 h2 %= CONNTRACK_LOCKS;
103 spin_unlock(&nf_conntrack_locks[h1]);
105 spin_unlock(&nf_conntrack_locks[h2]);
108 /* return true if we need to recompute hashes (in case hash table was resized) */
109 static bool nf_conntrack_double_lock(struct net *net, unsigned int h1,
110 unsigned int h2, unsigned int sequence)
112 h1 %= CONNTRACK_LOCKS;
113 h2 %= CONNTRACK_LOCKS;
115 nf_conntrack_lock(&nf_conntrack_locks[h1]);
117 spin_lock_nested(&nf_conntrack_locks[h2],
118 SINGLE_DEPTH_NESTING);
120 nf_conntrack_lock(&nf_conntrack_locks[h2]);
121 spin_lock_nested(&nf_conntrack_locks[h1],
122 SINGLE_DEPTH_NESTING);
124 if (read_seqcount_retry(&nf_conntrack_generation, sequence)) {
125 nf_conntrack_double_unlock(h1, h2);
131 static void nf_conntrack_all_lock(void)
135 spin_lock(&nf_conntrack_locks_all_lock);
136 nf_conntrack_locks_all = true;
139 * Order the above store of 'nf_conntrack_locks_all' against
140 * the spin_unlock_wait() loads below, such that if
141 * nf_conntrack_lock() observes 'nf_conntrack_locks_all'
142 * we must observe nf_conntrack_locks[] held:
144 smp_mb(); /* spin_lock(&nf_conntrack_locks_all_lock) */
146 for (i = 0; i < CONNTRACK_LOCKS; i++) {
147 spin_unlock_wait(&nf_conntrack_locks[i]);
151 static void nf_conntrack_all_unlock(void)
154 * All prior stores must be complete before we clear
155 * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock()
156 * might observe the false value but not the entire
159 smp_store_release(&nf_conntrack_locks_all, false);
160 spin_unlock(&nf_conntrack_locks_all_lock);
163 unsigned int nf_conntrack_htable_size __read_mostly;
164 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
166 unsigned int nf_conntrack_max __read_mostly;
167 EXPORT_SYMBOL_GPL(nf_conntrack_max);
169 DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
170 EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
172 static unsigned int nf_conntrack_hash_rnd __read_mostly;
174 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple,
175 const struct net *net)
180 get_random_once(&nf_conntrack_hash_rnd, sizeof(nf_conntrack_hash_rnd));
182 /* The direction must be ignored, so we hash everything up to the
183 * destination ports (which is a multiple of 4) and treat the last
184 * three bytes manually.
186 seed = nf_conntrack_hash_rnd ^ net_hash_mix(net);
187 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
188 return jhash2((u32 *)tuple, n, seed ^
189 (((__force __u16)tuple->dst.u.all << 16) |
190 tuple->dst.protonum));
193 static u32 scale_hash(u32 hash)
195 return reciprocal_scale(hash, nf_conntrack_htable_size);
198 static u32 __hash_conntrack(const struct net *net,
199 const struct nf_conntrack_tuple *tuple,
202 return reciprocal_scale(hash_conntrack_raw(tuple, net), size);
205 static u32 hash_conntrack(const struct net *net,
206 const struct nf_conntrack_tuple *tuple)
208 return scale_hash(hash_conntrack_raw(tuple, net));
212 nf_ct_get_tuple(const struct sk_buff *skb,
214 unsigned int dataoff,
218 struct nf_conntrack_tuple *tuple,
219 const struct nf_conntrack_l3proto *l3proto,
220 const struct nf_conntrack_l4proto *l4proto)
222 memset(tuple, 0, sizeof(*tuple));
224 tuple->src.l3num = l3num;
225 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
228 tuple->dst.protonum = protonum;
229 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
231 return l4proto->pkt_to_tuple(skb, dataoff, net, tuple);
233 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
235 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
237 struct net *net, struct nf_conntrack_tuple *tuple)
239 struct nf_conntrack_l3proto *l3proto;
240 struct nf_conntrack_l4proto *l4proto;
241 unsigned int protoff;
247 l3proto = __nf_ct_l3proto_find(l3num);
248 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
249 if (ret != NF_ACCEPT) {
254 l4proto = __nf_ct_l4proto_find(l3num, protonum);
256 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple,
262 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
265 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
266 const struct nf_conntrack_tuple *orig,
267 const struct nf_conntrack_l3proto *l3proto,
268 const struct nf_conntrack_l4proto *l4proto)
270 memset(inverse, 0, sizeof(*inverse));
272 inverse->src.l3num = orig->src.l3num;
273 if (l3proto->invert_tuple(inverse, orig) == 0)
276 inverse->dst.dir = !orig->dst.dir;
278 inverse->dst.protonum = orig->dst.protonum;
279 return l4proto->invert_tuple(inverse, orig);
281 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
284 clean_from_lists(struct nf_conn *ct)
286 pr_debug("clean_from_lists(%p)\n", ct);
287 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
288 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
290 /* Destroy all pending expectations */
291 nf_ct_remove_expectations(ct);
294 /* must be called with local_bh_disable */
295 static void nf_ct_add_to_dying_list(struct nf_conn *ct)
297 struct ct_pcpu *pcpu;
299 /* add this conntrack to the (per cpu) dying list */
300 ct->cpu = smp_processor_id();
301 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
303 spin_lock(&pcpu->lock);
304 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
306 spin_unlock(&pcpu->lock);
309 /* must be called with local_bh_disable */
310 static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct)
312 struct ct_pcpu *pcpu;
314 /* add this conntrack to the (per cpu) unconfirmed list */
315 ct->cpu = smp_processor_id();
316 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
318 spin_lock(&pcpu->lock);
319 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
321 spin_unlock(&pcpu->lock);
324 /* must be called with local_bh_disable */
325 static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct)
327 struct ct_pcpu *pcpu;
329 /* We overload first tuple to link into unconfirmed or dying list.*/
330 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
332 spin_lock(&pcpu->lock);
333 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
334 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
335 spin_unlock(&pcpu->lock);
338 /* Released via destroy_conntrack() */
339 struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
340 const struct nf_conntrack_zone *zone,
343 struct nf_conn *tmpl;
345 tmpl = kzalloc(sizeof(*tmpl), flags);
349 tmpl->status = IPS_TEMPLATE;
350 write_pnet(&tmpl->ct_net, net);
351 nf_ct_zone_add(tmpl, zone);
352 atomic_set(&tmpl->ct_general.use, 0);
356 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
358 void nf_ct_tmpl_free(struct nf_conn *tmpl)
360 nf_ct_ext_destroy(tmpl);
361 nf_ct_ext_free(tmpl);
364 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free);
367 destroy_conntrack(struct nf_conntrack *nfct)
369 struct nf_conn *ct = (struct nf_conn *)nfct;
370 struct net *net = nf_ct_net(ct);
371 struct nf_conntrack_l4proto *l4proto;
373 pr_debug("destroy_conntrack(%p)\n", ct);
374 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
375 NF_CT_ASSERT(!timer_pending(&ct->timeout));
377 if (unlikely(nf_ct_is_template(ct))) {
382 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
383 if (l4proto->destroy)
384 l4proto->destroy(ct);
389 /* Expectations will have been removed in clean_from_lists,
390 * except TFTP can create an expectation on the first packet,
391 * before connection is in the list, so we need to clean here,
394 nf_ct_remove_expectations(ct);
396 nf_ct_del_from_dying_or_unconfirmed_list(ct);
398 NF_CT_STAT_INC(net, delete);
402 nf_ct_put(ct->master);
404 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
405 nf_conntrack_free(ct);
408 static void nf_ct_delete_from_lists(struct nf_conn *ct)
410 struct net *net = nf_ct_net(ct);
411 unsigned int hash, reply_hash;
412 unsigned int sequence;
414 nf_ct_helper_destroy(ct);
418 sequence = read_seqcount_begin(&nf_conntrack_generation);
419 hash = hash_conntrack(net,
420 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
421 reply_hash = hash_conntrack(net,
422 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
423 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
425 clean_from_lists(ct);
426 nf_conntrack_double_unlock(hash, reply_hash);
428 nf_ct_add_to_dying_list(ct);
430 NF_CT_STAT_INC(net, delete_list);
434 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
436 struct nf_conn_tstamp *tstamp;
438 tstamp = nf_conn_tstamp_find(ct);
439 if (tstamp && tstamp->stop == 0)
440 tstamp->stop = ktime_get_real_ns();
442 if (nf_ct_is_dying(ct))
445 if (nf_conntrack_event_report(IPCT_DESTROY, ct,
446 portid, report) < 0) {
447 /* destroy event was not delivered */
448 nf_ct_delete_from_lists(ct);
449 nf_conntrack_ecache_delayed_work(nf_ct_net(ct));
453 nf_conntrack_ecache_work(nf_ct_net(ct));
454 set_bit(IPS_DYING_BIT, &ct->status);
456 nf_ct_delete_from_lists(ct);
460 EXPORT_SYMBOL_GPL(nf_ct_delete);
462 static void death_by_timeout(unsigned long ul_conntrack)
464 nf_ct_delete((struct nf_conn *)ul_conntrack, 0, 0);
468 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
469 const struct nf_conntrack_tuple *tuple,
470 const struct nf_conntrack_zone *zone,
471 const struct net *net)
473 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
475 /* A conntrack can be recreated with the equal tuple,
476 * so we need to check that the conntrack is confirmed
478 return nf_ct_tuple_equal(tuple, &h->tuple) &&
479 nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) &&
480 nf_ct_is_confirmed(ct) &&
481 net_eq(net, nf_ct_net(ct));
484 /* must be called with rcu read lock held */
485 void nf_conntrack_get_ht(struct hlist_nulls_head **hash, unsigned int *hsize)
487 struct hlist_nulls_head *hptr;
488 unsigned int sequence, hsz;
491 sequence = read_seqcount_begin(&nf_conntrack_generation);
492 hsz = nf_conntrack_htable_size;
493 hptr = nf_conntrack_hash;
494 } while (read_seqcount_retry(&nf_conntrack_generation, sequence));
499 EXPORT_SYMBOL_GPL(nf_conntrack_get_ht);
503 * - Caller must take a reference on returned object
504 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
506 static struct nf_conntrack_tuple_hash *
507 ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone,
508 const struct nf_conntrack_tuple *tuple, u32 hash)
510 struct nf_conntrack_tuple_hash *h;
511 struct hlist_nulls_head *ct_hash;
512 struct hlist_nulls_node *n;
513 unsigned int bucket, sequence;
517 sequence = read_seqcount_begin(&nf_conntrack_generation);
518 bucket = scale_hash(hash);
519 ct_hash = nf_conntrack_hash;
520 } while (read_seqcount_retry(&nf_conntrack_generation, sequence));
522 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[bucket], hnnode) {
523 if (nf_ct_key_equal(h, tuple, zone, net)) {
524 NF_CT_STAT_INC_ATOMIC(net, found);
527 NF_CT_STAT_INC_ATOMIC(net, searched);
530 * if the nulls value we got at the end of this lookup is
531 * not the expected one, we must restart lookup.
532 * We probably met an item that was moved to another chain.
534 if (get_nulls_value(n) != bucket) {
535 NF_CT_STAT_INC_ATOMIC(net, search_restart);
542 /* Find a connection corresponding to a tuple. */
543 static struct nf_conntrack_tuple_hash *
544 __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
545 const struct nf_conntrack_tuple *tuple, u32 hash)
547 struct nf_conntrack_tuple_hash *h;
552 h = ____nf_conntrack_find(net, zone, tuple, hash);
554 ct = nf_ct_tuplehash_to_ctrack(h);
555 if (unlikely(nf_ct_is_dying(ct) ||
556 !atomic_inc_not_zero(&ct->ct_general.use)))
559 if (unlikely(!nf_ct_key_equal(h, tuple, zone, net))) {
570 struct nf_conntrack_tuple_hash *
571 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
572 const struct nf_conntrack_tuple *tuple)
574 return __nf_conntrack_find_get(net, zone, tuple,
575 hash_conntrack_raw(tuple, net));
577 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
579 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
581 unsigned int reply_hash)
583 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
584 &nf_conntrack_hash[hash]);
585 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
586 &nf_conntrack_hash[reply_hash]);
590 nf_conntrack_hash_check_insert(struct nf_conn *ct)
592 const struct nf_conntrack_zone *zone;
593 struct net *net = nf_ct_net(ct);
594 unsigned int hash, reply_hash;
595 struct nf_conntrack_tuple_hash *h;
596 struct hlist_nulls_node *n;
597 unsigned int sequence;
599 zone = nf_ct_zone(ct);
603 sequence = read_seqcount_begin(&nf_conntrack_generation);
604 hash = hash_conntrack(net,
605 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
606 reply_hash = hash_conntrack(net,
607 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
608 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
610 /* See if there's one in the list already, including reverse */
611 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
612 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
616 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
617 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
621 add_timer(&ct->timeout);
623 /* The caller holds a reference to this object */
624 atomic_set(&ct->ct_general.use, 2);
625 __nf_conntrack_hash_insert(ct, hash, reply_hash);
626 nf_conntrack_double_unlock(hash, reply_hash);
627 NF_CT_STAT_INC(net, insert);
632 nf_conntrack_double_unlock(hash, reply_hash);
633 NF_CT_STAT_INC(net, insert_failed);
637 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
639 static inline void nf_ct_acct_update(struct nf_conn *ct,
640 enum ip_conntrack_info ctinfo,
643 struct nf_conn_acct *acct;
645 acct = nf_conn_acct_find(ct);
647 struct nf_conn_counter *counter = acct->counter;
649 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
650 atomic64_add(len, &counter[CTINFO2DIR(ctinfo)].bytes);
654 static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
655 const struct nf_conn *loser_ct)
657 struct nf_conn_acct *acct;
659 acct = nf_conn_acct_find(loser_ct);
661 struct nf_conn_counter *counter = acct->counter;
664 /* u32 should be fine since we must have seen one packet. */
665 bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes);
666 nf_ct_acct_update(ct, ctinfo, bytes);
670 /* Resolve race on insertion if this protocol allows this. */
671 static int nf_ct_resolve_clash(struct net *net, struct sk_buff *skb,
672 enum ip_conntrack_info ctinfo,
673 struct nf_conntrack_tuple_hash *h)
675 /* This is the conntrack entry already in hashes that won race. */
676 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
677 struct nf_conntrack_l4proto *l4proto;
679 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
680 if (l4proto->allow_clash &&
682 !nf_ct_is_dying(ct) &&
683 atomic_inc_not_zero(&ct->ct_general.use)) {
684 nf_ct_acct_merge(ct, ctinfo, (struct nf_conn *)skb->nfct);
685 nf_conntrack_put(skb->nfct);
686 /* Assign conntrack already in hashes to this skbuff. Don't
687 * modify skb->nfctinfo to ensure consistent stateful filtering.
689 skb->nfct = &ct->ct_general;
692 NF_CT_STAT_INC(net, drop);
696 /* Confirm a connection given skb; places it in hash table */
698 __nf_conntrack_confirm(struct sk_buff *skb)
700 const struct nf_conntrack_zone *zone;
701 unsigned int hash, reply_hash;
702 struct nf_conntrack_tuple_hash *h;
704 struct nf_conn_help *help;
705 struct nf_conn_tstamp *tstamp;
706 struct hlist_nulls_node *n;
707 enum ip_conntrack_info ctinfo;
709 unsigned int sequence;
712 ct = nf_ct_get(skb, &ctinfo);
715 /* ipt_REJECT uses nf_conntrack_attach to attach related
716 ICMP/TCP RST packets in other direction. Actual packet
717 which created connection will be IP_CT_NEW or for an
718 expected connection, IP_CT_RELATED. */
719 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
722 zone = nf_ct_zone(ct);
726 sequence = read_seqcount_begin(&nf_conntrack_generation);
727 /* reuse the hash saved before */
728 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
729 hash = scale_hash(hash);
730 reply_hash = hash_conntrack(net,
731 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
733 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
735 /* We're not in hash table, and we refuse to set up related
736 * connections for unconfirmed conns. But packet copies and
737 * REJECT will give spurious warnings here.
739 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
741 /* No external references means no one else could have
744 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
745 pr_debug("Confirming conntrack %p\n", ct);
746 /* We have to check the DYING flag after unlink to prevent
747 * a race against nf_ct_get_next_corpse() possibly called from
748 * user context, else we insert an already 'dead' hash, blocking
749 * further use of that particular connection -JM.
751 nf_ct_del_from_dying_or_unconfirmed_list(ct);
753 if (unlikely(nf_ct_is_dying(ct))) {
754 nf_ct_add_to_dying_list(ct);
758 /* See if there's one in the list already, including reverse:
759 NAT could have grabbed it without realizing, since we're
760 not in the hash. If there is, we lost race. */
761 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
762 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
766 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
767 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
771 /* Timer relative to confirmation time, not original
772 setting time, otherwise we'd get timer wrap in
773 weird delay cases. */
774 ct->timeout.expires += jiffies;
775 add_timer(&ct->timeout);
776 atomic_inc(&ct->ct_general.use);
777 ct->status |= IPS_CONFIRMED;
779 /* set conntrack timestamp, if enabled. */
780 tstamp = nf_conn_tstamp_find(ct);
782 if (skb->tstamp.tv64 == 0)
783 __net_timestamp(skb);
785 tstamp->start = ktime_to_ns(skb->tstamp);
787 /* Since the lookup is lockless, hash insertion must be done after
788 * starting the timer and setting the CONFIRMED bit. The RCU barriers
789 * guarantee that no other CPU can find the conntrack before the above
790 * stores are visible.
792 __nf_conntrack_hash_insert(ct, hash, reply_hash);
793 nf_conntrack_double_unlock(hash, reply_hash);
794 NF_CT_STAT_INC(net, insert);
797 help = nfct_help(ct);
798 if (help && help->helper)
799 nf_conntrack_event_cache(IPCT_HELPER, ct);
801 nf_conntrack_event_cache(master_ct(ct) ?
802 IPCT_RELATED : IPCT_NEW, ct);
806 nf_ct_add_to_dying_list(ct);
807 ret = nf_ct_resolve_clash(net, skb, ctinfo, h);
809 nf_conntrack_double_unlock(hash, reply_hash);
810 NF_CT_STAT_INC(net, insert_failed);
814 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
816 /* Returns true if a connection correspondings to the tuple (required
819 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
820 const struct nf_conn *ignored_conntrack)
822 struct net *net = nf_ct_net(ignored_conntrack);
823 const struct nf_conntrack_zone *zone;
824 struct nf_conntrack_tuple_hash *h;
825 struct hlist_nulls_head *ct_hash;
826 unsigned int hash, sequence;
827 struct hlist_nulls_node *n;
830 zone = nf_ct_zone(ignored_conntrack);
834 sequence = read_seqcount_begin(&nf_conntrack_generation);
835 hash = hash_conntrack(net, tuple);
836 ct_hash = nf_conntrack_hash;
837 } while (read_seqcount_retry(&nf_conntrack_generation, sequence));
839 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) {
840 ct = nf_ct_tuplehash_to_ctrack(h);
841 if (ct != ignored_conntrack &&
842 nf_ct_key_equal(h, tuple, zone, net)) {
843 NF_CT_STAT_INC_ATOMIC(net, found);
847 NF_CT_STAT_INC_ATOMIC(net, searched);
853 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
855 #define NF_CT_EVICTION_RANGE 8
857 /* There's a small race here where we may free a just-assured
858 connection. Too bad: we're in trouble anyway. */
859 static unsigned int early_drop_list(struct net *net,
860 struct hlist_nulls_head *head)
862 struct nf_conntrack_tuple_hash *h;
863 struct hlist_nulls_node *n;
864 unsigned int drops = 0;
867 hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) {
868 tmp = nf_ct_tuplehash_to_ctrack(h);
870 if (test_bit(IPS_ASSURED_BIT, &tmp->status) ||
871 !net_eq(nf_ct_net(tmp), net) ||
875 if (!atomic_inc_not_zero(&tmp->ct_general.use))
878 /* kill only if still in same netns -- might have moved due to
879 * SLAB_DESTROY_BY_RCU rules.
881 * We steal the timer reference. If that fails timer has
882 * already fired or someone else deleted it. Just drop ref
883 * and move to next entry.
885 if (net_eq(nf_ct_net(tmp), net) &&
886 nf_ct_is_confirmed(tmp) &&
887 del_timer(&tmp->timeout) &&
888 nf_ct_delete(tmp, 0, 0))
897 static noinline int early_drop(struct net *net, unsigned int _hash)
901 for (i = 0; i < NF_CT_EVICTION_RANGE; i++) {
902 struct hlist_nulls_head *ct_hash;
903 unsigned hash, sequence, drops;
907 sequence = read_seqcount_begin(&nf_conntrack_generation);
908 hash = scale_hash(_hash++);
909 ct_hash = nf_conntrack_hash;
910 } while (read_seqcount_retry(&nf_conntrack_generation, sequence));
912 drops = early_drop_list(net, &ct_hash[hash]);
916 NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops);
924 static struct nf_conn *
925 __nf_conntrack_alloc(struct net *net,
926 const struct nf_conntrack_zone *zone,
927 const struct nf_conntrack_tuple *orig,
928 const struct nf_conntrack_tuple *repl,
933 /* We don't want any race condition at early drop stage */
934 atomic_inc(&net->ct.count);
936 if (nf_conntrack_max &&
937 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
938 if (!early_drop(net, hash)) {
939 atomic_dec(&net->ct.count);
940 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
941 return ERR_PTR(-ENOMEM);
946 * Do not use kmem_cache_zalloc(), as this cache uses
947 * SLAB_DESTROY_BY_RCU.
949 ct = kmem_cache_alloc(nf_conntrack_cachep, gfp);
953 spin_lock_init(&ct->lock);
954 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
955 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
956 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
957 /* save hash for reusing when confirming */
958 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
960 /* Don't set timer yet: wait for confirmation */
961 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
962 write_pnet(&ct->ct_net, net);
963 memset(&ct->__nfct_init_offset[0], 0,
964 offsetof(struct nf_conn, proto) -
965 offsetof(struct nf_conn, __nfct_init_offset[0]));
967 nf_ct_zone_add(ct, zone);
969 /* Because we use RCU lookups, we set ct_general.use to zero before
970 * this is inserted in any list.
972 atomic_set(&ct->ct_general.use, 0);
975 atomic_dec(&net->ct.count);
976 return ERR_PTR(-ENOMEM);
979 struct nf_conn *nf_conntrack_alloc(struct net *net,
980 const struct nf_conntrack_zone *zone,
981 const struct nf_conntrack_tuple *orig,
982 const struct nf_conntrack_tuple *repl,
985 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
987 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
989 void nf_conntrack_free(struct nf_conn *ct)
991 struct net *net = nf_ct_net(ct);
993 /* A freed object has refcnt == 0, that's
994 * the golden rule for SLAB_DESTROY_BY_RCU
996 NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0);
998 nf_ct_ext_destroy(ct);
1000 kmem_cache_free(nf_conntrack_cachep, ct);
1001 smp_mb__before_atomic();
1002 atomic_dec(&net->ct.count);
1004 EXPORT_SYMBOL_GPL(nf_conntrack_free);
1007 /* Allocate a new conntrack: we return -ENOMEM if classification
1008 failed due to stress. Otherwise it really is unclassifiable. */
1009 static struct nf_conntrack_tuple_hash *
1010 init_conntrack(struct net *net, struct nf_conn *tmpl,
1011 const struct nf_conntrack_tuple *tuple,
1012 struct nf_conntrack_l3proto *l3proto,
1013 struct nf_conntrack_l4proto *l4proto,
1014 struct sk_buff *skb,
1015 unsigned int dataoff, u32 hash)
1018 struct nf_conn_help *help;
1019 struct nf_conntrack_tuple repl_tuple;
1020 struct nf_conntrack_ecache *ecache;
1021 struct nf_conntrack_expect *exp = NULL;
1022 const struct nf_conntrack_zone *zone;
1023 struct nf_conn_timeout *timeout_ext;
1024 struct nf_conntrack_zone tmp;
1025 unsigned int *timeouts;
1027 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
1028 pr_debug("Can't invert tuple.\n");
1032 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1033 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
1036 return (struct nf_conntrack_tuple_hash *)ct;
1038 if (tmpl && nfct_synproxy(tmpl)) {
1039 nfct_seqadj_ext_add(ct);
1040 nfct_synproxy_ext_add(ct);
1043 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
1045 timeouts = nf_ct_timeout_data(timeout_ext);
1046 if (unlikely(!timeouts))
1047 timeouts = l4proto->get_timeouts(net);
1049 timeouts = l4proto->get_timeouts(net);
1052 if (!l4proto->new(ct, skb, dataoff, timeouts)) {
1053 nf_conntrack_free(ct);
1054 pr_debug("can't track with proto module\n");
1059 nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout),
1062 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
1063 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
1064 nf_ct_labels_ext_add(ct);
1066 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
1067 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
1068 ecache ? ecache->expmask : 0,
1072 if (net->ct.expect_count) {
1073 spin_lock(&nf_conntrack_expect_lock);
1074 exp = nf_ct_find_expectation(net, zone, tuple);
1076 pr_debug("expectation arrives ct=%p exp=%p\n",
1078 /* Welcome, Mr. Bond. We've been expecting you... */
1079 __set_bit(IPS_EXPECTED_BIT, &ct->status);
1080 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1081 ct->master = exp->master;
1083 help = nf_ct_helper_ext_add(ct, exp->helper,
1086 rcu_assign_pointer(help->helper, exp->helper);
1089 #ifdef CONFIG_NF_CONNTRACK_MARK
1090 ct->mark = exp->master->mark;
1092 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1093 ct->secmark = exp->master->secmark;
1095 NF_CT_STAT_INC(net, expect_new);
1097 spin_unlock(&nf_conntrack_expect_lock);
1100 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
1101 NF_CT_STAT_INC(net, new);
1104 /* Now it is inserted into the unconfirmed list, bump refcount */
1105 nf_conntrack_get(&ct->ct_general);
1106 nf_ct_add_to_unconfirmed_list(ct);
1112 exp->expectfn(ct, exp);
1113 nf_ct_expect_put(exp);
1116 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1119 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
1120 static inline struct nf_conn *
1121 resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
1122 struct sk_buff *skb,
1123 unsigned int dataoff,
1126 struct nf_conntrack_l3proto *l3proto,
1127 struct nf_conntrack_l4proto *l4proto,
1129 enum ip_conntrack_info *ctinfo)
1131 const struct nf_conntrack_zone *zone;
1132 struct nf_conntrack_tuple tuple;
1133 struct nf_conntrack_tuple_hash *h;
1134 struct nf_conntrack_zone tmp;
1138 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1139 dataoff, l3num, protonum, net, &tuple, l3proto,
1141 pr_debug("Can't get tuple\n");
1145 /* look for tuple match */
1146 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1147 hash = hash_conntrack_raw(&tuple, net);
1148 h = __nf_conntrack_find_get(net, zone, &tuple, hash);
1150 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
1151 skb, dataoff, hash);
1157 ct = nf_ct_tuplehash_to_ctrack(h);
1159 /* It exists; we have (non-exclusive) reference. */
1160 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1161 *ctinfo = IP_CT_ESTABLISHED_REPLY;
1162 /* Please set reply bit if this packet OK */
1165 /* Once we've had two way comms, always ESTABLISHED. */
1166 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1167 pr_debug("normal packet for %p\n", ct);
1168 *ctinfo = IP_CT_ESTABLISHED;
1169 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1170 pr_debug("related packet for %p\n", ct);
1171 *ctinfo = IP_CT_RELATED;
1173 pr_debug("new packet for %p\n", ct);
1174 *ctinfo = IP_CT_NEW;
1178 skb->nfct = &ct->ct_general;
1179 skb->nfctinfo = *ctinfo;
1184 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
1185 struct sk_buff *skb)
1187 struct nf_conn *ct, *tmpl = NULL;
1188 enum ip_conntrack_info ctinfo;
1189 struct nf_conntrack_l3proto *l3proto;
1190 struct nf_conntrack_l4proto *l4proto;
1191 unsigned int *timeouts;
1192 unsigned int dataoff;
1198 /* Previously seen (loopback or untracked)? Ignore. */
1199 tmpl = (struct nf_conn *)skb->nfct;
1200 if (!nf_ct_is_template(tmpl)) {
1201 NF_CT_STAT_INC_ATOMIC(net, ignore);
1207 /* rcu_read_lock()ed by nf_hook_slow */
1208 l3proto = __nf_ct_l3proto_find(pf);
1209 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
1210 &dataoff, &protonum);
1212 pr_debug("not prepared to track yet or error occurred\n");
1213 NF_CT_STAT_INC_ATOMIC(net, error);
1214 NF_CT_STAT_INC_ATOMIC(net, invalid);
1219 l4proto = __nf_ct_l4proto_find(pf, protonum);
1221 /* It may be an special packet, error, unclean...
1222 * inverse of the return code tells to the netfilter
1223 * core what to do with the packet. */
1224 if (l4proto->error != NULL) {
1225 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
1228 NF_CT_STAT_INC_ATOMIC(net, error);
1229 NF_CT_STAT_INC_ATOMIC(net, invalid);
1233 /* ICMP[v6] protocol trackers may assign one conntrack. */
1238 ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
1239 l3proto, l4proto, &set_reply, &ctinfo);
1241 /* Not valid part of a connection */
1242 NF_CT_STAT_INC_ATOMIC(net, invalid);
1248 /* Too stressed to deal. */
1249 NF_CT_STAT_INC_ATOMIC(net, drop);
1254 NF_CT_ASSERT(skb->nfct);
1256 /* Decide what timeout policy we want to apply to this flow. */
1257 timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1259 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1261 /* Invalid: inverse of the return code tells
1262 * the netfilter core what to do */
1263 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1264 nf_conntrack_put(skb->nfct);
1266 NF_CT_STAT_INC_ATOMIC(net, invalid);
1267 if (ret == -NF_DROP)
1268 NF_CT_STAT_INC_ATOMIC(net, drop);
1273 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1274 nf_conntrack_event_cache(IPCT_REPLY, ct);
1277 /* Special case: we have to repeat this hook, assign the
1278 * template again to this packet. We assume that this packet
1279 * has no conntrack assigned. This is used by nf_ct_tcp. */
1280 if (ret == NF_REPEAT)
1281 skb->nfct = (struct nf_conntrack *)tmpl;
1288 EXPORT_SYMBOL_GPL(nf_conntrack_in);
1290 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1291 const struct nf_conntrack_tuple *orig)
1296 ret = nf_ct_invert_tuple(inverse, orig,
1297 __nf_ct_l3proto_find(orig->src.l3num),
1298 __nf_ct_l4proto_find(orig->src.l3num,
1299 orig->dst.protonum));
1303 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1305 /* Alter reply tuple (maybe alter helper). This is for NAT, and is
1306 implicitly racy: see __nf_conntrack_confirm */
1307 void nf_conntrack_alter_reply(struct nf_conn *ct,
1308 const struct nf_conntrack_tuple *newreply)
1310 struct nf_conn_help *help = nfct_help(ct);
1312 /* Should be unconfirmed, so not in hash table yet */
1313 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1315 pr_debug("Altering reply tuple of %p to ", ct);
1316 nf_ct_dump_tuple(newreply);
1318 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1319 if (ct->master || (help && !hlist_empty(&help->expectations)))
1323 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1326 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1328 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
1329 void __nf_ct_refresh_acct(struct nf_conn *ct,
1330 enum ip_conntrack_info ctinfo,
1331 const struct sk_buff *skb,
1332 unsigned long extra_jiffies,
1335 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
1338 /* Only update if this is not a fixed timeout */
1339 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1342 /* If not in hash table, timer will not be active yet */
1343 if (!nf_ct_is_confirmed(ct)) {
1344 ct->timeout.expires = extra_jiffies;
1346 unsigned long newtime = jiffies + extra_jiffies;
1348 /* Only update the timeout if the new timeout is at least
1349 HZ jiffies from the old timeout. Need del_timer for race
1350 avoidance (may already be dying). */
1351 if (newtime - ct->timeout.expires >= HZ)
1352 mod_timer_pending(&ct->timeout, newtime);
1357 nf_ct_acct_update(ct, ctinfo, skb->len);
1359 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1361 bool __nf_ct_kill_acct(struct nf_conn *ct,
1362 enum ip_conntrack_info ctinfo,
1363 const struct sk_buff *skb,
1367 nf_ct_acct_update(ct, ctinfo, skb->len);
1369 if (del_timer(&ct->timeout)) {
1370 ct->timeout.function((unsigned long)ct);
1375 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
1377 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1379 #include <linux/netfilter/nfnetlink.h>
1380 #include <linux/netfilter/nfnetlink_conntrack.h>
1381 #include <linux/mutex.h>
1383 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
1384 * in ip_conntrack_core, since we don't want the protocols to autoload
1385 * or depend on ctnetlink */
1386 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1387 const struct nf_conntrack_tuple *tuple)
1389 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1390 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1391 goto nla_put_failure;
1397 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1399 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1400 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
1401 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
1403 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1405 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1406 struct nf_conntrack_tuple *t)
1408 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1411 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1412 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1416 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1418 int nf_ct_port_nlattr_tuple_size(void)
1420 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1422 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1425 /* Used by ipt_REJECT and ip6t_REJECT. */
1426 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1429 enum ip_conntrack_info ctinfo;
1431 /* This ICMP is in reverse direction to the packet which caused it */
1432 ct = nf_ct_get(skb, &ctinfo);
1433 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1434 ctinfo = IP_CT_RELATED_REPLY;
1436 ctinfo = IP_CT_RELATED;
1438 /* Attach to new skbuff, and increment count */
1439 nskb->nfct = &ct->ct_general;
1440 nskb->nfctinfo = ctinfo;
1441 nf_conntrack_get(nskb->nfct);
1444 /* Bring out ya dead! */
1445 static struct nf_conn *
1446 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1447 void *data, unsigned int *bucket)
1449 struct nf_conntrack_tuple_hash *h;
1451 struct hlist_nulls_node *n;
1455 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
1456 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
1458 nf_conntrack_lock(lockp);
1459 if (*bucket < nf_conntrack_htable_size) {
1460 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) {
1461 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1463 ct = nf_ct_tuplehash_to_ctrack(h);
1464 if (net_eq(nf_ct_net(ct), net) &&
1474 for_each_possible_cpu(cpu) {
1475 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1477 spin_lock_bh(&pcpu->lock);
1478 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
1479 ct = nf_ct_tuplehash_to_ctrack(h);
1481 set_bit(IPS_DYING_BIT, &ct->status);
1483 spin_unlock_bh(&pcpu->lock);
1488 atomic_inc(&ct->ct_general.use);
1494 void nf_ct_iterate_cleanup(struct net *net,
1495 int (*iter)(struct nf_conn *i, void *data),
1496 void *data, u32 portid, int report)
1499 unsigned int bucket = 0;
1503 if (atomic_read(&net->ct.count) == 0)
1506 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1507 /* Time to push up daises... */
1508 if (del_timer(&ct->timeout))
1509 nf_ct_delete(ct, portid, report);
1511 /* ... else the timer will get him soon. */
1517 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1519 static int kill_all(struct nf_conn *i, void *data)
1524 void nf_ct_free_hashtable(void *hash, unsigned int size)
1526 if (is_vmalloc_addr(hash))
1529 free_pages((unsigned long)hash,
1530 get_order(sizeof(struct hlist_head) * size));
1532 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1534 static int untrack_refs(void)
1538 for_each_possible_cpu(cpu) {
1539 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1541 cnt += atomic_read(&ct->ct_general.use) - 1;
1546 void nf_conntrack_cleanup_start(void)
1548 RCU_INIT_POINTER(ip_ct_attach, NULL);
1551 void nf_conntrack_cleanup_end(void)
1553 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1554 while (untrack_refs() > 0)
1557 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size);
1559 nf_conntrack_proto_fini();
1560 nf_conntrack_seqadj_fini();
1561 nf_conntrack_labels_fini();
1562 nf_conntrack_helper_fini();
1563 nf_conntrack_timeout_fini();
1564 nf_conntrack_ecache_fini();
1565 nf_conntrack_tstamp_fini();
1566 nf_conntrack_acct_fini();
1567 nf_conntrack_expect_fini();
1569 kmem_cache_destroy(nf_conntrack_cachep);
1573 * Mishearing the voices in his head, our hero wonders how he's
1574 * supposed to kill the mall.
1576 void nf_conntrack_cleanup_net(struct net *net)
1580 list_add(&net->exit_list, &single);
1581 nf_conntrack_cleanup_net_list(&single);
1584 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1590 * This makes sure all current packets have passed through
1591 * netfilter framework. Roll on, two-stage module
1597 list_for_each_entry(net, net_exit_list, exit_list) {
1598 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1599 if (atomic_read(&net->ct.count) != 0)
1604 goto i_see_dead_people;
1607 list_for_each_entry(net, net_exit_list, exit_list) {
1608 nf_conntrack_proto_pernet_fini(net);
1609 nf_conntrack_helper_pernet_fini(net);
1610 nf_conntrack_ecache_pernet_fini(net);
1611 nf_conntrack_tstamp_pernet_fini(net);
1612 nf_conntrack_acct_pernet_fini(net);
1613 nf_conntrack_expect_pernet_fini(net);
1614 free_percpu(net->ct.stat);
1615 free_percpu(net->ct.pcpu_lists);
1619 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1621 struct hlist_nulls_head *hash;
1622 unsigned int nr_slots, i;
1625 if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head)))
1628 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1629 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1631 if (nr_slots > (UINT_MAX / sizeof(struct hlist_nulls_head)))
1634 sz = nr_slots * sizeof(struct hlist_nulls_head);
1635 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1641 for (i = 0; i < nr_slots; i++)
1642 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1646 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1648 int nf_conntrack_hash_resize(unsigned int hashsize)
1651 unsigned int old_size;
1652 struct hlist_nulls_head *hash, *old_hash;
1653 struct nf_conntrack_tuple_hash *h;
1659 hash = nf_ct_alloc_hashtable(&hashsize, 1);
1663 old_size = nf_conntrack_htable_size;
1664 if (old_size == hashsize) {
1665 nf_ct_free_hashtable(hash, hashsize);
1670 nf_conntrack_all_lock();
1671 write_seqcount_begin(&nf_conntrack_generation);
1673 /* Lookups in the old hash might happen in parallel, which means we
1674 * might get false negatives during connection lookup. New connections
1675 * created because of a false negative won't make it into the hash
1676 * though since that required taking the locks.
1679 for (i = 0; i < nf_conntrack_htable_size; i++) {
1680 while (!hlist_nulls_empty(&nf_conntrack_hash[i])) {
1681 h = hlist_nulls_entry(nf_conntrack_hash[i].first,
1682 struct nf_conntrack_tuple_hash, hnnode);
1683 ct = nf_ct_tuplehash_to_ctrack(h);
1684 hlist_nulls_del_rcu(&h->hnnode);
1685 bucket = __hash_conntrack(nf_ct_net(ct),
1686 &h->tuple, hashsize);
1687 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1690 old_size = nf_conntrack_htable_size;
1691 old_hash = nf_conntrack_hash;
1693 nf_conntrack_hash = hash;
1694 nf_conntrack_htable_size = hashsize;
1696 write_seqcount_end(&nf_conntrack_generation);
1697 nf_conntrack_all_unlock();
1701 nf_ct_free_hashtable(old_hash, old_size);
1705 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1707 unsigned int hashsize;
1710 if (current->nsproxy->net_ns != &init_net)
1713 /* On boot, we can set this without any fancy locking. */
1714 if (!nf_conntrack_htable_size)
1715 return param_set_uint(val, kp);
1717 rc = kstrtouint(val, 0, &hashsize);
1721 return nf_conntrack_hash_resize(hashsize);
1723 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1725 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1726 &nf_conntrack_htable_size, 0600);
1728 void nf_ct_untracked_status_or(unsigned long bits)
1732 for_each_possible_cpu(cpu)
1733 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1735 EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1737 int nf_conntrack_init_start(void)
1743 seqcount_init(&nf_conntrack_generation);
1745 for (i = 0; i < CONNTRACK_LOCKS; i++)
1746 spin_lock_init(&nf_conntrack_locks[i]);
1748 if (!nf_conntrack_htable_size) {
1749 /* Idea from tcp.c: use 1/16384 of memory.
1750 * On i386: 32MB machine has 512 buckets.
1751 * >= 1GB machines have 16384 buckets.
1752 * >= 4GB machines have 65536 buckets.
1754 nf_conntrack_htable_size
1755 = (((totalram_pages << PAGE_SHIFT) / 16384)
1756 / sizeof(struct hlist_head));
1757 if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
1758 nf_conntrack_htable_size = 65536;
1759 else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1760 nf_conntrack_htable_size = 16384;
1761 if (nf_conntrack_htable_size < 32)
1762 nf_conntrack_htable_size = 32;
1764 /* Use a max. factor of four by default to get the same max as
1765 * with the old struct list_heads. When a table size is given
1766 * we use the old value of 8 to avoid reducing the max.
1771 nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1);
1772 if (!nf_conntrack_hash)
1775 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1777 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1778 sizeof(struct nf_conn), 0,
1779 SLAB_DESTROY_BY_RCU | SLAB_HWCACHE_ALIGN, NULL);
1780 if (!nf_conntrack_cachep)
1783 printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1784 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1787 ret = nf_conntrack_expect_init();
1791 ret = nf_conntrack_acct_init();
1795 ret = nf_conntrack_tstamp_init();
1799 ret = nf_conntrack_ecache_init();
1803 ret = nf_conntrack_timeout_init();
1807 ret = nf_conntrack_helper_init();
1811 ret = nf_conntrack_labels_init();
1815 ret = nf_conntrack_seqadj_init();
1819 ret = nf_conntrack_proto_init();
1823 /* Set up fake conntrack: to never be deleted, not in any hashes */
1824 for_each_possible_cpu(cpu) {
1825 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1826 write_pnet(&ct->ct_net, &init_net);
1827 atomic_set(&ct->ct_general.use, 1);
1829 /* - and look it like as a confirmed connection */
1830 nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1834 nf_conntrack_seqadj_fini();
1836 nf_conntrack_labels_fini();
1838 nf_conntrack_helper_fini();
1840 nf_conntrack_timeout_fini();
1842 nf_conntrack_ecache_fini();
1844 nf_conntrack_tstamp_fini();
1846 nf_conntrack_acct_fini();
1848 nf_conntrack_expect_fini();
1850 kmem_cache_destroy(nf_conntrack_cachep);
1852 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size);
1856 void nf_conntrack_init_end(void)
1858 /* For use by REJECT target */
1859 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1860 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1864 * We need to use special "null" values, not used in hash table
1866 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1867 #define DYING_NULLS_VAL ((1<<30)+1)
1868 #define TEMPLATE_NULLS_VAL ((1<<30)+2)
1870 int nf_conntrack_init_net(struct net *net)
1875 atomic_set(&net->ct.count, 0);
1877 net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
1878 if (!net->ct.pcpu_lists)
1881 for_each_possible_cpu(cpu) {
1882 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1884 spin_lock_init(&pcpu->lock);
1885 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
1886 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
1889 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1891 goto err_pcpu_lists;
1893 ret = nf_conntrack_expect_pernet_init(net);
1896 ret = nf_conntrack_acct_pernet_init(net);
1899 ret = nf_conntrack_tstamp_pernet_init(net);
1902 ret = nf_conntrack_ecache_pernet_init(net);
1905 ret = nf_conntrack_helper_pernet_init(net);
1908 ret = nf_conntrack_proto_pernet_init(net);
1914 nf_conntrack_helper_pernet_fini(net);
1916 nf_conntrack_ecache_pernet_fini(net);
1918 nf_conntrack_tstamp_pernet_fini(net);
1920 nf_conntrack_acct_pernet_fini(net);
1922 nf_conntrack_expect_pernet_fini(net);
1924 free_percpu(net->ct.stat);
1926 free_percpu(net->ct.pcpu_lists);