1
2
3
4
5
6
7
8
9
10
11
12
13
14
15#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
16
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>
35#include <linux/mm.h>
36#include <linux/nsproxy.h>
37#include <linux/rculist_nulls.h>
38
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>
58
59#define NF_CONNTRACK_VERSION "0.5.0"
60
61int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
62 enum nf_nat_manip_type manip,
63 const struct nlattr *attr) __read_mostly;
64EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
65
66__cacheline_aligned_in_smp spinlock_t nf_conntrack_locks[CONNTRACK_LOCKS];
67EXPORT_SYMBOL_GPL(nf_conntrack_locks);
68
69__cacheline_aligned_in_smp DEFINE_SPINLOCK(nf_conntrack_expect_lock);
70EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock);
71
72struct hlist_nulls_head *nf_conntrack_hash __read_mostly;
73EXPORT_SYMBOL_GPL(nf_conntrack_hash);
74
75struct conntrack_gc_work {
76 struct delayed_work dwork;
77 u32 last_bucket;
78 bool exiting;
79 long next_gc_run;
80};
81
82static __read_mostly struct kmem_cache *nf_conntrack_cachep;
83static __read_mostly spinlock_t nf_conntrack_locks_all_lock;
84static __read_mostly DEFINE_SPINLOCK(nf_conntrack_locks_all_lock);
85static __read_mostly bool nf_conntrack_locks_all;
86
87
88#define GC_MAX_BUCKETS_DIV 64u
89
90#define GC_INTERVAL_MAX (2 * HZ)
91
92#define GC_MAX_EVICTS 256u
93
94static struct conntrack_gc_work conntrack_gc_work;
95
96void nf_conntrack_lock(spinlock_t *lock) __acquires(lock)
97{
98 spin_lock(lock);
99 while (unlikely(nf_conntrack_locks_all)) {
100 spin_unlock(lock);
101
102
103
104
105
106
107 smp_rmb();
108 spin_unlock_wait(&nf_conntrack_locks_all_lock);
109 spin_lock(lock);
110 }
111}
112EXPORT_SYMBOL_GPL(nf_conntrack_lock);
113
114static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2)
115{
116 h1 %= CONNTRACK_LOCKS;
117 h2 %= CONNTRACK_LOCKS;
118 spin_unlock(&nf_conntrack_locks[h1]);
119 if (h1 != h2)
120 spin_unlock(&nf_conntrack_locks[h2]);
121}
122
123
124static bool nf_conntrack_double_lock(struct net *net, unsigned int h1,
125 unsigned int h2, unsigned int sequence)
126{
127 h1 %= CONNTRACK_LOCKS;
128 h2 %= CONNTRACK_LOCKS;
129 if (h1 <= h2) {
130 nf_conntrack_lock(&nf_conntrack_locks[h1]);
131 if (h1 != h2)
132 spin_lock_nested(&nf_conntrack_locks[h2],
133 SINGLE_DEPTH_NESTING);
134 } else {
135 nf_conntrack_lock(&nf_conntrack_locks[h2]);
136 spin_lock_nested(&nf_conntrack_locks[h1],
137 SINGLE_DEPTH_NESTING);
138 }
139 if (read_seqcount_retry(&nf_conntrack_generation, sequence)) {
140 nf_conntrack_double_unlock(h1, h2);
141 return true;
142 }
143 return false;
144}
145
146static void nf_conntrack_all_lock(void)
147{
148 int i;
149
150 spin_lock(&nf_conntrack_locks_all_lock);
151 nf_conntrack_locks_all = true;
152
153
154
155
156
157
158
159 smp_mb();
160
161 for (i = 0; i < CONNTRACK_LOCKS; i++) {
162 spin_unlock_wait(&nf_conntrack_locks[i]);
163 }
164}
165
166static void nf_conntrack_all_unlock(void)
167{
168
169
170
171
172
173
174 smp_store_release(&nf_conntrack_locks_all, false);
175 spin_unlock(&nf_conntrack_locks_all_lock);
176}
177
178unsigned int nf_conntrack_htable_size __read_mostly;
179EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
180
181unsigned int nf_conntrack_max __read_mostly;
182seqcount_t nf_conntrack_generation __read_mostly;
183
184DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
185EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked);
186
187static unsigned int nf_conntrack_hash_rnd __read_mostly;
188
189static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple,
190 const struct net *net)
191{
192 unsigned int n;
193 u32 seed;
194
195 get_random_once(&nf_conntrack_hash_rnd, sizeof(nf_conntrack_hash_rnd));
196
197
198
199
200
201 seed = nf_conntrack_hash_rnd ^ net_hash_mix(net);
202 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
203 return jhash2((u32 *)tuple, n, seed ^
204 (((__force __u16)tuple->dst.u.all << 16) |
205 tuple->dst.protonum));
206}
207
208static u32 scale_hash(u32 hash)
209{
210 return reciprocal_scale(hash, nf_conntrack_htable_size);
211}
212
213static u32 __hash_conntrack(const struct net *net,
214 const struct nf_conntrack_tuple *tuple,
215 unsigned int size)
216{
217 return reciprocal_scale(hash_conntrack_raw(tuple, net), size);
218}
219
220static u32 hash_conntrack(const struct net *net,
221 const struct nf_conntrack_tuple *tuple)
222{
223 return scale_hash(hash_conntrack_raw(tuple, net));
224}
225
226bool
227nf_ct_get_tuple(const struct sk_buff *skb,
228 unsigned int nhoff,
229 unsigned int dataoff,
230 u_int16_t l3num,
231 u_int8_t protonum,
232 struct net *net,
233 struct nf_conntrack_tuple *tuple,
234 const struct nf_conntrack_l3proto *l3proto,
235 const struct nf_conntrack_l4proto *l4proto)
236{
237 memset(tuple, 0, sizeof(*tuple));
238
239 tuple->src.l3num = l3num;
240 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
241 return false;
242
243 tuple->dst.protonum = protonum;
244 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
245
246 return l4proto->pkt_to_tuple(skb, dataoff, net, tuple);
247}
248EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
249
250bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
251 u_int16_t l3num,
252 struct net *net, struct nf_conntrack_tuple *tuple)
253{
254 struct nf_conntrack_l3proto *l3proto;
255 struct nf_conntrack_l4proto *l4proto;
256 unsigned int protoff;
257 u_int8_t protonum;
258 int ret;
259
260 rcu_read_lock();
261
262 l3proto = __nf_ct_l3proto_find(l3num);
263 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
264 if (ret != NF_ACCEPT) {
265 rcu_read_unlock();
266 return false;
267 }
268
269 l4proto = __nf_ct_l4proto_find(l3num, protonum);
270
271 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple,
272 l3proto, l4proto);
273
274 rcu_read_unlock();
275 return ret;
276}
277EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
278
279bool
280nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
281 const struct nf_conntrack_tuple *orig,
282 const struct nf_conntrack_l3proto *l3proto,
283 const struct nf_conntrack_l4proto *l4proto)
284{
285 memset(inverse, 0, sizeof(*inverse));
286
287 inverse->src.l3num = orig->src.l3num;
288 if (l3proto->invert_tuple(inverse, orig) == 0)
289 return false;
290
291 inverse->dst.dir = !orig->dst.dir;
292
293 inverse->dst.protonum = orig->dst.protonum;
294 return l4proto->invert_tuple(inverse, orig);
295}
296EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
297
298static void
299clean_from_lists(struct nf_conn *ct)
300{
301 pr_debug("clean_from_lists(%p)\n", ct);
302 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
303 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
304
305
306 nf_ct_remove_expectations(ct);
307}
308
309
310static void nf_ct_add_to_dying_list(struct nf_conn *ct)
311{
312 struct ct_pcpu *pcpu;
313
314
315 ct->cpu = smp_processor_id();
316 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
317
318 spin_lock(&pcpu->lock);
319 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
320 &pcpu->dying);
321 spin_unlock(&pcpu->lock);
322}
323
324
325static void nf_ct_add_to_unconfirmed_list(struct nf_conn *ct)
326{
327 struct ct_pcpu *pcpu;
328
329
330 ct->cpu = smp_processor_id();
331 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
332
333 spin_lock(&pcpu->lock);
334 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
335 &pcpu->unconfirmed);
336 spin_unlock(&pcpu->lock);
337}
338
339
340static void nf_ct_del_from_dying_or_unconfirmed_list(struct nf_conn *ct)
341{
342 struct ct_pcpu *pcpu;
343
344
345 pcpu = per_cpu_ptr(nf_ct_net(ct)->ct.pcpu_lists, ct->cpu);
346
347 spin_lock(&pcpu->lock);
348 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
349 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
350 spin_unlock(&pcpu->lock);
351}
352
353
354struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
355 const struct nf_conntrack_zone *zone,
356 gfp_t flags)
357{
358 struct nf_conn *tmpl;
359
360 tmpl = kzalloc(sizeof(*tmpl), flags);
361 if (tmpl == NULL)
362 return NULL;
363
364 tmpl->status = IPS_TEMPLATE;
365 write_pnet(&tmpl->ct_net, net);
366 nf_ct_zone_add(tmpl, zone);
367 atomic_set(&tmpl->ct_general.use, 0);
368
369 return tmpl;
370}
371EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
372
373void nf_ct_tmpl_free(struct nf_conn *tmpl)
374{
375 nf_ct_ext_destroy(tmpl);
376 nf_ct_ext_free(tmpl);
377 kfree(tmpl);
378}
379EXPORT_SYMBOL_GPL(nf_ct_tmpl_free);
380
381static void
382destroy_conntrack(struct nf_conntrack *nfct)
383{
384 struct nf_conn *ct = (struct nf_conn *)nfct;
385 struct nf_conntrack_l4proto *l4proto;
386
387 pr_debug("destroy_conntrack(%p)\n", ct);
388 NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
389
390 if (unlikely(nf_ct_is_template(ct))) {
391 nf_ct_tmpl_free(ct);
392 return;
393 }
394 rcu_read_lock();
395 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
396 if (l4proto->destroy)
397 l4proto->destroy(ct);
398
399 rcu_read_unlock();
400
401 local_bh_disable();
402
403
404
405
406
407 nf_ct_remove_expectations(ct);
408
409 nf_ct_del_from_dying_or_unconfirmed_list(ct);
410
411 local_bh_enable();
412
413 if (ct->master)
414 nf_ct_put(ct->master);
415
416 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
417 nf_conntrack_free(ct);
418}
419
420static void nf_ct_delete_from_lists(struct nf_conn *ct)
421{
422 struct net *net = nf_ct_net(ct);
423 unsigned int hash, reply_hash;
424 unsigned int sequence;
425
426 nf_ct_helper_destroy(ct);
427
428 local_bh_disable();
429 do {
430 sequence = read_seqcount_begin(&nf_conntrack_generation);
431 hash = hash_conntrack(net,
432 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
433 reply_hash = hash_conntrack(net,
434 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
435 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
436
437 clean_from_lists(ct);
438 nf_conntrack_double_unlock(hash, reply_hash);
439
440 nf_ct_add_to_dying_list(ct);
441
442 local_bh_enable();
443}
444
445bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
446{
447 struct nf_conn_tstamp *tstamp;
448
449 if (test_and_set_bit(IPS_DYING_BIT, &ct->status))
450 return false;
451
452 tstamp = nf_conn_tstamp_find(ct);
453 if (tstamp && tstamp->stop == 0)
454 tstamp->stop = ktime_get_real_ns();
455
456 if (nf_conntrack_event_report(IPCT_DESTROY, ct,
457 portid, report) < 0) {
458
459
460
461 nf_ct_delete_from_lists(ct);
462 nf_conntrack_ecache_delayed_work(nf_ct_net(ct));
463 return false;
464 }
465
466 nf_conntrack_ecache_work(nf_ct_net(ct));
467 nf_ct_delete_from_lists(ct);
468 nf_ct_put(ct);
469 return true;
470}
471EXPORT_SYMBOL_GPL(nf_ct_delete);
472
473static inline bool
474nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
475 const struct nf_conntrack_tuple *tuple,
476 const struct nf_conntrack_zone *zone,
477 const struct net *net)
478{
479 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
480
481
482
483
484 return nf_ct_tuple_equal(tuple, &h->tuple) &&
485 nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) &&
486 nf_ct_is_confirmed(ct) &&
487 net_eq(net, nf_ct_net(ct));
488}
489
490
491static void nf_ct_gc_expired(struct nf_conn *ct)
492{
493 if (!atomic_inc_not_zero(&ct->ct_general.use))
494 return;
495
496 if (nf_ct_should_gc(ct))
497 nf_ct_kill(ct);
498
499 nf_ct_put(ct);
500}
501
502
503
504
505
506
507static struct nf_conntrack_tuple_hash *
508____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone,
509 const struct nf_conntrack_tuple *tuple, u32 hash)
510{
511 struct nf_conntrack_tuple_hash *h;
512 struct hlist_nulls_head *ct_hash;
513 struct hlist_nulls_node *n;
514 unsigned int bucket, hsize;
515
516begin:
517 nf_conntrack_get_ht(&ct_hash, &hsize);
518 bucket = reciprocal_scale(hash, hsize);
519
520 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[bucket], hnnode) {
521 struct nf_conn *ct;
522
523 ct = nf_ct_tuplehash_to_ctrack(h);
524 if (nf_ct_is_expired(ct)) {
525 nf_ct_gc_expired(ct);
526 continue;
527 }
528
529 if (nf_ct_is_dying(ct))
530 continue;
531
532 if (nf_ct_key_equal(h, tuple, zone, net))
533 return h;
534 }
535
536
537
538
539
540 if (get_nulls_value(n) != bucket) {
541 NF_CT_STAT_INC_ATOMIC(net, search_restart);
542 goto begin;
543 }
544
545 return NULL;
546}
547
548
549static struct nf_conntrack_tuple_hash *
550__nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
551 const struct nf_conntrack_tuple *tuple, u32 hash)
552{
553 struct nf_conntrack_tuple_hash *h;
554 struct nf_conn *ct;
555
556 rcu_read_lock();
557begin:
558 h = ____nf_conntrack_find(net, zone, tuple, hash);
559 if (h) {
560 ct = nf_ct_tuplehash_to_ctrack(h);
561 if (unlikely(nf_ct_is_dying(ct) ||
562 !atomic_inc_not_zero(&ct->ct_general.use)))
563 h = NULL;
564 else {
565 if (unlikely(!nf_ct_key_equal(h, tuple, zone, net))) {
566 nf_ct_put(ct);
567 goto begin;
568 }
569 }
570 }
571 rcu_read_unlock();
572
573 return h;
574}
575
576struct nf_conntrack_tuple_hash *
577nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
578 const struct nf_conntrack_tuple *tuple)
579{
580 return __nf_conntrack_find_get(net, zone, tuple,
581 hash_conntrack_raw(tuple, net));
582}
583EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
584
585static void __nf_conntrack_hash_insert(struct nf_conn *ct,
586 unsigned int hash,
587 unsigned int reply_hash)
588{
589 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
590 &nf_conntrack_hash[hash]);
591 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
592 &nf_conntrack_hash[reply_hash]);
593}
594
595int
596nf_conntrack_hash_check_insert(struct nf_conn *ct)
597{
598 const struct nf_conntrack_zone *zone;
599 struct net *net = nf_ct_net(ct);
600 unsigned int hash, reply_hash;
601 struct nf_conntrack_tuple_hash *h;
602 struct hlist_nulls_node *n;
603 unsigned int sequence;
604
605 zone = nf_ct_zone(ct);
606
607 local_bh_disable();
608 do {
609 sequence = read_seqcount_begin(&nf_conntrack_generation);
610 hash = hash_conntrack(net,
611 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
612 reply_hash = hash_conntrack(net,
613 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
614 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
615
616
617 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
618 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
619 zone, net))
620 goto out;
621
622 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
623 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
624 zone, net))
625 goto out;
626
627 smp_wmb();
628
629 atomic_set(&ct->ct_general.use, 2);
630 __nf_conntrack_hash_insert(ct, hash, reply_hash);
631 nf_conntrack_double_unlock(hash, reply_hash);
632 NF_CT_STAT_INC(net, insert);
633 local_bh_enable();
634 return 0;
635
636out:
637 nf_conntrack_double_unlock(hash, reply_hash);
638 NF_CT_STAT_INC(net, insert_failed);
639 local_bh_enable();
640 return -EEXIST;
641}
642EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
643
644static inline void nf_ct_acct_update(struct nf_conn *ct,
645 enum ip_conntrack_info ctinfo,
646 unsigned int len)
647{
648 struct nf_conn_acct *acct;
649
650 acct = nf_conn_acct_find(ct);
651 if (acct) {
652 struct nf_conn_counter *counter = acct->counter;
653
654 atomic64_inc(&counter[CTINFO2DIR(ctinfo)].packets);
655 atomic64_add(len, &counter[CTINFO2DIR(ctinfo)].bytes);
656 }
657}
658
659static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
660 const struct nf_conn *loser_ct)
661{
662 struct nf_conn_acct *acct;
663
664 acct = nf_conn_acct_find(loser_ct);
665 if (acct) {
666 struct nf_conn_counter *counter = acct->counter;
667 unsigned int bytes;
668
669
670 bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes);
671 nf_ct_acct_update(ct, ctinfo, bytes);
672 }
673}
674
675
676static int nf_ct_resolve_clash(struct net *net, struct sk_buff *skb,
677 enum ip_conntrack_info ctinfo,
678 struct nf_conntrack_tuple_hash *h)
679{
680
681 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
682 struct nf_conntrack_l4proto *l4proto;
683
684 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
685 if (l4proto->allow_clash &&
686 !nfct_nat(ct) &&
687 !nf_ct_is_dying(ct) &&
688 atomic_inc_not_zero(&ct->ct_general.use)) {
689 nf_ct_acct_merge(ct, ctinfo, (struct nf_conn *)skb->nfct);
690 nf_conntrack_put(skb->nfct);
691
692
693
694 skb->nfct = &ct->ct_general;
695 return NF_ACCEPT;
696 }
697 NF_CT_STAT_INC(net, drop);
698 return NF_DROP;
699}
700
701
702int
703__nf_conntrack_confirm(struct sk_buff *skb)
704{
705 const struct nf_conntrack_zone *zone;
706 unsigned int hash, reply_hash;
707 struct nf_conntrack_tuple_hash *h;
708 struct nf_conn *ct;
709 struct nf_conn_help *help;
710 struct nf_conn_tstamp *tstamp;
711 struct hlist_nulls_node *n;
712 enum ip_conntrack_info ctinfo;
713 struct net *net;
714 unsigned int sequence;
715 int ret = NF_DROP;
716
717 ct = nf_ct_get(skb, &ctinfo);
718 net = nf_ct_net(ct);
719
720
721
722
723
724 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
725 return NF_ACCEPT;
726
727 zone = nf_ct_zone(ct);
728 local_bh_disable();
729
730 do {
731 sequence = read_seqcount_begin(&nf_conntrack_generation);
732
733 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
734 hash = scale_hash(hash);
735 reply_hash = hash_conntrack(net,
736 &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
737
738 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
739
740
741
742
743
744
745
746
747
748
749 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
750 pr_debug("Confirming conntrack %p\n", ct);
751
752
753
754
755
756 nf_ct_del_from_dying_or_unconfirmed_list(ct);
757
758 if (unlikely(nf_ct_is_dying(ct))) {
759 nf_ct_add_to_dying_list(ct);
760 goto dying;
761 }
762
763
764
765
766 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode)
767 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
768 zone, net))
769 goto out;
770
771 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode)
772 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
773 zone, net))
774 goto out;
775
776
777
778
779 ct->timeout += nfct_time_stamp;
780 atomic_inc(&ct->ct_general.use);
781 ct->status |= IPS_CONFIRMED;
782
783
784 tstamp = nf_conn_tstamp_find(ct);
785 if (tstamp) {
786 if (skb->tstamp.tv64 == 0)
787 __net_timestamp(skb);
788
789 tstamp->start = ktime_to_ns(skb->tstamp);
790 }
791
792
793
794
795
796 __nf_conntrack_hash_insert(ct, hash, reply_hash);
797 nf_conntrack_double_unlock(hash, reply_hash);
798 local_bh_enable();
799
800 help = nfct_help(ct);
801 if (help && help->helper)
802 nf_conntrack_event_cache(IPCT_HELPER, ct);
803
804 nf_conntrack_event_cache(master_ct(ct) ?
805 IPCT_RELATED : IPCT_NEW, ct);
806 return NF_ACCEPT;
807
808out:
809 nf_ct_add_to_dying_list(ct);
810 ret = nf_ct_resolve_clash(net, skb, ctinfo, h);
811dying:
812 nf_conntrack_double_unlock(hash, reply_hash);
813 NF_CT_STAT_INC(net, insert_failed);
814 local_bh_enable();
815 return ret;
816}
817EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
818
819
820
821int
822nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
823 const struct nf_conn *ignored_conntrack)
824{
825 struct net *net = nf_ct_net(ignored_conntrack);
826 const struct nf_conntrack_zone *zone;
827 struct nf_conntrack_tuple_hash *h;
828 struct hlist_nulls_head *ct_hash;
829 unsigned int hash, hsize;
830 struct hlist_nulls_node *n;
831 struct nf_conn *ct;
832
833 zone = nf_ct_zone(ignored_conntrack);
834
835 rcu_read_lock();
836 begin:
837 nf_conntrack_get_ht(&ct_hash, &hsize);
838 hash = __hash_conntrack(net, tuple, hsize);
839
840 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) {
841 ct = nf_ct_tuplehash_to_ctrack(h);
842
843 if (ct == ignored_conntrack)
844 continue;
845
846 if (nf_ct_is_expired(ct)) {
847 nf_ct_gc_expired(ct);
848 continue;
849 }
850
851 if (nf_ct_key_equal(h, tuple, zone, net)) {
852 NF_CT_STAT_INC_ATOMIC(net, found);
853 rcu_read_unlock();
854 return 1;
855 }
856 }
857
858 if (get_nulls_value(n) != hash) {
859 NF_CT_STAT_INC_ATOMIC(net, search_restart);
860 goto begin;
861 }
862
863 rcu_read_unlock();
864
865 return 0;
866}
867EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
868
869#define NF_CT_EVICTION_RANGE 8
870
871
872
873static unsigned int early_drop_list(struct net *net,
874 struct hlist_nulls_head *head)
875{
876 struct nf_conntrack_tuple_hash *h;
877 struct hlist_nulls_node *n;
878 unsigned int drops = 0;
879 struct nf_conn *tmp;
880
881 hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) {
882 tmp = nf_ct_tuplehash_to_ctrack(h);
883
884 if (nf_ct_is_expired(tmp)) {
885 nf_ct_gc_expired(tmp);
886 continue;
887 }
888
889 if (test_bit(IPS_ASSURED_BIT, &tmp->status) ||
890 !net_eq(nf_ct_net(tmp), net) ||
891 nf_ct_is_dying(tmp))
892 continue;
893
894 if (!atomic_inc_not_zero(&tmp->ct_general.use))
895 continue;
896
897
898
899
900
901
902
903
904 if (net_eq(nf_ct_net(tmp), net) &&
905 nf_ct_is_confirmed(tmp) &&
906 nf_ct_delete(tmp, 0, 0))
907 drops++;
908
909 nf_ct_put(tmp);
910 }
911
912 return drops;
913}
914
915static noinline int early_drop(struct net *net, unsigned int _hash)
916{
917 unsigned int i;
918
919 for (i = 0; i < NF_CT_EVICTION_RANGE; i++) {
920 struct hlist_nulls_head *ct_hash;
921 unsigned int hash, hsize, drops;
922
923 rcu_read_lock();
924 nf_conntrack_get_ht(&ct_hash, &hsize);
925 hash = reciprocal_scale(_hash++, hsize);
926
927 drops = early_drop_list(net, &ct_hash[hash]);
928 rcu_read_unlock();
929
930 if (drops) {
931 NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops);
932 return true;
933 }
934 }
935
936 return false;
937}
938
939static void gc_worker(struct work_struct *work)
940{
941 unsigned int i, goal, buckets = 0, expired_count = 0;
942 struct conntrack_gc_work *gc_work;
943 unsigned int ratio, scanned = 0;
944 unsigned long next_run;
945
946 gc_work = container_of(work, struct conntrack_gc_work, dwork.work);
947
948 goal = nf_conntrack_htable_size / GC_MAX_BUCKETS_DIV;
949 i = gc_work->last_bucket;
950
951 do {
952 struct nf_conntrack_tuple_hash *h;
953 struct hlist_nulls_head *ct_hash;
954 struct hlist_nulls_node *n;
955 unsigned int hashsz;
956 struct nf_conn *tmp;
957
958 i++;
959 rcu_read_lock();
960
961 nf_conntrack_get_ht(&ct_hash, &hashsz);
962 if (i >= hashsz)
963 i = 0;
964
965 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[i], hnnode) {
966 tmp = nf_ct_tuplehash_to_ctrack(h);
967
968 scanned++;
969 if (nf_ct_is_expired(tmp)) {
970 nf_ct_gc_expired(tmp);
971 expired_count++;
972 continue;
973 }
974 }
975
976
977
978
979
980 rcu_read_unlock();
981 cond_resched_rcu_qs();
982 } while (++buckets < goal &&
983 expired_count < GC_MAX_EVICTS);
984
985 if (gc_work->exiting)
986 return;
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007 ratio = scanned ? expired_count * 100 / scanned : 0;
1008 if (ratio >= 90 || expired_count == GC_MAX_EVICTS) {
1009 gc_work->next_gc_run = 0;
1010 next_run = 0;
1011 } else if (expired_count) {
1012 gc_work->next_gc_run /= 2U;
1013 next_run = msecs_to_jiffies(1);
1014 } else {
1015 if (gc_work->next_gc_run < GC_INTERVAL_MAX)
1016 gc_work->next_gc_run += msecs_to_jiffies(1);
1017
1018 next_run = gc_work->next_gc_run;
1019 }
1020
1021 gc_work->last_bucket = i;
1022 queue_delayed_work(system_long_wq, &gc_work->dwork, next_run);
1023}
1024
1025static void conntrack_gc_work_init(struct conntrack_gc_work *gc_work)
1026{
1027 INIT_DELAYED_WORK(&gc_work->dwork, gc_worker);
1028 gc_work->next_gc_run = GC_INTERVAL_MAX;
1029 gc_work->exiting = false;
1030}
1031
1032static struct nf_conn *
1033__nf_conntrack_alloc(struct net *net,
1034 const struct nf_conntrack_zone *zone,
1035 const struct nf_conntrack_tuple *orig,
1036 const struct nf_conntrack_tuple *repl,
1037 gfp_t gfp, u32 hash)
1038{
1039 struct nf_conn *ct;
1040
1041
1042 atomic_inc(&net->ct.count);
1043
1044 if (nf_conntrack_max &&
1045 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
1046 if (!early_drop(net, hash)) {
1047 atomic_dec(&net->ct.count);
1048 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
1049 return ERR_PTR(-ENOMEM);
1050 }
1051 }
1052
1053
1054
1055
1056
1057 ct = kmem_cache_alloc(nf_conntrack_cachep, gfp);
1058 if (ct == NULL)
1059 goto out;
1060
1061 spin_lock_init(&ct->lock);
1062 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
1063 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
1064 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
1065
1066 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
1067 ct->status = 0;
1068 write_pnet(&ct->ct_net, net);
1069 memset(&ct->__nfct_init_offset[0], 0,
1070 offsetof(struct nf_conn, proto) -
1071 offsetof(struct nf_conn, __nfct_init_offset[0]));
1072
1073 nf_ct_zone_add(ct, zone);
1074
1075
1076
1077
1078 atomic_set(&ct->ct_general.use, 0);
1079 return ct;
1080out:
1081 atomic_dec(&net->ct.count);
1082 return ERR_PTR(-ENOMEM);
1083}
1084
1085struct nf_conn *nf_conntrack_alloc(struct net *net,
1086 const struct nf_conntrack_zone *zone,
1087 const struct nf_conntrack_tuple *orig,
1088 const struct nf_conntrack_tuple *repl,
1089 gfp_t gfp)
1090{
1091 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
1092}
1093EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
1094
1095void nf_conntrack_free(struct nf_conn *ct)
1096{
1097 struct net *net = nf_ct_net(ct);
1098
1099
1100
1101
1102 NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 0);
1103
1104 nf_ct_ext_destroy(ct);
1105 nf_ct_ext_free(ct);
1106 kmem_cache_free(nf_conntrack_cachep, ct);
1107 smp_mb__before_atomic();
1108 atomic_dec(&net->ct.count);
1109}
1110EXPORT_SYMBOL_GPL(nf_conntrack_free);
1111
1112
1113
1114
1115static struct nf_conntrack_tuple_hash *
1116init_conntrack(struct net *net, struct nf_conn *tmpl,
1117 const struct nf_conntrack_tuple *tuple,
1118 struct nf_conntrack_l3proto *l3proto,
1119 struct nf_conntrack_l4proto *l4proto,
1120 struct sk_buff *skb,
1121 unsigned int dataoff, u32 hash)
1122{
1123 struct nf_conn *ct;
1124 struct nf_conn_help *help;
1125 struct nf_conntrack_tuple repl_tuple;
1126 struct nf_conntrack_ecache *ecache;
1127 struct nf_conntrack_expect *exp = NULL;
1128 const struct nf_conntrack_zone *zone;
1129 struct nf_conn_timeout *timeout_ext;
1130 struct nf_conntrack_zone tmp;
1131 unsigned int *timeouts;
1132
1133 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
1134 pr_debug("Can't invert tuple.\n");
1135 return NULL;
1136 }
1137
1138 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1139 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
1140 hash);
1141 if (IS_ERR(ct))
1142 return (struct nf_conntrack_tuple_hash *)ct;
1143
1144 if (!nf_ct_add_synproxy(ct, tmpl)) {
1145 nf_conntrack_free(ct);
1146 return ERR_PTR(-ENOMEM);
1147 }
1148
1149 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
1150 if (timeout_ext) {
1151 timeouts = nf_ct_timeout_data(timeout_ext);
1152 if (unlikely(!timeouts))
1153 timeouts = l4proto->get_timeouts(net);
1154 } else {
1155 timeouts = l4proto->get_timeouts(net);
1156 }
1157
1158 if (!l4proto->new(ct, skb, dataoff, timeouts)) {
1159 nf_conntrack_free(ct);
1160 pr_debug("can't track with proto module\n");
1161 return NULL;
1162 }
1163
1164 if (timeout_ext)
1165 nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout),
1166 GFP_ATOMIC);
1167
1168 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
1169 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
1170 nf_ct_labels_ext_add(ct);
1171
1172 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
1173 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
1174 ecache ? ecache->expmask : 0,
1175 GFP_ATOMIC);
1176
1177 local_bh_disable();
1178 if (net->ct.expect_count) {
1179 spin_lock(&nf_conntrack_expect_lock);
1180 exp = nf_ct_find_expectation(net, zone, tuple);
1181 if (exp) {
1182 pr_debug("expectation arrives ct=%p exp=%p\n",
1183 ct, exp);
1184
1185 __set_bit(IPS_EXPECTED_BIT, &ct->status);
1186
1187 ct->master = exp->master;
1188 if (exp->helper) {
1189 help = nf_ct_helper_ext_add(ct, exp->helper,
1190 GFP_ATOMIC);
1191 if (help)
1192 rcu_assign_pointer(help->helper, exp->helper);
1193 }
1194
1195#ifdef CONFIG_NF_CONNTRACK_MARK
1196 ct->mark = exp->master->mark;
1197#endif
1198#ifdef CONFIG_NF_CONNTRACK_SECMARK
1199 ct->secmark = exp->master->secmark;
1200#endif
1201 NF_CT_STAT_INC(net, expect_new);
1202 }
1203 spin_unlock(&nf_conntrack_expect_lock);
1204 }
1205 if (!exp)
1206 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
1207
1208
1209 nf_conntrack_get(&ct->ct_general);
1210 nf_ct_add_to_unconfirmed_list(ct);
1211
1212 local_bh_enable();
1213
1214 if (exp) {
1215 if (exp->expectfn)
1216 exp->expectfn(ct, exp);
1217 nf_ct_expect_put(exp);
1218 }
1219
1220 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1221}
1222
1223
1224static inline struct nf_conn *
1225resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
1226 struct sk_buff *skb,
1227 unsigned int dataoff,
1228 u_int16_t l3num,
1229 u_int8_t protonum,
1230 struct nf_conntrack_l3proto *l3proto,
1231 struct nf_conntrack_l4proto *l4proto,
1232 int *set_reply,
1233 enum ip_conntrack_info *ctinfo)
1234{
1235 const struct nf_conntrack_zone *zone;
1236 struct nf_conntrack_tuple tuple;
1237 struct nf_conntrack_tuple_hash *h;
1238 struct nf_conntrack_zone tmp;
1239 struct nf_conn *ct;
1240 u32 hash;
1241
1242 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1243 dataoff, l3num, protonum, net, &tuple, l3proto,
1244 l4proto)) {
1245 pr_debug("Can't get tuple\n");
1246 return NULL;
1247 }
1248
1249
1250 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1251 hash = hash_conntrack_raw(&tuple, net);
1252 h = __nf_conntrack_find_get(net, zone, &tuple, hash);
1253 if (!h) {
1254 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
1255 skb, dataoff, hash);
1256 if (!h)
1257 return NULL;
1258 if (IS_ERR(h))
1259 return (void *)h;
1260 }
1261 ct = nf_ct_tuplehash_to_ctrack(h);
1262
1263
1264 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1265 *ctinfo = IP_CT_ESTABLISHED_REPLY;
1266
1267 *set_reply = 1;
1268 } else {
1269
1270 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1271 pr_debug("normal packet for %p\n", ct);
1272 *ctinfo = IP_CT_ESTABLISHED;
1273 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
1274 pr_debug("related packet for %p\n", ct);
1275 *ctinfo = IP_CT_RELATED;
1276 } else {
1277 pr_debug("new packet for %p\n", ct);
1278 *ctinfo = IP_CT_NEW;
1279 }
1280 *set_reply = 0;
1281 }
1282 skb->nfct = &ct->ct_general;
1283 skb->nfctinfo = *ctinfo;
1284 return ct;
1285}
1286
1287unsigned int
1288nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
1289 struct sk_buff *skb)
1290{
1291 struct nf_conn *ct, *tmpl = NULL;
1292 enum ip_conntrack_info ctinfo;
1293 struct nf_conntrack_l3proto *l3proto;
1294 struct nf_conntrack_l4proto *l4proto;
1295 unsigned int *timeouts;
1296 unsigned int dataoff;
1297 u_int8_t protonum;
1298 int set_reply = 0;
1299 int ret;
1300
1301 if (skb->nfct) {
1302
1303 tmpl = (struct nf_conn *)skb->nfct;
1304 if (!nf_ct_is_template(tmpl)) {
1305 NF_CT_STAT_INC_ATOMIC(net, ignore);
1306 return NF_ACCEPT;
1307 }
1308 skb->nfct = NULL;
1309 }
1310
1311
1312 l3proto = __nf_ct_l3proto_find(pf);
1313 ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
1314 &dataoff, &protonum);
1315 if (ret <= 0) {
1316 pr_debug("not prepared to track yet or error occurred\n");
1317 NF_CT_STAT_INC_ATOMIC(net, error);
1318 NF_CT_STAT_INC_ATOMIC(net, invalid);
1319 ret = -ret;
1320 goto out;
1321 }
1322
1323 l4proto = __nf_ct_l4proto_find(pf, protonum);
1324
1325
1326
1327
1328 if (l4proto->error != NULL) {
1329 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo,
1330 pf, hooknum);
1331 if (ret <= 0) {
1332 NF_CT_STAT_INC_ATOMIC(net, error);
1333 NF_CT_STAT_INC_ATOMIC(net, invalid);
1334 ret = -ret;
1335 goto out;
1336 }
1337
1338 if (skb->nfct)
1339 goto out;
1340 }
1341
1342 ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum,
1343 l3proto, l4proto, &set_reply, &ctinfo);
1344 if (!ct) {
1345
1346 NF_CT_STAT_INC_ATOMIC(net, invalid);
1347 ret = NF_ACCEPT;
1348 goto out;
1349 }
1350
1351 if (IS_ERR(ct)) {
1352
1353 NF_CT_STAT_INC_ATOMIC(net, drop);
1354 ret = NF_DROP;
1355 goto out;
1356 }
1357
1358 NF_CT_ASSERT(skb->nfct);
1359
1360
1361 timeouts = nf_ct_timeout_lookup(net, ct, l4proto);
1362
1363 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts);
1364 if (ret <= 0) {
1365
1366
1367 pr_debug("nf_conntrack_in: Can't track with proto module\n");
1368 nf_conntrack_put(skb->nfct);
1369 skb->nfct = NULL;
1370 NF_CT_STAT_INC_ATOMIC(net, invalid);
1371 if (ret == -NF_DROP)
1372 NF_CT_STAT_INC_ATOMIC(net, drop);
1373 ret = -ret;
1374 goto out;
1375 }
1376
1377 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
1378 nf_conntrack_event_cache(IPCT_REPLY, ct);
1379out:
1380 if (tmpl) {
1381
1382
1383
1384 if (ret == NF_REPEAT)
1385 skb->nfct = (struct nf_conntrack *)tmpl;
1386 else
1387 nf_ct_put(tmpl);
1388 }
1389
1390 return ret;
1391}
1392EXPORT_SYMBOL_GPL(nf_conntrack_in);
1393
1394bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
1395 const struct nf_conntrack_tuple *orig)
1396{
1397 bool ret;
1398
1399 rcu_read_lock();
1400 ret = nf_ct_invert_tuple(inverse, orig,
1401 __nf_ct_l3proto_find(orig->src.l3num),
1402 __nf_ct_l4proto_find(orig->src.l3num,
1403 orig->dst.protonum));
1404 rcu_read_unlock();
1405 return ret;
1406}
1407EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
1408
1409
1410
1411void nf_conntrack_alter_reply(struct nf_conn *ct,
1412 const struct nf_conntrack_tuple *newreply)
1413{
1414 struct nf_conn_help *help = nfct_help(ct);
1415
1416
1417 NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
1418
1419 pr_debug("Altering reply tuple of %p to ", ct);
1420 nf_ct_dump_tuple(newreply);
1421
1422 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
1423 if (ct->master || (help && !hlist_empty(&help->expectations)))
1424 return;
1425
1426 rcu_read_lock();
1427 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC);
1428 rcu_read_unlock();
1429}
1430EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
1431
1432
1433void __nf_ct_refresh_acct(struct nf_conn *ct,
1434 enum ip_conntrack_info ctinfo,
1435 const struct sk_buff *skb,
1436 unsigned long extra_jiffies,
1437 int do_acct)
1438{
1439 NF_CT_ASSERT(skb);
1440
1441
1442 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
1443 goto acct;
1444
1445
1446 if (nf_ct_is_confirmed(ct))
1447 extra_jiffies += nfct_time_stamp;
1448
1449 ct->timeout = extra_jiffies;
1450acct:
1451 if (do_acct)
1452 nf_ct_acct_update(ct, ctinfo, skb->len);
1453}
1454EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
1455
1456bool nf_ct_kill_acct(struct nf_conn *ct,
1457 enum ip_conntrack_info ctinfo,
1458 const struct sk_buff *skb)
1459{
1460 nf_ct_acct_update(ct, ctinfo, skb->len);
1461
1462 return nf_ct_delete(ct, 0, 0);
1463}
1464EXPORT_SYMBOL_GPL(nf_ct_kill_acct);
1465
1466#if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1467
1468#include <linux/netfilter/nfnetlink.h>
1469#include <linux/netfilter/nfnetlink_conntrack.h>
1470#include <linux/mutex.h>
1471
1472
1473
1474
1475int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
1476 const struct nf_conntrack_tuple *tuple)
1477{
1478 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
1479 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
1480 goto nla_put_failure;
1481 return 0;
1482
1483nla_put_failure:
1484 return -1;
1485}
1486EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
1487
1488const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
1489 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
1490 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
1491};
1492EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
1493
1494int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
1495 struct nf_conntrack_tuple *t)
1496{
1497 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
1498 return -EINVAL;
1499
1500 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
1501 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
1502
1503 return 0;
1504}
1505EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
1506
1507int nf_ct_port_nlattr_tuple_size(void)
1508{
1509 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1510}
1511EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
1512#endif
1513
1514
1515static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
1516{
1517 struct nf_conn *ct;
1518 enum ip_conntrack_info ctinfo;
1519
1520
1521 ct = nf_ct_get(skb, &ctinfo);
1522 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
1523 ctinfo = IP_CT_RELATED_REPLY;
1524 else
1525 ctinfo = IP_CT_RELATED;
1526
1527
1528 nskb->nfct = &ct->ct_general;
1529 nskb->nfctinfo = ctinfo;
1530 nf_conntrack_get(nskb->nfct);
1531}
1532
1533
1534static struct nf_conn *
1535get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
1536 void *data, unsigned int *bucket)
1537{
1538 struct nf_conntrack_tuple_hash *h;
1539 struct nf_conn *ct;
1540 struct hlist_nulls_node *n;
1541 int cpu;
1542 spinlock_t *lockp;
1543
1544 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
1545 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
1546 local_bh_disable();
1547 nf_conntrack_lock(lockp);
1548 if (*bucket < nf_conntrack_htable_size) {
1549 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnnode) {
1550 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1551 continue;
1552 ct = nf_ct_tuplehash_to_ctrack(h);
1553 if (net_eq(nf_ct_net(ct), net) &&
1554 iter(ct, data))
1555 goto found;
1556 }
1557 }
1558 spin_unlock(lockp);
1559 local_bh_enable();
1560 cond_resched();
1561 }
1562
1563 for_each_possible_cpu(cpu) {
1564 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1565
1566 spin_lock_bh(&pcpu->lock);
1567 hlist_nulls_for_each_entry(h, n, &pcpu->unconfirmed, hnnode) {
1568 ct = nf_ct_tuplehash_to_ctrack(h);
1569 if (iter(ct, data))
1570 set_bit(IPS_DYING_BIT, &ct->status);
1571 }
1572 spin_unlock_bh(&pcpu->lock);
1573 cond_resched();
1574 }
1575 return NULL;
1576found:
1577 atomic_inc(&ct->ct_general.use);
1578 spin_unlock(lockp);
1579 local_bh_enable();
1580 return ct;
1581}
1582
1583void nf_ct_iterate_cleanup(struct net *net,
1584 int (*iter)(struct nf_conn *i, void *data),
1585 void *data, u32 portid, int report)
1586{
1587 struct nf_conn *ct;
1588 unsigned int bucket = 0;
1589
1590 might_sleep();
1591
1592 if (atomic_read(&net->ct.count) == 0)
1593 return;
1594
1595 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
1596
1597
1598 nf_ct_delete(ct, portid, report);
1599 nf_ct_put(ct);
1600 cond_resched();
1601 }
1602}
1603EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
1604
1605static int kill_all(struct nf_conn *i, void *data)
1606{
1607 return 1;
1608}
1609
1610void nf_ct_free_hashtable(void *hash, unsigned int size)
1611{
1612 if (is_vmalloc_addr(hash))
1613 vfree(hash);
1614 else
1615 free_pages((unsigned long)hash,
1616 get_order(sizeof(struct hlist_head) * size));
1617}
1618EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1619
1620static int untrack_refs(void)
1621{
1622 int cnt = 0, cpu;
1623
1624 for_each_possible_cpu(cpu) {
1625 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1626
1627 cnt += atomic_read(&ct->ct_general.use) - 1;
1628 }
1629 return cnt;
1630}
1631
1632void nf_conntrack_cleanup_start(void)
1633{
1634 conntrack_gc_work.exiting = true;
1635 RCU_INIT_POINTER(ip_ct_attach, NULL);
1636}
1637
1638void nf_conntrack_cleanup_end(void)
1639{
1640 RCU_INIT_POINTER(nf_ct_destroy, NULL);
1641 while (untrack_refs() > 0)
1642 schedule();
1643
1644 cancel_delayed_work_sync(&conntrack_gc_work.dwork);
1645 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size);
1646
1647 nf_conntrack_proto_fini();
1648 nf_conntrack_seqadj_fini();
1649 nf_conntrack_labels_fini();
1650 nf_conntrack_helper_fini();
1651 nf_conntrack_timeout_fini();
1652 nf_conntrack_ecache_fini();
1653 nf_conntrack_tstamp_fini();
1654 nf_conntrack_acct_fini();
1655 nf_conntrack_expect_fini();
1656
1657 kmem_cache_destroy(nf_conntrack_cachep);
1658}
1659
1660
1661
1662
1663
1664void nf_conntrack_cleanup_net(struct net *net)
1665{
1666 LIST_HEAD(single);
1667
1668 list_add(&net->exit_list, &single);
1669 nf_conntrack_cleanup_net_list(&single);
1670}
1671
1672void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
1673{
1674 int busy;
1675 struct net *net;
1676
1677
1678
1679
1680
1681
1682 synchronize_net();
1683i_see_dead_people:
1684 busy = 0;
1685 list_for_each_entry(net, net_exit_list, exit_list) {
1686 nf_ct_iterate_cleanup(net, kill_all, NULL, 0, 0);
1687 if (atomic_read(&net->ct.count) != 0)
1688 busy = 1;
1689 }
1690 if (busy) {
1691 schedule();
1692 goto i_see_dead_people;
1693 }
1694
1695 list_for_each_entry(net, net_exit_list, exit_list) {
1696 nf_conntrack_proto_pernet_fini(net);
1697 nf_conntrack_helper_pernet_fini(net);
1698 nf_conntrack_ecache_pernet_fini(net);
1699 nf_conntrack_tstamp_pernet_fini(net);
1700 nf_conntrack_acct_pernet_fini(net);
1701 nf_conntrack_expect_pernet_fini(net);
1702 free_percpu(net->ct.stat);
1703 free_percpu(net->ct.pcpu_lists);
1704 }
1705}
1706
1707void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
1708{
1709 struct hlist_nulls_head *hash;
1710 unsigned int nr_slots, i;
1711 size_t sz;
1712
1713 if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head)))
1714 return NULL;
1715
1716 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1717 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1718
1719 if (nr_slots > (UINT_MAX / sizeof(struct hlist_nulls_head)))
1720 return NULL;
1721
1722 sz = nr_slots * sizeof(struct hlist_nulls_head);
1723 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1724 get_order(sz));
1725 if (!hash)
1726 hash = vzalloc(sz);
1727
1728 if (hash && nulls)
1729 for (i = 0; i < nr_slots; i++)
1730 INIT_HLIST_NULLS_HEAD(&hash[i], i);
1731
1732 return hash;
1733}
1734EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1735
1736int nf_conntrack_hash_resize(unsigned int hashsize)
1737{
1738 int i, bucket;
1739 unsigned int old_size;
1740 struct hlist_nulls_head *hash, *old_hash;
1741 struct nf_conntrack_tuple_hash *h;
1742 struct nf_conn *ct;
1743
1744 if (!hashsize)
1745 return -EINVAL;
1746
1747 hash = nf_ct_alloc_hashtable(&hashsize, 1);
1748 if (!hash)
1749 return -ENOMEM;
1750
1751 old_size = nf_conntrack_htable_size;
1752 if (old_size == hashsize) {
1753 nf_ct_free_hashtable(hash, hashsize);
1754 return 0;
1755 }
1756
1757 local_bh_disable();
1758 nf_conntrack_all_lock();
1759 write_seqcount_begin(&nf_conntrack_generation);
1760
1761
1762
1763
1764
1765
1766
1767 for (i = 0; i < nf_conntrack_htable_size; i++) {
1768 while (!hlist_nulls_empty(&nf_conntrack_hash[i])) {
1769 h = hlist_nulls_entry(nf_conntrack_hash[i].first,
1770 struct nf_conntrack_tuple_hash, hnnode);
1771 ct = nf_ct_tuplehash_to_ctrack(h);
1772 hlist_nulls_del_rcu(&h->hnnode);
1773 bucket = __hash_conntrack(nf_ct_net(ct),
1774 &h->tuple, hashsize);
1775 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1776 }
1777 }
1778 old_size = nf_conntrack_htable_size;
1779 old_hash = nf_conntrack_hash;
1780
1781 nf_conntrack_hash = hash;
1782 nf_conntrack_htable_size = hashsize;
1783
1784 write_seqcount_end(&nf_conntrack_generation);
1785 nf_conntrack_all_unlock();
1786 local_bh_enable();
1787
1788 synchronize_net();
1789 nf_ct_free_hashtable(old_hash, old_size);
1790 return 0;
1791}
1792
1793int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1794{
1795 unsigned int hashsize;
1796 int rc;
1797
1798 if (current->nsproxy->net_ns != &init_net)
1799 return -EOPNOTSUPP;
1800
1801
1802 if (!nf_conntrack_htable_size)
1803 return param_set_uint(val, kp);
1804
1805 rc = kstrtouint(val, 0, &hashsize);
1806 if (rc)
1807 return rc;
1808
1809 return nf_conntrack_hash_resize(hashsize);
1810}
1811EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1812
1813module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1814 &nf_conntrack_htable_size, 0600);
1815
1816void nf_ct_untracked_status_or(unsigned long bits)
1817{
1818 int cpu;
1819
1820 for_each_possible_cpu(cpu)
1821 per_cpu(nf_conntrack_untracked, cpu).status |= bits;
1822}
1823EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or);
1824
1825int nf_conntrack_init_start(void)
1826{
1827 int max_factor = 8;
1828 int ret = -ENOMEM;
1829 int i, cpu;
1830
1831 seqcount_init(&nf_conntrack_generation);
1832
1833 for (i = 0; i < CONNTRACK_LOCKS; i++)
1834 spin_lock_init(&nf_conntrack_locks[i]);
1835
1836 if (!nf_conntrack_htable_size) {
1837
1838
1839
1840
1841
1842 nf_conntrack_htable_size
1843 = (((totalram_pages << PAGE_SHIFT) / 16384)
1844 / sizeof(struct hlist_head));
1845 if (totalram_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
1846 nf_conntrack_htable_size = 65536;
1847 else if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
1848 nf_conntrack_htable_size = 16384;
1849 if (nf_conntrack_htable_size < 32)
1850 nf_conntrack_htable_size = 32;
1851
1852
1853
1854
1855
1856 max_factor = 4;
1857 }
1858
1859 nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1);
1860 if (!nf_conntrack_hash)
1861 return -ENOMEM;
1862
1863 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1864
1865 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1866 sizeof(struct nf_conn), 0,
1867 SLAB_DESTROY_BY_RCU | SLAB_HWCACHE_ALIGN, NULL);
1868 if (!nf_conntrack_cachep)
1869 goto err_cachep;
1870
1871 printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n",
1872 NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1873 nf_conntrack_max);
1874
1875 ret = nf_conntrack_expect_init();
1876 if (ret < 0)
1877 goto err_expect;
1878
1879 ret = nf_conntrack_acct_init();
1880 if (ret < 0)
1881 goto err_acct;
1882
1883 ret = nf_conntrack_tstamp_init();
1884 if (ret < 0)
1885 goto err_tstamp;
1886
1887 ret = nf_conntrack_ecache_init();
1888 if (ret < 0)
1889 goto err_ecache;
1890
1891 ret = nf_conntrack_timeout_init();
1892 if (ret < 0)
1893 goto err_timeout;
1894
1895 ret = nf_conntrack_helper_init();
1896 if (ret < 0)
1897 goto err_helper;
1898
1899 ret = nf_conntrack_labels_init();
1900 if (ret < 0)
1901 goto err_labels;
1902
1903 ret = nf_conntrack_seqadj_init();
1904 if (ret < 0)
1905 goto err_seqadj;
1906
1907 ret = nf_conntrack_proto_init();
1908 if (ret < 0)
1909 goto err_proto;
1910
1911
1912 for_each_possible_cpu(cpu) {
1913 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu);
1914 write_pnet(&ct->ct_net, &init_net);
1915 atomic_set(&ct->ct_general.use, 1);
1916 }
1917
1918 nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED);
1919
1920 conntrack_gc_work_init(&conntrack_gc_work);
1921 queue_delayed_work(system_long_wq, &conntrack_gc_work.dwork, GC_INTERVAL_MAX);
1922
1923 return 0;
1924
1925err_proto:
1926 nf_conntrack_seqadj_fini();
1927err_seqadj:
1928 nf_conntrack_labels_fini();
1929err_labels:
1930 nf_conntrack_helper_fini();
1931err_helper:
1932 nf_conntrack_timeout_fini();
1933err_timeout:
1934 nf_conntrack_ecache_fini();
1935err_ecache:
1936 nf_conntrack_tstamp_fini();
1937err_tstamp:
1938 nf_conntrack_acct_fini();
1939err_acct:
1940 nf_conntrack_expect_fini();
1941err_expect:
1942 kmem_cache_destroy(nf_conntrack_cachep);
1943err_cachep:
1944 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_htable_size);
1945 return ret;
1946}
1947
1948void nf_conntrack_init_end(void)
1949{
1950
1951 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach);
1952 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack);
1953}
1954
1955
1956
1957
1958#define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
1959#define DYING_NULLS_VAL ((1<<30)+1)
1960#define TEMPLATE_NULLS_VAL ((1<<30)+2)
1961
1962int nf_conntrack_init_net(struct net *net)
1963{
1964 int ret = -ENOMEM;
1965 int cpu;
1966
1967 atomic_set(&net->ct.count, 0);
1968
1969 net->ct.pcpu_lists = alloc_percpu(struct ct_pcpu);
1970 if (!net->ct.pcpu_lists)
1971 goto err_stat;
1972
1973 for_each_possible_cpu(cpu) {
1974 struct ct_pcpu *pcpu = per_cpu_ptr(net->ct.pcpu_lists, cpu);
1975
1976 spin_lock_init(&pcpu->lock);
1977 INIT_HLIST_NULLS_HEAD(&pcpu->unconfirmed, UNCONFIRMED_NULLS_VAL);
1978 INIT_HLIST_NULLS_HEAD(&pcpu->dying, DYING_NULLS_VAL);
1979 }
1980
1981 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1982 if (!net->ct.stat)
1983 goto err_pcpu_lists;
1984
1985 ret = nf_conntrack_expect_pernet_init(net);
1986 if (ret < 0)
1987 goto err_expect;
1988 ret = nf_conntrack_acct_pernet_init(net);
1989 if (ret < 0)
1990 goto err_acct;
1991 ret = nf_conntrack_tstamp_pernet_init(net);
1992 if (ret < 0)
1993 goto err_tstamp;
1994 ret = nf_conntrack_ecache_pernet_init(net);
1995 if (ret < 0)
1996 goto err_ecache;
1997 ret = nf_conntrack_helper_pernet_init(net);
1998 if (ret < 0)
1999 goto err_helper;
2000 ret = nf_conntrack_proto_pernet_init(net);
2001 if (ret < 0)
2002 goto err_proto;
2003 return 0;
2004
2005err_proto:
2006 nf_conntrack_helper_pernet_fini(net);
2007err_helper:
2008 nf_conntrack_ecache_pernet_fini(net);
2009err_ecache:
2010 nf_conntrack_tstamp_pernet_fini(net);
2011err_tstamp:
2012 nf_conntrack_acct_pernet_fini(net);
2013err_acct:
2014 nf_conntrack_expect_pernet_fini(net);
2015err_expect:
2016 free_percpu(net->ct.stat);
2017err_pcpu_lists:
2018 free_percpu(net->ct.pcpu_lists);
2019err_stat:
2020 return ret;
2021}
2022