1 /* Expectation handling for nf_conntrack. */
2 
3 /* (C) 1999-2001 Paul `Rusty' Russell
4  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
5  * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
6  * (c) 2005-2012 Patrick McHardy <kaber@trash.net>
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License version 2 as
10  * published by the Free Software Foundation.
11  */
12 
13 #include <linux/types.h>
14 #include <linux/netfilter.h>
15 #include <linux/skbuff.h>
16 #include <linux/proc_fs.h>
17 #include <linux/seq_file.h>
18 #include <linux/stddef.h>
19 #include <linux/slab.h>
20 #include <linux/err.h>
21 #include <linux/percpu.h>
22 #include <linux/kernel.h>
23 #include <linux/jhash.h>
24 #include <linux/moduleparam.h>
25 #include <linux/export.h>
26 #include <net/net_namespace.h>
27 #include <net/netns/hash.h>
28 
29 #include <net/netfilter/nf_conntrack.h>
30 #include <net/netfilter/nf_conntrack_core.h>
31 #include <net/netfilter/nf_conntrack_expect.h>
32 #include <net/netfilter/nf_conntrack_helper.h>
33 #include <net/netfilter/nf_conntrack_tuple.h>
34 #include <net/netfilter/nf_conntrack_zones.h>
35 
36 unsigned int nf_ct_expect_hsize __read_mostly;
37 EXPORT_SYMBOL_GPL(nf_ct_expect_hsize);
38 
39 struct hlist_head *nf_ct_expect_hash __read_mostly;
40 EXPORT_SYMBOL_GPL(nf_ct_expect_hash);
41 
42 unsigned int nf_ct_expect_max __read_mostly;
43 
44 static struct kmem_cache *nf_ct_expect_cachep __read_mostly;
45 static unsigned int nf_ct_expect_hashrnd __read_mostly;
46 
47 /* nf_conntrack_expect helper functions */
48 void nf_ct_unlink_expect_report(struct nf_conntrack_expect *exp,
49 				u32 portid, int report)
50 {
51 	struct nf_conn_help *master_help = nfct_help(exp->master);
52 	struct net *net = nf_ct_exp_net(exp);
53 
54 	WARN_ON(!master_help);
55 	WARN_ON(timer_pending(&exp->timeout));
56 
57 	hlist_del_rcu(&exp->hnode);
58 	net->ct.expect_count--;
59 
60 	hlist_del_rcu(&exp->lnode);
61 	master_help->expecting[exp->class]--;
62 
63 	nf_ct_expect_event_report(IPEXP_DESTROY, exp, portid, report);
64 	nf_ct_expect_put(exp);
65 
66 	NF_CT_STAT_INC(net, expect_delete);
67 }
68 EXPORT_SYMBOL_GPL(nf_ct_unlink_expect_report);
69 
70 static void nf_ct_expectation_timed_out(unsigned long ul_expect)
71 {
72 	struct nf_conntrack_expect *exp = (void *)ul_expect;
73 
74 	spin_lock_bh(&nf_conntrack_expect_lock);
75 	nf_ct_unlink_expect(exp);
76 	spin_unlock_bh(&nf_conntrack_expect_lock);
77 	nf_ct_expect_put(exp);
78 }
79 
80 static unsigned int nf_ct_expect_dst_hash(const struct net *n, const struct nf_conntrack_tuple *tuple)
81 {
82 	unsigned int hash, seed;
83 
84 	get_random_once(&nf_ct_expect_hashrnd, sizeof(nf_ct_expect_hashrnd));
85 
86 	seed = nf_ct_expect_hashrnd ^ net_hash_mix(n);
87 
88 	hash = jhash2(tuple->dst.u3.all, ARRAY_SIZE(tuple->dst.u3.all),
89 		      (((tuple->dst.protonum ^ tuple->src.l3num) << 16) |
90 		       (__force __u16)tuple->dst.u.all) ^ seed);
91 
92 	return reciprocal_scale(hash, nf_ct_expect_hsize);
93 }
94 
95 static bool
96 nf_ct_exp_equal(const struct nf_conntrack_tuple *tuple,
97 		const struct nf_conntrack_expect *i,
98 		const struct nf_conntrack_zone *zone,
99 		const struct net *net)
100 {
101 	return nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask) &&
102 	       net_eq(net, nf_ct_net(i->master)) &&
103 	       nf_ct_zone_equal_any(i->master, zone);
104 }
105 
106 bool nf_ct_remove_expect(struct nf_conntrack_expect *exp)
107 {
108 	if (del_timer(&exp->timeout)) {
109 		nf_ct_unlink_expect(exp);
110 		nf_ct_expect_put(exp);
111 		return true;
112 	}
113 	return false;
114 }
115 EXPORT_SYMBOL_GPL(nf_ct_remove_expect);
116 
117 struct nf_conntrack_expect *
118 __nf_ct_expect_find(struct net *net,
119 		    const struct nf_conntrack_zone *zone,
120 		    const struct nf_conntrack_tuple *tuple)
121 {
122 	struct nf_conntrack_expect *i;
123 	unsigned int h;
124 
125 	if (!net->ct.expect_count)
126 		return NULL;
127 
128 	h = nf_ct_expect_dst_hash(net, tuple);
129 	hlist_for_each_entry_rcu(i, &nf_ct_expect_hash[h], hnode) {
130 		if (nf_ct_exp_equal(tuple, i, zone, net))
131 			return i;
132 	}
133 	return NULL;
134 }
135 EXPORT_SYMBOL_GPL(__nf_ct_expect_find);
136 
137 /* Just find a expectation corresponding to a tuple. */
138 struct nf_conntrack_expect *
139 nf_ct_expect_find_get(struct net *net,
140 		      const struct nf_conntrack_zone *zone,
141 		      const struct nf_conntrack_tuple *tuple)
142 {
143 	struct nf_conntrack_expect *i;
144 
145 	rcu_read_lock();
146 	i = __nf_ct_expect_find(net, zone, tuple);
147 	if (i && !refcount_inc_not_zero(&i->use))
148 		i = NULL;
149 	rcu_read_unlock();
150 
151 	return i;
152 }
153 EXPORT_SYMBOL_GPL(nf_ct_expect_find_get);
154 
155 /* If an expectation for this connection is found, it gets delete from
156  * global list then returned. */
157 struct nf_conntrack_expect *
158 nf_ct_find_expectation(struct net *net,
159 		       const struct nf_conntrack_zone *zone,
160 		       const struct nf_conntrack_tuple *tuple)
161 {
162 	struct nf_conntrack_expect *i, *exp = NULL;
163 	unsigned int h;
164 
165 	if (!net->ct.expect_count)
166 		return NULL;
167 
168 	h = nf_ct_expect_dst_hash(net, tuple);
169 	hlist_for_each_entry(i, &nf_ct_expect_hash[h], hnode) {
170 		if (!(i->flags & NF_CT_EXPECT_INACTIVE) &&
171 		    nf_ct_exp_equal(tuple, i, zone, net)) {
172 			exp = i;
173 			break;
174 		}
175 	}
176 	if (!exp)
177 		return NULL;
178 
179 	/* If master is not in hash table yet (ie. packet hasn't left
180 	   this machine yet), how can other end know about expected?
181 	   Hence these are not the droids you are looking for (if
182 	   master ct never got confirmed, we'd hold a reference to it
183 	   and weird things would happen to future packets). */
184 	if (!nf_ct_is_confirmed(exp->master))
185 		return NULL;
186 
187 	/* Avoid race with other CPUs, that for exp->master ct, is
188 	 * about to invoke ->destroy(), or nf_ct_delete() via timeout
189 	 * or early_drop().
190 	 *
191 	 * The atomic_inc_not_zero() check tells:  If that fails, we
192 	 * know that the ct is being destroyed.  If it succeeds, we
193 	 * can be sure the ct cannot disappear underneath.
194 	 */
195 	if (unlikely(nf_ct_is_dying(exp->master) ||
196 		     !atomic_inc_not_zero(&exp->master->ct_general.use)))
197 		return NULL;
198 
199 	if (exp->flags & NF_CT_EXPECT_PERMANENT) {
200 		refcount_inc(&exp->use);
201 		return exp;
202 	} else if (del_timer(&exp->timeout)) {
203 		nf_ct_unlink_expect(exp);
204 		return exp;
205 	}
206 	/* Undo exp->master refcnt increase, if del_timer() failed */
207 	nf_ct_put(exp->master);
208 
209 	return NULL;
210 }
211 
212 /* delete all expectations for this conntrack */
213 void nf_ct_remove_expectations(struct nf_conn *ct)
214 {
215 	struct nf_conn_help *help = nfct_help(ct);
216 	struct nf_conntrack_expect *exp;
217 	struct hlist_node *next;
218 
219 	/* Optimization: most connection never expect any others. */
220 	if (!help)
221 		return;
222 
223 	spin_lock_bh(&nf_conntrack_expect_lock);
224 	hlist_for_each_entry_safe(exp, next, &help->expectations, lnode) {
225 		nf_ct_remove_expect(exp);
226 	}
227 	spin_unlock_bh(&nf_conntrack_expect_lock);
228 }
229 EXPORT_SYMBOL_GPL(nf_ct_remove_expectations);
230 
231 /* Would two expected things clash? */
232 static inline int expect_clash(const struct nf_conntrack_expect *a,
233 			       const struct nf_conntrack_expect *b)
234 {
235 	/* Part covered by intersection of masks must be unequal,
236 	   otherwise they clash */
237 	struct nf_conntrack_tuple_mask intersect_mask;
238 	int count;
239 
240 	intersect_mask.src.u.all = a->mask.src.u.all & b->mask.src.u.all;
241 
242 	for (count = 0; count < NF_CT_TUPLE_L3SIZE; count++){
243 		intersect_mask.src.u3.all[count] =
244 			a->mask.src.u3.all[count] & b->mask.src.u3.all[count];
245 	}
246 
247 	return nf_ct_tuple_mask_cmp(&a->tuple, &b->tuple, &intersect_mask) &&
248 	       net_eq(nf_ct_net(a->master), nf_ct_net(b->master)) &&
249 	       nf_ct_zone_equal_any(a->master, nf_ct_zone(b->master));
250 }
251 
252 static inline int expect_matches(const struct nf_conntrack_expect *a,
253 				 const struct nf_conntrack_expect *b)
254 {
255 	return a->master == b->master && a->class == b->class &&
256 	       nf_ct_tuple_equal(&a->tuple, &b->tuple) &&
257 	       nf_ct_tuple_mask_equal(&a->mask, &b->mask) &&
258 	       net_eq(nf_ct_net(a->master), nf_ct_net(b->master)) &&
259 	       nf_ct_zone_equal_any(a->master, nf_ct_zone(b->master));
260 }
261 
262 /* Generally a bad idea to call this: could have matched already. */
263 void nf_ct_unexpect_related(struct nf_conntrack_expect *exp)
264 {
265 	spin_lock_bh(&nf_conntrack_expect_lock);
266 	nf_ct_remove_expect(exp);
267 	spin_unlock_bh(&nf_conntrack_expect_lock);
268 }
269 EXPORT_SYMBOL_GPL(nf_ct_unexpect_related);
270 
271 /* We don't increase the master conntrack refcount for non-fulfilled
272  * conntracks. During the conntrack destruction, the expectations are
273  * always killed before the conntrack itself */
274 struct nf_conntrack_expect *nf_ct_expect_alloc(struct nf_conn *me)
275 {
276 	struct nf_conntrack_expect *new;
277 
278 	new = kmem_cache_alloc(nf_ct_expect_cachep, GFP_ATOMIC);
279 	if (!new)
280 		return NULL;
281 
282 	new->master = me;
283 	refcount_set(&new->use, 1);
284 	return new;
285 }
286 EXPORT_SYMBOL_GPL(nf_ct_expect_alloc);
287 
288 void nf_ct_expect_init(struct nf_conntrack_expect *exp, unsigned int class,
289 		       u_int8_t family,
290 		       const union nf_inet_addr *saddr,
291 		       const union nf_inet_addr *daddr,
292 		       u_int8_t proto, const __be16 *src, const __be16 *dst)
293 {
294 	int len;
295 
296 	if (family == AF_INET)
297 		len = 4;
298 	else
299 		len = 16;
300 
301 	exp->flags = 0;
302 	exp->class = class;
303 	exp->expectfn = NULL;
304 	exp->helper = NULL;
305 	exp->tuple.src.l3num = family;
306 	exp->tuple.dst.protonum = proto;
307 
308 	if (saddr) {
309 		memcpy(&exp->tuple.src.u3, saddr, len);
310 		if (sizeof(exp->tuple.src.u3) > len)
311 			/* address needs to be cleared for nf_ct_tuple_equal */
312 			memset((void *)&exp->tuple.src.u3 + len, 0x00,
313 			       sizeof(exp->tuple.src.u3) - len);
314 		memset(&exp->mask.src.u3, 0xFF, len);
315 		if (sizeof(exp->mask.src.u3) > len)
316 			memset((void *)&exp->mask.src.u3 + len, 0x00,
317 			       sizeof(exp->mask.src.u3) - len);
318 	} else {
319 		memset(&exp->tuple.src.u3, 0x00, sizeof(exp->tuple.src.u3));
320 		memset(&exp->mask.src.u3, 0x00, sizeof(exp->mask.src.u3));
321 	}
322 
323 	if (src) {
324 		exp->tuple.src.u.all = *src;
325 		exp->mask.src.u.all = htons(0xFFFF);
326 	} else {
327 		exp->tuple.src.u.all = 0;
328 		exp->mask.src.u.all = 0;
329 	}
330 
331 	memcpy(&exp->tuple.dst.u3, daddr, len);
332 	if (sizeof(exp->tuple.dst.u3) > len)
333 		/* address needs to be cleared for nf_ct_tuple_equal */
334 		memset((void *)&exp->tuple.dst.u3 + len, 0x00,
335 		       sizeof(exp->tuple.dst.u3) - len);
336 
337 	exp->tuple.dst.u.all = *dst;
338 
339 #ifdef CONFIG_NF_NAT_NEEDED
340 	memset(&exp->saved_addr, 0, sizeof(exp->saved_addr));
341 	memset(&exp->saved_proto, 0, sizeof(exp->saved_proto));
342 #endif
343 }
344 EXPORT_SYMBOL_GPL(nf_ct_expect_init);
345 
346 static void nf_ct_expect_free_rcu(struct rcu_head *head)
347 {
348 	struct nf_conntrack_expect *exp;
349 
350 	exp = container_of(head, struct nf_conntrack_expect, rcu);
351 	kmem_cache_free(nf_ct_expect_cachep, exp);
352 }
353 
354 void nf_ct_expect_put(struct nf_conntrack_expect *exp)
355 {
356 	if (refcount_dec_and_test(&exp->use))
357 		call_rcu(&exp->rcu, nf_ct_expect_free_rcu);
358 }
359 EXPORT_SYMBOL_GPL(nf_ct_expect_put);
360 
361 static void nf_ct_expect_insert(struct nf_conntrack_expect *exp)
362 {
363 	struct nf_conn_help *master_help = nfct_help(exp->master);
364 	struct nf_conntrack_helper *helper;
365 	struct net *net = nf_ct_exp_net(exp);
366 	unsigned int h = nf_ct_expect_dst_hash(net, &exp->tuple);
367 
368 	/* two references : one for hash insert, one for the timer */
369 	refcount_add(2, &exp->use);
370 
371 	setup_timer(&exp->timeout, nf_ct_expectation_timed_out,
372 		    (unsigned long)exp);
373 	helper = rcu_dereference_protected(master_help->helper,
374 					   lockdep_is_held(&nf_conntrack_expect_lock));
375 	if (helper) {
376 		exp->timeout.expires = jiffies +
377 			helper->expect_policy[exp->class].timeout * HZ;
378 	}
379 	add_timer(&exp->timeout);
380 
381 	hlist_add_head_rcu(&exp->lnode, &master_help->expectations);
382 	master_help->expecting[exp->class]++;
383 
384 	hlist_add_head_rcu(&exp->hnode, &nf_ct_expect_hash[h]);
385 	net->ct.expect_count++;
386 
387 	NF_CT_STAT_INC(net, expect_create);
388 }
389 
390 /* Race with expectations being used means we could have none to find; OK. */
391 static void evict_oldest_expect(struct nf_conn *master,
392 				struct nf_conntrack_expect *new)
393 {
394 	struct nf_conn_help *master_help = nfct_help(master);
395 	struct nf_conntrack_expect *exp, *last = NULL;
396 
397 	hlist_for_each_entry(exp, &master_help->expectations, lnode) {
398 		if (exp->class == new->class)
399 			last = exp;
400 	}
401 
402 	if (last)
403 		nf_ct_remove_expect(last);
404 }
405 
406 static inline int __nf_ct_expect_check(struct nf_conntrack_expect *expect)
407 {
408 	const struct nf_conntrack_expect_policy *p;
409 	struct nf_conntrack_expect *i;
410 	struct nf_conn *master = expect->master;
411 	struct nf_conn_help *master_help = nfct_help(master);
412 	struct nf_conntrack_helper *helper;
413 	struct net *net = nf_ct_exp_net(expect);
414 	struct hlist_node *next;
415 	unsigned int h;
416 	int ret = 0;
417 
418 	if (!master_help) {
419 		ret = -ESHUTDOWN;
420 		goto out;
421 	}
422 	h = nf_ct_expect_dst_hash(net, &expect->tuple);
423 	hlist_for_each_entry_safe(i, next, &nf_ct_expect_hash[h], hnode) {
424 		if (expect_matches(i, expect)) {
425 			if (nf_ct_remove_expect(i))
426 				break;
427 		} else if (expect_clash(i, expect)) {
428 			ret = -EBUSY;
429 			goto out;
430 		}
431 	}
432 	/* Will be over limit? */
433 	helper = rcu_dereference_protected(master_help->helper,
434 					   lockdep_is_held(&nf_conntrack_expect_lock));
435 	if (helper) {
436 		p = &helper->expect_policy[expect->class];
437 		if (p->max_expected &&
438 		    master_help->expecting[expect->class] >= p->max_expected) {
439 			evict_oldest_expect(master, expect);
440 			if (master_help->expecting[expect->class]
441 						>= p->max_expected) {
442 				ret = -EMFILE;
443 				goto out;
444 			}
445 		}
446 	}
447 
448 	if (net->ct.expect_count >= nf_ct_expect_max) {
449 		net_warn_ratelimited("nf_conntrack: expectation table full\n");
450 		ret = -EMFILE;
451 	}
452 out:
453 	return ret;
454 }
455 
456 int nf_ct_expect_related_report(struct nf_conntrack_expect *expect,
457 				u32 portid, int report)
458 {
459 	int ret;
460 
461 	spin_lock_bh(&nf_conntrack_expect_lock);
462 	ret = __nf_ct_expect_check(expect);
463 	if (ret < 0)
464 		goto out;
465 
466 	nf_ct_expect_insert(expect);
467 
468 	spin_unlock_bh(&nf_conntrack_expect_lock);
469 	nf_ct_expect_event_report(IPEXP_NEW, expect, portid, report);
470 	return 0;
471 out:
472 	spin_unlock_bh(&nf_conntrack_expect_lock);
473 	return ret;
474 }
475 EXPORT_SYMBOL_GPL(nf_ct_expect_related_report);
476 
477 void nf_ct_expect_iterate_destroy(bool (*iter)(struct nf_conntrack_expect *e, void *data),
478 				  void *data)
479 {
480 	struct nf_conntrack_expect *exp;
481 	const struct hlist_node *next;
482 	unsigned int i;
483 
484 	spin_lock_bh(&nf_conntrack_expect_lock);
485 
486 	for (i = 0; i < nf_ct_expect_hsize; i++) {
487 		hlist_for_each_entry_safe(exp, next,
488 					  &nf_ct_expect_hash[i],
489 					  hnode) {
490 			if (iter(exp, data) && del_timer(&exp->timeout)) {
491 				nf_ct_unlink_expect(exp);
492 				nf_ct_expect_put(exp);
493 			}
494 		}
495 	}
496 
497 	spin_unlock_bh(&nf_conntrack_expect_lock);
498 }
499 EXPORT_SYMBOL_GPL(nf_ct_expect_iterate_destroy);
500 
501 void nf_ct_expect_iterate_net(struct net *net,
502 			      bool (*iter)(struct nf_conntrack_expect *e, void *data),
503 			      void *data,
504 			      u32 portid, int report)
505 {
506 	struct nf_conntrack_expect *exp;
507 	const struct hlist_node *next;
508 	unsigned int i;
509 
510 	spin_lock_bh(&nf_conntrack_expect_lock);
511 
512 	for (i = 0; i < nf_ct_expect_hsize; i++) {
513 		hlist_for_each_entry_safe(exp, next,
514 					  &nf_ct_expect_hash[i],
515 					  hnode) {
516 
517 			if (!net_eq(nf_ct_exp_net(exp), net))
518 				continue;
519 
520 			if (iter(exp, data) && del_timer(&exp->timeout)) {
521 				nf_ct_unlink_expect_report(exp, portid, report);
522 				nf_ct_expect_put(exp);
523 			}
524 		}
525 	}
526 
527 	spin_unlock_bh(&nf_conntrack_expect_lock);
528 }
529 EXPORT_SYMBOL_GPL(nf_ct_expect_iterate_net);
530 
531 #ifdef CONFIG_NF_CONNTRACK_PROCFS
532 struct ct_expect_iter_state {
533 	struct seq_net_private p;
534 	unsigned int bucket;
535 };
536 
537 static struct hlist_node *ct_expect_get_first(struct seq_file *seq)
538 {
539 	struct ct_expect_iter_state *st = seq->private;
540 	struct hlist_node *n;
541 
542 	for (st->bucket = 0; st->bucket < nf_ct_expect_hsize; st->bucket++) {
543 		n = rcu_dereference(hlist_first_rcu(&nf_ct_expect_hash[st->bucket]));
544 		if (n)
545 			return n;
546 	}
547 	return NULL;
548 }
549 
550 static struct hlist_node *ct_expect_get_next(struct seq_file *seq,
551 					     struct hlist_node *head)
552 {
553 	struct ct_expect_iter_state *st = seq->private;
554 
555 	head = rcu_dereference(hlist_next_rcu(head));
556 	while (head == NULL) {
557 		if (++st->bucket >= nf_ct_expect_hsize)
558 			return NULL;
559 		head = rcu_dereference(hlist_first_rcu(&nf_ct_expect_hash[st->bucket]));
560 	}
561 	return head;
562 }
563 
564 static struct hlist_node *ct_expect_get_idx(struct seq_file *seq, loff_t pos)
565 {
566 	struct hlist_node *head = ct_expect_get_first(seq);
567 
568 	if (head)
569 		while (pos && (head = ct_expect_get_next(seq, head)))
570 			pos--;
571 	return pos ? NULL : head;
572 }
573 
574 static void *exp_seq_start(struct seq_file *seq, loff_t *pos)
575 	__acquires(RCU)
576 {
577 	rcu_read_lock();
578 	return ct_expect_get_idx(seq, *pos);
579 }
580 
581 static void *exp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
582 {
583 	(*pos)++;
584 	return ct_expect_get_next(seq, v);
585 }
586 
587 static void exp_seq_stop(struct seq_file *seq, void *v)
588 	__releases(RCU)
589 {
590 	rcu_read_unlock();
591 }
592 
593 static int exp_seq_show(struct seq_file *s, void *v)
594 {
595 	struct nf_conntrack_expect *expect;
596 	struct nf_conntrack_helper *helper;
597 	struct hlist_node *n = v;
598 	char *delim = "";
599 
600 	expect = hlist_entry(n, struct nf_conntrack_expect, hnode);
601 
602 	if (expect->timeout.function)
603 		seq_printf(s, "%ld ", timer_pending(&expect->timeout)
604 			   ? (long)(expect->timeout.expires - jiffies)/HZ : 0);
605 	else
606 		seq_puts(s, "- ");
607 	seq_printf(s, "l3proto = %u proto=%u ",
608 		   expect->tuple.src.l3num,
609 		   expect->tuple.dst.protonum);
610 	print_tuple(s, &expect->tuple,
611 		    __nf_ct_l3proto_find(expect->tuple.src.l3num),
612 		    __nf_ct_l4proto_find(expect->tuple.src.l3num,
613 				       expect->tuple.dst.protonum));
614 
615 	if (expect->flags & NF_CT_EXPECT_PERMANENT) {
616 		seq_puts(s, "PERMANENT");
617 		delim = ",";
618 	}
619 	if (expect->flags & NF_CT_EXPECT_INACTIVE) {
620 		seq_printf(s, "%sINACTIVE", delim);
621 		delim = ",";
622 	}
623 	if (expect->flags & NF_CT_EXPECT_USERSPACE)
624 		seq_printf(s, "%sUSERSPACE", delim);
625 
626 	helper = rcu_dereference(nfct_help(expect->master)->helper);
627 	if (helper) {
628 		seq_printf(s, "%s%s", expect->flags ? " " : "", helper->name);
629 		if (helper->expect_policy[expect->class].name[0])
630 			seq_printf(s, "/%s",
631 				   helper->expect_policy[expect->class].name);
632 	}
633 
634 	seq_putc(s, '\n');
635 
636 	return 0;
637 }
638 
639 static const struct seq_operations exp_seq_ops = {
640 	.start = exp_seq_start,
641 	.next = exp_seq_next,
642 	.stop = exp_seq_stop,
643 	.show = exp_seq_show
644 };
645 
646 static int exp_open(struct inode *inode, struct file *file)
647 {
648 	return seq_open_net(inode, file, &exp_seq_ops,
649 			sizeof(struct ct_expect_iter_state));
650 }
651 
652 static const struct file_operations exp_file_ops = {
653 	.owner   = THIS_MODULE,
654 	.open    = exp_open,
655 	.read    = seq_read,
656 	.llseek  = seq_lseek,
657 	.release = seq_release_net,
658 };
659 #endif /* CONFIG_NF_CONNTRACK_PROCFS */
660 
661 static int exp_proc_init(struct net *net)
662 {
663 #ifdef CONFIG_NF_CONNTRACK_PROCFS
664 	struct proc_dir_entry *proc;
665 	kuid_t root_uid;
666 	kgid_t root_gid;
667 
668 	proc = proc_create("nf_conntrack_expect", 0440, net->proc_net,
669 			   &exp_file_ops);
670 	if (!proc)
671 		return -ENOMEM;
672 
673 	root_uid = make_kuid(net->user_ns, 0);
674 	root_gid = make_kgid(net->user_ns, 0);
675 	if (uid_valid(root_uid) && gid_valid(root_gid))
676 		proc_set_user(proc, root_uid, root_gid);
677 #endif /* CONFIG_NF_CONNTRACK_PROCFS */
678 	return 0;
679 }
680 
681 static void exp_proc_remove(struct net *net)
682 {
683 #ifdef CONFIG_NF_CONNTRACK_PROCFS
684 	remove_proc_entry("nf_conntrack_expect", net->proc_net);
685 #endif /* CONFIG_NF_CONNTRACK_PROCFS */
686 }
687 
688 module_param_named(expect_hashsize, nf_ct_expect_hsize, uint, 0400);
689 
690 int nf_conntrack_expect_pernet_init(struct net *net)
691 {
692 	net->ct.expect_count = 0;
693 	return exp_proc_init(net);
694 }
695 
696 void nf_conntrack_expect_pernet_fini(struct net *net)
697 {
698 	exp_proc_remove(net);
699 }
700 
701 int nf_conntrack_expect_init(void)
702 {
703 	if (!nf_ct_expect_hsize) {
704 		nf_ct_expect_hsize = nf_conntrack_htable_size / 256;
705 		if (!nf_ct_expect_hsize)
706 			nf_ct_expect_hsize = 1;
707 	}
708 	nf_ct_expect_max = nf_ct_expect_hsize * 4;
709 	nf_ct_expect_cachep = kmem_cache_create("nf_conntrack_expect",
710 				sizeof(struct nf_conntrack_expect),
711 				0, 0, NULL);
712 	if (!nf_ct_expect_cachep)
713 		return -ENOMEM;
714 
715 	nf_ct_expect_hash = nf_ct_alloc_hashtable(&nf_ct_expect_hsize, 0);
716 	if (!nf_ct_expect_hash) {
717 		kmem_cache_destroy(nf_ct_expect_cachep);
718 		return -ENOMEM;
719 	}
720 
721 	return 0;
722 }
723 
724 void nf_conntrack_expect_fini(void)
725 {
726 	rcu_barrier(); /* Wait for call_rcu() before destroy */
727 	kmem_cache_destroy(nf_ct_expect_cachep);
728 	nf_ct_free_hashtable(nf_ct_expect_hash, nf_ct_expect_hsize);
729 }
730