dm thin: wake worker when discard is prepared
[linux-2.6.git] / net / netfilter / nf_conntrack_expect.c
blob527651a53a45ded66c97162f4e072ca55a7d1efa
1 /* Expectation handling for nf_conntrack. */
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>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
12 #include <linux/types.h>
13 #include <linux/netfilter.h>
14 #include <linux/skbuff.h>
15 #include <linux/proc_fs.h>
16 #include <linux/seq_file.h>
17 #include <linux/stddef.h>
18 #include <linux/slab.h>
19 #include <linux/err.h>
20 #include <linux/percpu.h>
21 #include <linux/kernel.h>
22 #include <linux/jhash.h>
23 #include <linux/moduleparam.h>
24 #include <linux/export.h>
25 #include <net/net_namespace.h>
27 #include <net/netfilter/nf_conntrack.h>
28 #include <net/netfilter/nf_conntrack_core.h>
29 #include <net/netfilter/nf_conntrack_expect.h>
30 #include <net/netfilter/nf_conntrack_helper.h>
31 #include <net/netfilter/nf_conntrack_tuple.h>
32 #include <net/netfilter/nf_conntrack_zones.h>
34 unsigned int nf_ct_expect_hsize __read_mostly;
35 EXPORT_SYMBOL_GPL(nf_ct_expect_hsize);
37 unsigned int nf_ct_expect_max __read_mostly;
39 static struct kmem_cache *nf_ct_expect_cachep __read_mostly;
41 /* nf_conntrack_expect helper functions */
42 void nf_ct_unlink_expect_report(struct nf_conntrack_expect *exp,
43 u32 pid, int report)
45 struct nf_conn_help *master_help = nfct_help(exp->master);
46 struct net *net = nf_ct_exp_net(exp);
48 NF_CT_ASSERT(master_help);
49 NF_CT_ASSERT(!timer_pending(&exp->timeout));
51 hlist_del_rcu(&exp->hnode);
52 net->ct.expect_count--;
54 hlist_del(&exp->lnode);
55 master_help->expecting[exp->class]--;
57 nf_ct_expect_event_report(IPEXP_DESTROY, exp, pid, report);
58 nf_ct_expect_put(exp);
60 NF_CT_STAT_INC(net, expect_delete);
62 EXPORT_SYMBOL_GPL(nf_ct_unlink_expect_report);
64 static void nf_ct_expectation_timed_out(unsigned long ul_expect)
66 struct nf_conntrack_expect *exp = (void *)ul_expect;
68 spin_lock_bh(&nf_conntrack_lock);
69 nf_ct_unlink_expect(exp);
70 spin_unlock_bh(&nf_conntrack_lock);
71 nf_ct_expect_put(exp);
74 static unsigned int nf_ct_expect_dst_hash(const struct nf_conntrack_tuple *tuple)
76 unsigned int hash;
78 if (unlikely(!nf_conntrack_hash_rnd)) {
79 init_nf_conntrack_hash_rnd();
82 hash = jhash2(tuple->dst.u3.all, ARRAY_SIZE(tuple->dst.u3.all),
83 (((tuple->dst.protonum ^ tuple->src.l3num) << 16) |
84 (__force __u16)tuple->dst.u.all) ^ nf_conntrack_hash_rnd);
85 return ((u64)hash * nf_ct_expect_hsize) >> 32;
88 struct nf_conntrack_expect *
89 __nf_ct_expect_find(struct net *net, u16 zone,
90 const struct nf_conntrack_tuple *tuple)
92 struct nf_conntrack_expect *i;
93 struct hlist_node *n;
94 unsigned int h;
96 if (!net->ct.expect_count)
97 return NULL;
99 h = nf_ct_expect_dst_hash(tuple);
100 hlist_for_each_entry_rcu(i, n, &net->ct.expect_hash[h], hnode) {
101 if (nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask) &&
102 nf_ct_zone(i->master) == zone)
103 return i;
105 return NULL;
107 EXPORT_SYMBOL_GPL(__nf_ct_expect_find);
109 /* Just find a expectation corresponding to a tuple. */
110 struct nf_conntrack_expect *
111 nf_ct_expect_find_get(struct net *net, u16 zone,
112 const struct nf_conntrack_tuple *tuple)
114 struct nf_conntrack_expect *i;
116 rcu_read_lock();
117 i = __nf_ct_expect_find(net, zone, tuple);
118 if (i && !atomic_inc_not_zero(&i->use))
119 i = NULL;
120 rcu_read_unlock();
122 return i;
124 EXPORT_SYMBOL_GPL(nf_ct_expect_find_get);
126 /* If an expectation for this connection is found, it gets delete from
127 * global list then returned. */
128 struct nf_conntrack_expect *
129 nf_ct_find_expectation(struct net *net, u16 zone,
130 const struct nf_conntrack_tuple *tuple)
132 struct nf_conntrack_expect *i, *exp = NULL;
133 struct hlist_node *n;
134 unsigned int h;
136 if (!net->ct.expect_count)
137 return NULL;
139 h = nf_ct_expect_dst_hash(tuple);
140 hlist_for_each_entry(i, n, &net->ct.expect_hash[h], hnode) {
141 if (!(i->flags & NF_CT_EXPECT_INACTIVE) &&
142 nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask) &&
143 nf_ct_zone(i->master) == zone) {
144 exp = i;
145 break;
148 if (!exp)
149 return NULL;
151 /* If master is not in hash table yet (ie. packet hasn't left
152 this machine yet), how can other end know about expected?
153 Hence these are not the droids you are looking for (if
154 master ct never got confirmed, we'd hold a reference to it
155 and weird things would happen to future packets). */
156 if (!nf_ct_is_confirmed(exp->master))
157 return NULL;
159 if (exp->flags & NF_CT_EXPECT_PERMANENT) {
160 atomic_inc(&exp->use);
161 return exp;
162 } else if (del_timer(&exp->timeout)) {
163 nf_ct_unlink_expect(exp);
164 return exp;
167 return NULL;
170 /* delete all expectations for this conntrack */
171 void nf_ct_remove_expectations(struct nf_conn *ct)
173 struct nf_conn_help *help = nfct_help(ct);
174 struct nf_conntrack_expect *exp;
175 struct hlist_node *n, *next;
177 /* Optimization: most connection never expect any others. */
178 if (!help)
179 return;
181 hlist_for_each_entry_safe(exp, n, next, &help->expectations, lnode) {
182 if (del_timer(&exp->timeout)) {
183 nf_ct_unlink_expect(exp);
184 nf_ct_expect_put(exp);
188 EXPORT_SYMBOL_GPL(nf_ct_remove_expectations);
190 /* Would two expected things clash? */
191 static inline int expect_clash(const struct nf_conntrack_expect *a,
192 const struct nf_conntrack_expect *b)
194 /* Part covered by intersection of masks must be unequal,
195 otherwise they clash */
196 struct nf_conntrack_tuple_mask intersect_mask;
197 int count;
199 intersect_mask.src.u.all = a->mask.src.u.all & b->mask.src.u.all;
201 for (count = 0; count < NF_CT_TUPLE_L3SIZE; count++){
202 intersect_mask.src.u3.all[count] =
203 a->mask.src.u3.all[count] & b->mask.src.u3.all[count];
206 return nf_ct_tuple_mask_cmp(&a->tuple, &b->tuple, &intersect_mask);
209 static inline int expect_matches(const struct nf_conntrack_expect *a,
210 const struct nf_conntrack_expect *b)
212 return a->master == b->master && a->class == b->class &&
213 nf_ct_tuple_equal(&a->tuple, &b->tuple) &&
214 nf_ct_tuple_mask_equal(&a->mask, &b->mask) &&
215 nf_ct_zone(a->master) == nf_ct_zone(b->master);
218 /* Generally a bad idea to call this: could have matched already. */
219 void nf_ct_unexpect_related(struct nf_conntrack_expect *exp)
221 spin_lock_bh(&nf_conntrack_lock);
222 if (del_timer(&exp->timeout)) {
223 nf_ct_unlink_expect(exp);
224 nf_ct_expect_put(exp);
226 spin_unlock_bh(&nf_conntrack_lock);
228 EXPORT_SYMBOL_GPL(nf_ct_unexpect_related);
230 /* We don't increase the master conntrack refcount for non-fulfilled
231 * conntracks. During the conntrack destruction, the expectations are
232 * always killed before the conntrack itself */
233 struct nf_conntrack_expect *nf_ct_expect_alloc(struct nf_conn *me)
235 struct nf_conntrack_expect *new;
237 new = kmem_cache_alloc(nf_ct_expect_cachep, GFP_ATOMIC);
238 if (!new)
239 return NULL;
241 new->master = me;
242 atomic_set(&new->use, 1);
243 return new;
245 EXPORT_SYMBOL_GPL(nf_ct_expect_alloc);
247 void nf_ct_expect_init(struct nf_conntrack_expect *exp, unsigned int class,
248 u_int8_t family,
249 const union nf_inet_addr *saddr,
250 const union nf_inet_addr *daddr,
251 u_int8_t proto, const __be16 *src, const __be16 *dst)
253 int len;
255 if (family == AF_INET)
256 len = 4;
257 else
258 len = 16;
260 exp->flags = 0;
261 exp->class = class;
262 exp->expectfn = NULL;
263 exp->helper = NULL;
264 exp->tuple.src.l3num = family;
265 exp->tuple.dst.protonum = proto;
267 if (saddr) {
268 memcpy(&exp->tuple.src.u3, saddr, len);
269 if (sizeof(exp->tuple.src.u3) > len)
270 /* address needs to be cleared for nf_ct_tuple_equal */
271 memset((void *)&exp->tuple.src.u3 + len, 0x00,
272 sizeof(exp->tuple.src.u3) - len);
273 memset(&exp->mask.src.u3, 0xFF, len);
274 if (sizeof(exp->mask.src.u3) > len)
275 memset((void *)&exp->mask.src.u3 + len, 0x00,
276 sizeof(exp->mask.src.u3) - len);
277 } else {
278 memset(&exp->tuple.src.u3, 0x00, sizeof(exp->tuple.src.u3));
279 memset(&exp->mask.src.u3, 0x00, sizeof(exp->mask.src.u3));
282 if (src) {
283 exp->tuple.src.u.all = *src;
284 exp->mask.src.u.all = htons(0xFFFF);
285 } else {
286 exp->tuple.src.u.all = 0;
287 exp->mask.src.u.all = 0;
290 memcpy(&exp->tuple.dst.u3, daddr, len);
291 if (sizeof(exp->tuple.dst.u3) > len)
292 /* address needs to be cleared for nf_ct_tuple_equal */
293 memset((void *)&exp->tuple.dst.u3 + len, 0x00,
294 sizeof(exp->tuple.dst.u3) - len);
296 exp->tuple.dst.u.all = *dst;
298 EXPORT_SYMBOL_GPL(nf_ct_expect_init);
300 static void nf_ct_expect_free_rcu(struct rcu_head *head)
302 struct nf_conntrack_expect *exp;
304 exp = container_of(head, struct nf_conntrack_expect, rcu);
305 kmem_cache_free(nf_ct_expect_cachep, exp);
308 void nf_ct_expect_put(struct nf_conntrack_expect *exp)
310 if (atomic_dec_and_test(&exp->use))
311 call_rcu(&exp->rcu, nf_ct_expect_free_rcu);
313 EXPORT_SYMBOL_GPL(nf_ct_expect_put);
315 static int nf_ct_expect_insert(struct nf_conntrack_expect *exp)
317 struct nf_conn_help *master_help = nfct_help(exp->master);
318 struct nf_conntrack_helper *helper;
319 struct net *net = nf_ct_exp_net(exp);
320 unsigned int h = nf_ct_expect_dst_hash(&exp->tuple);
322 /* two references : one for hash insert, one for the timer */
323 atomic_add(2, &exp->use);
325 hlist_add_head(&exp->lnode, &master_help->expectations);
326 master_help->expecting[exp->class]++;
328 hlist_add_head_rcu(&exp->hnode, &net->ct.expect_hash[h]);
329 net->ct.expect_count++;
331 setup_timer(&exp->timeout, nf_ct_expectation_timed_out,
332 (unsigned long)exp);
333 helper = rcu_dereference_protected(master_help->helper,
334 lockdep_is_held(&nf_conntrack_lock));
335 if (helper) {
336 exp->timeout.expires = jiffies +
337 helper->expect_policy[exp->class].timeout * HZ;
339 add_timer(&exp->timeout);
341 NF_CT_STAT_INC(net, expect_create);
342 return 0;
345 /* Race with expectations being used means we could have none to find; OK. */
346 static void evict_oldest_expect(struct nf_conn *master,
347 struct nf_conntrack_expect *new)
349 struct nf_conn_help *master_help = nfct_help(master);
350 struct nf_conntrack_expect *exp, *last = NULL;
351 struct hlist_node *n;
353 hlist_for_each_entry(exp, n, &master_help->expectations, lnode) {
354 if (exp->class == new->class)
355 last = exp;
358 if (last && del_timer(&last->timeout)) {
359 nf_ct_unlink_expect(last);
360 nf_ct_expect_put(last);
364 static inline int __nf_ct_expect_check(struct nf_conntrack_expect *expect)
366 const struct nf_conntrack_expect_policy *p;
367 struct nf_conntrack_expect *i;
368 struct nf_conn *master = expect->master;
369 struct nf_conn_help *master_help = nfct_help(master);
370 struct nf_conntrack_helper *helper;
371 struct net *net = nf_ct_exp_net(expect);
372 struct hlist_node *n, *next;
373 unsigned int h;
374 int ret = 1;
376 if (!master_help) {
377 ret = -ESHUTDOWN;
378 goto out;
380 h = nf_ct_expect_dst_hash(&expect->tuple);
381 hlist_for_each_entry_safe(i, n, next, &net->ct.expect_hash[h], hnode) {
382 if (expect_matches(i, expect)) {
383 if (del_timer(&i->timeout)) {
384 nf_ct_unlink_expect(i);
385 nf_ct_expect_put(i);
386 break;
388 } else if (expect_clash(i, expect)) {
389 ret = -EBUSY;
390 goto out;
393 /* Will be over limit? */
394 helper = rcu_dereference_protected(master_help->helper,
395 lockdep_is_held(&nf_conntrack_lock));
396 if (helper) {
397 p = &helper->expect_policy[expect->class];
398 if (p->max_expected &&
399 master_help->expecting[expect->class] >= p->max_expected) {
400 evict_oldest_expect(master, expect);
401 if (master_help->expecting[expect->class]
402 >= p->max_expected) {
403 ret = -EMFILE;
404 goto out;
409 if (net->ct.expect_count >= nf_ct_expect_max) {
410 net_warn_ratelimited("nf_conntrack: expectation table full\n");
411 ret = -EMFILE;
413 out:
414 return ret;
417 int nf_ct_expect_related_report(struct nf_conntrack_expect *expect,
418 u32 pid, int report)
420 int ret;
422 spin_lock_bh(&nf_conntrack_lock);
423 ret = __nf_ct_expect_check(expect);
424 if (ret <= 0)
425 goto out;
427 ret = nf_ct_expect_insert(expect);
428 if (ret < 0)
429 goto out;
430 spin_unlock_bh(&nf_conntrack_lock);
431 nf_ct_expect_event_report(IPEXP_NEW, expect, pid, report);
432 return ret;
433 out:
434 spin_unlock_bh(&nf_conntrack_lock);
435 return ret;
437 EXPORT_SYMBOL_GPL(nf_ct_expect_related_report);
439 #ifdef CONFIG_NF_CONNTRACK_PROCFS
440 struct ct_expect_iter_state {
441 struct seq_net_private p;
442 unsigned int bucket;
445 static struct hlist_node *ct_expect_get_first(struct seq_file *seq)
447 struct net *net = seq_file_net(seq);
448 struct ct_expect_iter_state *st = seq->private;
449 struct hlist_node *n;
451 for (st->bucket = 0; st->bucket < nf_ct_expect_hsize; st->bucket++) {
452 n = rcu_dereference(hlist_first_rcu(&net->ct.expect_hash[st->bucket]));
453 if (n)
454 return n;
456 return NULL;
459 static struct hlist_node *ct_expect_get_next(struct seq_file *seq,
460 struct hlist_node *head)
462 struct net *net = seq_file_net(seq);
463 struct ct_expect_iter_state *st = seq->private;
465 head = rcu_dereference(hlist_next_rcu(head));
466 while (head == NULL) {
467 if (++st->bucket >= nf_ct_expect_hsize)
468 return NULL;
469 head = rcu_dereference(hlist_first_rcu(&net->ct.expect_hash[st->bucket]));
471 return head;
474 static struct hlist_node *ct_expect_get_idx(struct seq_file *seq, loff_t pos)
476 struct hlist_node *head = ct_expect_get_first(seq);
478 if (head)
479 while (pos && (head = ct_expect_get_next(seq, head)))
480 pos--;
481 return pos ? NULL : head;
484 static void *exp_seq_start(struct seq_file *seq, loff_t *pos)
485 __acquires(RCU)
487 rcu_read_lock();
488 return ct_expect_get_idx(seq, *pos);
491 static void *exp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
493 (*pos)++;
494 return ct_expect_get_next(seq, v);
497 static void exp_seq_stop(struct seq_file *seq, void *v)
498 __releases(RCU)
500 rcu_read_unlock();
503 static int exp_seq_show(struct seq_file *s, void *v)
505 struct nf_conntrack_expect *expect;
506 struct nf_conntrack_helper *helper;
507 struct hlist_node *n = v;
508 char *delim = "";
510 expect = hlist_entry(n, struct nf_conntrack_expect, hnode);
512 if (expect->timeout.function)
513 seq_printf(s, "%ld ", timer_pending(&expect->timeout)
514 ? (long)(expect->timeout.expires - jiffies)/HZ : 0);
515 else
516 seq_printf(s, "- ");
517 seq_printf(s, "l3proto = %u proto=%u ",
518 expect->tuple.src.l3num,
519 expect->tuple.dst.protonum);
520 print_tuple(s, &expect->tuple,
521 __nf_ct_l3proto_find(expect->tuple.src.l3num),
522 __nf_ct_l4proto_find(expect->tuple.src.l3num,
523 expect->tuple.dst.protonum));
525 if (expect->flags & NF_CT_EXPECT_PERMANENT) {
526 seq_printf(s, "PERMANENT");
527 delim = ",";
529 if (expect->flags & NF_CT_EXPECT_INACTIVE) {
530 seq_printf(s, "%sINACTIVE", delim);
531 delim = ",";
533 if (expect->flags & NF_CT_EXPECT_USERSPACE)
534 seq_printf(s, "%sUSERSPACE", delim);
536 helper = rcu_dereference(nfct_help(expect->master)->helper);
537 if (helper) {
538 seq_printf(s, "%s%s", expect->flags ? " " : "", helper->name);
539 if (helper->expect_policy[expect->class].name)
540 seq_printf(s, "/%s",
541 helper->expect_policy[expect->class].name);
544 return seq_putc(s, '\n');
547 static const struct seq_operations exp_seq_ops = {
548 .start = exp_seq_start,
549 .next = exp_seq_next,
550 .stop = exp_seq_stop,
551 .show = exp_seq_show
554 static int exp_open(struct inode *inode, struct file *file)
556 return seq_open_net(inode, file, &exp_seq_ops,
557 sizeof(struct ct_expect_iter_state));
560 static const struct file_operations exp_file_ops = {
561 .owner = THIS_MODULE,
562 .open = exp_open,
563 .read = seq_read,
564 .llseek = seq_lseek,
565 .release = seq_release_net,
567 #endif /* CONFIG_NF_CONNTRACK_PROCFS */
569 static int exp_proc_init(struct net *net)
571 #ifdef CONFIG_NF_CONNTRACK_PROCFS
572 struct proc_dir_entry *proc;
574 proc = proc_net_fops_create(net, "nf_conntrack_expect", 0440, &exp_file_ops);
575 if (!proc)
576 return -ENOMEM;
577 #endif /* CONFIG_NF_CONNTRACK_PROCFS */
578 return 0;
581 static void exp_proc_remove(struct net *net)
583 #ifdef CONFIG_NF_CONNTRACK_PROCFS
584 proc_net_remove(net, "nf_conntrack_expect");
585 #endif /* CONFIG_NF_CONNTRACK_PROCFS */
588 module_param_named(expect_hashsize, nf_ct_expect_hsize, uint, 0400);
590 int nf_conntrack_expect_init(struct net *net)
592 int err = -ENOMEM;
594 if (net_eq(net, &init_net)) {
595 if (!nf_ct_expect_hsize) {
596 nf_ct_expect_hsize = net->ct.htable_size / 256;
597 if (!nf_ct_expect_hsize)
598 nf_ct_expect_hsize = 1;
600 nf_ct_expect_max = nf_ct_expect_hsize * 4;
603 net->ct.expect_count = 0;
604 net->ct.expect_hash = nf_ct_alloc_hashtable(&nf_ct_expect_hsize, 0);
605 if (net->ct.expect_hash == NULL)
606 goto err1;
608 if (net_eq(net, &init_net)) {
609 nf_ct_expect_cachep = kmem_cache_create("nf_conntrack_expect",
610 sizeof(struct nf_conntrack_expect),
611 0, 0, NULL);
612 if (!nf_ct_expect_cachep)
613 goto err2;
616 err = exp_proc_init(net);
617 if (err < 0)
618 goto err3;
620 return 0;
622 err3:
623 if (net_eq(net, &init_net))
624 kmem_cache_destroy(nf_ct_expect_cachep);
625 err2:
626 nf_ct_free_hashtable(net->ct.expect_hash, nf_ct_expect_hsize);
627 err1:
628 return err;
631 void nf_conntrack_expect_fini(struct net *net)
633 exp_proc_remove(net);
634 if (net_eq(net, &init_net)) {
635 rcu_barrier(); /* Wait for call_rcu() before destroy */
636 kmem_cache_destroy(nf_ct_expect_cachep);
638 nf_ct_free_hashtable(net->ct.expect_hash, nf_ct_expect_hsize);