mm: Remove slab destructors from kmem_cache_create().
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / net / netfilter / nf_conntrack_expect.c
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 *
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.
10 */
11
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
24 #include <net/netfilter/nf_conntrack.h>
25 #include <net/netfilter/nf_conntrack_core.h>
26 #include <net/netfilter/nf_conntrack_expect.h>
27 #include <net/netfilter/nf_conntrack_helper.h>
28 #include <net/netfilter/nf_conntrack_tuple.h>
29
30 struct hlist_head *nf_ct_expect_hash __read_mostly;
31 EXPORT_SYMBOL_GPL(nf_ct_expect_hash);
32
33 unsigned int nf_ct_expect_hsize __read_mostly;
34 EXPORT_SYMBOL_GPL(nf_ct_expect_hsize);
35
36 static unsigned int nf_ct_expect_hash_rnd __read_mostly;
37 static unsigned int nf_ct_expect_count;
38 unsigned int nf_ct_expect_max __read_mostly;
39 static int nf_ct_expect_hash_rnd_initted __read_mostly;
40 static int nf_ct_expect_vmalloc;
41
42 static struct kmem_cache *nf_ct_expect_cachep __read_mostly;
43 static unsigned int nf_ct_expect_next_id;
44
45 /* nf_conntrack_expect helper functions */
46 void nf_ct_unlink_expect(struct nf_conntrack_expect *exp)
47 {
48 struct nf_conn_help *master_help = nfct_help(exp->master);
49
50 NF_CT_ASSERT(master_help);
51 NF_CT_ASSERT(!timer_pending(&exp->timeout));
52
53 hlist_del(&exp->hnode);
54 nf_ct_expect_count--;
55
56 hlist_del(&exp->lnode);
57 master_help->expecting--;
58 nf_ct_expect_put(exp);
59
60 NF_CT_STAT_INC(expect_delete);
61 }
62 EXPORT_SYMBOL_GPL(nf_ct_unlink_expect);
63
64 static void nf_ct_expectation_timed_out(unsigned long ul_expect)
65 {
66 struct nf_conntrack_expect *exp = (void *)ul_expect;
67
68 write_lock_bh(&nf_conntrack_lock);
69 nf_ct_unlink_expect(exp);
70 write_unlock_bh(&nf_conntrack_lock);
71 nf_ct_expect_put(exp);
72 }
73
74 static unsigned int nf_ct_expect_dst_hash(const struct nf_conntrack_tuple *tuple)
75 {
76 if (unlikely(!nf_ct_expect_hash_rnd_initted)) {
77 get_random_bytes(&nf_ct_expect_hash_rnd, 4);
78 nf_ct_expect_hash_rnd_initted = 1;
79 }
80
81 return jhash2(tuple->dst.u3.all, ARRAY_SIZE(tuple->dst.u3.all),
82 (((tuple->dst.protonum ^ tuple->src.l3num) << 16) |
83 tuple->dst.u.all) ^ nf_ct_expect_hash_rnd) %
84 nf_ct_expect_hsize;
85 }
86
87 struct nf_conntrack_expect *
88 __nf_ct_expect_find(const struct nf_conntrack_tuple *tuple)
89 {
90 struct nf_conntrack_expect *i;
91 struct hlist_node *n;
92 unsigned int h;
93
94 if (!nf_ct_expect_count)
95 return NULL;
96
97 h = nf_ct_expect_dst_hash(tuple);
98 hlist_for_each_entry(i, n, &nf_ct_expect_hash[h], hnode) {
99 if (nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask))
100 return i;
101 }
102 return NULL;
103 }
104 EXPORT_SYMBOL_GPL(__nf_ct_expect_find);
105
106 /* Just find a expectation corresponding to a tuple. */
107 struct nf_conntrack_expect *
108 nf_ct_expect_find_get(const struct nf_conntrack_tuple *tuple)
109 {
110 struct nf_conntrack_expect *i;
111
112 read_lock_bh(&nf_conntrack_lock);
113 i = __nf_ct_expect_find(tuple);
114 if (i)
115 atomic_inc(&i->use);
116 read_unlock_bh(&nf_conntrack_lock);
117
118 return i;
119 }
120 EXPORT_SYMBOL_GPL(nf_ct_expect_find_get);
121
122 /* If an expectation for this connection is found, it gets delete from
123 * global list then returned. */
124 struct nf_conntrack_expect *
125 nf_ct_find_expectation(const struct nf_conntrack_tuple *tuple)
126 {
127 struct nf_conntrack_expect *exp;
128
129 exp = __nf_ct_expect_find(tuple);
130 if (!exp)
131 return NULL;
132
133 /* If master is not in hash table yet (ie. packet hasn't left
134 this machine yet), how can other end know about expected?
135 Hence these are not the droids you are looking for (if
136 master ct never got confirmed, we'd hold a reference to it
137 and weird things would happen to future packets). */
138 if (!nf_ct_is_confirmed(exp->master))
139 return NULL;
140
141 if (exp->flags & NF_CT_EXPECT_PERMANENT) {
142 atomic_inc(&exp->use);
143 return exp;
144 } else if (del_timer(&exp->timeout)) {
145 nf_ct_unlink_expect(exp);
146 return exp;
147 }
148
149 return NULL;
150 }
151
152 /* delete all expectations for this conntrack */
153 void nf_ct_remove_expectations(struct nf_conn *ct)
154 {
155 struct nf_conn_help *help = nfct_help(ct);
156 struct nf_conntrack_expect *exp;
157 struct hlist_node *n, *next;
158
159 /* Optimization: most connection never expect any others. */
160 if (!help || help->expecting == 0)
161 return;
162
163 hlist_for_each_entry_safe(exp, n, next, &help->expectations, lnode) {
164 if (del_timer(&exp->timeout)) {
165 nf_ct_unlink_expect(exp);
166 nf_ct_expect_put(exp);
167 }
168 }
169 }
170 EXPORT_SYMBOL_GPL(nf_ct_remove_expectations);
171
172 /* Would two expected things clash? */
173 static inline int expect_clash(const struct nf_conntrack_expect *a,
174 const struct nf_conntrack_expect *b)
175 {
176 /* Part covered by intersection of masks must be unequal,
177 otherwise they clash */
178 struct nf_conntrack_tuple_mask intersect_mask;
179 int count;
180
181 intersect_mask.src.u.all = a->mask.src.u.all & b->mask.src.u.all;
182
183 for (count = 0; count < NF_CT_TUPLE_L3SIZE; count++){
184 intersect_mask.src.u3.all[count] =
185 a->mask.src.u3.all[count] & b->mask.src.u3.all[count];
186 }
187
188 return nf_ct_tuple_mask_cmp(&a->tuple, &b->tuple, &intersect_mask);
189 }
190
191 static inline int expect_matches(const struct nf_conntrack_expect *a,
192 const struct nf_conntrack_expect *b)
193 {
194 return a->master == b->master
195 && nf_ct_tuple_equal(&a->tuple, &b->tuple)
196 && nf_ct_tuple_mask_equal(&a->mask, &b->mask);
197 }
198
199 /* Generally a bad idea to call this: could have matched already. */
200 void nf_ct_unexpect_related(struct nf_conntrack_expect *exp)
201 {
202 write_lock_bh(&nf_conntrack_lock);
203 if (del_timer(&exp->timeout)) {
204 nf_ct_unlink_expect(exp);
205 nf_ct_expect_put(exp);
206 }
207 write_unlock_bh(&nf_conntrack_lock);
208 }
209 EXPORT_SYMBOL_GPL(nf_ct_unexpect_related);
210
211 /* We don't increase the master conntrack refcount for non-fulfilled
212 * conntracks. During the conntrack destruction, the expectations are
213 * always killed before the conntrack itself */
214 struct nf_conntrack_expect *nf_ct_expect_alloc(struct nf_conn *me)
215 {
216 struct nf_conntrack_expect *new;
217
218 new = kmem_cache_alloc(nf_ct_expect_cachep, GFP_ATOMIC);
219 if (!new)
220 return NULL;
221
222 new->master = me;
223 atomic_set(&new->use, 1);
224 return new;
225 }
226 EXPORT_SYMBOL_GPL(nf_ct_expect_alloc);
227
228 void nf_ct_expect_init(struct nf_conntrack_expect *exp, int family,
229 union nf_conntrack_address *saddr,
230 union nf_conntrack_address *daddr,
231 u_int8_t proto, __be16 *src, __be16 *dst)
232 {
233 int len;
234
235 if (family == AF_INET)
236 len = 4;
237 else
238 len = 16;
239
240 exp->flags = 0;
241 exp->expectfn = NULL;
242 exp->helper = NULL;
243 exp->tuple.src.l3num = family;
244 exp->tuple.dst.protonum = proto;
245
246 if (saddr) {
247 memcpy(&exp->tuple.src.u3, saddr, len);
248 if (sizeof(exp->tuple.src.u3) > len)
249 /* address needs to be cleared for nf_ct_tuple_equal */
250 memset((void *)&exp->tuple.src.u3 + len, 0x00,
251 sizeof(exp->tuple.src.u3) - len);
252 memset(&exp->mask.src.u3, 0xFF, len);
253 if (sizeof(exp->mask.src.u3) > len)
254 memset((void *)&exp->mask.src.u3 + len, 0x00,
255 sizeof(exp->mask.src.u3) - len);
256 } else {
257 memset(&exp->tuple.src.u3, 0x00, sizeof(exp->tuple.src.u3));
258 memset(&exp->mask.src.u3, 0x00, sizeof(exp->mask.src.u3));
259 }
260
261 if (src) {
262 exp->tuple.src.u.all = (__force u16)*src;
263 exp->mask.src.u.all = 0xFFFF;
264 } else {
265 exp->tuple.src.u.all = 0;
266 exp->mask.src.u.all = 0;
267 }
268
269 memcpy(&exp->tuple.dst.u3, daddr, len);
270 if (sizeof(exp->tuple.dst.u3) > len)
271 /* address needs to be cleared for nf_ct_tuple_equal */
272 memset((void *)&exp->tuple.dst.u3 + len, 0x00,
273 sizeof(exp->tuple.dst.u3) - len);
274
275 exp->tuple.dst.u.all = (__force u16)*dst;
276 }
277 EXPORT_SYMBOL_GPL(nf_ct_expect_init);
278
279 void nf_ct_expect_put(struct nf_conntrack_expect *exp)
280 {
281 if (atomic_dec_and_test(&exp->use))
282 kmem_cache_free(nf_ct_expect_cachep, exp);
283 }
284 EXPORT_SYMBOL_GPL(nf_ct_expect_put);
285
286 static void nf_ct_expect_insert(struct nf_conntrack_expect *exp)
287 {
288 struct nf_conn_help *master_help = nfct_help(exp->master);
289 unsigned int h = nf_ct_expect_dst_hash(&exp->tuple);
290
291 atomic_inc(&exp->use);
292
293 hlist_add_head(&exp->lnode, &master_help->expectations);
294 master_help->expecting++;
295
296 hlist_add_head(&exp->hnode, &nf_ct_expect_hash[h]);
297 nf_ct_expect_count++;
298
299 setup_timer(&exp->timeout, nf_ct_expectation_timed_out,
300 (unsigned long)exp);
301 exp->timeout.expires = jiffies + master_help->helper->timeout * HZ;
302 add_timer(&exp->timeout);
303
304 exp->id = ++nf_ct_expect_next_id;
305 atomic_inc(&exp->use);
306 NF_CT_STAT_INC(expect_create);
307 }
308
309 /* Race with expectations being used means we could have none to find; OK. */
310 static void evict_oldest_expect(struct nf_conn *master)
311 {
312 struct nf_conn_help *master_help = nfct_help(master);
313 struct nf_conntrack_expect *exp = NULL;
314 struct hlist_node *n;
315
316 hlist_for_each_entry(exp, n, &master_help->expectations, lnode)
317 ; /* nothing */
318
319 if (exp && del_timer(&exp->timeout)) {
320 nf_ct_unlink_expect(exp);
321 nf_ct_expect_put(exp);
322 }
323 }
324
325 static inline int refresh_timer(struct nf_conntrack_expect *i)
326 {
327 struct nf_conn_help *master_help = nfct_help(i->master);
328
329 if (!del_timer(&i->timeout))
330 return 0;
331
332 i->timeout.expires = jiffies + master_help->helper->timeout*HZ;
333 add_timer(&i->timeout);
334 return 1;
335 }
336
337 int nf_ct_expect_related(struct nf_conntrack_expect *expect)
338 {
339 struct nf_conntrack_expect *i;
340 struct nf_conn *master = expect->master;
341 struct nf_conn_help *master_help = nfct_help(master);
342 struct hlist_node *n;
343 unsigned int h;
344 int ret;
345
346 NF_CT_ASSERT(master_help);
347
348 write_lock_bh(&nf_conntrack_lock);
349 if (!master_help->helper) {
350 ret = -ESHUTDOWN;
351 goto out;
352 }
353 h = nf_ct_expect_dst_hash(&expect->tuple);
354 hlist_for_each_entry(i, n, &nf_ct_expect_hash[h], hnode) {
355 if (expect_matches(i, expect)) {
356 /* Refresh timer: if it's dying, ignore.. */
357 if (refresh_timer(i)) {
358 ret = 0;
359 goto out;
360 }
361 } else if (expect_clash(i, expect)) {
362 ret = -EBUSY;
363 goto out;
364 }
365 }
366 /* Will be over limit? */
367 if (master_help->helper->max_expected &&
368 master_help->expecting >= master_help->helper->max_expected)
369 evict_oldest_expect(master);
370
371 if (nf_ct_expect_count >= nf_ct_expect_max) {
372 if (net_ratelimit())
373 printk(KERN_WARNING
374 "nf_conntrack: expectation table full");
375 ret = -EMFILE;
376 goto out;
377 }
378
379 nf_ct_expect_insert(expect);
380 nf_ct_expect_event(IPEXP_NEW, expect);
381 ret = 0;
382 out:
383 write_unlock_bh(&nf_conntrack_lock);
384 return ret;
385 }
386 EXPORT_SYMBOL_GPL(nf_ct_expect_related);
387
388 #ifdef CONFIG_PROC_FS
389 struct ct_expect_iter_state {
390 unsigned int bucket;
391 };
392
393 static struct hlist_node *ct_expect_get_first(struct seq_file *seq)
394 {
395 struct ct_expect_iter_state *st = seq->private;
396
397 for (st->bucket = 0; st->bucket < nf_ct_expect_hsize; st->bucket++) {
398 if (!hlist_empty(&nf_ct_expect_hash[st->bucket]))
399 return nf_ct_expect_hash[st->bucket].first;
400 }
401 return NULL;
402 }
403
404 static struct hlist_node *ct_expect_get_next(struct seq_file *seq,
405 struct hlist_node *head)
406 {
407 struct ct_expect_iter_state *st = seq->private;
408
409 head = head->next;
410 while (head == NULL) {
411 if (++st->bucket >= nf_ct_expect_hsize)
412 return NULL;
413 head = nf_ct_expect_hash[st->bucket].first;
414 }
415 return head;
416 }
417
418 static struct hlist_node *ct_expect_get_idx(struct seq_file *seq, loff_t pos)
419 {
420 struct hlist_node *head = ct_expect_get_first(seq);
421
422 if (head)
423 while (pos && (head = ct_expect_get_next(seq, head)))
424 pos--;
425 return pos ? NULL : head;
426 }
427
428 static void *exp_seq_start(struct seq_file *seq, loff_t *pos)
429 {
430 read_lock_bh(&nf_conntrack_lock);
431 return ct_expect_get_idx(seq, *pos);
432 }
433
434 static void *exp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
435 {
436 (*pos)++;
437 return ct_expect_get_next(seq, v);
438 }
439
440 static void exp_seq_stop(struct seq_file *seq, void *v)
441 {
442 read_unlock_bh(&nf_conntrack_lock);
443 }
444
445 static int exp_seq_show(struct seq_file *s, void *v)
446 {
447 struct nf_conntrack_expect *expect;
448 struct hlist_node *n = v;
449
450 expect = hlist_entry(n, struct nf_conntrack_expect, hnode);
451
452 if (expect->timeout.function)
453 seq_printf(s, "%ld ", timer_pending(&expect->timeout)
454 ? (long)(expect->timeout.expires - jiffies)/HZ : 0);
455 else
456 seq_printf(s, "- ");
457 seq_printf(s, "l3proto = %u proto=%u ",
458 expect->tuple.src.l3num,
459 expect->tuple.dst.protonum);
460 print_tuple(s, &expect->tuple,
461 __nf_ct_l3proto_find(expect->tuple.src.l3num),
462 __nf_ct_l4proto_find(expect->tuple.src.l3num,
463 expect->tuple.dst.protonum));
464 return seq_putc(s, '\n');
465 }
466
467 static const struct seq_operations exp_seq_ops = {
468 .start = exp_seq_start,
469 .next = exp_seq_next,
470 .stop = exp_seq_stop,
471 .show = exp_seq_show
472 };
473
474 static int exp_open(struct inode *inode, struct file *file)
475 {
476 struct seq_file *seq;
477 struct ct_expect_iter_state *st;
478 int ret;
479
480 st = kmalloc(sizeof(struct ct_expect_iter_state), GFP_KERNEL);
481 if (st == NULL)
482 return -ENOMEM;
483 ret = seq_open(file, &exp_seq_ops);
484 if (ret)
485 goto out_free;
486 seq = file->private_data;
487 seq->private = st;
488 memset(st, 0, sizeof(struct ct_expect_iter_state));
489 return ret;
490 out_free:
491 kfree(st);
492 return ret;
493 }
494
495 static const struct file_operations exp_file_ops = {
496 .owner = THIS_MODULE,
497 .open = exp_open,
498 .read = seq_read,
499 .llseek = seq_lseek,
500 .release = seq_release_private,
501 };
502 #endif /* CONFIG_PROC_FS */
503
504 static int __init exp_proc_init(void)
505 {
506 #ifdef CONFIG_PROC_FS
507 struct proc_dir_entry *proc;
508
509 proc = proc_net_fops_create("nf_conntrack_expect", 0440, &exp_file_ops);
510 if (!proc)
511 return -ENOMEM;
512 #endif /* CONFIG_PROC_FS */
513 return 0;
514 }
515
516 static void exp_proc_remove(void)
517 {
518 #ifdef CONFIG_PROC_FS
519 proc_net_remove("nf_conntrack_expect");
520 #endif /* CONFIG_PROC_FS */
521 }
522
523 module_param_named(expect_hashsize, nf_ct_expect_hsize, uint, 0600);
524
525 int __init nf_conntrack_expect_init(void)
526 {
527 int err = -ENOMEM;
528
529 if (!nf_ct_expect_hsize) {
530 nf_ct_expect_hsize = nf_conntrack_htable_size / 256;
531 if (!nf_ct_expect_hsize)
532 nf_ct_expect_hsize = 1;
533 }
534 nf_ct_expect_max = nf_ct_expect_hsize * 4;
535
536 nf_ct_expect_hash = nf_ct_alloc_hashtable(&nf_ct_expect_hsize,
537 &nf_ct_expect_vmalloc);
538 if (nf_ct_expect_hash == NULL)
539 goto err1;
540
541 nf_ct_expect_cachep = kmem_cache_create("nf_conntrack_expect",
542 sizeof(struct nf_conntrack_expect),
543 0, 0, NULL);
544 if (!nf_ct_expect_cachep)
545 goto err2;
546
547 err = exp_proc_init();
548 if (err < 0)
549 goto err3;
550
551 return 0;
552
553 err3:
554 nf_ct_free_hashtable(nf_ct_expect_hash, nf_ct_expect_vmalloc,
555 nf_ct_expect_hsize);
556 err2:
557 kmem_cache_destroy(nf_ct_expect_cachep);
558 err1:
559 return err;
560 }
561
562 void nf_conntrack_expect_fini(void)
563 {
564 exp_proc_remove();
565 kmem_cache_destroy(nf_ct_expect_cachep);
566 nf_ct_free_hashtable(nf_ct_expect_hash, nf_ct_expect_vmalloc,
567 nf_ct_expect_hsize);
568 }