inotify: fix race
[linux-2.6.22.y-op.git] / net / netfilter / nf_conntrack_expect.c
blob504fb6c083f99bc30a9fac847943fa2dac4c51cb
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>
23 #include <net/netfilter/nf_conntrack.h>
24 #include <net/netfilter/nf_conntrack_core.h>
25 #include <net/netfilter/nf_conntrack_expect.h>
26 #include <net/netfilter/nf_conntrack_helper.h>
27 #include <net/netfilter/nf_conntrack_tuple.h>
29 LIST_HEAD(nf_conntrack_expect_list);
30 EXPORT_SYMBOL_GPL(nf_conntrack_expect_list);
32 struct kmem_cache *nf_conntrack_expect_cachep __read_mostly;
33 static unsigned int nf_conntrack_expect_next_id;
35 /* nf_conntrack_expect helper functions */
36 void nf_ct_unlink_expect(struct nf_conntrack_expect *exp)
38 struct nf_conn_help *master_help = nfct_help(exp->master);
40 NF_CT_ASSERT(master_help);
41 NF_CT_ASSERT(!timer_pending(&exp->timeout));
43 list_del(&exp->list);
44 NF_CT_STAT_INC(expect_delete);
45 master_help->expecting--;
46 nf_conntrack_expect_put(exp);
48 EXPORT_SYMBOL_GPL(nf_ct_unlink_expect);
50 static void expectation_timed_out(unsigned long ul_expect)
52 struct nf_conntrack_expect *exp = (void *)ul_expect;
54 write_lock_bh(&nf_conntrack_lock);
55 nf_ct_unlink_expect(exp);
56 write_unlock_bh(&nf_conntrack_lock);
57 nf_conntrack_expect_put(exp);
60 struct nf_conntrack_expect *
61 __nf_conntrack_expect_find(const struct nf_conntrack_tuple *tuple)
63 struct nf_conntrack_expect *i;
65 list_for_each_entry(i, &nf_conntrack_expect_list, list) {
66 if (nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask))
67 return i;
69 return NULL;
71 EXPORT_SYMBOL_GPL(__nf_conntrack_expect_find);
73 /* Just find a expectation corresponding to a tuple. */
74 struct nf_conntrack_expect *
75 nf_conntrack_expect_find_get(const struct nf_conntrack_tuple *tuple)
77 struct nf_conntrack_expect *i;
79 read_lock_bh(&nf_conntrack_lock);
80 i = __nf_conntrack_expect_find(tuple);
81 if (i)
82 atomic_inc(&i->use);
83 read_unlock_bh(&nf_conntrack_lock);
85 return i;
87 EXPORT_SYMBOL_GPL(nf_conntrack_expect_find_get);
89 /* If an expectation for this connection is found, it gets delete from
90 * global list then returned. */
91 struct nf_conntrack_expect *
92 find_expectation(const struct nf_conntrack_tuple *tuple)
94 struct nf_conntrack_expect *exp;
96 exp = __nf_conntrack_expect_find(tuple);
97 if (!exp)
98 return NULL;
100 /* If master is not in hash table yet (ie. packet hasn't left
101 this machine yet), how can other end know about expected?
102 Hence these are not the droids you are looking for (if
103 master ct never got confirmed, we'd hold a reference to it
104 and weird things would happen to future packets). */
105 if (!nf_ct_is_confirmed(exp->master))
106 return NULL;
108 if (exp->flags & NF_CT_EXPECT_PERMANENT) {
109 atomic_inc(&exp->use);
110 return exp;
111 } else if (del_timer(&exp->timeout)) {
112 nf_ct_unlink_expect(exp);
113 return exp;
116 return NULL;
119 /* delete all expectations for this conntrack */
120 void nf_ct_remove_expectations(struct nf_conn *ct)
122 struct nf_conntrack_expect *i, *tmp;
123 struct nf_conn_help *help = nfct_help(ct);
125 /* Optimization: most connection never expect any others. */
126 if (!help || help->expecting == 0)
127 return;
129 list_for_each_entry_safe(i, tmp, &nf_conntrack_expect_list, list) {
130 if (i->master == ct && del_timer(&i->timeout)) {
131 nf_ct_unlink_expect(i);
132 nf_conntrack_expect_put(i);
136 EXPORT_SYMBOL_GPL(nf_ct_remove_expectations);
138 /* Would two expected things clash? */
139 static inline int expect_clash(const struct nf_conntrack_expect *a,
140 const struct nf_conntrack_expect *b)
142 /* Part covered by intersection of masks must be unequal,
143 otherwise they clash */
144 struct nf_conntrack_tuple intersect_mask;
145 int count;
147 intersect_mask.src.l3num = a->mask.src.l3num & b->mask.src.l3num;
148 intersect_mask.src.u.all = a->mask.src.u.all & b->mask.src.u.all;
149 intersect_mask.dst.u.all = a->mask.dst.u.all & b->mask.dst.u.all;
150 intersect_mask.dst.protonum = a->mask.dst.protonum
151 & b->mask.dst.protonum;
153 for (count = 0; count < NF_CT_TUPLE_L3SIZE; count++){
154 intersect_mask.src.u3.all[count] =
155 a->mask.src.u3.all[count] & b->mask.src.u3.all[count];
158 for (count = 0; count < NF_CT_TUPLE_L3SIZE; count++){
159 intersect_mask.dst.u3.all[count] =
160 a->mask.dst.u3.all[count] & b->mask.dst.u3.all[count];
163 return nf_ct_tuple_mask_cmp(&a->tuple, &b->tuple, &intersect_mask);
166 static inline int expect_matches(const struct nf_conntrack_expect *a,
167 const struct nf_conntrack_expect *b)
169 return a->master == b->master
170 && nf_ct_tuple_equal(&a->tuple, &b->tuple)
171 && nf_ct_tuple_equal(&a->mask, &b->mask);
174 /* Generally a bad idea to call this: could have matched already. */
175 void nf_conntrack_unexpect_related(struct nf_conntrack_expect *exp)
177 struct nf_conntrack_expect *i;
179 write_lock_bh(&nf_conntrack_lock);
180 /* choose the oldest expectation to evict */
181 list_for_each_entry_reverse(i, &nf_conntrack_expect_list, list) {
182 if (expect_matches(i, exp) && del_timer(&i->timeout)) {
183 nf_ct_unlink_expect(i);
184 write_unlock_bh(&nf_conntrack_lock);
185 nf_conntrack_expect_put(i);
186 return;
189 write_unlock_bh(&nf_conntrack_lock);
191 EXPORT_SYMBOL_GPL(nf_conntrack_unexpect_related);
193 /* We don't increase the master conntrack refcount for non-fulfilled
194 * conntracks. During the conntrack destruction, the expectations are
195 * always killed before the conntrack itself */
196 struct nf_conntrack_expect *nf_conntrack_expect_alloc(struct nf_conn *me)
198 struct nf_conntrack_expect *new;
200 new = kmem_cache_alloc(nf_conntrack_expect_cachep, GFP_ATOMIC);
201 if (!new)
202 return NULL;
204 new->master = me;
205 atomic_set(&new->use, 1);
206 return new;
208 EXPORT_SYMBOL_GPL(nf_conntrack_expect_alloc);
210 void nf_conntrack_expect_init(struct nf_conntrack_expect *exp, int family,
211 union nf_conntrack_address *saddr,
212 union nf_conntrack_address *daddr,
213 u_int8_t proto, __be16 *src, __be16 *dst)
215 int len;
217 if (family == AF_INET)
218 len = 4;
219 else
220 len = 16;
222 exp->flags = 0;
223 exp->expectfn = NULL;
224 exp->helper = NULL;
225 exp->tuple.src.l3num = family;
226 exp->tuple.dst.protonum = proto;
227 exp->mask.src.l3num = 0xFFFF;
228 exp->mask.dst.protonum = 0xFF;
230 if (saddr) {
231 memcpy(&exp->tuple.src.u3, saddr, len);
232 if (sizeof(exp->tuple.src.u3) > len)
233 /* address needs to be cleared for nf_ct_tuple_equal */
234 memset((void *)&exp->tuple.src.u3 + len, 0x00,
235 sizeof(exp->tuple.src.u3) - len);
236 memset(&exp->mask.src.u3, 0xFF, len);
237 if (sizeof(exp->mask.src.u3) > len)
238 memset((void *)&exp->mask.src.u3 + len, 0x00,
239 sizeof(exp->mask.src.u3) - len);
240 } else {
241 memset(&exp->tuple.src.u3, 0x00, sizeof(exp->tuple.src.u3));
242 memset(&exp->mask.src.u3, 0x00, sizeof(exp->mask.src.u3));
245 if (daddr) {
246 memcpy(&exp->tuple.dst.u3, daddr, len);
247 if (sizeof(exp->tuple.dst.u3) > len)
248 /* address needs to be cleared for nf_ct_tuple_equal */
249 memset((void *)&exp->tuple.dst.u3 + len, 0x00,
250 sizeof(exp->tuple.dst.u3) - len);
251 memset(&exp->mask.dst.u3, 0xFF, len);
252 if (sizeof(exp->mask.dst.u3) > len)
253 memset((void *)&exp->mask.dst.u3 + len, 0x00,
254 sizeof(exp->mask.dst.u3) - len);
255 } else {
256 memset(&exp->tuple.dst.u3, 0x00, sizeof(exp->tuple.dst.u3));
257 memset(&exp->mask.dst.u3, 0x00, sizeof(exp->mask.dst.u3));
260 if (src) {
261 exp->tuple.src.u.all = (__force u16)*src;
262 exp->mask.src.u.all = 0xFFFF;
263 } else {
264 exp->tuple.src.u.all = 0;
265 exp->mask.src.u.all = 0;
268 if (dst) {
269 exp->tuple.dst.u.all = (__force u16)*dst;
270 exp->mask.dst.u.all = 0xFFFF;
271 } else {
272 exp->tuple.dst.u.all = 0;
273 exp->mask.dst.u.all = 0;
276 EXPORT_SYMBOL_GPL(nf_conntrack_expect_init);
278 void nf_conntrack_expect_put(struct nf_conntrack_expect *exp)
280 if (atomic_dec_and_test(&exp->use))
281 kmem_cache_free(nf_conntrack_expect_cachep, exp);
283 EXPORT_SYMBOL_GPL(nf_conntrack_expect_put);
285 static void nf_conntrack_expect_insert(struct nf_conntrack_expect *exp)
287 struct nf_conn_help *master_help = nfct_help(exp->master);
289 atomic_inc(&exp->use);
290 master_help->expecting++;
291 list_add(&exp->list, &nf_conntrack_expect_list);
293 setup_timer(&exp->timeout, expectation_timed_out, (unsigned long)exp);
294 exp->timeout.expires = jiffies + master_help->helper->timeout * HZ;
295 add_timer(&exp->timeout);
297 exp->id = ++nf_conntrack_expect_next_id;
298 atomic_inc(&exp->use);
299 NF_CT_STAT_INC(expect_create);
302 /* Race with expectations being used means we could have none to find; OK. */
303 static void evict_oldest_expect(struct nf_conn *master)
305 struct nf_conntrack_expect *i;
307 list_for_each_entry_reverse(i, &nf_conntrack_expect_list, list) {
308 if (i->master == master) {
309 if (del_timer(&i->timeout)) {
310 nf_ct_unlink_expect(i);
311 nf_conntrack_expect_put(i);
313 break;
318 static inline int refresh_timer(struct nf_conntrack_expect *i)
320 struct nf_conn_help *master_help = nfct_help(i->master);
322 if (!del_timer(&i->timeout))
323 return 0;
325 i->timeout.expires = jiffies + master_help->helper->timeout*HZ;
326 add_timer(&i->timeout);
327 return 1;
330 int nf_conntrack_expect_related(struct nf_conntrack_expect *expect)
332 struct nf_conntrack_expect *i;
333 struct nf_conn *master = expect->master;
334 struct nf_conn_help *master_help = nfct_help(master);
335 int ret;
337 NF_CT_ASSERT(master_help);
339 write_lock_bh(&nf_conntrack_lock);
340 if (!master_help->helper) {
341 ret = -ESHUTDOWN;
342 goto out;
344 list_for_each_entry(i, &nf_conntrack_expect_list, list) {
345 if (expect_matches(i, expect)) {
346 /* Refresh timer: if it's dying, ignore.. */
347 if (refresh_timer(i)) {
348 ret = 0;
349 goto out;
351 } else if (expect_clash(i, expect)) {
352 ret = -EBUSY;
353 goto out;
356 /* Will be over limit? */
357 if (master_help->helper->max_expected &&
358 master_help->expecting >= master_help->helper->max_expected)
359 evict_oldest_expect(master);
361 nf_conntrack_expect_insert(expect);
362 nf_conntrack_expect_event(IPEXP_NEW, expect);
363 ret = 0;
364 out:
365 write_unlock_bh(&nf_conntrack_lock);
366 return ret;
368 EXPORT_SYMBOL_GPL(nf_conntrack_expect_related);
370 #ifdef CONFIG_PROC_FS
371 static void *exp_seq_start(struct seq_file *s, loff_t *pos)
373 struct list_head *e = &nf_conntrack_expect_list;
374 loff_t i;
376 /* strange seq_file api calls stop even if we fail,
377 * thus we need to grab lock since stop unlocks */
378 read_lock_bh(&nf_conntrack_lock);
380 if (list_empty(e))
381 return NULL;
383 for (i = 0; i <= *pos; i++) {
384 e = e->next;
385 if (e == &nf_conntrack_expect_list)
386 return NULL;
388 return e;
391 static void *exp_seq_next(struct seq_file *s, void *v, loff_t *pos)
393 struct list_head *e = v;
395 ++*pos;
396 e = e->next;
398 if (e == &nf_conntrack_expect_list)
399 return NULL;
401 return e;
404 static void exp_seq_stop(struct seq_file *s, void *v)
406 read_unlock_bh(&nf_conntrack_lock);
409 static int exp_seq_show(struct seq_file *s, void *v)
411 struct nf_conntrack_expect *expect = v;
413 if (expect->timeout.function)
414 seq_printf(s, "%ld ", timer_pending(&expect->timeout)
415 ? (long)(expect->timeout.expires - jiffies)/HZ : 0);
416 else
417 seq_printf(s, "- ");
418 seq_printf(s, "l3proto = %u proto=%u ",
419 expect->tuple.src.l3num,
420 expect->tuple.dst.protonum);
421 print_tuple(s, &expect->tuple,
422 __nf_ct_l3proto_find(expect->tuple.src.l3num),
423 __nf_ct_l4proto_find(expect->tuple.src.l3num,
424 expect->tuple.dst.protonum));
425 return seq_putc(s, '\n');
428 static struct seq_operations exp_seq_ops = {
429 .start = exp_seq_start,
430 .next = exp_seq_next,
431 .stop = exp_seq_stop,
432 .show = exp_seq_show
435 static int exp_open(struct inode *inode, struct file *file)
437 return seq_open(file, &exp_seq_ops);
440 const struct file_operations exp_file_ops = {
441 .owner = THIS_MODULE,
442 .open = exp_open,
443 .read = seq_read,
444 .llseek = seq_lseek,
445 .release = seq_release
447 #endif /* CONFIG_PROC_FS */