1 /* Event cache for netfilter. */
4 * (C) 2005 Harald Welte <laforge@gnumonks.org>
5 * (C) 2005 Patrick McHardy <kaber@trash.net>
6 * (C) 2005-2006 Netfilter Core Team <coreteam@netfilter.org>
7 * (C) 2005 USAGI/WIDE Project <http://www.linux-ipv6.org>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
14 #include <linux/types.h>
15 #include <linux/netfilter.h>
16 #include <linux/skbuff.h>
17 #include <linux/vmalloc.h>
18 #include <linux/stddef.h>
19 #include <linux/err.h>
20 #include <linux/percpu.h>
21 #include <linux/kernel.h>
22 #include <linux/netdevice.h>
23 #include <linux/slab.h>
24 #include <linux/export.h>
26 #include <net/netfilter/nf_conntrack.h>
27 #include <net/netfilter/nf_conntrack_core.h>
28 #include <net/netfilter/nf_conntrack_extend.h>
30 static DEFINE_MUTEX(nf_ct_ecache_mutex);
32 #define ECACHE_RETRY_WAIT (HZ/10)
40 static enum retry_state ecache_work_evict_list(struct ct_pcpu *pcpu)
42 struct nf_conn *refs[16];
43 struct nf_conntrack_tuple_hash *h;
44 struct hlist_nulls_node *n;
45 unsigned int evicted = 0;
46 enum retry_state ret = STATE_DONE;
48 spin_lock(&pcpu->lock);
50 hlist_nulls_for_each_entry(h, n, &pcpu->dying, hnnode) {
51 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
53 if (nf_ct_is_dying(ct))
56 if (nf_conntrack_event(IPCT_DESTROY, ct)) {
57 ret = STATE_CONGESTED;
61 /* we've got the event delivered, now it's dying */
62 set_bit(IPS_DYING_BIT, &ct->status);
65 if (++evicted >= ARRAY_SIZE(refs)) {
71 spin_unlock(&pcpu->lock);
73 /* can't _put while holding lock */
75 nf_ct_put(refs[--evicted]);
80 static void ecache_work(struct work_struct *work)
82 struct netns_ct *ctnet =
83 container_of(work, struct netns_ct, ecache_dwork.work);
89 for_each_possible_cpu(cpu) {
92 pcpu = per_cpu_ptr(ctnet->pcpu_lists, cpu);
94 ret = ecache_work_evict_list(pcpu);
98 delay = ECACHE_RETRY_WAIT;
111 ctnet->ecache_dwork_pending = delay > 0;
113 schedule_delayed_work(&ctnet->ecache_dwork, delay);
116 int nf_conntrack_eventmask_report(unsigned int eventmask, struct nf_conn *ct,
117 u32 portid, int report)
120 struct net *net = nf_ct_net(ct);
121 struct nf_ct_event_notifier *notify;
122 struct nf_conntrack_ecache *e;
125 notify = rcu_dereference(net->ct.nf_conntrack_event_cb);
129 e = nf_ct_ecache_find(ct);
133 if (nf_ct_is_confirmed(ct) && !nf_ct_is_dying(ct)) {
134 struct nf_ct_event item = {
136 .portid = e->portid ? e->portid : portid,
139 /* This is a resent of a destroy event? If so, skip missed */
140 unsigned long missed = e->portid ? 0 : e->missed;
142 if (!((eventmask | missed) & e->ctmask))
145 ret = notify->fcn(eventmask | missed, &item);
146 if (unlikely(ret < 0 || missed)) {
147 spin_lock_bh(&ct->lock);
149 /* This is a destroy event that has been
150 * triggered by a process, we store the PORTID
151 * to include it in the retransmission.
153 if (eventmask & (1 << IPCT_DESTROY) &&
154 e->portid == 0 && portid != 0)
157 e->missed |= eventmask;
159 e->missed &= ~missed;
161 spin_unlock_bh(&ct->lock);
168 EXPORT_SYMBOL_GPL(nf_conntrack_eventmask_report);
170 /* deliver cached events and clear cache entry - must be called with locally
171 * disabled softirqs */
172 void nf_ct_deliver_cached_events(struct nf_conn *ct)
174 struct net *net = nf_ct_net(ct);
175 unsigned long events, missed;
176 struct nf_ct_event_notifier *notify;
177 struct nf_conntrack_ecache *e;
178 struct nf_ct_event item;
182 notify = rcu_dereference(net->ct.nf_conntrack_event_cb);
186 e = nf_ct_ecache_find(ct);
190 events = xchg(&e->cache, 0);
192 if (!nf_ct_is_confirmed(ct) || nf_ct_is_dying(ct) || !events)
195 /* We make a copy of the missed event cache without taking
196 * the lock, thus we may send missed events twice. However,
197 * this does not harm and it happens very rarely. */
200 if (!((events | missed) & e->ctmask))
207 ret = notify->fcn(events | missed, &item);
209 if (likely(ret >= 0 && !missed))
212 spin_lock_bh(&ct->lock);
216 e->missed &= ~missed;
217 spin_unlock_bh(&ct->lock);
222 EXPORT_SYMBOL_GPL(nf_ct_deliver_cached_events);
224 void nf_ct_expect_event_report(enum ip_conntrack_expect_events event,
225 struct nf_conntrack_expect *exp,
226 u32 portid, int report)
229 struct net *net = nf_ct_exp_net(exp);
230 struct nf_exp_event_notifier *notify;
231 struct nf_conntrack_ecache *e;
234 notify = rcu_dereference(net->ct.nf_expect_event_cb);
238 e = nf_ct_ecache_find(exp->master);
242 if (e->expmask & (1 << event)) {
243 struct nf_exp_event item = {
248 notify->fcn(1 << event, &item);
254 int nf_conntrack_register_notifier(struct net *net,
255 struct nf_ct_event_notifier *new)
258 struct nf_ct_event_notifier *notify;
260 mutex_lock(&nf_ct_ecache_mutex);
261 notify = rcu_dereference_protected(net->ct.nf_conntrack_event_cb,
262 lockdep_is_held(&nf_ct_ecache_mutex));
263 if (notify != NULL) {
267 rcu_assign_pointer(net->ct.nf_conntrack_event_cb, new);
271 mutex_unlock(&nf_ct_ecache_mutex);
274 EXPORT_SYMBOL_GPL(nf_conntrack_register_notifier);
276 void nf_conntrack_unregister_notifier(struct net *net,
277 struct nf_ct_event_notifier *new)
279 struct nf_ct_event_notifier *notify;
281 mutex_lock(&nf_ct_ecache_mutex);
282 notify = rcu_dereference_protected(net->ct.nf_conntrack_event_cb,
283 lockdep_is_held(&nf_ct_ecache_mutex));
284 BUG_ON(notify != new);
285 RCU_INIT_POINTER(net->ct.nf_conntrack_event_cb, NULL);
286 mutex_unlock(&nf_ct_ecache_mutex);
288 EXPORT_SYMBOL_GPL(nf_conntrack_unregister_notifier);
290 int nf_ct_expect_register_notifier(struct net *net,
291 struct nf_exp_event_notifier *new)
294 struct nf_exp_event_notifier *notify;
296 mutex_lock(&nf_ct_ecache_mutex);
297 notify = rcu_dereference_protected(net->ct.nf_expect_event_cb,
298 lockdep_is_held(&nf_ct_ecache_mutex));
299 if (notify != NULL) {
303 rcu_assign_pointer(net->ct.nf_expect_event_cb, new);
307 mutex_unlock(&nf_ct_ecache_mutex);
310 EXPORT_SYMBOL_GPL(nf_ct_expect_register_notifier);
312 void nf_ct_expect_unregister_notifier(struct net *net,
313 struct nf_exp_event_notifier *new)
315 struct nf_exp_event_notifier *notify;
317 mutex_lock(&nf_ct_ecache_mutex);
318 notify = rcu_dereference_protected(net->ct.nf_expect_event_cb,
319 lockdep_is_held(&nf_ct_ecache_mutex));
320 BUG_ON(notify != new);
321 RCU_INIT_POINTER(net->ct.nf_expect_event_cb, NULL);
322 mutex_unlock(&nf_ct_ecache_mutex);
324 EXPORT_SYMBOL_GPL(nf_ct_expect_unregister_notifier);
326 #define NF_CT_EVENTS_DEFAULT 1
327 static int nf_ct_events __read_mostly = NF_CT_EVENTS_DEFAULT;
330 static struct ctl_table event_sysctl_table[] = {
332 .procname = "nf_conntrack_events",
333 .data = &init_net.ct.sysctl_events,
334 .maxlen = sizeof(unsigned int),
336 .proc_handler = proc_dointvec,
340 #endif /* CONFIG_SYSCTL */
342 static struct nf_ct_ext_type event_extend __read_mostly = {
343 .len = sizeof(struct nf_conntrack_ecache),
344 .align = __alignof__(struct nf_conntrack_ecache),
345 .id = NF_CT_EXT_ECACHE,
349 static int nf_conntrack_event_init_sysctl(struct net *net)
351 struct ctl_table *table;
353 table = kmemdup(event_sysctl_table, sizeof(event_sysctl_table),
358 table[0].data = &net->ct.sysctl_events;
360 /* Don't export sysctls to unprivileged users */
361 if (net->user_ns != &init_user_ns)
362 table[0].procname = NULL;
364 net->ct.event_sysctl_header =
365 register_net_sysctl(net, "net/netfilter", table);
366 if (!net->ct.event_sysctl_header) {
367 printk(KERN_ERR "nf_ct_event: can't register to sysctl.\n");
378 static void nf_conntrack_event_fini_sysctl(struct net *net)
380 struct ctl_table *table;
382 table = net->ct.event_sysctl_header->ctl_table_arg;
383 unregister_net_sysctl_table(net->ct.event_sysctl_header);
387 static int nf_conntrack_event_init_sysctl(struct net *net)
392 static void nf_conntrack_event_fini_sysctl(struct net *net)
395 #endif /* CONFIG_SYSCTL */
397 int nf_conntrack_ecache_pernet_init(struct net *net)
399 net->ct.sysctl_events = nf_ct_events;
400 INIT_DELAYED_WORK(&net->ct.ecache_dwork, ecache_work);
401 return nf_conntrack_event_init_sysctl(net);
404 void nf_conntrack_ecache_pernet_fini(struct net *net)
406 cancel_delayed_work_sync(&net->ct.ecache_dwork);
407 nf_conntrack_event_fini_sysctl(net);
410 int nf_conntrack_ecache_init(void)
412 int ret = nf_ct_extend_register(&event_extend);
414 pr_err("nf_ct_event: Unable to register event extension.\n");
418 void nf_conntrack_ecache_fini(void)
420 nf_ct_extend_unregister(&event_extend);