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 /* deliver cached events and clear cache entry - must be called with locally
117 * disabled softirqs */
118 void nf_ct_deliver_cached_events(struct nf_conn *ct)
120 struct net *net = nf_ct_net(ct);
121 unsigned long events, missed;
122 struct nf_ct_event_notifier *notify;
123 struct nf_conntrack_ecache *e;
124 struct nf_ct_event item;
128 notify = rcu_dereference(net->ct.nf_conntrack_event_cb);
132 e = nf_ct_ecache_find(ct);
136 events = xchg(&e->cache, 0);
138 if (!nf_ct_is_confirmed(ct) || nf_ct_is_dying(ct) || !events)
141 /* We make a copy of the missed event cache without taking
142 * the lock, thus we may send missed events twice. However,
143 * this does not harm and it happens very rarely. */
146 if (!((events | missed) & e->ctmask))
153 ret = notify->fcn(events | missed, &item);
155 if (likely(ret >= 0 && !missed))
158 spin_lock_bh(&ct->lock);
162 e->missed &= ~missed;
163 spin_unlock_bh(&ct->lock);
168 EXPORT_SYMBOL_GPL(nf_ct_deliver_cached_events);
170 int nf_conntrack_register_notifier(struct net *net,
171 struct nf_ct_event_notifier *new)
174 struct nf_ct_event_notifier *notify;
176 mutex_lock(&nf_ct_ecache_mutex);
177 notify = rcu_dereference_protected(net->ct.nf_conntrack_event_cb,
178 lockdep_is_held(&nf_ct_ecache_mutex));
179 if (notify != NULL) {
183 rcu_assign_pointer(net->ct.nf_conntrack_event_cb, new);
187 mutex_unlock(&nf_ct_ecache_mutex);
190 EXPORT_SYMBOL_GPL(nf_conntrack_register_notifier);
192 void nf_conntrack_unregister_notifier(struct net *net,
193 struct nf_ct_event_notifier *new)
195 struct nf_ct_event_notifier *notify;
197 mutex_lock(&nf_ct_ecache_mutex);
198 notify = rcu_dereference_protected(net->ct.nf_conntrack_event_cb,
199 lockdep_is_held(&nf_ct_ecache_mutex));
200 BUG_ON(notify != new);
201 RCU_INIT_POINTER(net->ct.nf_conntrack_event_cb, NULL);
202 mutex_unlock(&nf_ct_ecache_mutex);
204 EXPORT_SYMBOL_GPL(nf_conntrack_unregister_notifier);
206 int nf_ct_expect_register_notifier(struct net *net,
207 struct nf_exp_event_notifier *new)
210 struct nf_exp_event_notifier *notify;
212 mutex_lock(&nf_ct_ecache_mutex);
213 notify = rcu_dereference_protected(net->ct.nf_expect_event_cb,
214 lockdep_is_held(&nf_ct_ecache_mutex));
215 if (notify != NULL) {
219 rcu_assign_pointer(net->ct.nf_expect_event_cb, new);
223 mutex_unlock(&nf_ct_ecache_mutex);
226 EXPORT_SYMBOL_GPL(nf_ct_expect_register_notifier);
228 void nf_ct_expect_unregister_notifier(struct net *net,
229 struct nf_exp_event_notifier *new)
231 struct nf_exp_event_notifier *notify;
233 mutex_lock(&nf_ct_ecache_mutex);
234 notify = rcu_dereference_protected(net->ct.nf_expect_event_cb,
235 lockdep_is_held(&nf_ct_ecache_mutex));
236 BUG_ON(notify != new);
237 RCU_INIT_POINTER(net->ct.nf_expect_event_cb, NULL);
238 mutex_unlock(&nf_ct_ecache_mutex);
240 EXPORT_SYMBOL_GPL(nf_ct_expect_unregister_notifier);
242 #define NF_CT_EVENTS_DEFAULT 1
243 static int nf_ct_events __read_mostly = NF_CT_EVENTS_DEFAULT;
246 static struct ctl_table event_sysctl_table[] = {
248 .procname = "nf_conntrack_events",
249 .data = &init_net.ct.sysctl_events,
250 .maxlen = sizeof(unsigned int),
252 .proc_handler = proc_dointvec,
256 #endif /* CONFIG_SYSCTL */
258 static struct nf_ct_ext_type event_extend __read_mostly = {
259 .len = sizeof(struct nf_conntrack_ecache),
260 .align = __alignof__(struct nf_conntrack_ecache),
261 .id = NF_CT_EXT_ECACHE,
265 static int nf_conntrack_event_init_sysctl(struct net *net)
267 struct ctl_table *table;
269 table = kmemdup(event_sysctl_table, sizeof(event_sysctl_table),
274 table[0].data = &net->ct.sysctl_events;
276 /* Don't export sysctls to unprivileged users */
277 if (net->user_ns != &init_user_ns)
278 table[0].procname = NULL;
280 net->ct.event_sysctl_header =
281 register_net_sysctl(net, "net/netfilter", table);
282 if (!net->ct.event_sysctl_header) {
283 printk(KERN_ERR "nf_ct_event: can't register to sysctl.\n");
294 static void nf_conntrack_event_fini_sysctl(struct net *net)
296 struct ctl_table *table;
298 table = net->ct.event_sysctl_header->ctl_table_arg;
299 unregister_net_sysctl_table(net->ct.event_sysctl_header);
303 static int nf_conntrack_event_init_sysctl(struct net *net)
308 static void nf_conntrack_event_fini_sysctl(struct net *net)
311 #endif /* CONFIG_SYSCTL */
313 int nf_conntrack_ecache_pernet_init(struct net *net)
315 net->ct.sysctl_events = nf_ct_events;
316 INIT_DELAYED_WORK(&net->ct.ecache_dwork, ecache_work);
317 return nf_conntrack_event_init_sysctl(net);
320 void nf_conntrack_ecache_pernet_fini(struct net *net)
322 cancel_delayed_work_sync(&net->ct.ecache_dwork);
323 nf_conntrack_event_fini_sysctl(net);
326 int nf_conntrack_ecache_init(void)
328 int ret = nf_ct_extend_register(&event_extend);
330 pr_err("nf_ct_event: Unable to register event extension.\n");
334 void nf_conntrack_ecache_fini(void)
336 nf_ct_extend_unregister(&event_extend);