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);
52 struct nf_conntrack_ecache *e;
54 if (!nf_ct_is_confirmed(ct))
57 e = nf_ct_ecache_find(ct);
58 if (!e || e->state != NFCT_ECACHE_DESTROY_FAIL)
61 if (nf_conntrack_event(IPCT_DESTROY, ct)) {
62 ret = STATE_CONGESTED;
66 e->state = NFCT_ECACHE_DESTROY_SENT;
69 if (++evicted >= ARRAY_SIZE(refs)) {
75 spin_unlock(&pcpu->lock);
77 /* can't _put while holding lock */
79 nf_ct_put(refs[--evicted]);
84 static void ecache_work(struct work_struct *work)
86 struct netns_ct *ctnet =
87 container_of(work, struct netns_ct, ecache_dwork.work);
93 for_each_possible_cpu(cpu) {
96 pcpu = per_cpu_ptr(ctnet->pcpu_lists, cpu);
98 ret = ecache_work_evict_list(pcpu);
101 case STATE_CONGESTED:
102 delay = ECACHE_RETRY_WAIT;
115 ctnet->ecache_dwork_pending = delay > 0;
117 schedule_delayed_work(&ctnet->ecache_dwork, delay);
120 int nf_conntrack_eventmask_report(unsigned int eventmask, struct nf_conn *ct,
121 u32 portid, int report)
124 struct net *net = nf_ct_net(ct);
125 struct nf_ct_event_notifier *notify;
126 struct nf_conntrack_ecache *e;
129 notify = rcu_dereference(net->ct.nf_conntrack_event_cb);
133 e = nf_ct_ecache_find(ct);
137 if (nf_ct_is_confirmed(ct)) {
138 struct nf_ct_event item = {
140 .portid = e->portid ? e->portid : portid,
143 /* This is a resent of a destroy event? If so, skip missed */
144 unsigned long missed = e->portid ? 0 : e->missed;
146 if (!((eventmask | missed) & e->ctmask))
149 ret = notify->fcn(eventmask | missed, &item);
150 if (unlikely(ret < 0 || missed)) {
151 spin_lock_bh(&ct->lock);
153 /* This is a destroy event that has been
154 * triggered by a process, we store the PORTID
155 * to include it in the retransmission.
157 if (eventmask & (1 << IPCT_DESTROY)) {
158 if (e->portid == 0 && portid != 0)
160 e->state = NFCT_ECACHE_DESTROY_FAIL;
162 e->missed |= eventmask;
165 e->missed &= ~missed;
167 spin_unlock_bh(&ct->lock);
174 EXPORT_SYMBOL_GPL(nf_conntrack_eventmask_report);
176 /* deliver cached events and clear cache entry - must be called with locally
177 * disabled softirqs */
178 void nf_ct_deliver_cached_events(struct nf_conn *ct)
180 struct net *net = nf_ct_net(ct);
181 unsigned long events, missed;
182 struct nf_ct_event_notifier *notify;
183 struct nf_conntrack_ecache *e;
184 struct nf_ct_event item;
188 notify = rcu_dereference(net->ct.nf_conntrack_event_cb);
192 e = nf_ct_ecache_find(ct);
196 events = xchg(&e->cache, 0);
198 if (!nf_ct_is_confirmed(ct) || nf_ct_is_dying(ct) || !events)
201 /* We make a copy of the missed event cache without taking
202 * the lock, thus we may send missed events twice. However,
203 * this does not harm and it happens very rarely. */
206 if (!((events | missed) & e->ctmask))
213 ret = notify->fcn(events | missed, &item);
215 if (likely(ret >= 0 && !missed))
218 spin_lock_bh(&ct->lock);
222 e->missed &= ~missed;
223 spin_unlock_bh(&ct->lock);
228 EXPORT_SYMBOL_GPL(nf_ct_deliver_cached_events);
230 void nf_ct_expect_event_report(enum ip_conntrack_expect_events event,
231 struct nf_conntrack_expect *exp,
232 u32 portid, int report)
235 struct net *net = nf_ct_exp_net(exp);
236 struct nf_exp_event_notifier *notify;
237 struct nf_conntrack_ecache *e;
240 notify = rcu_dereference(net->ct.nf_expect_event_cb);
244 e = nf_ct_ecache_find(exp->master);
248 if (e->expmask & (1 << event)) {
249 struct nf_exp_event item = {
254 notify->fcn(1 << event, &item);
260 int nf_conntrack_register_notifier(struct net *net,
261 struct nf_ct_event_notifier *new)
264 struct nf_ct_event_notifier *notify;
266 mutex_lock(&nf_ct_ecache_mutex);
267 notify = rcu_dereference_protected(net->ct.nf_conntrack_event_cb,
268 lockdep_is_held(&nf_ct_ecache_mutex));
269 if (notify != NULL) {
273 rcu_assign_pointer(net->ct.nf_conntrack_event_cb, new);
277 mutex_unlock(&nf_ct_ecache_mutex);
280 EXPORT_SYMBOL_GPL(nf_conntrack_register_notifier);
282 void nf_conntrack_unregister_notifier(struct net *net,
283 struct nf_ct_event_notifier *new)
285 struct nf_ct_event_notifier *notify;
287 mutex_lock(&nf_ct_ecache_mutex);
288 notify = rcu_dereference_protected(net->ct.nf_conntrack_event_cb,
289 lockdep_is_held(&nf_ct_ecache_mutex));
290 BUG_ON(notify != new);
291 RCU_INIT_POINTER(net->ct.nf_conntrack_event_cb, NULL);
292 mutex_unlock(&nf_ct_ecache_mutex);
293 /* synchronize_rcu() is called from ctnetlink_exit. */
295 EXPORT_SYMBOL_GPL(nf_conntrack_unregister_notifier);
297 int nf_ct_expect_register_notifier(struct net *net,
298 struct nf_exp_event_notifier *new)
301 struct nf_exp_event_notifier *notify;
303 mutex_lock(&nf_ct_ecache_mutex);
304 notify = rcu_dereference_protected(net->ct.nf_expect_event_cb,
305 lockdep_is_held(&nf_ct_ecache_mutex));
306 if (notify != NULL) {
310 rcu_assign_pointer(net->ct.nf_expect_event_cb, new);
314 mutex_unlock(&nf_ct_ecache_mutex);
317 EXPORT_SYMBOL_GPL(nf_ct_expect_register_notifier);
319 void nf_ct_expect_unregister_notifier(struct net *net,
320 struct nf_exp_event_notifier *new)
322 struct nf_exp_event_notifier *notify;
324 mutex_lock(&nf_ct_ecache_mutex);
325 notify = rcu_dereference_protected(net->ct.nf_expect_event_cb,
326 lockdep_is_held(&nf_ct_ecache_mutex));
327 BUG_ON(notify != new);
328 RCU_INIT_POINTER(net->ct.nf_expect_event_cb, NULL);
329 mutex_unlock(&nf_ct_ecache_mutex);
330 /* synchronize_rcu() is called from ctnetlink_exit. */
332 EXPORT_SYMBOL_GPL(nf_ct_expect_unregister_notifier);
334 #define NF_CT_EVENTS_DEFAULT 1
335 static int nf_ct_events __read_mostly = NF_CT_EVENTS_DEFAULT;
338 static struct ctl_table event_sysctl_table[] = {
340 .procname = "nf_conntrack_events",
341 .data = &init_net.ct.sysctl_events,
342 .maxlen = sizeof(unsigned int),
344 .proc_handler = proc_dointvec,
348 #endif /* CONFIG_SYSCTL */
350 static struct nf_ct_ext_type event_extend __read_mostly = {
351 .len = sizeof(struct nf_conntrack_ecache),
352 .align = __alignof__(struct nf_conntrack_ecache),
353 .id = NF_CT_EXT_ECACHE,
357 static int nf_conntrack_event_init_sysctl(struct net *net)
359 struct ctl_table *table;
361 table = kmemdup(event_sysctl_table, sizeof(event_sysctl_table),
366 table[0].data = &net->ct.sysctl_events;
368 /* Don't export sysctls to unprivileged users */
369 if (net->user_ns != &init_user_ns)
370 table[0].procname = NULL;
372 net->ct.event_sysctl_header =
373 register_net_sysctl(net, "net/netfilter", table);
374 if (!net->ct.event_sysctl_header) {
375 printk(KERN_ERR "nf_ct_event: can't register to sysctl.\n");
386 static void nf_conntrack_event_fini_sysctl(struct net *net)
388 struct ctl_table *table;
390 table = net->ct.event_sysctl_header->ctl_table_arg;
391 unregister_net_sysctl_table(net->ct.event_sysctl_header);
395 static int nf_conntrack_event_init_sysctl(struct net *net)
400 static void nf_conntrack_event_fini_sysctl(struct net *net)
403 #endif /* CONFIG_SYSCTL */
405 int nf_conntrack_ecache_pernet_init(struct net *net)
407 net->ct.sysctl_events = nf_ct_events;
408 INIT_DELAYED_WORK(&net->ct.ecache_dwork, ecache_work);
409 return nf_conntrack_event_init_sysctl(net);
412 void nf_conntrack_ecache_pernet_fini(struct net *net)
414 cancel_delayed_work_sync(&net->ct.ecache_dwork);
415 nf_conntrack_event_fini_sysctl(net);
418 int nf_conntrack_ecache_init(void)
420 int ret = nf_ct_extend_register(&event_extend);
422 pr_err("nf_ct_event: Unable to register event extension.\n");
426 void nf_conntrack_ecache_fini(void)
428 nf_ct_extend_unregister(&event_extend);