1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Event cache for netfilter. */
5 * (C) 2005 Harald Welte <laforge@gnumonks.org>
6 * (C) 2005 Patrick McHardy <kaber@trash.net>
7 * (C) 2005-2006 Netfilter Core Team <coreteam@netfilter.org>
8 * (C) 2005 USAGI/WIDE Project <http://www.linux-ipv6.org>
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/types.h>
14 #include <linux/netfilter.h>
15 #include <linux/skbuff.h>
16 #include <linux/vmalloc.h>
17 #include <linux/stddef.h>
18 #include <linux/err.h>
19 #include <linux/percpu.h>
20 #include <linux/kernel.h>
21 #include <linux/netdevice.h>
22 #include <linux/slab.h>
23 #include <linux/export.h>
25 #include <net/netfilter/nf_conntrack.h>
26 #include <net/netfilter/nf_conntrack_core.h>
27 #include <net/netfilter/nf_conntrack_extend.h>
29 static DEFINE_MUTEX(nf_ct_ecache_mutex
);
31 #define ECACHE_RETRY_WAIT (HZ/10)
39 static enum retry_state
ecache_work_evict_list(struct ct_pcpu
*pcpu
)
41 struct nf_conn
*refs
[16];
42 struct nf_conntrack_tuple_hash
*h
;
43 struct hlist_nulls_node
*n
;
44 unsigned int evicted
= 0;
45 enum retry_state ret
= STATE_DONE
;
47 spin_lock(&pcpu
->lock
);
49 hlist_nulls_for_each_entry(h
, n
, &pcpu
->dying
, hnnode
) {
50 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
51 struct nf_conntrack_ecache
*e
;
53 if (!nf_ct_is_confirmed(ct
))
56 e
= nf_ct_ecache_find(ct
);
57 if (!e
|| e
->state
!= NFCT_ECACHE_DESTROY_FAIL
)
60 if (nf_conntrack_event(IPCT_DESTROY
, ct
)) {
61 ret
= STATE_CONGESTED
;
65 e
->state
= NFCT_ECACHE_DESTROY_SENT
;
68 if (++evicted
>= ARRAY_SIZE(refs
)) {
74 spin_unlock(&pcpu
->lock
);
76 /* can't _put while holding lock */
78 nf_ct_put(refs
[--evicted
]);
83 static void ecache_work(struct work_struct
*work
)
85 struct netns_ct
*ctnet
=
86 container_of(work
, struct netns_ct
, ecache_dwork
.work
);
92 for_each_possible_cpu(cpu
) {
95 pcpu
= per_cpu_ptr(ctnet
->pcpu_lists
, cpu
);
97 ret
= ecache_work_evict_list(pcpu
);
100 case STATE_CONGESTED
:
101 delay
= ECACHE_RETRY_WAIT
;
114 ctnet
->ecache_dwork_pending
= delay
> 0;
116 schedule_delayed_work(&ctnet
->ecache_dwork
, delay
);
119 int nf_conntrack_eventmask_report(unsigned int eventmask
, struct nf_conn
*ct
,
120 u32 portid
, int report
)
123 struct net
*net
= nf_ct_net(ct
);
124 struct nf_ct_event_notifier
*notify
;
125 struct nf_conntrack_ecache
*e
;
128 notify
= rcu_dereference(net
->ct
.nf_conntrack_event_cb
);
132 e
= nf_ct_ecache_find(ct
);
136 if (nf_ct_is_confirmed(ct
)) {
137 struct nf_ct_event item
= {
139 .portid
= e
->portid
? e
->portid
: portid
,
142 /* This is a resent of a destroy event? If so, skip missed */
143 unsigned long missed
= e
->portid
? 0 : e
->missed
;
145 if (!((eventmask
| missed
) & e
->ctmask
))
148 ret
= notify
->fcn(eventmask
| missed
, &item
);
149 if (unlikely(ret
< 0 || missed
)) {
150 spin_lock_bh(&ct
->lock
);
152 /* This is a destroy event that has been
153 * triggered by a process, we store the PORTID
154 * to include it in the retransmission.
156 if (eventmask
& (1 << IPCT_DESTROY
)) {
157 if (e
->portid
== 0 && portid
!= 0)
159 e
->state
= NFCT_ECACHE_DESTROY_FAIL
;
161 e
->missed
|= eventmask
;
164 e
->missed
&= ~missed
;
166 spin_unlock_bh(&ct
->lock
);
173 EXPORT_SYMBOL_GPL(nf_conntrack_eventmask_report
);
175 /* deliver cached events and clear cache entry - must be called with locally
176 * disabled softirqs */
177 void nf_ct_deliver_cached_events(struct nf_conn
*ct
)
179 struct net
*net
= nf_ct_net(ct
);
180 unsigned long events
, missed
;
181 struct nf_ct_event_notifier
*notify
;
182 struct nf_conntrack_ecache
*e
;
183 struct nf_ct_event item
;
187 notify
= rcu_dereference(net
->ct
.nf_conntrack_event_cb
);
191 e
= nf_ct_ecache_find(ct
);
195 events
= xchg(&e
->cache
, 0);
197 if (!nf_ct_is_confirmed(ct
) || nf_ct_is_dying(ct
))
200 /* We make a copy of the missed event cache without taking
201 * the lock, thus we may send missed events twice. However,
202 * this does not harm and it happens very rarely. */
205 if (!((events
| missed
) & e
->ctmask
))
212 ret
= notify
->fcn(events
| missed
, &item
);
214 if (likely(ret
== 0 && !missed
))
217 spin_lock_bh(&ct
->lock
);
221 e
->missed
&= ~missed
;
222 spin_unlock_bh(&ct
->lock
);
227 EXPORT_SYMBOL_GPL(nf_ct_deliver_cached_events
);
229 void nf_ct_expect_event_report(enum ip_conntrack_expect_events event
,
230 struct nf_conntrack_expect
*exp
,
231 u32 portid
, int report
)
234 struct net
*net
= nf_ct_exp_net(exp
);
235 struct nf_exp_event_notifier
*notify
;
236 struct nf_conntrack_ecache
*e
;
239 notify
= rcu_dereference(net
->ct
.nf_expect_event_cb
);
243 e
= nf_ct_ecache_find(exp
->master
);
247 if (e
->expmask
& (1 << event
)) {
248 struct nf_exp_event item
= {
253 notify
->fcn(1 << event
, &item
);
259 int nf_conntrack_register_notifier(struct net
*net
,
260 struct nf_ct_event_notifier
*new)
263 struct nf_ct_event_notifier
*notify
;
265 mutex_lock(&nf_ct_ecache_mutex
);
266 notify
= rcu_dereference_protected(net
->ct
.nf_conntrack_event_cb
,
267 lockdep_is_held(&nf_ct_ecache_mutex
));
268 if (notify
!= NULL
) {
272 rcu_assign_pointer(net
->ct
.nf_conntrack_event_cb
, new);
276 mutex_unlock(&nf_ct_ecache_mutex
);
279 EXPORT_SYMBOL_GPL(nf_conntrack_register_notifier
);
281 void nf_conntrack_unregister_notifier(struct net
*net
,
282 struct nf_ct_event_notifier
*new)
284 struct nf_ct_event_notifier
*notify
;
286 mutex_lock(&nf_ct_ecache_mutex
);
287 notify
= rcu_dereference_protected(net
->ct
.nf_conntrack_event_cb
,
288 lockdep_is_held(&nf_ct_ecache_mutex
));
289 BUG_ON(notify
!= new);
290 RCU_INIT_POINTER(net
->ct
.nf_conntrack_event_cb
, NULL
);
291 mutex_unlock(&nf_ct_ecache_mutex
);
292 /* synchronize_rcu() is called from ctnetlink_exit. */
294 EXPORT_SYMBOL_GPL(nf_conntrack_unregister_notifier
);
296 int nf_ct_expect_register_notifier(struct net
*net
,
297 struct nf_exp_event_notifier
*new)
300 struct nf_exp_event_notifier
*notify
;
302 mutex_lock(&nf_ct_ecache_mutex
);
303 notify
= rcu_dereference_protected(net
->ct
.nf_expect_event_cb
,
304 lockdep_is_held(&nf_ct_ecache_mutex
));
305 if (notify
!= NULL
) {
309 rcu_assign_pointer(net
->ct
.nf_expect_event_cb
, new);
313 mutex_unlock(&nf_ct_ecache_mutex
);
316 EXPORT_SYMBOL_GPL(nf_ct_expect_register_notifier
);
318 void nf_ct_expect_unregister_notifier(struct net
*net
,
319 struct nf_exp_event_notifier
*new)
321 struct nf_exp_event_notifier
*notify
;
323 mutex_lock(&nf_ct_ecache_mutex
);
324 notify
= rcu_dereference_protected(net
->ct
.nf_expect_event_cb
,
325 lockdep_is_held(&nf_ct_ecache_mutex
));
326 BUG_ON(notify
!= new);
327 RCU_INIT_POINTER(net
->ct
.nf_expect_event_cb
, NULL
);
328 mutex_unlock(&nf_ct_ecache_mutex
);
329 /* synchronize_rcu() is called from ctnetlink_exit. */
331 EXPORT_SYMBOL_GPL(nf_ct_expect_unregister_notifier
);
333 #define NF_CT_EVENTS_DEFAULT 1
334 static int nf_ct_events __read_mostly
= NF_CT_EVENTS_DEFAULT
;
336 static const struct nf_ct_ext_type event_extend
= {
337 .len
= sizeof(struct nf_conntrack_ecache
),
338 .align
= __alignof__(struct nf_conntrack_ecache
),
339 .id
= NF_CT_EXT_ECACHE
,
342 void nf_conntrack_ecache_pernet_init(struct net
*net
)
344 net
->ct
.sysctl_events
= nf_ct_events
;
345 INIT_DELAYED_WORK(&net
->ct
.ecache_dwork
, ecache_work
);
348 void nf_conntrack_ecache_pernet_fini(struct net
*net
)
350 cancel_delayed_work_sync(&net
->ct
.ecache_dwork
);
353 int nf_conntrack_ecache_init(void)
355 int ret
= nf_ct_extend_register(&event_extend
);
357 pr_err("Unable to register event extension\n");
359 BUILD_BUG_ON(__IPCT_MAX
>= 16); /* ctmask, missed use u16 */
364 void nf_conntrack_ecache_fini(void)
366 nf_ct_extend_unregister(&event_extend
);