2 * (C) 2012 by Pablo Neira Ayuso <pablo@netfilter.org>
3 * (C) 2012 by Vyatta Inc. <http://www.vyatta.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License version 2 as
7 * published by the Free Software Foundation (or any later at your option).
9 #include <linux/init.h>
10 #include <linux/module.h>
11 #include <linux/kernel.h>
12 #include <linux/rculist.h>
13 #include <linux/rculist_nulls.h>
14 #include <linux/types.h>
15 #include <linux/timer.h>
16 #include <linux/security.h>
17 #include <linux/skbuff.h>
18 #include <linux/errno.h>
19 #include <linux/netlink.h>
20 #include <linux/spinlock.h>
21 #include <linux/interrupt.h>
22 #include <linux/slab.h>
24 #include <linux/netfilter.h>
25 #include <net/netlink.h>
27 #include <net/netfilter/nf_conntrack.h>
28 #include <net/netfilter/nf_conntrack_core.h>
29 #include <net/netfilter/nf_conntrack_l3proto.h>
30 #include <net/netfilter/nf_conntrack_l4proto.h>
31 #include <net/netfilter/nf_conntrack_tuple.h>
32 #include <net/netfilter/nf_conntrack_timeout.h>
34 #include <linux/netfilter/nfnetlink.h>
35 #include <linux/netfilter/nfnetlink_cttimeout.h>
37 MODULE_LICENSE("GPL");
38 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
39 MODULE_DESCRIPTION("cttimeout: Extended Netfilter Connection Tracking timeout tuning");
41 static LIST_HEAD(cttimeout_list
);
43 static const struct nla_policy cttimeout_nla_policy
[CTA_TIMEOUT_MAX
+1] = {
44 [CTA_TIMEOUT_NAME
] = { .type
= NLA_NUL_STRING
,
45 .len
= CTNL_TIMEOUT_NAME_MAX
- 1},
46 [CTA_TIMEOUT_L3PROTO
] = { .type
= NLA_U16
},
47 [CTA_TIMEOUT_L4PROTO
] = { .type
= NLA_U8
},
48 [CTA_TIMEOUT_DATA
] = { .type
= NLA_NESTED
},
52 ctnl_timeout_parse_policy(void *timeouts
, struct nf_conntrack_l4proto
*l4proto
,
53 struct net
*net
, const struct nlattr
*attr
)
57 if (likely(l4proto
->ctnl_timeout
.nlattr_to_obj
)) {
58 struct nlattr
*tb
[l4proto
->ctnl_timeout
.nlattr_max
+1];
60 ret
= nla_parse_nested(tb
, l4proto
->ctnl_timeout
.nlattr_max
,
61 attr
, l4proto
->ctnl_timeout
.nla_policy
);
65 ret
= l4proto
->ctnl_timeout
.nlattr_to_obj(tb
, net
, timeouts
);
71 cttimeout_new_timeout(struct sock
*ctnl
, struct sk_buff
*skb
,
72 const struct nlmsghdr
*nlh
,
73 const struct nlattr
* const cda
[])
77 struct nf_conntrack_l4proto
*l4proto
;
78 struct ctnl_timeout
*timeout
, *matching
= NULL
;
79 struct net
*net
= sock_net(skb
->sk
);
83 if (!cda
[CTA_TIMEOUT_NAME
] ||
84 !cda
[CTA_TIMEOUT_L3PROTO
] ||
85 !cda
[CTA_TIMEOUT_L4PROTO
] ||
86 !cda
[CTA_TIMEOUT_DATA
])
89 name
= nla_data(cda
[CTA_TIMEOUT_NAME
]);
90 l3num
= ntohs(nla_get_be16(cda
[CTA_TIMEOUT_L3PROTO
]));
91 l4num
= nla_get_u8(cda
[CTA_TIMEOUT_L4PROTO
]);
93 list_for_each_entry(timeout
, &cttimeout_list
, head
) {
94 if (strncmp(timeout
->name
, name
, CTNL_TIMEOUT_NAME_MAX
) != 0)
97 if (nlh
->nlmsg_flags
& NLM_F_EXCL
)
104 l4proto
= nf_ct_l4proto_find_get(l3num
, l4num
);
106 /* This protocol is not supportted, skip. */
107 if (l4proto
->l4proto
!= l4num
) {
113 if (nlh
->nlmsg_flags
& NLM_F_REPLACE
) {
114 /* You cannot replace one timeout policy by another of
115 * different kind, sorry.
117 if (matching
->l3num
!= l3num
||
118 matching
->l4proto
->l4proto
!= l4num
) {
123 ret
= ctnl_timeout_parse_policy(&matching
->data
,
125 cda
[CTA_TIMEOUT_DATA
]);
132 timeout
= kzalloc(sizeof(struct ctnl_timeout
) +
133 l4proto
->ctnl_timeout
.obj_size
, GFP_KERNEL
);
134 if (timeout
== NULL
) {
139 ret
= ctnl_timeout_parse_policy(&timeout
->data
, l4proto
, net
,
140 cda
[CTA_TIMEOUT_DATA
]);
144 strcpy(timeout
->name
, nla_data(cda
[CTA_TIMEOUT_NAME
]));
145 timeout
->l3num
= l3num
;
146 timeout
->l4proto
= l4proto
;
147 atomic_set(&timeout
->refcnt
, 1);
148 list_add_tail_rcu(&timeout
->head
, &cttimeout_list
);
154 nf_ct_l4proto_put(l4proto
);
159 ctnl_timeout_fill_info(struct sk_buff
*skb
, u32 portid
, u32 seq
, u32 type
,
160 int event
, struct ctnl_timeout
*timeout
)
162 struct nlmsghdr
*nlh
;
163 struct nfgenmsg
*nfmsg
;
164 unsigned int flags
= portid
? NLM_F_MULTI
: 0;
165 struct nf_conntrack_l4proto
*l4proto
= timeout
->l4proto
;
167 event
|= NFNL_SUBSYS_CTNETLINK_TIMEOUT
<< 8;
168 nlh
= nlmsg_put(skb
, portid
, seq
, event
, sizeof(*nfmsg
), flags
);
172 nfmsg
= nlmsg_data(nlh
);
173 nfmsg
->nfgen_family
= AF_UNSPEC
;
174 nfmsg
->version
= NFNETLINK_V0
;
177 if (nla_put_string(skb
, CTA_TIMEOUT_NAME
, timeout
->name
) ||
178 nla_put_be16(skb
, CTA_TIMEOUT_L3PROTO
, htons(timeout
->l3num
)) ||
179 nla_put_u8(skb
, CTA_TIMEOUT_L4PROTO
, timeout
->l4proto
->l4proto
) ||
180 nla_put_be32(skb
, CTA_TIMEOUT_USE
,
181 htonl(atomic_read(&timeout
->refcnt
))))
182 goto nla_put_failure
;
184 if (likely(l4proto
->ctnl_timeout
.obj_to_nlattr
)) {
185 struct nlattr
*nest_parms
;
188 nest_parms
= nla_nest_start(skb
,
189 CTA_TIMEOUT_DATA
| NLA_F_NESTED
);
191 goto nla_put_failure
;
193 ret
= l4proto
->ctnl_timeout
.obj_to_nlattr(skb
, &timeout
->data
);
195 goto nla_put_failure
;
197 nla_nest_end(skb
, nest_parms
);
205 nlmsg_cancel(skb
, nlh
);
210 ctnl_timeout_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
212 struct ctnl_timeout
*cur
, *last
;
217 last
= (struct ctnl_timeout
*)cb
->args
[1];
222 list_for_each_entry_rcu(cur
, &cttimeout_list
, head
) {
229 if (ctnl_timeout_fill_info(skb
, NETLINK_CB(cb
->skb
).portid
,
231 NFNL_MSG_TYPE(cb
->nlh
->nlmsg_type
),
232 IPCTNL_MSG_TIMEOUT_NEW
, cur
) < 0) {
233 cb
->args
[1] = (unsigned long)cur
;
244 cttimeout_get_timeout(struct sock
*ctnl
, struct sk_buff
*skb
,
245 const struct nlmsghdr
*nlh
,
246 const struct nlattr
* const cda
[])
250 struct ctnl_timeout
*cur
;
252 if (nlh
->nlmsg_flags
& NLM_F_DUMP
) {
253 struct netlink_dump_control c
= {
254 .dump
= ctnl_timeout_dump
,
256 return netlink_dump_start(ctnl
, skb
, nlh
, &c
);
259 if (!cda
[CTA_TIMEOUT_NAME
])
261 name
= nla_data(cda
[CTA_TIMEOUT_NAME
]);
263 list_for_each_entry(cur
, &cttimeout_list
, head
) {
264 struct sk_buff
*skb2
;
266 if (strncmp(cur
->name
, name
, CTNL_TIMEOUT_NAME_MAX
) != 0)
269 skb2
= nlmsg_new(NLMSG_DEFAULT_SIZE
, GFP_KERNEL
);
275 ret
= ctnl_timeout_fill_info(skb2
, NETLINK_CB(skb
).portid
,
277 NFNL_MSG_TYPE(nlh
->nlmsg_type
),
278 IPCTNL_MSG_TIMEOUT_NEW
, cur
);
283 ret
= netlink_unicast(ctnl
, skb2
, NETLINK_CB(skb
).portid
,
288 /* this avoids a loop in nfnetlink. */
289 return ret
== -EAGAIN
? -ENOBUFS
: ret
;
294 static void untimeout(struct nf_conntrack_tuple_hash
*i
,
295 struct ctnl_timeout
*timeout
)
297 struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(i
);
298 struct nf_conn_timeout
*timeout_ext
= nf_ct_timeout_find(ct
);
300 if (timeout_ext
&& (!timeout
|| timeout_ext
->timeout
== timeout
))
301 RCU_INIT_POINTER(timeout_ext
->timeout
, NULL
);
304 static void ctnl_untimeout(struct ctnl_timeout
*timeout
)
306 struct nf_conntrack_tuple_hash
*h
;
307 const struct hlist_nulls_node
*nn
;
311 for (i
= 0; i
< init_net
.ct
.htable_size
; i
++) {
312 spin_lock(&nf_conntrack_locks
[i
% CONNTRACK_LOCKS
]);
313 if (i
< init_net
.ct
.htable_size
) {
314 hlist_nulls_for_each_entry(h
, nn
, &init_net
.ct
.hash
[i
], hnnode
)
315 untimeout(h
, timeout
);
317 spin_unlock(&nf_conntrack_locks
[i
% CONNTRACK_LOCKS
]);
322 /* try to delete object, fail if it is still in use. */
323 static int ctnl_timeout_try_del(struct ctnl_timeout
*timeout
)
327 /* we want to avoid races with nf_ct_timeout_find_get. */
328 if (atomic_dec_and_test(&timeout
->refcnt
)) {
329 /* We are protected by nfnl mutex. */
330 list_del_rcu(&timeout
->head
);
331 nf_ct_l4proto_put(timeout
->l4proto
);
332 ctnl_untimeout(timeout
);
333 kfree_rcu(timeout
, rcu_head
);
335 /* still in use, restore reference counter. */
336 atomic_inc(&timeout
->refcnt
);
343 cttimeout_del_timeout(struct sock
*ctnl
, struct sk_buff
*skb
,
344 const struct nlmsghdr
*nlh
,
345 const struct nlattr
* const cda
[])
348 struct ctnl_timeout
*cur
;
351 if (!cda
[CTA_TIMEOUT_NAME
]) {
352 list_for_each_entry(cur
, &cttimeout_list
, head
)
353 ctnl_timeout_try_del(cur
);
357 name
= nla_data(cda
[CTA_TIMEOUT_NAME
]);
359 list_for_each_entry(cur
, &cttimeout_list
, head
) {
360 if (strncmp(cur
->name
, name
, CTNL_TIMEOUT_NAME_MAX
) != 0)
363 ret
= ctnl_timeout_try_del(cur
);
373 cttimeout_default_set(struct sock
*ctnl
, struct sk_buff
*skb
,
374 const struct nlmsghdr
*nlh
,
375 const struct nlattr
* const cda
[])
379 struct nf_conntrack_l4proto
*l4proto
;
380 struct net
*net
= sock_net(skb
->sk
);
381 unsigned int *timeouts
;
384 if (!cda
[CTA_TIMEOUT_L3PROTO
] ||
385 !cda
[CTA_TIMEOUT_L4PROTO
] ||
386 !cda
[CTA_TIMEOUT_DATA
])
389 l3num
= ntohs(nla_get_be16(cda
[CTA_TIMEOUT_L3PROTO
]));
390 l4num
= nla_get_u8(cda
[CTA_TIMEOUT_L4PROTO
]);
391 l4proto
= nf_ct_l4proto_find_get(l3num
, l4num
);
393 /* This protocol is not supported, skip. */
394 if (l4proto
->l4proto
!= l4num
) {
399 timeouts
= l4proto
->get_timeouts(net
);
401 ret
= ctnl_timeout_parse_policy(timeouts
, l4proto
, net
,
402 cda
[CTA_TIMEOUT_DATA
]);
406 nf_ct_l4proto_put(l4proto
);
409 nf_ct_l4proto_put(l4proto
);
414 cttimeout_default_fill_info(struct net
*net
, struct sk_buff
*skb
, u32 portid
,
415 u32 seq
, u32 type
, int event
,
416 struct nf_conntrack_l4proto
*l4proto
)
418 struct nlmsghdr
*nlh
;
419 struct nfgenmsg
*nfmsg
;
420 unsigned int flags
= portid
? NLM_F_MULTI
: 0;
422 event
|= NFNL_SUBSYS_CTNETLINK_TIMEOUT
<< 8;
423 nlh
= nlmsg_put(skb
, portid
, seq
, event
, sizeof(*nfmsg
), flags
);
427 nfmsg
= nlmsg_data(nlh
);
428 nfmsg
->nfgen_family
= AF_UNSPEC
;
429 nfmsg
->version
= NFNETLINK_V0
;
432 if (nla_put_be16(skb
, CTA_TIMEOUT_L3PROTO
, htons(l4proto
->l3proto
)) ||
433 nla_put_u8(skb
, CTA_TIMEOUT_L4PROTO
, l4proto
->l4proto
))
434 goto nla_put_failure
;
436 if (likely(l4proto
->ctnl_timeout
.obj_to_nlattr
)) {
437 struct nlattr
*nest_parms
;
438 unsigned int *timeouts
= l4proto
->get_timeouts(net
);
441 nest_parms
= nla_nest_start(skb
,
442 CTA_TIMEOUT_DATA
| NLA_F_NESTED
);
444 goto nla_put_failure
;
446 ret
= l4proto
->ctnl_timeout
.obj_to_nlattr(skb
, timeouts
);
448 goto nla_put_failure
;
450 nla_nest_end(skb
, nest_parms
);
458 nlmsg_cancel(skb
, nlh
);
462 static int cttimeout_default_get(struct sock
*ctnl
, struct sk_buff
*skb
,
463 const struct nlmsghdr
*nlh
,
464 const struct nlattr
* const cda
[])
468 struct nf_conntrack_l4proto
*l4proto
;
469 struct net
*net
= sock_net(skb
->sk
);
470 struct sk_buff
*skb2
;
473 if (!cda
[CTA_TIMEOUT_L3PROTO
] || !cda
[CTA_TIMEOUT_L4PROTO
])
476 l3num
= ntohs(nla_get_be16(cda
[CTA_TIMEOUT_L3PROTO
]));
477 l4num
= nla_get_u8(cda
[CTA_TIMEOUT_L4PROTO
]);
478 l4proto
= nf_ct_l4proto_find_get(l3num
, l4num
);
480 /* This protocol is not supported, skip. */
481 if (l4proto
->l4proto
!= l4num
) {
486 skb2
= nlmsg_new(NLMSG_DEFAULT_SIZE
, GFP_KERNEL
);
492 ret
= cttimeout_default_fill_info(net
, skb2
, NETLINK_CB(skb
).portid
,
494 NFNL_MSG_TYPE(nlh
->nlmsg_type
),
495 IPCTNL_MSG_TIMEOUT_DEFAULT_SET
,
502 ret
= netlink_unicast(ctnl
, skb2
, NETLINK_CB(skb
).portid
, MSG_DONTWAIT
);
506 /* this avoids a loop in nfnetlink. */
507 return ret
== -EAGAIN
? -ENOBUFS
: ret
;
509 nf_ct_l4proto_put(l4proto
);
513 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
514 static struct ctnl_timeout
*ctnl_timeout_find_get(const char *name
)
516 struct ctnl_timeout
*timeout
, *matching
= NULL
;
519 list_for_each_entry_rcu(timeout
, &cttimeout_list
, head
) {
520 if (strncmp(timeout
->name
, name
, CTNL_TIMEOUT_NAME_MAX
) != 0)
523 if (!try_module_get(THIS_MODULE
))
526 if (!atomic_inc_not_zero(&timeout
->refcnt
)) {
527 module_put(THIS_MODULE
);
538 static void ctnl_timeout_put(struct ctnl_timeout
*timeout
)
540 atomic_dec(&timeout
->refcnt
);
541 module_put(THIS_MODULE
);
543 #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
545 static const struct nfnl_callback cttimeout_cb
[IPCTNL_MSG_TIMEOUT_MAX
] = {
546 [IPCTNL_MSG_TIMEOUT_NEW
] = { .call
= cttimeout_new_timeout
,
547 .attr_count
= CTA_TIMEOUT_MAX
,
548 .policy
= cttimeout_nla_policy
},
549 [IPCTNL_MSG_TIMEOUT_GET
] = { .call
= cttimeout_get_timeout
,
550 .attr_count
= CTA_TIMEOUT_MAX
,
551 .policy
= cttimeout_nla_policy
},
552 [IPCTNL_MSG_TIMEOUT_DELETE
] = { .call
= cttimeout_del_timeout
,
553 .attr_count
= CTA_TIMEOUT_MAX
,
554 .policy
= cttimeout_nla_policy
},
555 [IPCTNL_MSG_TIMEOUT_DEFAULT_SET
]= { .call
= cttimeout_default_set
,
556 .attr_count
= CTA_TIMEOUT_MAX
,
557 .policy
= cttimeout_nla_policy
},
558 [IPCTNL_MSG_TIMEOUT_DEFAULT_GET
]= { .call
= cttimeout_default_get
,
559 .attr_count
= CTA_TIMEOUT_MAX
,
560 .policy
= cttimeout_nla_policy
},
563 static const struct nfnetlink_subsystem cttimeout_subsys
= {
564 .name
= "conntrack_timeout",
565 .subsys_id
= NFNL_SUBSYS_CTNETLINK_TIMEOUT
,
566 .cb_count
= IPCTNL_MSG_TIMEOUT_MAX
,
570 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_CTNETLINK_TIMEOUT
);
572 static int __init
cttimeout_init(void)
576 ret
= nfnetlink_subsys_register(&cttimeout_subsys
);
578 pr_err("cttimeout_init: cannot register cttimeout with "
582 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
583 RCU_INIT_POINTER(nf_ct_timeout_find_get_hook
, ctnl_timeout_find_get
);
584 RCU_INIT_POINTER(nf_ct_timeout_put_hook
, ctnl_timeout_put
);
585 #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
592 static void __exit
cttimeout_exit(void)
594 struct ctnl_timeout
*cur
, *tmp
;
596 pr_info("cttimeout: unregistering from nfnetlink.\n");
598 nfnetlink_subsys_unregister(&cttimeout_subsys
);
600 /* Make sure no conntrack objects refer to custom timeouts anymore. */
601 ctnl_untimeout(NULL
);
603 list_for_each_entry_safe(cur
, tmp
, &cttimeout_list
, head
) {
604 list_del_rcu(&cur
->head
);
605 /* We are sure that our objects have no clients at this point,
606 * it's safe to release them all without checking refcnt.
608 nf_ct_l4proto_put(cur
->l4proto
);
609 kfree_rcu(cur
, rcu_head
);
611 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
612 RCU_INIT_POINTER(nf_ct_timeout_find_get_hook
, NULL
);
613 RCU_INIT_POINTER(nf_ct_timeout_put_hook
, NULL
);
614 #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
618 module_init(cttimeout_init
);
619 module_exit(cttimeout_exit
);