netfilter: add cttimeout infrastructure for fine timeout tuning
[deliverable/linux.git] / net / netfilter / nfnetlink_cttimeout.c
CommitLineData
50978462
PNA
1/*
2 * (C) 2012 by Pablo Neira Ayuso <pablo@netfilter.org>
3 * (C) 2012 by Vyatta Inc. <http://www.vyatta.com>
4 *
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).
8 */
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>
23
24#include <linux/netfilter.h>
25#include <net/netlink.h>
26#include <net/sock.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
33#include <linux/netfilter/nfnetlink.h>
34#include <linux/netfilter/nfnetlink_cttimeout.h>
35
36MODULE_LICENSE("GPL");
37MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
38MODULE_DESCRIPTION("cttimeout: Extended Netfilter Connection Tracking timeout tuning");
39
40struct ctnl_timeout {
41 struct list_head head;
42 struct rcu_head rcu_head;
43 atomic_t refcnt;
44 char name[CTNL_TIMEOUT_NAME_MAX];
45 __u16 l3num;
46 __u8 l4num;
47 char data[0];
48};
49
50static LIST_HEAD(cttimeout_list);
51
52static const struct nla_policy cttimeout_nla_policy[CTA_TIMEOUT_MAX+1] = {
53 [CTA_TIMEOUT_NAME] = { .type = NLA_NUL_STRING },
54 [CTA_TIMEOUT_L3PROTO] = { .type = NLA_U16 },
55 [CTA_TIMEOUT_L4PROTO] = { .type = NLA_U8 },
56 [CTA_TIMEOUT_DATA] = { .type = NLA_NESTED },
57};
58
59static int
60ctnl_timeout_parse_policy(struct ctnl_timeout *timeout,
61 struct nf_conntrack_l4proto *l4proto,
62 const struct nlattr *attr)
63{
64 int ret = 0;
65
66 if (likely(l4proto->ctnl_timeout.nlattr_to_obj)) {
67 struct nlattr *tb[l4proto->ctnl_timeout.nlattr_max+1];
68
69 nla_parse_nested(tb, l4proto->ctnl_timeout.nlattr_max,
70 attr, l4proto->ctnl_timeout.nla_policy);
71
72 ret = l4proto->ctnl_timeout.nlattr_to_obj(tb, &timeout->data);
73 }
74 return ret;
75}
76
77static int
78cttimeout_new_timeout(struct sock *ctnl, struct sk_buff *skb,
79 const struct nlmsghdr *nlh,
80 const struct nlattr * const cda[])
81{
82 __u16 l3num;
83 __u8 l4num;
84 struct nf_conntrack_l4proto *l4proto;
85 struct ctnl_timeout *timeout, *matching = NULL;
86 char *name;
87 int ret;
88
89 if (!cda[CTA_TIMEOUT_NAME] ||
90 !cda[CTA_TIMEOUT_L3PROTO] ||
91 !cda[CTA_TIMEOUT_L4PROTO] ||
92 !cda[CTA_TIMEOUT_DATA])
93 return -EINVAL;
94
95 name = nla_data(cda[CTA_TIMEOUT_NAME]);
96 l3num = ntohs(nla_get_be16(cda[CTA_TIMEOUT_L3PROTO]));
97 l4num = nla_get_u8(cda[CTA_TIMEOUT_L4PROTO]);
98
99 list_for_each_entry(timeout, &cttimeout_list, head) {
100 if (strncmp(timeout->name, name, CTNL_TIMEOUT_NAME_MAX) != 0)
101 continue;
102
103 if (nlh->nlmsg_flags & NLM_F_EXCL)
104 return -EEXIST;
105
106 matching = timeout;
107 break;
108 }
109
110 l4proto = __nf_ct_l4proto_find(l3num, l4num);
111
112 /* This protocol is not supportted, skip. */
113 if (l4proto->l4proto != l4num)
114 return -EOPNOTSUPP;
115
116 if (matching) {
117 if (nlh->nlmsg_flags & NLM_F_REPLACE) {
118 /* You cannot replace one timeout policy by another of
119 * different kind, sorry.
120 */
121 if (matching->l3num != l3num ||
122 matching->l4num != l4num)
123 return -EINVAL;
124
125 ret = ctnl_timeout_parse_policy(matching, l4proto,
126 cda[CTA_TIMEOUT_DATA]);
127 return ret;
128 }
129 return -EBUSY;
130 }
131
132 timeout = kzalloc(sizeof(struct ctnl_timeout) +
133 l4proto->ctnl_timeout.obj_size, GFP_KERNEL);
134 if (timeout == NULL)
135 return -ENOMEM;
136
137 ret = ctnl_timeout_parse_policy(timeout, l4proto,
138 cda[CTA_TIMEOUT_DATA]);
139 if (ret < 0)
140 goto err;
141
142 strcpy(timeout->name, nla_data(cda[CTA_TIMEOUT_NAME]));
143 timeout->l3num = l3num;
144 timeout->l4num = l4num;
145 atomic_set(&timeout->refcnt, 1);
146 list_add_tail_rcu(&timeout->head, &cttimeout_list);
147
148 return 0;
149err:
150 kfree(timeout);
151 return ret;
152}
153
154static int
155ctnl_timeout_fill_info(struct sk_buff *skb, u32 pid, u32 seq, u32 type,
156 int event, struct ctnl_timeout *timeout)
157{
158 struct nlmsghdr *nlh;
159 struct nfgenmsg *nfmsg;
160 unsigned int flags = pid ? NLM_F_MULTI : 0;
161 struct nf_conntrack_l4proto *l4proto;
162
163 event |= NFNL_SUBSYS_CTNETLINK_TIMEOUT << 8;
164 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*nfmsg), flags);
165 if (nlh == NULL)
166 goto nlmsg_failure;
167
168 nfmsg = nlmsg_data(nlh);
169 nfmsg->nfgen_family = AF_UNSPEC;
170 nfmsg->version = NFNETLINK_V0;
171 nfmsg->res_id = 0;
172
173 NLA_PUT_STRING(skb, CTA_TIMEOUT_NAME, timeout->name);
174 NLA_PUT_BE16(skb, CTA_TIMEOUT_L3PROTO, htons(timeout->l3num));
175 NLA_PUT_U8(skb, CTA_TIMEOUT_L4PROTO, timeout->l4num);
176 NLA_PUT_BE32(skb, CTA_TIMEOUT_USE,
177 htonl(atomic_read(&timeout->refcnt)));
178
179 l4proto = __nf_ct_l4proto_find(timeout->l3num, timeout->l4num);
180
181 /* If the timeout object does not match the layer 4 protocol tracker,
182 * then skip dumping the data part since we don't know how to
183 * interpret it. This may happen for UPDlite, SCTP and DCCP since
184 * you can unload the module.
185 */
186 if (timeout->l4num != l4proto->l4proto)
187 goto out;
188
189 if (likely(l4proto->ctnl_timeout.obj_to_nlattr)) {
190 struct nlattr *nest_parms;
191 int ret;
192
193 nest_parms = nla_nest_start(skb,
194 CTA_TIMEOUT_DATA | NLA_F_NESTED);
195 if (!nest_parms)
196 goto nla_put_failure;
197
198 ret = l4proto->ctnl_timeout.obj_to_nlattr(skb, &timeout->data);
199 if (ret < 0)
200 goto nla_put_failure;
201
202 nla_nest_end(skb, nest_parms);
203 }
204out:
205 nlmsg_end(skb, nlh);
206 return skb->len;
207
208nlmsg_failure:
209nla_put_failure:
210 nlmsg_cancel(skb, nlh);
211 return -1;
212}
213
214static int
215ctnl_timeout_dump(struct sk_buff *skb, struct netlink_callback *cb)
216{
217 struct ctnl_timeout *cur, *last;
218
219 if (cb->args[2])
220 return 0;
221
222 last = (struct ctnl_timeout *)cb->args[1];
223 if (cb->args[1])
224 cb->args[1] = 0;
225
226 rcu_read_lock();
227 list_for_each_entry_rcu(cur, &cttimeout_list, head) {
228 if (last && cur != last)
229 continue;
230
231 if (ctnl_timeout_fill_info(skb, NETLINK_CB(cb->skb).pid,
232 cb->nlh->nlmsg_seq,
233 NFNL_MSG_TYPE(cb->nlh->nlmsg_type),
234 IPCTNL_MSG_TIMEOUT_NEW, cur) < 0) {
235 cb->args[1] = (unsigned long)cur;
236 break;
237 }
238 }
239 if (!cb->args[1])
240 cb->args[2] = 1;
241 rcu_read_unlock();
242 return skb->len;
243}
244
245static int
246cttimeout_get_timeout(struct sock *ctnl, struct sk_buff *skb,
247 const struct nlmsghdr *nlh,
248 const struct nlattr * const cda[])
249{
250 int ret = -ENOENT;
251 char *name;
252 struct ctnl_timeout *cur;
253
254 if (nlh->nlmsg_flags & NLM_F_DUMP) {
255 struct netlink_dump_control c = {
256 .dump = ctnl_timeout_dump,
257 };
258 return netlink_dump_start(ctnl, skb, nlh, &c);
259 }
260
261 if (!cda[CTA_TIMEOUT_NAME])
262 return -EINVAL;
263 name = nla_data(cda[CTA_TIMEOUT_NAME]);
264
265 list_for_each_entry(cur, &cttimeout_list, head) {
266 struct sk_buff *skb2;
267
268 if (strncmp(cur->name, name, CTNL_TIMEOUT_NAME_MAX) != 0)
269 continue;
270
271 skb2 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
272 if (skb2 == NULL) {
273 ret = -ENOMEM;
274 break;
275 }
276
277 ret = ctnl_timeout_fill_info(skb2, NETLINK_CB(skb).pid,
278 nlh->nlmsg_seq,
279 NFNL_MSG_TYPE(nlh->nlmsg_type),
280 IPCTNL_MSG_TIMEOUT_NEW, cur);
281 if (ret <= 0) {
282 kfree_skb(skb2);
283 break;
284 }
285 ret = netlink_unicast(ctnl, skb2, NETLINK_CB(skb).pid,
286 MSG_DONTWAIT);
287 if (ret > 0)
288 ret = 0;
289
290 /* this avoids a loop in nfnetlink. */
291 return ret == -EAGAIN ? -ENOBUFS : ret;
292 }
293 return ret;
294}
295
296/* try to delete object, fail if it is still in use. */
297static int ctnl_timeout_try_del(struct ctnl_timeout *timeout)
298{
299 int ret = 0;
300
301 /* we want to avoid races with nf_ct_timeout_find_get. */
302 if (atomic_dec_and_test(&timeout->refcnt)) {
303 /* We are protected by nfnl mutex. */
304 list_del_rcu(&timeout->head);
305 kfree_rcu(timeout, rcu_head);
306 } else {
307 /* still in use, restore reference counter. */
308 atomic_inc(&timeout->refcnt);
309 ret = -EBUSY;
310 }
311 return ret;
312}
313
314static int
315cttimeout_del_timeout(struct sock *ctnl, struct sk_buff *skb,
316 const struct nlmsghdr *nlh,
317 const struct nlattr * const cda[])
318{
319 char *name;
320 struct ctnl_timeout *cur;
321 int ret = -ENOENT;
322
323 if (!cda[CTA_TIMEOUT_NAME]) {
324 list_for_each_entry(cur, &cttimeout_list, head)
325 ctnl_timeout_try_del(cur);
326
327 return 0;
328 }
329 name = nla_data(cda[CTA_TIMEOUT_NAME]);
330
331 list_for_each_entry(cur, &cttimeout_list, head) {
332 if (strncmp(cur->name, name, CTNL_TIMEOUT_NAME_MAX) != 0)
333 continue;
334
335 ret = ctnl_timeout_try_del(cur);
336 if (ret < 0)
337 return ret;
338
339 break;
340 }
341 return ret;
342}
343
344static const struct nfnl_callback cttimeout_cb[IPCTNL_MSG_TIMEOUT_MAX] = {
345 [IPCTNL_MSG_TIMEOUT_NEW] = { .call = cttimeout_new_timeout,
346 .attr_count = CTA_TIMEOUT_MAX,
347 .policy = cttimeout_nla_policy },
348 [IPCTNL_MSG_TIMEOUT_GET] = { .call = cttimeout_get_timeout,
349 .attr_count = CTA_TIMEOUT_MAX,
350 .policy = cttimeout_nla_policy },
351 [IPCTNL_MSG_TIMEOUT_DELETE] = { .call = cttimeout_del_timeout,
352 .attr_count = CTA_TIMEOUT_MAX,
353 .policy = cttimeout_nla_policy },
354};
355
356static const struct nfnetlink_subsystem cttimeout_subsys = {
357 .name = "conntrack_timeout",
358 .subsys_id = NFNL_SUBSYS_CTNETLINK_TIMEOUT,
359 .cb_count = IPCTNL_MSG_TIMEOUT_MAX,
360 .cb = cttimeout_cb,
361};
362
363MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_CTNETLINK_TIMEOUT);
364
365static int __init cttimeout_init(void)
366{
367 int ret;
368
369 ret = nfnetlink_subsys_register(&cttimeout_subsys);
370 if (ret < 0) {
371 pr_err("cttimeout_init: cannot register cttimeout with "
372 "nfnetlink.\n");
373 goto err_out;
374 }
375 return 0;
376
377err_out:
378 return ret;
379}
380
381static void __exit cttimeout_exit(void)
382{
383 struct ctnl_timeout *cur, *tmp;
384
385 pr_info("cttimeout: unregistering from nfnetlink.\n");
386
387 nfnetlink_subsys_unregister(&cttimeout_subsys);
388 list_for_each_entry_safe(cur, tmp, &cttimeout_list, head) {
389 list_del_rcu(&cur->head);
390 /* We are sure that our objects have no clients at this point,
391 * it's safe to release them all without checking refcnt.
392 */
393 kfree_rcu(cur, rcu_head);
394 }
395}
396
397module_init(cttimeout_init);
398module_exit(cttimeout_exit);
This page took 0.054374 seconds and 5 git commands to generate.