tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1/*
2 * (C) 2011 Pablo Neira Ayuso <pablo@netfilter.org>
3 * (C) 2011 Intra2net AG <http://www.intra2net.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/skbuff.h>
13#include <linux/atomic.h>
14#include <linux/netlink.h>
15#include <linux/rculist.h>
16#include <linux/slab.h>
17#include <linux/types.h>
18#include <linux/errno.h>
19#include <net/netlink.h>
20#include <net/sock.h>
21
22#include <linux/netfilter.h>
23#include <linux/netfilter/nfnetlink.h>
24#include <linux/netfilter/nfnetlink_acct.h>
25
26MODULE_LICENSE("GPL");
27MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
28MODULE_DESCRIPTION("nfacct: Extended Netfilter accounting infrastructure");
29
30static LIST_HEAD(nfnl_acct_list);
31
32struct nf_acct {
33 atomic64_t pkts;
34 atomic64_t bytes;
35 unsigned long flags;
36 struct list_head head;
37 atomic_t refcnt;
38 char name[NFACCT_NAME_MAX];
39 struct rcu_head rcu_head;
40 char data[0];
41};
42
43#define NFACCT_F_QUOTA (NFACCT_F_QUOTA_PKTS | NFACCT_F_QUOTA_BYTES)
44
45static int
46nfnl_acct_new(struct sock *nfnl, struct sk_buff *skb,
47 const struct nlmsghdr *nlh, const struct nlattr * const tb[])
48{
49 struct nf_acct *nfacct, *matching = NULL;
50 char *acct_name;
51 unsigned int size = 0;
52 u32 flags = 0;
53
54 if (!tb[NFACCT_NAME])
55 return -EINVAL;
56
57 acct_name = nla_data(tb[NFACCT_NAME]);
58 if (strlen(acct_name) == 0)
59 return -EINVAL;
60
61 list_for_each_entry(nfacct, &nfnl_acct_list, head) {
62 if (strncmp(nfacct->name, acct_name, NFACCT_NAME_MAX) != 0)
63 continue;
64
65 if (nlh->nlmsg_flags & NLM_F_EXCL)
66 return -EEXIST;
67
68 matching = nfacct;
69 break;
70 }
71
72 if (matching) {
73 if (nlh->nlmsg_flags & NLM_F_REPLACE) {
74 /* reset counters if you request a replacement. */
75 atomic64_set(&matching->pkts, 0);
76 atomic64_set(&matching->bytes, 0);
77 smp_mb__before_atomic();
78 /* reset overquota flag if quota is enabled. */
79 if ((matching->flags & NFACCT_F_QUOTA))
80 clear_bit(NFACCT_F_OVERQUOTA, &matching->flags);
81 return 0;
82 }
83 return -EBUSY;
84 }
85
86 if (tb[NFACCT_FLAGS]) {
87 flags = ntohl(nla_get_be32(tb[NFACCT_FLAGS]));
88 if (flags & ~NFACCT_F_QUOTA)
89 return -EOPNOTSUPP;
90 if ((flags & NFACCT_F_QUOTA) == NFACCT_F_QUOTA)
91 return -EINVAL;
92 if (flags & NFACCT_F_OVERQUOTA)
93 return -EINVAL;
94
95 size += sizeof(u64);
96 }
97
98 nfacct = kzalloc(sizeof(struct nf_acct) + size, GFP_KERNEL);
99 if (nfacct == NULL)
100 return -ENOMEM;
101
102 if (flags & NFACCT_F_QUOTA) {
103 u64 *quota = (u64 *)nfacct->data;
104
105 *quota = be64_to_cpu(nla_get_be64(tb[NFACCT_QUOTA]));
106 nfacct->flags = flags;
107 }
108
109 strncpy(nfacct->name, nla_data(tb[NFACCT_NAME]), NFACCT_NAME_MAX);
110
111 if (tb[NFACCT_BYTES]) {
112 atomic64_set(&nfacct->bytes,
113 be64_to_cpu(nla_get_be64(tb[NFACCT_BYTES])));
114 }
115 if (tb[NFACCT_PKTS]) {
116 atomic64_set(&nfacct->pkts,
117 be64_to_cpu(nla_get_be64(tb[NFACCT_PKTS])));
118 }
119 atomic_set(&nfacct->refcnt, 1);
120 list_add_tail_rcu(&nfacct->head, &nfnl_acct_list);
121 return 0;
122}
123
124static int
125nfnl_acct_fill_info(struct sk_buff *skb, u32 portid, u32 seq, u32 type,
126 int event, struct nf_acct *acct)
127{
128 struct nlmsghdr *nlh;
129 struct nfgenmsg *nfmsg;
130 unsigned int flags = portid ? NLM_F_MULTI : 0;
131 u64 pkts, bytes;
132
133 event |= NFNL_SUBSYS_ACCT << 8;
134 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nfmsg), flags);
135 if (nlh == NULL)
136 goto nlmsg_failure;
137
138 nfmsg = nlmsg_data(nlh);
139 nfmsg->nfgen_family = AF_UNSPEC;
140 nfmsg->version = NFNETLINK_V0;
141 nfmsg->res_id = 0;
142
143 if (nla_put_string(skb, NFACCT_NAME, acct->name))
144 goto nla_put_failure;
145
146 if (type == NFNL_MSG_ACCT_GET_CTRZERO) {
147 pkts = atomic64_xchg(&acct->pkts, 0);
148 bytes = atomic64_xchg(&acct->bytes, 0);
149 smp_mb__before_atomic();
150 if (acct->flags & NFACCT_F_QUOTA)
151 clear_bit(NFACCT_F_OVERQUOTA, &acct->flags);
152 } else {
153 pkts = atomic64_read(&acct->pkts);
154 bytes = atomic64_read(&acct->bytes);
155 }
156 if (nla_put_be64(skb, NFACCT_PKTS, cpu_to_be64(pkts)) ||
157 nla_put_be64(skb, NFACCT_BYTES, cpu_to_be64(bytes)) ||
158 nla_put_be32(skb, NFACCT_USE, htonl(atomic_read(&acct->refcnt))))
159 goto nla_put_failure;
160 if (acct->flags & NFACCT_F_QUOTA) {
161 u64 *quota = (u64 *)acct->data;
162
163 if (nla_put_be32(skb, NFACCT_FLAGS, htonl(acct->flags)) ||
164 nla_put_be64(skb, NFACCT_QUOTA, cpu_to_be64(*quota)))
165 goto nla_put_failure;
166 }
167 nlmsg_end(skb, nlh);
168 return skb->len;
169
170nlmsg_failure:
171nla_put_failure:
172 nlmsg_cancel(skb, nlh);
173 return -1;
174}
175
176static int
177nfnl_acct_dump(struct sk_buff *skb, struct netlink_callback *cb)
178{
179 struct nf_acct *cur, *last;
180
181 if (cb->args[2])
182 return 0;
183
184 last = (struct nf_acct *)cb->args[1];
185 if (cb->args[1])
186 cb->args[1] = 0;
187
188 rcu_read_lock();
189 list_for_each_entry_rcu(cur, &nfnl_acct_list, head) {
190 if (last) {
191 if (cur != last)
192 continue;
193
194 last = NULL;
195 }
196 if (nfnl_acct_fill_info(skb, NETLINK_CB(cb->skb).portid,
197 cb->nlh->nlmsg_seq,
198 NFNL_MSG_TYPE(cb->nlh->nlmsg_type),
199 NFNL_MSG_ACCT_NEW, cur) < 0) {
200 cb->args[1] = (unsigned long)cur;
201 break;
202 }
203 }
204 if (!cb->args[1])
205 cb->args[2] = 1;
206 rcu_read_unlock();
207 return skb->len;
208}
209
210static int
211nfnl_acct_get(struct sock *nfnl, struct sk_buff *skb,
212 const struct nlmsghdr *nlh, const struct nlattr * const tb[])
213{
214 int ret = -ENOENT;
215 struct nf_acct *cur;
216 char *acct_name;
217
218 if (nlh->nlmsg_flags & NLM_F_DUMP) {
219 struct netlink_dump_control c = {
220 .dump = nfnl_acct_dump,
221 };
222 return netlink_dump_start(nfnl, skb, nlh, &c);
223 }
224
225 if (!tb[NFACCT_NAME])
226 return -EINVAL;
227 acct_name = nla_data(tb[NFACCT_NAME]);
228
229 list_for_each_entry(cur, &nfnl_acct_list, head) {
230 struct sk_buff *skb2;
231
232 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
233 continue;
234
235 skb2 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
236 if (skb2 == NULL) {
237 ret = -ENOMEM;
238 break;
239 }
240
241 ret = nfnl_acct_fill_info(skb2, NETLINK_CB(skb).portid,
242 nlh->nlmsg_seq,
243 NFNL_MSG_TYPE(nlh->nlmsg_type),
244 NFNL_MSG_ACCT_NEW, cur);
245 if (ret <= 0) {
246 kfree_skb(skb2);
247 break;
248 }
249 ret = netlink_unicast(nfnl, skb2, NETLINK_CB(skb).portid,
250 MSG_DONTWAIT);
251 if (ret > 0)
252 ret = 0;
253
254 /* this avoids a loop in nfnetlink. */
255 return ret == -EAGAIN ? -ENOBUFS : ret;
256 }
257 return ret;
258}
259
260/* try to delete object, fail if it is still in use. */
261static int nfnl_acct_try_del(struct nf_acct *cur)
262{
263 int ret = 0;
264
265 /* we want to avoid races with nfnl_acct_find_get. */
266 if (atomic_dec_and_test(&cur->refcnt)) {
267 /* We are protected by nfnl mutex. */
268 list_del_rcu(&cur->head);
269 kfree_rcu(cur, rcu_head);
270 } else {
271 /* still in use, restore reference counter. */
272 atomic_inc(&cur->refcnt);
273 ret = -EBUSY;
274 }
275 return ret;
276}
277
278static int
279nfnl_acct_del(struct sock *nfnl, struct sk_buff *skb,
280 const struct nlmsghdr *nlh, const struct nlattr * const tb[])
281{
282 char *acct_name;
283 struct nf_acct *cur;
284 int ret = -ENOENT;
285
286 if (!tb[NFACCT_NAME]) {
287 list_for_each_entry(cur, &nfnl_acct_list, head)
288 nfnl_acct_try_del(cur);
289
290 return 0;
291 }
292 acct_name = nla_data(tb[NFACCT_NAME]);
293
294 list_for_each_entry(cur, &nfnl_acct_list, head) {
295 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX) != 0)
296 continue;
297
298 ret = nfnl_acct_try_del(cur);
299 if (ret < 0)
300 return ret;
301
302 break;
303 }
304 return ret;
305}
306
307static const struct nla_policy nfnl_acct_policy[NFACCT_MAX+1] = {
308 [NFACCT_NAME] = { .type = NLA_NUL_STRING, .len = NFACCT_NAME_MAX-1 },
309 [NFACCT_BYTES] = { .type = NLA_U64 },
310 [NFACCT_PKTS] = { .type = NLA_U64 },
311 [NFACCT_FLAGS] = { .type = NLA_U32 },
312 [NFACCT_QUOTA] = { .type = NLA_U64 },
313};
314
315static const struct nfnl_callback nfnl_acct_cb[NFNL_MSG_ACCT_MAX] = {
316 [NFNL_MSG_ACCT_NEW] = { .call = nfnl_acct_new,
317 .attr_count = NFACCT_MAX,
318 .policy = nfnl_acct_policy },
319 [NFNL_MSG_ACCT_GET] = { .call = nfnl_acct_get,
320 .attr_count = NFACCT_MAX,
321 .policy = nfnl_acct_policy },
322 [NFNL_MSG_ACCT_GET_CTRZERO] = { .call = nfnl_acct_get,
323 .attr_count = NFACCT_MAX,
324 .policy = nfnl_acct_policy },
325 [NFNL_MSG_ACCT_DEL] = { .call = nfnl_acct_del,
326 .attr_count = NFACCT_MAX,
327 .policy = nfnl_acct_policy },
328};
329
330static const struct nfnetlink_subsystem nfnl_acct_subsys = {
331 .name = "acct",
332 .subsys_id = NFNL_SUBSYS_ACCT,
333 .cb_count = NFNL_MSG_ACCT_MAX,
334 .cb = nfnl_acct_cb,
335};
336
337MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ACCT);
338
339struct nf_acct *nfnl_acct_find_get(const char *acct_name)
340{
341 struct nf_acct *cur, *acct = NULL;
342
343 rcu_read_lock();
344 list_for_each_entry_rcu(cur, &nfnl_acct_list, head) {
345 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
346 continue;
347
348 if (!try_module_get(THIS_MODULE))
349 goto err;
350
351 if (!atomic_inc_not_zero(&cur->refcnt)) {
352 module_put(THIS_MODULE);
353 goto err;
354 }
355
356 acct = cur;
357 break;
358 }
359err:
360 rcu_read_unlock();
361 return acct;
362}
363EXPORT_SYMBOL_GPL(nfnl_acct_find_get);
364
365void nfnl_acct_put(struct nf_acct *acct)
366{
367 atomic_dec(&acct->refcnt);
368 module_put(THIS_MODULE);
369}
370EXPORT_SYMBOL_GPL(nfnl_acct_put);
371
372void nfnl_acct_update(const struct sk_buff *skb, struct nf_acct *nfacct)
373{
374 atomic64_inc(&nfacct->pkts);
375 atomic64_add(skb->len, &nfacct->bytes);
376}
377EXPORT_SYMBOL_GPL(nfnl_acct_update);
378
379static void nfnl_overquota_report(struct nf_acct *nfacct)
380{
381 int ret;
382 struct sk_buff *skb;
383
384 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
385 if (skb == NULL)
386 return;
387
388 ret = nfnl_acct_fill_info(skb, 0, 0, NFNL_MSG_ACCT_OVERQUOTA, 0,
389 nfacct);
390 if (ret <= 0) {
391 kfree_skb(skb);
392 return;
393 }
394 netlink_broadcast(init_net.nfnl, skb, 0, NFNLGRP_ACCT_QUOTA,
395 GFP_ATOMIC);
396}
397
398int nfnl_acct_overquota(const struct sk_buff *skb, struct nf_acct *nfacct)
399{
400 u64 now;
401 u64 *quota;
402 int ret = NFACCT_UNDERQUOTA;
403
404 /* no place here if we don't have a quota */
405 if (!(nfacct->flags & NFACCT_F_QUOTA))
406 return NFACCT_NO_QUOTA;
407
408 quota = (u64 *)nfacct->data;
409 now = (nfacct->flags & NFACCT_F_QUOTA_PKTS) ?
410 atomic64_read(&nfacct->pkts) : atomic64_read(&nfacct->bytes);
411
412 ret = now > *quota;
413
414 if (now >= *quota &&
415 !test_and_set_bit(NFACCT_F_OVERQUOTA, &nfacct->flags)) {
416 nfnl_overquota_report(nfacct);
417 }
418
419 return ret;
420}
421EXPORT_SYMBOL_GPL(nfnl_acct_overquota);
422
423static int __init nfnl_acct_init(void)
424{
425 int ret;
426
427 pr_info("nfnl_acct: registering with nfnetlink.\n");
428 ret = nfnetlink_subsys_register(&nfnl_acct_subsys);
429 if (ret < 0) {
430 pr_err("nfnl_acct_init: cannot register with nfnetlink.\n");
431 goto err_out;
432 }
433 return 0;
434err_out:
435 return ret;
436}
437
438static void __exit nfnl_acct_exit(void)
439{
440 struct nf_acct *cur, *tmp;
441
442 pr_info("nfnl_acct: unregistering from nfnetlink.\n");
443 nfnetlink_subsys_unregister(&nfnl_acct_subsys);
444
445 list_for_each_entry_safe(cur, tmp, &nfnl_acct_list, head) {
446 list_del_rcu(&cur->head);
447 /* We are sure that our objects have no clients at this point,
448 * it's safe to release them all without checking refcnt. */
449 kfree_rcu(cur, rcu_head);
450 }
451}
452
453module_init(nfnl_acct_init);
454module_exit(nfnl_acct_exit);