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/refcount.h>
15 #include <linux/netlink.h>
16 #include <linux/rculist.h>
17 #include <linux/slab.h>
18 #include <linux/types.h>
19 #include <linux/errno.h>
20 #include <net/netlink.h>
21 #include <net/sock.h>
22
23 #include <linux/netfilter.h>
24 #include <linux/netfilter/nfnetlink.h>
25 #include <linux/netfilter/nfnetlink_acct.h>
26
27 MODULE_LICENSE("GPL");
28 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
29 MODULE_DESCRIPTION("nfacct: Extended Netfilter accounting infrastructure");
30
31 struct nf_acct {
32 atomic64_t pkts;
33 atomic64_t bytes;
34 unsigned long flags;
35 struct list_head head;
36 refcount_t refcnt;
37 char name[NFACCT_NAME_MAX];
38 struct rcu_head rcu_head;
39 char data[0];
40 };
41
42 struct nfacct_filter {
43 u32 value;
44 u32 mask;
45 };
46
47 #define NFACCT_F_QUOTA (NFACCT_F_QUOTA_PKTS | NFACCT_F_QUOTA_BYTES)
48 #define NFACCT_OVERQUOTA_BIT 2 /* NFACCT_F_OVERQUOTA */
49
nfnl_acct_new(struct net * net,struct sock * nfnl,struct sk_buff * skb,const struct nlmsghdr * nlh,const struct nlattr * const tb[],struct netlink_ext_ack * extack)50 static int nfnl_acct_new(struct net *net, struct sock *nfnl,
51 struct sk_buff *skb, const struct nlmsghdr *nlh,
52 const struct nlattr * const tb[],
53 struct netlink_ext_ack *extack)
54 {
55 struct nf_acct *nfacct, *matching = NULL;
56 char *acct_name;
57 unsigned int size = 0;
58 u32 flags = 0;
59
60 if (!tb[NFACCT_NAME])
61 return -EINVAL;
62
63 acct_name = nla_data(tb[NFACCT_NAME]);
64 if (strlen(acct_name) == 0)
65 return -EINVAL;
66
67 list_for_each_entry(nfacct, &net->nfnl_acct_list, head) {
68 if (strncmp(nfacct->name, acct_name, NFACCT_NAME_MAX) != 0)
69 continue;
70
71 if (nlh->nlmsg_flags & NLM_F_EXCL)
72 return -EEXIST;
73
74 matching = nfacct;
75 break;
76 }
77
78 if (matching) {
79 if (nlh->nlmsg_flags & NLM_F_REPLACE) {
80 /* reset counters if you request a replacement. */
81 atomic64_set(&matching->pkts, 0);
82 atomic64_set(&matching->bytes, 0);
83 smp_mb__before_atomic();
84 /* reset overquota flag if quota is enabled. */
85 if ((matching->flags & NFACCT_F_QUOTA))
86 clear_bit(NFACCT_OVERQUOTA_BIT,
87 &matching->flags);
88 return 0;
89 }
90 return -EBUSY;
91 }
92
93 if (tb[NFACCT_FLAGS]) {
94 flags = ntohl(nla_get_be32(tb[NFACCT_FLAGS]));
95 if (flags & ~NFACCT_F_QUOTA)
96 return -EOPNOTSUPP;
97 if ((flags & NFACCT_F_QUOTA) == NFACCT_F_QUOTA)
98 return -EINVAL;
99 if (flags & NFACCT_F_OVERQUOTA)
100 return -EINVAL;
101 if ((flags & NFACCT_F_QUOTA) && !tb[NFACCT_QUOTA])
102 return -EINVAL;
103
104 size += sizeof(u64);
105 }
106
107 nfacct = kzalloc(sizeof(struct nf_acct) + size, GFP_KERNEL);
108 if (nfacct == NULL)
109 return -ENOMEM;
110
111 if (flags & NFACCT_F_QUOTA) {
112 u64 *quota = (u64 *)nfacct->data;
113
114 *quota = be64_to_cpu(nla_get_be64(tb[NFACCT_QUOTA]));
115 nfacct->flags = flags;
116 }
117
118 nla_strlcpy(nfacct->name, tb[NFACCT_NAME], NFACCT_NAME_MAX);
119
120 if (tb[NFACCT_BYTES]) {
121 atomic64_set(&nfacct->bytes,
122 be64_to_cpu(nla_get_be64(tb[NFACCT_BYTES])));
123 }
124 if (tb[NFACCT_PKTS]) {
125 atomic64_set(&nfacct->pkts,
126 be64_to_cpu(nla_get_be64(tb[NFACCT_PKTS])));
127 }
128 refcount_set(&nfacct->refcnt, 1);
129 list_add_tail_rcu(&nfacct->head, &net->nfnl_acct_list);
130 return 0;
131 }
132
133 static int
nfnl_acct_fill_info(struct sk_buff * skb,u32 portid,u32 seq,u32 type,int event,struct nf_acct * acct)134 nfnl_acct_fill_info(struct sk_buff *skb, u32 portid, u32 seq, u32 type,
135 int event, struct nf_acct *acct)
136 {
137 struct nlmsghdr *nlh;
138 struct nfgenmsg *nfmsg;
139 unsigned int flags = portid ? NLM_F_MULTI : 0;
140 u64 pkts, bytes;
141 u32 old_flags;
142
143 event = nfnl_msg_type(NFNL_SUBSYS_ACCT, event);
144 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nfmsg), flags);
145 if (nlh == NULL)
146 goto nlmsg_failure;
147
148 nfmsg = nlmsg_data(nlh);
149 nfmsg->nfgen_family = AF_UNSPEC;
150 nfmsg->version = NFNETLINK_V0;
151 nfmsg->res_id = 0;
152
153 if (nla_put_string(skb, NFACCT_NAME, acct->name))
154 goto nla_put_failure;
155
156 old_flags = acct->flags;
157 if (type == NFNL_MSG_ACCT_GET_CTRZERO) {
158 pkts = atomic64_xchg(&acct->pkts, 0);
159 bytes = atomic64_xchg(&acct->bytes, 0);
160 smp_mb__before_atomic();
161 if (acct->flags & NFACCT_F_QUOTA)
162 clear_bit(NFACCT_OVERQUOTA_BIT, &acct->flags);
163 } else {
164 pkts = atomic64_read(&acct->pkts);
165 bytes = atomic64_read(&acct->bytes);
166 }
167 if (nla_put_be64(skb, NFACCT_PKTS, cpu_to_be64(pkts),
168 NFACCT_PAD) ||
169 nla_put_be64(skb, NFACCT_BYTES, cpu_to_be64(bytes),
170 NFACCT_PAD) ||
171 nla_put_be32(skb, NFACCT_USE, htonl(refcount_read(&acct->refcnt))))
172 goto nla_put_failure;
173 if (acct->flags & NFACCT_F_QUOTA) {
174 u64 *quota = (u64 *)acct->data;
175
176 if (nla_put_be32(skb, NFACCT_FLAGS, htonl(old_flags)) ||
177 nla_put_be64(skb, NFACCT_QUOTA, cpu_to_be64(*quota),
178 NFACCT_PAD))
179 goto nla_put_failure;
180 }
181 nlmsg_end(skb, nlh);
182 return skb->len;
183
184 nlmsg_failure:
185 nla_put_failure:
186 nlmsg_cancel(skb, nlh);
187 return -1;
188 }
189
190 static int
nfnl_acct_dump(struct sk_buff * skb,struct netlink_callback * cb)191 nfnl_acct_dump(struct sk_buff *skb, struct netlink_callback *cb)
192 {
193 struct net *net = sock_net(skb->sk);
194 struct nf_acct *cur, *last;
195 const struct nfacct_filter *filter = cb->data;
196
197 if (cb->args[2])
198 return 0;
199
200 last = (struct nf_acct *)cb->args[1];
201 if (cb->args[1])
202 cb->args[1] = 0;
203
204 rcu_read_lock();
205 list_for_each_entry_rcu(cur, &net->nfnl_acct_list, head) {
206 if (last) {
207 if (cur != last)
208 continue;
209
210 last = NULL;
211 }
212
213 if (filter && (cur->flags & filter->mask) != filter->value)
214 continue;
215
216 if (nfnl_acct_fill_info(skb, NETLINK_CB(cb->skb).portid,
217 cb->nlh->nlmsg_seq,
218 NFNL_MSG_TYPE(cb->nlh->nlmsg_type),
219 NFNL_MSG_ACCT_NEW, cur) < 0) {
220 cb->args[1] = (unsigned long)cur;
221 break;
222 }
223 }
224 if (!cb->args[1])
225 cb->args[2] = 1;
226 rcu_read_unlock();
227 return skb->len;
228 }
229
nfnl_acct_done(struct netlink_callback * cb)230 static int nfnl_acct_done(struct netlink_callback *cb)
231 {
232 kfree(cb->data);
233 return 0;
234 }
235
236 static const struct nla_policy filter_policy[NFACCT_FILTER_MAX + 1] = {
237 [NFACCT_FILTER_MASK] = { .type = NLA_U32 },
238 [NFACCT_FILTER_VALUE] = { .type = NLA_U32 },
239 };
240
nfnl_acct_start(struct netlink_callback * cb)241 static int nfnl_acct_start(struct netlink_callback *cb)
242 {
243 const struct nlattr *const attr = cb->data;
244 struct nlattr *tb[NFACCT_FILTER_MAX + 1];
245 struct nfacct_filter *filter;
246 int err;
247
248 if (!attr)
249 return 0;
250
251 err = nla_parse_nested(tb, NFACCT_FILTER_MAX, attr, filter_policy,
252 NULL);
253 if (err < 0)
254 return err;
255
256 if (!tb[NFACCT_FILTER_MASK] || !tb[NFACCT_FILTER_VALUE])
257 return -EINVAL;
258
259 filter = kzalloc(sizeof(struct nfacct_filter), GFP_KERNEL);
260 if (!filter)
261 return -ENOMEM;
262
263 filter->mask = ntohl(nla_get_be32(tb[NFACCT_FILTER_MASK]));
264 filter->value = ntohl(nla_get_be32(tb[NFACCT_FILTER_VALUE]));
265 cb->data = filter;
266
267 return 0;
268 }
269
nfnl_acct_get(struct net * net,struct sock * nfnl,struct sk_buff * skb,const struct nlmsghdr * nlh,const struct nlattr * const tb[],struct netlink_ext_ack * extack)270 static int nfnl_acct_get(struct net *net, struct sock *nfnl,
271 struct sk_buff *skb, const struct nlmsghdr *nlh,
272 const struct nlattr * const tb[],
273 struct netlink_ext_ack *extack)
274 {
275 int ret = -ENOENT;
276 struct nf_acct *cur;
277 char *acct_name;
278
279 if (nlh->nlmsg_flags & NLM_F_DUMP) {
280 struct netlink_dump_control c = {
281 .dump = nfnl_acct_dump,
282 .start = nfnl_acct_start,
283 .done = nfnl_acct_done,
284 .data = (void *)tb[NFACCT_FILTER],
285 };
286
287 return netlink_dump_start(nfnl, skb, nlh, &c);
288 }
289
290 if (!tb[NFACCT_NAME])
291 return -EINVAL;
292 acct_name = nla_data(tb[NFACCT_NAME]);
293
294 list_for_each_entry(cur, &net->nfnl_acct_list, head) {
295 struct sk_buff *skb2;
296
297 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
298 continue;
299
300 skb2 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
301 if (skb2 == NULL) {
302 ret = -ENOMEM;
303 break;
304 }
305
306 ret = nfnl_acct_fill_info(skb2, NETLINK_CB(skb).portid,
307 nlh->nlmsg_seq,
308 NFNL_MSG_TYPE(nlh->nlmsg_type),
309 NFNL_MSG_ACCT_NEW, cur);
310 if (ret <= 0) {
311 kfree_skb(skb2);
312 break;
313 }
314 ret = netlink_unicast(nfnl, skb2, NETLINK_CB(skb).portid,
315 MSG_DONTWAIT);
316 if (ret > 0)
317 ret = 0;
318
319 /* this avoids a loop in nfnetlink. */
320 return ret == -EAGAIN ? -ENOBUFS : ret;
321 }
322 return ret;
323 }
324
325 /* try to delete object, fail if it is still in use. */
nfnl_acct_try_del(struct nf_acct * cur)326 static int nfnl_acct_try_del(struct nf_acct *cur)
327 {
328 int ret = 0;
329
330 /* We want to avoid races with nfnl_acct_put. So only when the current
331 * refcnt is 1, we decrease it to 0.
332 */
333 if (refcount_dec_if_one(&cur->refcnt)) {
334 /* We are protected by nfnl mutex. */
335 list_del_rcu(&cur->head);
336 kfree_rcu(cur, rcu_head);
337 } else {
338 ret = -EBUSY;
339 }
340 return ret;
341 }
342
nfnl_acct_del(struct net * net,struct sock * nfnl,struct sk_buff * skb,const struct nlmsghdr * nlh,const struct nlattr * const tb[],struct netlink_ext_ack * extack)343 static int nfnl_acct_del(struct net *net, struct sock *nfnl,
344 struct sk_buff *skb, const struct nlmsghdr *nlh,
345 const struct nlattr * const tb[],
346 struct netlink_ext_ack *extack)
347 {
348 struct nf_acct *cur, *tmp;
349 int ret = -ENOENT;
350 char *acct_name;
351
352 if (!tb[NFACCT_NAME]) {
353 list_for_each_entry_safe(cur, tmp, &net->nfnl_acct_list, head)
354 nfnl_acct_try_del(cur);
355
356 return 0;
357 }
358 acct_name = nla_data(tb[NFACCT_NAME]);
359
360 list_for_each_entry(cur, &net->nfnl_acct_list, head) {
361 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX) != 0)
362 continue;
363
364 ret = nfnl_acct_try_del(cur);
365 if (ret < 0)
366 return ret;
367
368 break;
369 }
370 return ret;
371 }
372
373 static const struct nla_policy nfnl_acct_policy[NFACCT_MAX+1] = {
374 [NFACCT_NAME] = { .type = NLA_NUL_STRING, .len = NFACCT_NAME_MAX-1 },
375 [NFACCT_BYTES] = { .type = NLA_U64 },
376 [NFACCT_PKTS] = { .type = NLA_U64 },
377 [NFACCT_FLAGS] = { .type = NLA_U32 },
378 [NFACCT_QUOTA] = { .type = NLA_U64 },
379 [NFACCT_FILTER] = {.type = NLA_NESTED },
380 };
381
382 static const struct nfnl_callback nfnl_acct_cb[NFNL_MSG_ACCT_MAX] = {
383 [NFNL_MSG_ACCT_NEW] = { .call = nfnl_acct_new,
384 .attr_count = NFACCT_MAX,
385 .policy = nfnl_acct_policy },
386 [NFNL_MSG_ACCT_GET] = { .call = nfnl_acct_get,
387 .attr_count = NFACCT_MAX,
388 .policy = nfnl_acct_policy },
389 [NFNL_MSG_ACCT_GET_CTRZERO] = { .call = nfnl_acct_get,
390 .attr_count = NFACCT_MAX,
391 .policy = nfnl_acct_policy },
392 [NFNL_MSG_ACCT_DEL] = { .call = nfnl_acct_del,
393 .attr_count = NFACCT_MAX,
394 .policy = nfnl_acct_policy },
395 };
396
397 static const struct nfnetlink_subsystem nfnl_acct_subsys = {
398 .name = "acct",
399 .subsys_id = NFNL_SUBSYS_ACCT,
400 .cb_count = NFNL_MSG_ACCT_MAX,
401 .cb = nfnl_acct_cb,
402 };
403
404 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ACCT);
405
nfnl_acct_find_get(struct net * net,const char * acct_name)406 struct nf_acct *nfnl_acct_find_get(struct net *net, const char *acct_name)
407 {
408 struct nf_acct *cur, *acct = NULL;
409
410 rcu_read_lock();
411 list_for_each_entry_rcu(cur, &net->nfnl_acct_list, head) {
412 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
413 continue;
414
415 if (!try_module_get(THIS_MODULE))
416 goto err;
417
418 if (!refcount_inc_not_zero(&cur->refcnt)) {
419 module_put(THIS_MODULE);
420 goto err;
421 }
422
423 acct = cur;
424 break;
425 }
426 err:
427 rcu_read_unlock();
428 return acct;
429 }
430 EXPORT_SYMBOL_GPL(nfnl_acct_find_get);
431
nfnl_acct_put(struct nf_acct * acct)432 void nfnl_acct_put(struct nf_acct *acct)
433 {
434 if (refcount_dec_and_test(&acct->refcnt))
435 kfree_rcu(acct, rcu_head);
436
437 module_put(THIS_MODULE);
438 }
439 EXPORT_SYMBOL_GPL(nfnl_acct_put);
440
nfnl_acct_update(const struct sk_buff * skb,struct nf_acct * nfacct)441 void nfnl_acct_update(const struct sk_buff *skb, struct nf_acct *nfacct)
442 {
443 atomic64_inc(&nfacct->pkts);
444 atomic64_add(skb->len, &nfacct->bytes);
445 }
446 EXPORT_SYMBOL_GPL(nfnl_acct_update);
447
nfnl_overquota_report(struct net * net,struct nf_acct * nfacct)448 static void nfnl_overquota_report(struct net *net, struct nf_acct *nfacct)
449 {
450 int ret;
451 struct sk_buff *skb;
452
453 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
454 if (skb == NULL)
455 return;
456
457 ret = nfnl_acct_fill_info(skb, 0, 0, NFNL_MSG_ACCT_OVERQUOTA, 0,
458 nfacct);
459 if (ret <= 0) {
460 kfree_skb(skb);
461 return;
462 }
463 netlink_broadcast(net->nfnl, skb, 0, NFNLGRP_ACCT_QUOTA,
464 GFP_ATOMIC);
465 }
466
nfnl_acct_overquota(struct net * net,struct nf_acct * nfacct)467 int nfnl_acct_overquota(struct net *net, struct nf_acct *nfacct)
468 {
469 u64 now;
470 u64 *quota;
471 int ret = NFACCT_UNDERQUOTA;
472
473 /* no place here if we don't have a quota */
474 if (!(nfacct->flags & NFACCT_F_QUOTA))
475 return NFACCT_NO_QUOTA;
476
477 quota = (u64 *)nfacct->data;
478 now = (nfacct->flags & NFACCT_F_QUOTA_PKTS) ?
479 atomic64_read(&nfacct->pkts) : atomic64_read(&nfacct->bytes);
480
481 ret = now > *quota;
482
483 if (now >= *quota &&
484 !test_and_set_bit(NFACCT_OVERQUOTA_BIT, &nfacct->flags)) {
485 nfnl_overquota_report(net, nfacct);
486 }
487
488 return ret;
489 }
490 EXPORT_SYMBOL_GPL(nfnl_acct_overquota);
491
nfnl_acct_net_init(struct net * net)492 static int __net_init nfnl_acct_net_init(struct net *net)
493 {
494 INIT_LIST_HEAD(&net->nfnl_acct_list);
495
496 return 0;
497 }
498
nfnl_acct_net_exit(struct net * net)499 static void __net_exit nfnl_acct_net_exit(struct net *net)
500 {
501 struct nf_acct *cur, *tmp;
502
503 list_for_each_entry_safe(cur, tmp, &net->nfnl_acct_list, head) {
504 list_del_rcu(&cur->head);
505
506 if (refcount_dec_and_test(&cur->refcnt))
507 kfree_rcu(cur, rcu_head);
508 }
509 }
510
511 static struct pernet_operations nfnl_acct_ops = {
512 .init = nfnl_acct_net_init,
513 .exit = nfnl_acct_net_exit,
514 };
515
nfnl_acct_init(void)516 static int __init nfnl_acct_init(void)
517 {
518 int ret;
519
520 ret = register_pernet_subsys(&nfnl_acct_ops);
521 if (ret < 0) {
522 pr_err("nfnl_acct_init: failed to register pernet ops\n");
523 goto err_out;
524 }
525
526 ret = nfnetlink_subsys_register(&nfnl_acct_subsys);
527 if (ret < 0) {
528 pr_err("nfnl_acct_init: cannot register with nfnetlink.\n");
529 goto cleanup_pernet;
530 }
531 return 0;
532
533 cleanup_pernet:
534 unregister_pernet_subsys(&nfnl_acct_ops);
535 err_out:
536 return ret;
537 }
538
nfnl_acct_exit(void)539 static void __exit nfnl_acct_exit(void)
540 {
541 nfnetlink_subsys_unregister(&nfnl_acct_subsys);
542 unregister_pernet_subsys(&nfnl_acct_ops);
543 }
544
545 module_init(nfnl_acct_init);
546 module_exit(nfnl_acct_exit);
547
