1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * (C) 2011 Pablo Neira Ayuso <pablo@netfilter.org>
4 * (C) 2011 Intra2net AG <http://www.intra2net.com>
5 */
6 #include <linux/init.h>
7 #include <linux/module.h>
8 #include <linux/kernel.h>
9 #include <linux/skbuff.h>
10 #include <linux/atomic.h>
11 #include <linux/refcount.h>
12 #include <linux/netlink.h>
13 #include <linux/rculist.h>
14 #include <linux/slab.h>
15 #include <linux/types.h>
16 #include <linux/errno.h>
17 #include <net/netlink.h>
18 #include <net/sock.h>
19
20 #include <linux/netfilter.h>
21 #include <linux/netfilter/nfnetlink.h>
22 #include <linux/netfilter/nfnetlink_acct.h>
23
24 MODULE_LICENSE("GPL");
25 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
26 MODULE_DESCRIPTION("nfacct: Extended Netfilter accounting infrastructure");
27
28 struct nf_acct {
29 atomic64_t pkts;
30 atomic64_t bytes;
31 unsigned long flags;
32 struct list_head head;
33 refcount_t refcnt;
34 char name[NFACCT_NAME_MAX];
35 struct rcu_head rcu_head;
36 char data[0];
37 };
38
39 struct nfacct_filter {
40 u32 value;
41 u32 mask;
42 };
43
44 #define NFACCT_F_QUOTA (NFACCT_F_QUOTA_PKTS | NFACCT_F_QUOTA_BYTES)
45 #define NFACCT_OVERQUOTA_BIT 2 /* NFACCT_F_OVERQUOTA */
46
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)47 static int nfnl_acct_new(struct net *net, struct sock *nfnl,
48 struct sk_buff *skb, const struct nlmsghdr *nlh,
49 const struct nlattr * const tb[],
50 struct netlink_ext_ack *extack)
51 {
52 struct nf_acct *nfacct, *matching = NULL;
53 char *acct_name;
54 unsigned int size = 0;
55 u32 flags = 0;
56
57 if (!tb[NFACCT_NAME])
58 return -EINVAL;
59
60 acct_name = nla_data(tb[NFACCT_NAME]);
61 if (strlen(acct_name) == 0)
62 return -EINVAL;
63
64 list_for_each_entry(nfacct, &net->nfnl_acct_list, head) {
65 if (strncmp(nfacct->name, acct_name, NFACCT_NAME_MAX) != 0)
66 continue;
67
68 if (nlh->nlmsg_flags & NLM_F_EXCL)
69 return -EEXIST;
70
71 matching = nfacct;
72 break;
73 }
74
75 if (matching) {
76 if (nlh->nlmsg_flags & NLM_F_REPLACE) {
77 /* reset counters if you request a replacement. */
78 atomic64_set(&matching->pkts, 0);
79 atomic64_set(&matching->bytes, 0);
80 smp_mb__before_atomic();
81 /* reset overquota flag if quota is enabled. */
82 if ((matching->flags & NFACCT_F_QUOTA))
83 clear_bit(NFACCT_OVERQUOTA_BIT,
84 &matching->flags);
85 return 0;
86 }
87 return -EBUSY;
88 }
89
90 if (tb[NFACCT_FLAGS]) {
91 flags = ntohl(nla_get_be32(tb[NFACCT_FLAGS]));
92 if (flags & ~NFACCT_F_QUOTA)
93 return -EOPNOTSUPP;
94 if ((flags & NFACCT_F_QUOTA) == NFACCT_F_QUOTA)
95 return -EINVAL;
96 if (flags & NFACCT_F_OVERQUOTA)
97 return -EINVAL;
98 if ((flags & NFACCT_F_QUOTA) && !tb[NFACCT_QUOTA])
99 return -EINVAL;
100
101 size += sizeof(u64);
102 }
103
104 nfacct = kzalloc(sizeof(struct nf_acct) + size, GFP_KERNEL);
105 if (nfacct == NULL)
106 return -ENOMEM;
107
108 if (flags & NFACCT_F_QUOTA) {
109 u64 *quota = (u64 *)nfacct->data;
110
111 *quota = be64_to_cpu(nla_get_be64(tb[NFACCT_QUOTA]));
112 nfacct->flags = flags;
113 }
114
115 nla_strlcpy(nfacct->name, tb[NFACCT_NAME], NFACCT_NAME_MAX);
116
117 if (tb[NFACCT_BYTES]) {
118 atomic64_set(&nfacct->bytes,
119 be64_to_cpu(nla_get_be64(tb[NFACCT_BYTES])));
120 }
121 if (tb[NFACCT_PKTS]) {
122 atomic64_set(&nfacct->pkts,
123 be64_to_cpu(nla_get_be64(tb[NFACCT_PKTS])));
124 }
125 refcount_set(&nfacct->refcnt, 1);
126 list_add_tail_rcu(&nfacct->head, &net->nfnl_acct_list);
127 return 0;
128 }
129
130 static int
nfnl_acct_fill_info(struct sk_buff * skb,u32 portid,u32 seq,u32 type,int event,struct nf_acct * acct)131 nfnl_acct_fill_info(struct sk_buff *skb, u32 portid, u32 seq, u32 type,
132 int event, struct nf_acct *acct)
133 {
134 struct nlmsghdr *nlh;
135 struct nfgenmsg *nfmsg;
136 unsigned int flags = portid ? NLM_F_MULTI : 0;
137 u64 pkts, bytes;
138 u32 old_flags;
139
140 event = nfnl_msg_type(NFNL_SUBSYS_ACCT, event);
141 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nfmsg), flags);
142 if (nlh == NULL)
143 goto nlmsg_failure;
144
145 nfmsg = nlmsg_data(nlh);
146 nfmsg->nfgen_family = AF_UNSPEC;
147 nfmsg->version = NFNETLINK_V0;
148 nfmsg->res_id = 0;
149
150 if (nla_put_string(skb, NFACCT_NAME, acct->name))
151 goto nla_put_failure;
152
153 old_flags = acct->flags;
154 if (type == NFNL_MSG_ACCT_GET_CTRZERO) {
155 pkts = atomic64_xchg(&acct->pkts, 0);
156 bytes = atomic64_xchg(&acct->bytes, 0);
157 smp_mb__before_atomic();
158 if (acct->flags & NFACCT_F_QUOTA)
159 clear_bit(NFACCT_OVERQUOTA_BIT, &acct->flags);
160 } else {
161 pkts = atomic64_read(&acct->pkts);
162 bytes = atomic64_read(&acct->bytes);
163 }
164 if (nla_put_be64(skb, NFACCT_PKTS, cpu_to_be64(pkts),
165 NFACCT_PAD) ||
166 nla_put_be64(skb, NFACCT_BYTES, cpu_to_be64(bytes),
167 NFACCT_PAD) ||
168 nla_put_be32(skb, NFACCT_USE, htonl(refcount_read(&acct->refcnt))))
169 goto nla_put_failure;
170 if (acct->flags & NFACCT_F_QUOTA) {
171 u64 *quota = (u64 *)acct->data;
172
173 if (nla_put_be32(skb, NFACCT_FLAGS, htonl(old_flags)) ||
174 nla_put_be64(skb, NFACCT_QUOTA, cpu_to_be64(*quota),
175 NFACCT_PAD))
176 goto nla_put_failure;
177 }
178 nlmsg_end(skb, nlh);
179 return skb->len;
180
181 nlmsg_failure:
182 nla_put_failure:
183 nlmsg_cancel(skb, nlh);
184 return -1;
185 }
186
187 static int
nfnl_acct_dump(struct sk_buff * skb,struct netlink_callback * cb)188 nfnl_acct_dump(struct sk_buff *skb, struct netlink_callback *cb)
189 {
190 struct net *net = sock_net(skb->sk);
191 struct nf_acct *cur, *last;
192 const struct nfacct_filter *filter = cb->data;
193
194 if (cb->args[2])
195 return 0;
196
197 last = (struct nf_acct *)cb->args[1];
198 if (cb->args[1])
199 cb->args[1] = 0;
200
201 rcu_read_lock();
202 list_for_each_entry_rcu(cur, &net->nfnl_acct_list, head) {
203 if (last) {
204 if (cur != last)
205 continue;
206
207 last = NULL;
208 }
209
210 if (filter && (cur->flags & filter->mask) != filter->value)
211 continue;
212
213 if (nfnl_acct_fill_info(skb, NETLINK_CB(cb->skb).portid,
214 cb->nlh->nlmsg_seq,
215 NFNL_MSG_TYPE(cb->nlh->nlmsg_type),
216 NFNL_MSG_ACCT_NEW, cur) < 0) {
217 cb->args[1] = (unsigned long)cur;
218 break;
219 }
220 }
221 if (!cb->args[1])
222 cb->args[2] = 1;
223 rcu_read_unlock();
224 return skb->len;
225 }
226
nfnl_acct_done(struct netlink_callback * cb)227 static int nfnl_acct_done(struct netlink_callback *cb)
228 {
229 kfree(cb->data);
230 return 0;
231 }
232
233 static const struct nla_policy filter_policy[NFACCT_FILTER_MAX + 1] = {
234 [NFACCT_FILTER_MASK] = { .type = NLA_U32 },
235 [NFACCT_FILTER_VALUE] = { .type = NLA_U32 },
236 };
237
nfnl_acct_start(struct netlink_callback * cb)238 static int nfnl_acct_start(struct netlink_callback *cb)
239 {
240 const struct nlattr *const attr = cb->data;
241 struct nlattr *tb[NFACCT_FILTER_MAX + 1];
242 struct nfacct_filter *filter;
243 int err;
244
245 if (!attr)
246 return 0;
247
248 err = nla_parse_nested_deprecated(tb, NFACCT_FILTER_MAX, attr,
249 filter_policy, NULL);
250 if (err < 0)
251 return err;
252
253 if (!tb[NFACCT_FILTER_MASK] || !tb[NFACCT_FILTER_VALUE])
254 return -EINVAL;
255
256 filter = kzalloc(sizeof(struct nfacct_filter), GFP_KERNEL);
257 if (!filter)
258 return -ENOMEM;
259
260 filter->mask = ntohl(nla_get_be32(tb[NFACCT_FILTER_MASK]));
261 filter->value = ntohl(nla_get_be32(tb[NFACCT_FILTER_VALUE]));
262 cb->data = filter;
263
264 return 0;
265 }
266
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)267 static int nfnl_acct_get(struct net *net, struct sock *nfnl,
268 struct sk_buff *skb, const struct nlmsghdr *nlh,
269 const struct nlattr * const tb[],
270 struct netlink_ext_ack *extack)
271 {
272 int ret = -ENOENT;
273 struct nf_acct *cur;
274 char *acct_name;
275
276 if (nlh->nlmsg_flags & NLM_F_DUMP) {
277 struct netlink_dump_control c = {
278 .dump = nfnl_acct_dump,
279 .start = nfnl_acct_start,
280 .done = nfnl_acct_done,
281 .data = (void *)tb[NFACCT_FILTER],
282 };
283
284 return netlink_dump_start(nfnl, skb, nlh, &c);
285 }
286
287 if (!tb[NFACCT_NAME])
288 return -EINVAL;
289 acct_name = nla_data(tb[NFACCT_NAME]);
290
291 list_for_each_entry(cur, &net->nfnl_acct_list, head) {
292 struct sk_buff *skb2;
293
294 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
295 continue;
296
297 skb2 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
298 if (skb2 == NULL) {
299 ret = -ENOMEM;
300 break;
301 }
302
303 ret = nfnl_acct_fill_info(skb2, NETLINK_CB(skb).portid,
304 nlh->nlmsg_seq,
305 NFNL_MSG_TYPE(nlh->nlmsg_type),
306 NFNL_MSG_ACCT_NEW, cur);
307 if (ret <= 0) {
308 kfree_skb(skb2);
309 break;
310 }
311 ret = netlink_unicast(nfnl, skb2, NETLINK_CB(skb).portid,
312 MSG_DONTWAIT);
313 if (ret > 0)
314 ret = 0;
315
316 /* this avoids a loop in nfnetlink. */
317 return ret == -EAGAIN ? -ENOBUFS : ret;
318 }
319 return ret;
320 }
321
322 /* try to delete object, fail if it is still in use. */
nfnl_acct_try_del(struct nf_acct * cur)323 static int nfnl_acct_try_del(struct nf_acct *cur)
324 {
325 int ret = 0;
326
327 /* We want to avoid races with nfnl_acct_put. So only when the current
328 * refcnt is 1, we decrease it to 0.
329 */
330 if (refcount_dec_if_one(&cur->refcnt)) {
331 /* We are protected by nfnl mutex. */
332 list_del_rcu(&cur->head);
333 kfree_rcu(cur, rcu_head);
334 } else {
335 ret = -EBUSY;
336 }
337 return ret;
338 }
339
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)340 static int nfnl_acct_del(struct net *net, struct sock *nfnl,
341 struct sk_buff *skb, const struct nlmsghdr *nlh,
342 const struct nlattr * const tb[],
343 struct netlink_ext_ack *extack)
344 {
345 struct nf_acct *cur, *tmp;
346 int ret = -ENOENT;
347 char *acct_name;
348
349 if (!tb[NFACCT_NAME]) {
350 list_for_each_entry_safe(cur, tmp, &net->nfnl_acct_list, head)
351 nfnl_acct_try_del(cur);
352
353 return 0;
354 }
355 acct_name = nla_data(tb[NFACCT_NAME]);
356
357 list_for_each_entry(cur, &net->nfnl_acct_list, head) {
358 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX) != 0)
359 continue;
360
361 ret = nfnl_acct_try_del(cur);
362 if (ret < 0)
363 return ret;
364
365 break;
366 }
367 return ret;
368 }
369
370 static const struct nla_policy nfnl_acct_policy[NFACCT_MAX+1] = {
371 [NFACCT_NAME] = { .type = NLA_NUL_STRING, .len = NFACCT_NAME_MAX-1 },
372 [NFACCT_BYTES] = { .type = NLA_U64 },
373 [NFACCT_PKTS] = { .type = NLA_U64 },
374 [NFACCT_FLAGS] = { .type = NLA_U32 },
375 [NFACCT_QUOTA] = { .type = NLA_U64 },
376 [NFACCT_FILTER] = {.type = NLA_NESTED },
377 };
378
379 static const struct nfnl_callback nfnl_acct_cb[NFNL_MSG_ACCT_MAX] = {
380 [NFNL_MSG_ACCT_NEW] = { .call = nfnl_acct_new,
381 .attr_count = NFACCT_MAX,
382 .policy = nfnl_acct_policy },
383 [NFNL_MSG_ACCT_GET] = { .call = nfnl_acct_get,
384 .attr_count = NFACCT_MAX,
385 .policy = nfnl_acct_policy },
386 [NFNL_MSG_ACCT_GET_CTRZERO] = { .call = nfnl_acct_get,
387 .attr_count = NFACCT_MAX,
388 .policy = nfnl_acct_policy },
389 [NFNL_MSG_ACCT_DEL] = { .call = nfnl_acct_del,
390 .attr_count = NFACCT_MAX,
391 .policy = nfnl_acct_policy },
392 };
393
394 static const struct nfnetlink_subsystem nfnl_acct_subsys = {
395 .name = "acct",
396 .subsys_id = NFNL_SUBSYS_ACCT,
397 .cb_count = NFNL_MSG_ACCT_MAX,
398 .cb = nfnl_acct_cb,
399 };
400
401 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ACCT);
402
nfnl_acct_find_get(struct net * net,const char * acct_name)403 struct nf_acct *nfnl_acct_find_get(struct net *net, const char *acct_name)
404 {
405 struct nf_acct *cur, *acct = NULL;
406
407 rcu_read_lock();
408 list_for_each_entry_rcu(cur, &net->nfnl_acct_list, head) {
409 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
410 continue;
411
412 if (!try_module_get(THIS_MODULE))
413 goto err;
414
415 if (!refcount_inc_not_zero(&cur->refcnt)) {
416 module_put(THIS_MODULE);
417 goto err;
418 }
419
420 acct = cur;
421 break;
422 }
423 err:
424 rcu_read_unlock();
425 return acct;
426 }
427 EXPORT_SYMBOL_GPL(nfnl_acct_find_get);
428
nfnl_acct_put(struct nf_acct * acct)429 void nfnl_acct_put(struct nf_acct *acct)
430 {
431 if (refcount_dec_and_test(&acct->refcnt))
432 kfree_rcu(acct, rcu_head);
433
434 module_put(THIS_MODULE);
435 }
436 EXPORT_SYMBOL_GPL(nfnl_acct_put);
437
nfnl_acct_update(const struct sk_buff * skb,struct nf_acct * nfacct)438 void nfnl_acct_update(const struct sk_buff *skb, struct nf_acct *nfacct)
439 {
440 atomic64_inc(&nfacct->pkts);
441 atomic64_add(skb->len, &nfacct->bytes);
442 }
443 EXPORT_SYMBOL_GPL(nfnl_acct_update);
444
nfnl_overquota_report(struct net * net,struct nf_acct * nfacct)445 static void nfnl_overquota_report(struct net *net, struct nf_acct *nfacct)
446 {
447 int ret;
448 struct sk_buff *skb;
449
450 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
451 if (skb == NULL)
452 return;
453
454 ret = nfnl_acct_fill_info(skb, 0, 0, NFNL_MSG_ACCT_OVERQUOTA, 0,
455 nfacct);
456 if (ret <= 0) {
457 kfree_skb(skb);
458 return;
459 }
460 netlink_broadcast(net->nfnl, skb, 0, NFNLGRP_ACCT_QUOTA,
461 GFP_ATOMIC);
462 }
463
nfnl_acct_overquota(struct net * net,struct nf_acct * nfacct)464 int nfnl_acct_overquota(struct net *net, struct nf_acct *nfacct)
465 {
466 u64 now;
467 u64 *quota;
468 int ret = NFACCT_UNDERQUOTA;
469
470 /* no place here if we don't have a quota */
471 if (!(nfacct->flags & NFACCT_F_QUOTA))
472 return NFACCT_NO_QUOTA;
473
474 quota = (u64 *)nfacct->data;
475 now = (nfacct->flags & NFACCT_F_QUOTA_PKTS) ?
476 atomic64_read(&nfacct->pkts) : atomic64_read(&nfacct->bytes);
477
478 ret = now > *quota;
479
480 if (now >= *quota &&
481 !test_and_set_bit(NFACCT_OVERQUOTA_BIT, &nfacct->flags)) {
482 nfnl_overquota_report(net, nfacct);
483 }
484
485 return ret;
486 }
487 EXPORT_SYMBOL_GPL(nfnl_acct_overquota);
488
nfnl_acct_net_init(struct net * net)489 static int __net_init nfnl_acct_net_init(struct net *net)
490 {
491 INIT_LIST_HEAD(&net->nfnl_acct_list);
492
493 return 0;
494 }
495
nfnl_acct_net_exit(struct net * net)496 static void __net_exit nfnl_acct_net_exit(struct net *net)
497 {
498 struct nf_acct *cur, *tmp;
499
500 list_for_each_entry_safe(cur, tmp, &net->nfnl_acct_list, head) {
501 list_del_rcu(&cur->head);
502
503 if (refcount_dec_and_test(&cur->refcnt))
504 kfree_rcu(cur, rcu_head);
505 }
506 }
507
508 static struct pernet_operations nfnl_acct_ops = {
509 .init = nfnl_acct_net_init,
510 .exit = nfnl_acct_net_exit,
511 };
512
nfnl_acct_init(void)513 static int __init nfnl_acct_init(void)
514 {
515 int ret;
516
517 ret = register_pernet_subsys(&nfnl_acct_ops);
518 if (ret < 0) {
519 pr_err("nfnl_acct_init: failed to register pernet ops\n");
520 goto err_out;
521 }
522
523 ret = nfnetlink_subsys_register(&nfnl_acct_subsys);
524 if (ret < 0) {
525 pr_err("nfnl_acct_init: cannot register with nfnetlink.\n");
526 goto cleanup_pernet;
527 }
528 return 0;
529
530 cleanup_pernet:
531 unregister_pernet_subsys(&nfnl_acct_ops);
532 err_out:
533 return ret;
534 }
535
nfnl_acct_exit(void)536 static void __exit nfnl_acct_exit(void)
537 {
538 nfnetlink_subsys_unregister(&nfnl_acct_subsys);
539 unregister_pernet_subsys(&nfnl_acct_ops);
540 }
541
542 module_init(nfnl_acct_init);
543 module_exit(nfnl_acct_exit);
544