1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * (C) 2011 Pablo Neira Ayuso <pablo@netfilter.org>
4 * (C) 2011 Intra2net AG <https://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[];
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 unsigned int flags = portid ? NLM_F_MULTI : 0;
136 u64 pkts, bytes;
137 u32 old_flags;
138
139 event = nfnl_msg_type(NFNL_SUBSYS_ACCT, event);
140 nlh = nfnl_msg_put(skb, portid, seq, event, flags, AF_UNSPEC,
141 NFNETLINK_V0, 0);
142 if (!nlh)
143 goto nlmsg_failure;
144
145 if (nla_put_string(skb, NFACCT_NAME, acct->name))
146 goto nla_put_failure;
147
148 old_flags = acct->flags;
149 if (type == NFNL_MSG_ACCT_GET_CTRZERO) {
150 pkts = atomic64_xchg(&acct->pkts, 0);
151 bytes = atomic64_xchg(&acct->bytes, 0);
152 smp_mb__before_atomic();
153 if (acct->flags & NFACCT_F_QUOTA)
154 clear_bit(NFACCT_OVERQUOTA_BIT, &acct->flags);
155 } else {
156 pkts = atomic64_read(&acct->pkts);
157 bytes = atomic64_read(&acct->bytes);
158 }
159 if (nla_put_be64(skb, NFACCT_PKTS, cpu_to_be64(pkts),
160 NFACCT_PAD) ||
161 nla_put_be64(skb, NFACCT_BYTES, cpu_to_be64(bytes),
162 NFACCT_PAD) ||
163 nla_put_be32(skb, NFACCT_USE, htonl(refcount_read(&acct->refcnt))))
164 goto nla_put_failure;
165 if (acct->flags & NFACCT_F_QUOTA) {
166 u64 *quota = (u64 *)acct->data;
167
168 if (nla_put_be32(skb, NFACCT_FLAGS, htonl(old_flags)) ||
169 nla_put_be64(skb, NFACCT_QUOTA, cpu_to_be64(*quota),
170 NFACCT_PAD))
171 goto nla_put_failure;
172 }
173 nlmsg_end(skb, nlh);
174 return skb->len;
175
176 nlmsg_failure:
177 nla_put_failure:
178 nlmsg_cancel(skb, nlh);
179 return -1;
180 }
181
182 static int
nfnl_acct_dump(struct sk_buff * skb,struct netlink_callback * cb)183 nfnl_acct_dump(struct sk_buff *skb, struct netlink_callback *cb)
184 {
185 struct net *net = sock_net(skb->sk);
186 struct nf_acct *cur, *last;
187 const struct nfacct_filter *filter = cb->data;
188
189 if (cb->args[2])
190 return 0;
191
192 last = (struct nf_acct *)cb->args[1];
193 if (cb->args[1])
194 cb->args[1] = 0;
195
196 rcu_read_lock();
197 list_for_each_entry_rcu(cur, &net->nfnl_acct_list, head) {
198 if (last) {
199 if (cur != last)
200 continue;
201
202 last = NULL;
203 }
204
205 if (filter && (cur->flags & filter->mask) != filter->value)
206 continue;
207
208 if (nfnl_acct_fill_info(skb, NETLINK_CB(cb->skb).portid,
209 cb->nlh->nlmsg_seq,
210 NFNL_MSG_TYPE(cb->nlh->nlmsg_type),
211 NFNL_MSG_ACCT_NEW, cur) < 0) {
212 cb->args[1] = (unsigned long)cur;
213 break;
214 }
215 }
216 if (!cb->args[1])
217 cb->args[2] = 1;
218 rcu_read_unlock();
219 return skb->len;
220 }
221
nfnl_acct_done(struct netlink_callback * cb)222 static int nfnl_acct_done(struct netlink_callback *cb)
223 {
224 kfree(cb->data);
225 return 0;
226 }
227
228 static const struct nla_policy filter_policy[NFACCT_FILTER_MAX + 1] = {
229 [NFACCT_FILTER_MASK] = { .type = NLA_U32 },
230 [NFACCT_FILTER_VALUE] = { .type = NLA_U32 },
231 };
232
nfnl_acct_start(struct netlink_callback * cb)233 static int nfnl_acct_start(struct netlink_callback *cb)
234 {
235 const struct nlattr *const attr = cb->data;
236 struct nlattr *tb[NFACCT_FILTER_MAX + 1];
237 struct nfacct_filter *filter;
238 int err;
239
240 if (!attr)
241 return 0;
242
243 err = nla_parse_nested_deprecated(tb, NFACCT_FILTER_MAX, attr,
244 filter_policy, NULL);
245 if (err < 0)
246 return err;
247
248 if (!tb[NFACCT_FILTER_MASK] || !tb[NFACCT_FILTER_VALUE])
249 return -EINVAL;
250
251 filter = kzalloc(sizeof(struct nfacct_filter), GFP_KERNEL);
252 if (!filter)
253 return -ENOMEM;
254
255 filter->mask = ntohl(nla_get_be32(tb[NFACCT_FILTER_MASK]));
256 filter->value = ntohl(nla_get_be32(tb[NFACCT_FILTER_VALUE]));
257 cb->data = filter;
258
259 return 0;
260 }
261
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)262 static int nfnl_acct_get(struct net *net, struct sock *nfnl,
263 struct sk_buff *skb, const struct nlmsghdr *nlh,
264 const struct nlattr * const tb[],
265 struct netlink_ext_ack *extack)
266 {
267 int ret = -ENOENT;
268 struct nf_acct *cur;
269 char *acct_name;
270
271 if (nlh->nlmsg_flags & NLM_F_DUMP) {
272 struct netlink_dump_control c = {
273 .dump = nfnl_acct_dump,
274 .start = nfnl_acct_start,
275 .done = nfnl_acct_done,
276 .data = (void *)tb[NFACCT_FILTER],
277 };
278
279 return netlink_dump_start(nfnl, skb, nlh, &c);
280 }
281
282 if (!tb[NFACCT_NAME])
283 return -EINVAL;
284 acct_name = nla_data(tb[NFACCT_NAME]);
285
286 list_for_each_entry(cur, &net->nfnl_acct_list, head) {
287 struct sk_buff *skb2;
288
289 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
290 continue;
291
292 skb2 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
293 if (skb2 == NULL) {
294 ret = -ENOMEM;
295 break;
296 }
297
298 ret = nfnl_acct_fill_info(skb2, NETLINK_CB(skb).portid,
299 nlh->nlmsg_seq,
300 NFNL_MSG_TYPE(nlh->nlmsg_type),
301 NFNL_MSG_ACCT_NEW, cur);
302 if (ret <= 0) {
303 kfree_skb(skb2);
304 break;
305 }
306 ret = netlink_unicast(nfnl, skb2, NETLINK_CB(skb).portid,
307 MSG_DONTWAIT);
308 if (ret > 0)
309 ret = 0;
310
311 /* this avoids a loop in nfnetlink. */
312 return ret == -EAGAIN ? -ENOBUFS : ret;
313 }
314 return ret;
315 }
316
317 /* try to delete object, fail if it is still in use. */
nfnl_acct_try_del(struct nf_acct * cur)318 static int nfnl_acct_try_del(struct nf_acct *cur)
319 {
320 int ret = 0;
321
322 /* We want to avoid races with nfnl_acct_put. So only when the current
323 * refcnt is 1, we decrease it to 0.
324 */
325 if (refcount_dec_if_one(&cur->refcnt)) {
326 /* We are protected by nfnl mutex. */
327 list_del_rcu(&cur->head);
328 kfree_rcu(cur, rcu_head);
329 } else {
330 ret = -EBUSY;
331 }
332 return ret;
333 }
334
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)335 static int nfnl_acct_del(struct net *net, struct sock *nfnl,
336 struct sk_buff *skb, const struct nlmsghdr *nlh,
337 const struct nlattr * const tb[],
338 struct netlink_ext_ack *extack)
339 {
340 struct nf_acct *cur, *tmp;
341 int ret = -ENOENT;
342 char *acct_name;
343
344 if (!tb[NFACCT_NAME]) {
345 list_for_each_entry_safe(cur, tmp, &net->nfnl_acct_list, head)
346 nfnl_acct_try_del(cur);
347
348 return 0;
349 }
350 acct_name = nla_data(tb[NFACCT_NAME]);
351
352 list_for_each_entry(cur, &net->nfnl_acct_list, head) {
353 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX) != 0)
354 continue;
355
356 ret = nfnl_acct_try_del(cur);
357 if (ret < 0)
358 return ret;
359
360 break;
361 }
362 return ret;
363 }
364
365 static const struct nla_policy nfnl_acct_policy[NFACCT_MAX+1] = {
366 [NFACCT_NAME] = { .type = NLA_NUL_STRING, .len = NFACCT_NAME_MAX-1 },
367 [NFACCT_BYTES] = { .type = NLA_U64 },
368 [NFACCT_PKTS] = { .type = NLA_U64 },
369 [NFACCT_FLAGS] = { .type = NLA_U32 },
370 [NFACCT_QUOTA] = { .type = NLA_U64 },
371 [NFACCT_FILTER] = {.type = NLA_NESTED },
372 };
373
374 static const struct nfnl_callback nfnl_acct_cb[NFNL_MSG_ACCT_MAX] = {
375 [NFNL_MSG_ACCT_NEW] = { .call = nfnl_acct_new,
376 .attr_count = NFACCT_MAX,
377 .policy = nfnl_acct_policy },
378 [NFNL_MSG_ACCT_GET] = { .call = nfnl_acct_get,
379 .attr_count = NFACCT_MAX,
380 .policy = nfnl_acct_policy },
381 [NFNL_MSG_ACCT_GET_CTRZERO] = { .call = nfnl_acct_get,
382 .attr_count = NFACCT_MAX,
383 .policy = nfnl_acct_policy },
384 [NFNL_MSG_ACCT_DEL] = { .call = nfnl_acct_del,
385 .attr_count = NFACCT_MAX,
386 .policy = nfnl_acct_policy },
387 };
388
389 static const struct nfnetlink_subsystem nfnl_acct_subsys = {
390 .name = "acct",
391 .subsys_id = NFNL_SUBSYS_ACCT,
392 .cb_count = NFNL_MSG_ACCT_MAX,
393 .cb = nfnl_acct_cb,
394 };
395
396 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ACCT);
397
nfnl_acct_find_get(struct net * net,const char * acct_name)398 struct nf_acct *nfnl_acct_find_get(struct net *net, const char *acct_name)
399 {
400 struct nf_acct *cur, *acct = NULL;
401
402 rcu_read_lock();
403 list_for_each_entry_rcu(cur, &net->nfnl_acct_list, head) {
404 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
405 continue;
406
407 if (!try_module_get(THIS_MODULE))
408 goto err;
409
410 if (!refcount_inc_not_zero(&cur->refcnt)) {
411 module_put(THIS_MODULE);
412 goto err;
413 }
414
415 acct = cur;
416 break;
417 }
418 err:
419 rcu_read_unlock();
420 return acct;
421 }
422 EXPORT_SYMBOL_GPL(nfnl_acct_find_get);
423
nfnl_acct_put(struct nf_acct * acct)424 void nfnl_acct_put(struct nf_acct *acct)
425 {
426 if (refcount_dec_and_test(&acct->refcnt))
427 kfree_rcu(acct, rcu_head);
428
429 module_put(THIS_MODULE);
430 }
431 EXPORT_SYMBOL_GPL(nfnl_acct_put);
432
nfnl_acct_update(const struct sk_buff * skb,struct nf_acct * nfacct)433 void nfnl_acct_update(const struct sk_buff *skb, struct nf_acct *nfacct)
434 {
435 atomic64_inc(&nfacct->pkts);
436 atomic64_add(skb->len, &nfacct->bytes);
437 }
438 EXPORT_SYMBOL_GPL(nfnl_acct_update);
439
nfnl_overquota_report(struct net * net,struct nf_acct * nfacct)440 static void nfnl_overquota_report(struct net *net, struct nf_acct *nfacct)
441 {
442 int ret;
443 struct sk_buff *skb;
444
445 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
446 if (skb == NULL)
447 return;
448
449 ret = nfnl_acct_fill_info(skb, 0, 0, NFNL_MSG_ACCT_OVERQUOTA, 0,
450 nfacct);
451 if (ret <= 0) {
452 kfree_skb(skb);
453 return;
454 }
455 netlink_broadcast(net->nfnl, skb, 0, NFNLGRP_ACCT_QUOTA,
456 GFP_ATOMIC);
457 }
458
nfnl_acct_overquota(struct net * net,struct nf_acct * nfacct)459 int nfnl_acct_overquota(struct net *net, struct nf_acct *nfacct)
460 {
461 u64 now;
462 u64 *quota;
463 int ret = NFACCT_UNDERQUOTA;
464
465 /* no place here if we don't have a quota */
466 if (!(nfacct->flags & NFACCT_F_QUOTA))
467 return NFACCT_NO_QUOTA;
468
469 quota = (u64 *)nfacct->data;
470 now = (nfacct->flags & NFACCT_F_QUOTA_PKTS) ?
471 atomic64_read(&nfacct->pkts) : atomic64_read(&nfacct->bytes);
472
473 ret = now > *quota;
474
475 if (now >= *quota &&
476 !test_and_set_bit(NFACCT_OVERQUOTA_BIT, &nfacct->flags)) {
477 nfnl_overquota_report(net, nfacct);
478 }
479
480 return ret;
481 }
482 EXPORT_SYMBOL_GPL(nfnl_acct_overquota);
483
nfnl_acct_net_init(struct net * net)484 static int __net_init nfnl_acct_net_init(struct net *net)
485 {
486 INIT_LIST_HEAD(&net->nfnl_acct_list);
487
488 return 0;
489 }
490
nfnl_acct_net_exit(struct net * net)491 static void __net_exit nfnl_acct_net_exit(struct net *net)
492 {
493 struct nf_acct *cur, *tmp;
494
495 list_for_each_entry_safe(cur, tmp, &net->nfnl_acct_list, head) {
496 list_del_rcu(&cur->head);
497
498 if (refcount_dec_and_test(&cur->refcnt))
499 kfree_rcu(cur, rcu_head);
500 }
501 }
502
503 static struct pernet_operations nfnl_acct_ops = {
504 .init = nfnl_acct_net_init,
505 .exit = nfnl_acct_net_exit,
506 };
507
nfnl_acct_init(void)508 static int __init nfnl_acct_init(void)
509 {
510 int ret;
511
512 ret = register_pernet_subsys(&nfnl_acct_ops);
513 if (ret < 0) {
514 pr_err("nfnl_acct_init: failed to register pernet ops\n");
515 goto err_out;
516 }
517
518 ret = nfnetlink_subsys_register(&nfnl_acct_subsys);
519 if (ret < 0) {
520 pr_err("nfnl_acct_init: cannot register with nfnetlink.\n");
521 goto cleanup_pernet;
522 }
523 return 0;
524
525 cleanup_pernet:
526 unregister_pernet_subsys(&nfnl_acct_ops);
527 err_out:
528 return ret;
529 }
530
nfnl_acct_exit(void)531 static void __exit nfnl_acct_exit(void)
532 {
533 nfnetlink_subsys_unregister(&nfnl_acct_subsys);
534 unregister_pernet_subsys(&nfnl_acct_ops);
535 }
536
537 module_init(nfnl_acct_init);
538 module_exit(nfnl_acct_exit);
539