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#include <net/netns/generic.h>
20
21#include <linux/netfilter.h>
22#include <linux/netfilter/nfnetlink.h>
23#include <linux/netfilter/nfnetlink_acct.h>
24
25MODULE_LICENSE("GPL");
26MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
27MODULE_DESCRIPTION("nfacct: Extended Netfilter accounting infrastructure");
28
29struct nf_acct {
30 atomic64_t pkts;
31 atomic64_t bytes;
32 unsigned long flags;
33 struct list_head head;
34 refcount_t refcnt;
35 char name[NFACCT_NAME_MAX];
36 struct rcu_head rcu_head;
37 char data[];
38};
39
40struct nfacct_filter {
41 u32 value;
42 u32 mask;
43};
44
45struct nfnl_acct_net {
46 struct list_head nfnl_acct_list;
47};
48
49static unsigned int nfnl_acct_net_id __read_mostly;
50
51static inline struct nfnl_acct_net *nfnl_acct_pernet(struct net *net)
52{
53 return net_generic(net, id: nfnl_acct_net_id);
54}
55
56#define NFACCT_F_QUOTA (NFACCT_F_QUOTA_PKTS | NFACCT_F_QUOTA_BYTES)
57#define NFACCT_OVERQUOTA_BIT 2 /* NFACCT_F_OVERQUOTA */
58
59static int nfnl_acct_new(struct sk_buff *skb, const struct nfnl_info *info,
60 const struct nlattr * const tb[])
61{
62 struct nfnl_acct_net *nfnl_acct_net = nfnl_acct_pernet(net: info->net);
63 struct nf_acct *nfacct, *matching = NULL;
64 unsigned int size = 0;
65 char *acct_name;
66 u32 flags = 0;
67
68 if (!tb[NFACCT_NAME])
69 return -EINVAL;
70
71 acct_name = nla_data(nla: tb[NFACCT_NAME]);
72 if (strlen(acct_name) == 0)
73 return -EINVAL;
74
75 list_for_each_entry(nfacct, &nfnl_acct_net->nfnl_acct_list, head) {
76 if (strncmp(nfacct->name, acct_name, NFACCT_NAME_MAX) != 0)
77 continue;
78
79 if (info->nlh->nlmsg_flags & NLM_F_EXCL)
80 return -EEXIST;
81
82 matching = nfacct;
83 break;
84 }
85
86 if (matching) {
87 if (info->nlh->nlmsg_flags & NLM_F_REPLACE) {
88 /* reset counters if you request a replacement. */
89 atomic64_set(v: &matching->pkts, i: 0);
90 atomic64_set(v: &matching->bytes, i: 0);
91 smp_mb__before_atomic();
92 /* reset overquota flag if quota is enabled. */
93 if ((matching->flags & NFACCT_F_QUOTA))
94 clear_bit(NFACCT_OVERQUOTA_BIT,
95 addr: &matching->flags);
96 return 0;
97 }
98 return -EBUSY;
99 }
100
101 if (tb[NFACCT_FLAGS]) {
102 flags = ntohl(nla_get_be32(tb[NFACCT_FLAGS]));
103 if (flags & ~NFACCT_F_QUOTA)
104 return -EOPNOTSUPP;
105 if ((flags & NFACCT_F_QUOTA) == NFACCT_F_QUOTA)
106 return -EINVAL;
107 if (flags & NFACCT_F_OVERQUOTA)
108 return -EINVAL;
109 if ((flags & NFACCT_F_QUOTA) && !tb[NFACCT_QUOTA])
110 return -EINVAL;
111
112 size += sizeof(u64);
113 }
114
115 nfacct = kzalloc(sizeof(struct nf_acct) + size, GFP_KERNEL);
116 if (nfacct == NULL)
117 return -ENOMEM;
118
119 if (flags & NFACCT_F_QUOTA) {
120 u64 *quota = (u64 *)nfacct->data;
121
122 *quota = be64_to_cpu(nla_get_be64(tb[NFACCT_QUOTA]));
123 nfacct->flags = flags;
124 }
125
126 nla_strscpy(dst: nfacct->name, nla: tb[NFACCT_NAME], NFACCT_NAME_MAX);
127
128 if (tb[NFACCT_BYTES]) {
129 atomic64_set(v: &nfacct->bytes,
130 be64_to_cpu(nla_get_be64(tb[NFACCT_BYTES])));
131 }
132 if (tb[NFACCT_PKTS]) {
133 atomic64_set(v: &nfacct->pkts,
134 be64_to_cpu(nla_get_be64(tb[NFACCT_PKTS])));
135 }
136 refcount_set(r: &nfacct->refcnt, n: 1);
137 list_add_tail_rcu(new: &nfacct->head, head: &nfnl_acct_net->nfnl_acct_list);
138 return 0;
139}
140
141static int
142nfnl_acct_fill_info(struct sk_buff *skb, u32 portid, u32 seq, u32 type,
143 int event, struct nf_acct *acct)
144{
145 struct nlmsghdr *nlh;
146 unsigned int flags = portid ? NLM_F_MULTI : 0;
147 u64 pkts, bytes;
148 u32 old_flags;
149
150 event = nfnl_msg_type(NFNL_SUBSYS_ACCT, msg_type: event);
151 nlh = nfnl_msg_put(skb, portid, seq, type: event, flags, AF_UNSPEC,
152 NFNETLINK_V0, res_id: 0);
153 if (!nlh)
154 goto nlmsg_failure;
155
156 if (nla_put_string(skb, attrtype: NFACCT_NAME, str: acct->name))
157 goto nla_put_failure;
158
159 old_flags = acct->flags;
160 if (type == NFNL_MSG_ACCT_GET_CTRZERO) {
161 pkts = atomic64_xchg(v: &acct->pkts, new: 0);
162 bytes = atomic64_xchg(v: &acct->bytes, new: 0);
163 smp_mb__before_atomic();
164 if (acct->flags & NFACCT_F_QUOTA)
165 clear_bit(NFACCT_OVERQUOTA_BIT, addr: &acct->flags);
166 } else {
167 pkts = atomic64_read(v: &acct->pkts);
168 bytes = atomic64_read(v: &acct->bytes);
169 }
170 if (nla_put_be64(skb, attrtype: NFACCT_PKTS, cpu_to_be64(pkts),
171 padattr: NFACCT_PAD) ||
172 nla_put_be64(skb, attrtype: NFACCT_BYTES, cpu_to_be64(bytes),
173 padattr: NFACCT_PAD) ||
174 nla_put_be32(skb, attrtype: NFACCT_USE, htonl(refcount_read(&acct->refcnt))))
175 goto nla_put_failure;
176 if (acct->flags & NFACCT_F_QUOTA) {
177 u64 *quota = (u64 *)acct->data;
178
179 if (nla_put_be32(skb, attrtype: NFACCT_FLAGS, htonl(old_flags)) ||
180 nla_put_be64(skb, attrtype: NFACCT_QUOTA, cpu_to_be64(*quota),
181 padattr: NFACCT_PAD))
182 goto nla_put_failure;
183 }
184 nlmsg_end(skb, nlh);
185 return skb->len;
186
187nlmsg_failure:
188nla_put_failure:
189 nlmsg_cancel(skb, nlh);
190 return -1;
191}
192
193static int
194nfnl_acct_dump(struct sk_buff *skb, struct netlink_callback *cb)
195{
196 struct net *net = sock_net(sk: skb->sk);
197 struct nfnl_acct_net *nfnl_acct_net = nfnl_acct_pernet(net);
198 struct nf_acct *cur, *last;
199 const struct nfacct_filter *filter = cb->data;
200
201 if (cb->args[2])
202 return 0;
203
204 last = (struct nf_acct *)cb->args[1];
205 if (cb->args[1])
206 cb->args[1] = 0;
207
208 rcu_read_lock();
209 list_for_each_entry_rcu(cur, &nfnl_acct_net->nfnl_acct_list, head) {
210 if (last) {
211 if (cur != last)
212 continue;
213
214 last = NULL;
215 }
216
217 if (filter && (cur->flags & filter->mask) != filter->value)
218 continue;
219
220 if (nfnl_acct_fill_info(skb, NETLINK_CB(cb->skb).portid,
221 seq: cb->nlh->nlmsg_seq,
222 NFNL_MSG_TYPE(cb->nlh->nlmsg_type),
223 event: NFNL_MSG_ACCT_NEW, acct: cur) < 0) {
224 cb->args[1] = (unsigned long)cur;
225 break;
226 }
227 }
228 if (!cb->args[1])
229 cb->args[2] = 1;
230 rcu_read_unlock();
231 return skb->len;
232}
233
234static int nfnl_acct_done(struct netlink_callback *cb)
235{
236 kfree(objp: cb->data);
237 return 0;
238}
239
240static const struct nla_policy filter_policy[NFACCT_FILTER_MAX + 1] = {
241 [NFACCT_FILTER_MASK] = { .type = NLA_U32 },
242 [NFACCT_FILTER_VALUE] = { .type = NLA_U32 },
243};
244
245static int nfnl_acct_start(struct netlink_callback *cb)
246{
247 const struct nlattr *const attr = cb->data;
248 struct nlattr *tb[NFACCT_FILTER_MAX + 1];
249 struct nfacct_filter *filter;
250 int err;
251
252 if (!attr)
253 return 0;
254
255 err = nla_parse_nested_deprecated(tb, NFACCT_FILTER_MAX, nla: attr,
256 policy: filter_policy, NULL);
257 if (err < 0)
258 return err;
259
260 if (!tb[NFACCT_FILTER_MASK] || !tb[NFACCT_FILTER_VALUE])
261 return -EINVAL;
262
263 filter = kzalloc(sizeof(struct nfacct_filter), GFP_KERNEL);
264 if (!filter)
265 return -ENOMEM;
266
267 filter->mask = ntohl(nla_get_be32(tb[NFACCT_FILTER_MASK]));
268 filter->value = ntohl(nla_get_be32(tb[NFACCT_FILTER_VALUE]));
269 cb->data = filter;
270
271 return 0;
272}
273
274static int nfnl_acct_get(struct sk_buff *skb, const struct nfnl_info *info,
275 const struct nlattr * const tb[])
276{
277 struct nfnl_acct_net *nfnl_acct_net = nfnl_acct_pernet(net: info->net);
278 int ret = -ENOENT;
279 struct nf_acct *cur;
280 char *acct_name;
281
282 if (info->nlh->nlmsg_flags & NLM_F_DUMP) {
283 struct netlink_dump_control c = {
284 .dump = nfnl_acct_dump,
285 .start = nfnl_acct_start,
286 .done = nfnl_acct_done,
287 .data = (void *)tb[NFACCT_FILTER],
288 };
289
290 return netlink_dump_start(ssk: info->sk, skb, nlh: info->nlh, control: &c);
291 }
292
293 if (!tb[NFACCT_NAME])
294 return -EINVAL;
295 acct_name = nla_data(nla: tb[NFACCT_NAME]);
296
297 list_for_each_entry(cur, &nfnl_acct_net->nfnl_acct_list, head) {
298 struct sk_buff *skb2;
299
300 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
301 continue;
302
303 skb2 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
304 if (skb2 == NULL) {
305 ret = -ENOMEM;
306 break;
307 }
308
309 ret = nfnl_acct_fill_info(skb: skb2, NETLINK_CB(skb).portid,
310 seq: info->nlh->nlmsg_seq,
311 NFNL_MSG_TYPE(info->nlh->nlmsg_type),
312 event: NFNL_MSG_ACCT_NEW, acct: cur);
313 if (ret <= 0) {
314 kfree_skb(skb: skb2);
315 break;
316 }
317
318 ret = nfnetlink_unicast(skb: skb2, net: info->net, NETLINK_CB(skb).portid);
319 break;
320 }
321
322 return ret;
323}
324
325/* try to delete object, fail if it is still in use. */
326static 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(r: &cur->refcnt)) {
334 /* We are protected by nfnl mutex. */
335 list_del_rcu(entry: &cur->head);
336 kfree_rcu(cur, rcu_head);
337 } else {
338 ret = -EBUSY;
339 }
340 return ret;
341}
342
343static int nfnl_acct_del(struct sk_buff *skb, const struct nfnl_info *info,
344 const struct nlattr * const tb[])
345{
346 struct nfnl_acct_net *nfnl_acct_net = nfnl_acct_pernet(net: info->net);
347 struct nf_acct *cur, *tmp;
348 int ret = -ENOENT;
349 char *acct_name;
350
351 if (!tb[NFACCT_NAME]) {
352 list_for_each_entry_safe(cur, tmp, &nfnl_acct_net->nfnl_acct_list, head)
353 nfnl_acct_try_del(cur);
354
355 return 0;
356 }
357 acct_name = nla_data(nla: tb[NFACCT_NAME]);
358
359 list_for_each_entry(cur, &nfnl_acct_net->nfnl_acct_list, head) {
360 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX) != 0)
361 continue;
362
363 ret = nfnl_acct_try_del(cur);
364 if (ret < 0)
365 return ret;
366
367 break;
368 }
369 return ret;
370}
371
372static const struct nla_policy nfnl_acct_policy[NFACCT_MAX+1] = {
373 [NFACCT_NAME] = { .type = NLA_NUL_STRING, .len = NFACCT_NAME_MAX-1 },
374 [NFACCT_BYTES] = { .type = NLA_U64 },
375 [NFACCT_PKTS] = { .type = NLA_U64 },
376 [NFACCT_FLAGS] = { .type = NLA_U32 },
377 [NFACCT_QUOTA] = { .type = NLA_U64 },
378 [NFACCT_FILTER] = {.type = NLA_NESTED },
379};
380
381static const struct nfnl_callback nfnl_acct_cb[NFNL_MSG_ACCT_MAX] = {
382 [NFNL_MSG_ACCT_NEW] = {
383 .call = nfnl_acct_new,
384 .type = NFNL_CB_MUTEX,
385 .attr_count = NFACCT_MAX,
386 .policy = nfnl_acct_policy
387 },
388 [NFNL_MSG_ACCT_GET] = {
389 .call = nfnl_acct_get,
390 .type = NFNL_CB_MUTEX,
391 .attr_count = NFACCT_MAX,
392 .policy = nfnl_acct_policy
393 },
394 [NFNL_MSG_ACCT_GET_CTRZERO] = {
395 .call = nfnl_acct_get,
396 .type = NFNL_CB_MUTEX,
397 .attr_count = NFACCT_MAX,
398 .policy = nfnl_acct_policy
399 },
400 [NFNL_MSG_ACCT_DEL] = {
401 .call = nfnl_acct_del,
402 .type = NFNL_CB_MUTEX,
403 .attr_count = NFACCT_MAX,
404 .policy = nfnl_acct_policy
405 },
406};
407
408static const struct nfnetlink_subsystem nfnl_acct_subsys = {
409 .name = "acct",
410 .subsys_id = NFNL_SUBSYS_ACCT,
411 .cb_count = NFNL_MSG_ACCT_MAX,
412 .cb = nfnl_acct_cb,
413};
414
415MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ACCT);
416
417struct nf_acct *nfnl_acct_find_get(struct net *net, const char *acct_name)
418{
419 struct nfnl_acct_net *nfnl_acct_net = nfnl_acct_pernet(net);
420 struct nf_acct *cur, *acct = NULL;
421
422 rcu_read_lock();
423 list_for_each_entry_rcu(cur, &nfnl_acct_net->nfnl_acct_list, head) {
424 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
425 continue;
426
427 if (!try_module_get(THIS_MODULE))
428 goto err;
429
430 if (!refcount_inc_not_zero(r: &cur->refcnt)) {
431 module_put(THIS_MODULE);
432 goto err;
433 }
434
435 acct = cur;
436 break;
437 }
438err:
439 rcu_read_unlock();
440 return acct;
441}
442EXPORT_SYMBOL_GPL(nfnl_acct_find_get);
443
444void nfnl_acct_put(struct nf_acct *acct)
445{
446 if (refcount_dec_and_test(r: &acct->refcnt))
447 kfree_rcu(acct, rcu_head);
448
449 module_put(THIS_MODULE);
450}
451EXPORT_SYMBOL_GPL(nfnl_acct_put);
452
453void nfnl_acct_update(const struct sk_buff *skb, struct nf_acct *nfacct)
454{
455 atomic64_inc(v: &nfacct->pkts);
456 atomic64_add(i: skb->len, v: &nfacct->bytes);
457}
458EXPORT_SYMBOL_GPL(nfnl_acct_update);
459
460static void nfnl_overquota_report(struct net *net, struct nf_acct *nfacct)
461{
462 int ret;
463 struct sk_buff *skb;
464
465 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
466 if (skb == NULL)
467 return;
468
469 ret = nfnl_acct_fill_info(skb, portid: 0, seq: 0, type: NFNL_MSG_ACCT_OVERQUOTA, event: 0,
470 acct: nfacct);
471 if (ret <= 0) {
472 kfree_skb(skb);
473 return;
474 }
475 nfnetlink_broadcast(net, skb, portid: 0, NFNLGRP_ACCT_QUOTA, GFP_ATOMIC);
476}
477
478int nfnl_acct_overquota(struct net *net, struct nf_acct *nfacct)
479{
480 u64 now;
481 u64 *quota;
482 int ret = NFACCT_UNDERQUOTA;
483
484 /* no place here if we don't have a quota */
485 if (!(nfacct->flags & NFACCT_F_QUOTA))
486 return NFACCT_NO_QUOTA;
487
488 quota = (u64 *)nfacct->data;
489 now = (nfacct->flags & NFACCT_F_QUOTA_PKTS) ?
490 atomic64_read(v: &nfacct->pkts) : atomic64_read(v: &nfacct->bytes);
491
492 ret = now > *quota;
493
494 if (now >= *quota &&
495 !test_and_set_bit(NFACCT_OVERQUOTA_BIT, addr: &nfacct->flags)) {
496 nfnl_overquota_report(net, nfacct);
497 }
498
499 return ret;
500}
501EXPORT_SYMBOL_GPL(nfnl_acct_overquota);
502
503static int __net_init nfnl_acct_net_init(struct net *net)
504{
505 INIT_LIST_HEAD(list: &nfnl_acct_pernet(net)->nfnl_acct_list);
506
507 return 0;
508}
509
510static void __net_exit nfnl_acct_net_exit(struct net *net)
511{
512 struct nfnl_acct_net *nfnl_acct_net = nfnl_acct_pernet(net);
513 struct nf_acct *cur, *tmp;
514
515 list_for_each_entry_safe(cur, tmp, &nfnl_acct_net->nfnl_acct_list, head) {
516 list_del_rcu(entry: &cur->head);
517
518 if (refcount_dec_and_test(r: &cur->refcnt))
519 kfree_rcu(cur, rcu_head);
520 }
521}
522
523static struct pernet_operations nfnl_acct_ops = {
524 .init = nfnl_acct_net_init,
525 .exit = nfnl_acct_net_exit,
526 .id = &nfnl_acct_net_id,
527 .size = sizeof(struct nfnl_acct_net),
528};
529
530static int __init nfnl_acct_init(void)
531{
532 int ret;
533
534 ret = register_pernet_subsys(&nfnl_acct_ops);
535 if (ret < 0) {
536 pr_err("nfnl_acct_init: failed to register pernet ops\n");
537 goto err_out;
538 }
539
540 ret = nfnetlink_subsys_register(n: &nfnl_acct_subsys);
541 if (ret < 0) {
542 pr_err("nfnl_acct_init: cannot register with nfnetlink.\n");
543 goto cleanup_pernet;
544 }
545 return 0;
546
547cleanup_pernet:
548 unregister_pernet_subsys(&nfnl_acct_ops);
549err_out:
550 return ret;
551}
552
553static void __exit nfnl_acct_exit(void)
554{
555 nfnetlink_subsys_unregister(n: &nfnl_acct_subsys);
556 unregister_pernet_subsys(&nfnl_acct_ops);
557}
558
559module_init(nfnl_acct_init);
560module_exit(nfnl_acct_exit);
561

source code of linux/net/netfilter/nfnetlink_acct.c