1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * (C) 2011 Pablo Neira Ayuso <pablo@netfilter.org>
4 * (C) 2011 Intra2net AG <https://www.intra2net.com>
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>
19 #include <net/netns/generic.h>
21 #include <linux/netfilter.h>
22 #include <linux/netfilter/nfnetlink.h>
23 #include <linux/netfilter/nfnetlink_acct.h>
25 MODULE_LICENSE("GPL");
26 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
27 MODULE_DESCRIPTION("nfacct: Extended Netfilter accounting infrastructure");
33 struct list_head head;
35 char name[NFACCT_NAME_MAX];
36 struct rcu_head rcu_head;
40 struct nfacct_filter {
45 struct nfnl_acct_net {
46 struct list_head nfnl_acct_list;
49 static unsigned int nfnl_acct_net_id __read_mostly;
51 static inline struct nfnl_acct_net *nfnl_acct_pernet(struct net *net)
53 return net_generic(net, nfnl_acct_net_id);
56 #define NFACCT_F_QUOTA (NFACCT_F_QUOTA_PKTS | NFACCT_F_QUOTA_BYTES)
57 #define NFACCT_OVERQUOTA_BIT 2 /* NFACCT_F_OVERQUOTA */
59 static int nfnl_acct_new(struct sk_buff *skb, const struct nfnl_info *info,
60 const struct nlattr * const tb[])
62 struct nfnl_acct_net *nfnl_acct_net = nfnl_acct_pernet(info->net);
63 struct nf_acct *nfacct, *matching = NULL;
64 unsigned int size = 0;
71 acct_name = nla_data(tb[NFACCT_NAME]);
72 if (strlen(acct_name) == 0)
75 list_for_each_entry(nfacct, &nfnl_acct_net->nfnl_acct_list, head) {
76 if (strncmp(nfacct->name, acct_name, NFACCT_NAME_MAX) != 0)
79 if (info->nlh->nlmsg_flags & NLM_F_EXCL)
87 if (info->nlh->nlmsg_flags & NLM_F_REPLACE) {
88 /* reset counters if you request a replacement. */
89 atomic64_set(&matching->pkts, 0);
90 atomic64_set(&matching->bytes, 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,
101 if (tb[NFACCT_FLAGS]) {
102 flags = ntohl(nla_get_be32(tb[NFACCT_FLAGS]));
103 if (flags & ~NFACCT_F_QUOTA)
105 if ((flags & NFACCT_F_QUOTA) == NFACCT_F_QUOTA)
107 if (flags & NFACCT_F_OVERQUOTA)
109 if ((flags & NFACCT_F_QUOTA) && !tb[NFACCT_QUOTA])
115 nfacct = kzalloc(sizeof(struct nf_acct) + size, GFP_KERNEL);
119 if (flags & NFACCT_F_QUOTA) {
120 u64 *quota = (u64 *)nfacct->data;
122 *quota = be64_to_cpu(nla_get_be64(tb[NFACCT_QUOTA]));
123 nfacct->flags = flags;
126 nla_strscpy(nfacct->name, tb[NFACCT_NAME], NFACCT_NAME_MAX);
128 if (tb[NFACCT_BYTES]) {
129 atomic64_set(&nfacct->bytes,
130 be64_to_cpu(nla_get_be64(tb[NFACCT_BYTES])));
132 if (tb[NFACCT_PKTS]) {
133 atomic64_set(&nfacct->pkts,
134 be64_to_cpu(nla_get_be64(tb[NFACCT_PKTS])));
136 refcount_set(&nfacct->refcnt, 1);
137 list_add_tail_rcu(&nfacct->head, &nfnl_acct_net->nfnl_acct_list);
142 nfnl_acct_fill_info(struct sk_buff *skb, u32 portid, u32 seq, u32 type,
143 int event, struct nf_acct *acct)
145 struct nlmsghdr *nlh;
146 unsigned int flags = portid ? NLM_F_MULTI : 0;
150 event = nfnl_msg_type(NFNL_SUBSYS_ACCT, event);
151 nlh = nfnl_msg_put(skb, portid, seq, event, flags, AF_UNSPEC,
156 if (nla_put_string(skb, NFACCT_NAME, acct->name))
157 goto nla_put_failure;
159 old_flags = acct->flags;
160 if (type == NFNL_MSG_ACCT_GET_CTRZERO) {
161 pkts = atomic64_xchg(&acct->pkts, 0);
162 bytes = atomic64_xchg(&acct->bytes, 0);
163 smp_mb__before_atomic();
164 if (acct->flags & NFACCT_F_QUOTA)
165 clear_bit(NFACCT_OVERQUOTA_BIT, &acct->flags);
167 pkts = atomic64_read(&acct->pkts);
168 bytes = atomic64_read(&acct->bytes);
170 if (nla_put_be64(skb, NFACCT_PKTS, cpu_to_be64(pkts),
172 nla_put_be64(skb, NFACCT_BYTES, cpu_to_be64(bytes),
174 nla_put_be32(skb, 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;
179 if (nla_put_be32(skb, NFACCT_FLAGS, htonl(old_flags)) ||
180 nla_put_be64(skb, NFACCT_QUOTA, cpu_to_be64(*quota),
182 goto nla_put_failure;
189 nlmsg_cancel(skb, nlh);
194 nfnl_acct_dump(struct sk_buff *skb, struct netlink_callback *cb)
196 struct net *net = sock_net(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;
204 last = (struct nf_acct *)cb->args[1];
209 list_for_each_entry_rcu(cur, &nfnl_acct_net->nfnl_acct_list, head) {
217 if (filter && (cur->flags & filter->mask) != filter->value)
220 if (nfnl_acct_fill_info(skb, NETLINK_CB(cb->skb).portid,
222 NFNL_MSG_TYPE(cb->nlh->nlmsg_type),
223 NFNL_MSG_ACCT_NEW, cur) < 0) {
224 cb->args[1] = (unsigned long)cur;
234 static int nfnl_acct_done(struct netlink_callback *cb)
240 static const struct nla_policy filter_policy[NFACCT_FILTER_MAX + 1] = {
241 [NFACCT_FILTER_MASK] = { .type = NLA_U32 },
242 [NFACCT_FILTER_VALUE] = { .type = NLA_U32 },
245 static int nfnl_acct_start(struct netlink_callback *cb)
247 const struct nlattr *const attr = cb->data;
248 struct nlattr *tb[NFACCT_FILTER_MAX + 1];
249 struct nfacct_filter *filter;
255 err = nla_parse_nested_deprecated(tb, NFACCT_FILTER_MAX, attr,
256 filter_policy, NULL);
260 if (!tb[NFACCT_FILTER_MASK] || !tb[NFACCT_FILTER_VALUE])
263 filter = kzalloc(sizeof(struct nfacct_filter), GFP_KERNEL);
267 filter->mask = ntohl(nla_get_be32(tb[NFACCT_FILTER_MASK]));
268 filter->value = ntohl(nla_get_be32(tb[NFACCT_FILTER_VALUE]));
274 static int nfnl_acct_get(struct sk_buff *skb, const struct nfnl_info *info,
275 const struct nlattr * const tb[])
277 struct nfnl_acct_net *nfnl_acct_net = nfnl_acct_pernet(info->net);
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],
290 return netlink_dump_start(info->sk, skb, info->nlh, &c);
293 if (!tb[NFACCT_NAME])
295 acct_name = nla_data(tb[NFACCT_NAME]);
297 list_for_each_entry(cur, &nfnl_acct_net->nfnl_acct_list, head) {
298 struct sk_buff *skb2;
300 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
303 skb2 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
309 ret = nfnl_acct_fill_info(skb2, NETLINK_CB(skb).portid,
310 info->nlh->nlmsg_seq,
311 NFNL_MSG_TYPE(info->nlh->nlmsg_type),
312 NFNL_MSG_ACCT_NEW, cur);
317 ret = netlink_unicast(info->sk, skb2, NETLINK_CB(skb).portid,
322 /* this avoids a loop in nfnetlink. */
323 return ret == -EAGAIN ? -ENOBUFS : ret;
328 /* try to delete object, fail if it is still in use. */
329 static int nfnl_acct_try_del(struct nf_acct *cur)
333 /* We want to avoid races with nfnl_acct_put. So only when the current
334 * refcnt is 1, we decrease it to 0.
336 if (refcount_dec_if_one(&cur->refcnt)) {
337 /* We are protected by nfnl mutex. */
338 list_del_rcu(&cur->head);
339 kfree_rcu(cur, rcu_head);
346 static int nfnl_acct_del(struct sk_buff *skb, const struct nfnl_info *info,
347 const struct nlattr * const tb[])
349 struct nfnl_acct_net *nfnl_acct_net = nfnl_acct_pernet(info->net);
350 struct nf_acct *cur, *tmp;
354 if (!tb[NFACCT_NAME]) {
355 list_for_each_entry_safe(cur, tmp, &nfnl_acct_net->nfnl_acct_list, head)
356 nfnl_acct_try_del(cur);
360 acct_name = nla_data(tb[NFACCT_NAME]);
362 list_for_each_entry(cur, &nfnl_acct_net->nfnl_acct_list, head) {
363 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX) != 0)
366 ret = nfnl_acct_try_del(cur);
375 static const struct nla_policy nfnl_acct_policy[NFACCT_MAX+1] = {
376 [NFACCT_NAME] = { .type = NLA_NUL_STRING, .len = NFACCT_NAME_MAX-1 },
377 [NFACCT_BYTES] = { .type = NLA_U64 },
378 [NFACCT_PKTS] = { .type = NLA_U64 },
379 [NFACCT_FLAGS] = { .type = NLA_U32 },
380 [NFACCT_QUOTA] = { .type = NLA_U64 },
381 [NFACCT_FILTER] = {.type = NLA_NESTED },
384 static const struct nfnl_callback nfnl_acct_cb[NFNL_MSG_ACCT_MAX] = {
385 [NFNL_MSG_ACCT_NEW] = {
386 .call = nfnl_acct_new,
387 .type = NFNL_CB_MUTEX,
388 .attr_count = NFACCT_MAX,
389 .policy = nfnl_acct_policy
391 [NFNL_MSG_ACCT_GET] = {
392 .call = nfnl_acct_get,
393 .type = NFNL_CB_MUTEX,
394 .attr_count = NFACCT_MAX,
395 .policy = nfnl_acct_policy
397 [NFNL_MSG_ACCT_GET_CTRZERO] = {
398 .call = nfnl_acct_get,
399 .type = NFNL_CB_MUTEX,
400 .attr_count = NFACCT_MAX,
401 .policy = nfnl_acct_policy
403 [NFNL_MSG_ACCT_DEL] = {
404 .call = nfnl_acct_del,
405 .type = NFNL_CB_MUTEX,
406 .attr_count = NFACCT_MAX,
407 .policy = nfnl_acct_policy
411 static const struct nfnetlink_subsystem nfnl_acct_subsys = {
413 .subsys_id = NFNL_SUBSYS_ACCT,
414 .cb_count = NFNL_MSG_ACCT_MAX,
418 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ACCT);
420 struct nf_acct *nfnl_acct_find_get(struct net *net, const char *acct_name)
422 struct nfnl_acct_net *nfnl_acct_net = nfnl_acct_pernet(net);
423 struct nf_acct *cur, *acct = NULL;
426 list_for_each_entry_rcu(cur, &nfnl_acct_net->nfnl_acct_list, head) {
427 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
430 if (!try_module_get(THIS_MODULE))
433 if (!refcount_inc_not_zero(&cur->refcnt)) {
434 module_put(THIS_MODULE);
445 EXPORT_SYMBOL_GPL(nfnl_acct_find_get);
447 void nfnl_acct_put(struct nf_acct *acct)
449 if (refcount_dec_and_test(&acct->refcnt))
450 kfree_rcu(acct, rcu_head);
452 module_put(THIS_MODULE);
454 EXPORT_SYMBOL_GPL(nfnl_acct_put);
456 void nfnl_acct_update(const struct sk_buff *skb, struct nf_acct *nfacct)
458 atomic64_inc(&nfacct->pkts);
459 atomic64_add(skb->len, &nfacct->bytes);
461 EXPORT_SYMBOL_GPL(nfnl_acct_update);
463 static void nfnl_overquota_report(struct net *net, struct nf_acct *nfacct)
468 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
472 ret = nfnl_acct_fill_info(skb, 0, 0, NFNL_MSG_ACCT_OVERQUOTA, 0,
478 nfnetlink_broadcast(net, skb, 0, NFNLGRP_ACCT_QUOTA, GFP_ATOMIC);
481 int nfnl_acct_overquota(struct net *net, struct nf_acct *nfacct)
485 int ret = NFACCT_UNDERQUOTA;
487 /* no place here if we don't have a quota */
488 if (!(nfacct->flags & NFACCT_F_QUOTA))
489 return NFACCT_NO_QUOTA;
491 quota = (u64 *)nfacct->data;
492 now = (nfacct->flags & NFACCT_F_QUOTA_PKTS) ?
493 atomic64_read(&nfacct->pkts) : atomic64_read(&nfacct->bytes);
498 !test_and_set_bit(NFACCT_OVERQUOTA_BIT, &nfacct->flags)) {
499 nfnl_overquota_report(net, nfacct);
504 EXPORT_SYMBOL_GPL(nfnl_acct_overquota);
506 static int __net_init nfnl_acct_net_init(struct net *net)
508 INIT_LIST_HEAD(&nfnl_acct_pernet(net)->nfnl_acct_list);
513 static void __net_exit nfnl_acct_net_exit(struct net *net)
515 struct nfnl_acct_net *nfnl_acct_net = nfnl_acct_pernet(net);
516 struct nf_acct *cur, *tmp;
518 list_for_each_entry_safe(cur, tmp, &nfnl_acct_net->nfnl_acct_list, head) {
519 list_del_rcu(&cur->head);
521 if (refcount_dec_and_test(&cur->refcnt))
522 kfree_rcu(cur, rcu_head);
526 static struct pernet_operations nfnl_acct_ops = {
527 .init = nfnl_acct_net_init,
528 .exit = nfnl_acct_net_exit,
529 .id = &nfnl_acct_net_id,
530 .size = sizeof(struct nfnl_acct_net),
533 static int __init nfnl_acct_init(void)
537 ret = register_pernet_subsys(&nfnl_acct_ops);
539 pr_err("nfnl_acct_init: failed to register pernet ops\n");
543 ret = nfnetlink_subsys_register(&nfnl_acct_subsys);
545 pr_err("nfnl_acct_init: cannot register with nfnetlink.\n");
551 unregister_pernet_subsys(&nfnl_acct_ops);
556 static void __exit nfnl_acct_exit(void)
558 nfnetlink_subsys_unregister(&nfnl_acct_subsys);
559 unregister_pernet_subsys(&nfnl_acct_ops);
562 module_init(nfnl_acct_init);
563 module_exit(nfnl_acct_exit);