2 * (C) 2011 Pablo Neira Ayuso <pablo@netfilter.org>
3 * (C) 2011 Intra2net AG <http://www.intra2net.com>
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).
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>
23 #include <linux/netfilter.h>
24 #include <linux/netfilter/nfnetlink.h>
25 #include <linux/netfilter/nfnetlink_acct.h>
27 MODULE_LICENSE("GPL");
28 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
29 MODULE_DESCRIPTION("nfacct: Extended Netfilter accounting infrastructure");
35 struct list_head head;
37 char name[NFACCT_NAME_MAX];
38 struct rcu_head rcu_head;
42 struct nfacct_filter {
47 #define NFACCT_F_QUOTA (NFACCT_F_QUOTA_PKTS | NFACCT_F_QUOTA_BYTES)
48 #define NFACCT_OVERQUOTA_BIT 2 /* NFACCT_F_OVERQUOTA */
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)
55 struct nf_acct *nfacct, *matching = NULL;
57 unsigned int size = 0;
63 acct_name = nla_data(tb[NFACCT_NAME]);
64 if (strlen(acct_name) == 0)
67 list_for_each_entry(nfacct, &net->nfnl_acct_list, head) {
68 if (strncmp(nfacct->name, acct_name, NFACCT_NAME_MAX) != 0)
71 if (nlh->nlmsg_flags & NLM_F_EXCL)
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,
93 if (tb[NFACCT_FLAGS]) {
94 flags = ntohl(nla_get_be32(tb[NFACCT_FLAGS]));
95 if (flags & ~NFACCT_F_QUOTA)
97 if ((flags & NFACCT_F_QUOTA) == NFACCT_F_QUOTA)
99 if (flags & NFACCT_F_OVERQUOTA)
101 if ((flags & NFACCT_F_QUOTA) && !tb[NFACCT_QUOTA])
107 nfacct = kzalloc(sizeof(struct nf_acct) + size, GFP_KERNEL);
111 if (flags & NFACCT_F_QUOTA) {
112 u64 *quota = (u64 *)nfacct->data;
114 *quota = be64_to_cpu(nla_get_be64(tb[NFACCT_QUOTA]));
115 nfacct->flags = flags;
118 strncpy(nfacct->name, nla_data(tb[NFACCT_NAME]), NFACCT_NAME_MAX);
120 if (tb[NFACCT_BYTES]) {
121 atomic64_set(&nfacct->bytes,
122 be64_to_cpu(nla_get_be64(tb[NFACCT_BYTES])));
124 if (tb[NFACCT_PKTS]) {
125 atomic64_set(&nfacct->pkts,
126 be64_to_cpu(nla_get_be64(tb[NFACCT_PKTS])));
128 refcount_set(&nfacct->refcnt, 1);
129 list_add_tail_rcu(&nfacct->head, &net->nfnl_acct_list);
134 nfnl_acct_fill_info(struct sk_buff *skb, u32 portid, u32 seq, u32 type,
135 int event, struct nf_acct *acct)
137 struct nlmsghdr *nlh;
138 struct nfgenmsg *nfmsg;
139 unsigned int flags = portid ? NLM_F_MULTI : 0;
143 event = nfnl_msg_type(NFNL_SUBSYS_ACCT, event);
144 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nfmsg), flags);
148 nfmsg = nlmsg_data(nlh);
149 nfmsg->nfgen_family = AF_UNSPEC;
150 nfmsg->version = NFNETLINK_V0;
153 if (nla_put_string(skb, NFACCT_NAME, acct->name))
154 goto nla_put_failure;
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);
164 pkts = atomic64_read(&acct->pkts);
165 bytes = atomic64_read(&acct->bytes);
167 if (nla_put_be64(skb, NFACCT_PKTS, cpu_to_be64(pkts),
169 nla_put_be64(skb, NFACCT_BYTES, cpu_to_be64(bytes),
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;
176 if (nla_put_be32(skb, NFACCT_FLAGS, htonl(old_flags)) ||
177 nla_put_be64(skb, NFACCT_QUOTA, cpu_to_be64(*quota),
179 goto nla_put_failure;
186 nlmsg_cancel(skb, nlh);
191 nfnl_acct_dump(struct sk_buff *skb, struct netlink_callback *cb)
193 struct net *net = sock_net(skb->sk);
194 struct nf_acct *cur, *last;
195 const struct nfacct_filter *filter = cb->data;
200 last = (struct nf_acct *)cb->args[1];
205 list_for_each_entry_rcu(cur, &net->nfnl_acct_list, head) {
213 if (filter && (cur->flags & filter->mask) != filter->value)
216 if (nfnl_acct_fill_info(skb, NETLINK_CB(cb->skb).portid,
218 NFNL_MSG_TYPE(cb->nlh->nlmsg_type),
219 NFNL_MSG_ACCT_NEW, cur) < 0) {
220 cb->args[1] = (unsigned long)cur;
230 static int nfnl_acct_done(struct netlink_callback *cb)
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 },
241 static struct nfacct_filter *
242 nfacct_filter_alloc(const struct nlattr * const attr)
244 struct nfacct_filter *filter;
245 struct nlattr *tb[NFACCT_FILTER_MAX + 1];
248 err = nla_parse_nested(tb, NFACCT_FILTER_MAX, attr, filter_policy,
253 if (!tb[NFACCT_FILTER_MASK] || !tb[NFACCT_FILTER_VALUE])
254 return ERR_PTR(-EINVAL);
256 filter = kzalloc(sizeof(struct nfacct_filter), GFP_KERNEL);
258 return ERR_PTR(-ENOMEM);
260 filter->mask = ntohl(nla_get_be32(tb[NFACCT_FILTER_MASK]));
261 filter->value = ntohl(nla_get_be32(tb[NFACCT_FILTER_VALUE]));
266 static int nfnl_acct_get(struct net *net, struct sock *nfnl,
267 struct sk_buff *skb, const struct nlmsghdr *nlh,
268 const struct nlattr * const tb[],
269 struct netlink_ext_ack *extack)
275 if (nlh->nlmsg_flags & NLM_F_DUMP) {
276 struct netlink_dump_control c = {
277 .dump = nfnl_acct_dump,
278 .done = nfnl_acct_done,
281 if (tb[NFACCT_FILTER]) {
282 struct nfacct_filter *filter;
284 filter = nfacct_filter_alloc(tb[NFACCT_FILTER]);
286 return PTR_ERR(filter);
290 return netlink_dump_start(nfnl, skb, nlh, &c);
293 if (!tb[NFACCT_NAME])
295 acct_name = nla_data(tb[NFACCT_NAME]);
297 list_for_each_entry(cur, &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,
311 NFNL_MSG_TYPE(nlh->nlmsg_type),
312 NFNL_MSG_ACCT_NEW, cur);
317 ret = netlink_unicast(nfnl, 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 net *net, struct sock *nfnl,
347 struct sk_buff *skb, const struct nlmsghdr *nlh,
348 const struct nlattr * const tb[],
349 struct netlink_ext_ack *extack)
351 struct nf_acct *cur, *tmp;
355 if (!tb[NFACCT_NAME]) {
356 list_for_each_entry_safe(cur, tmp, &net->nfnl_acct_list, head)
357 nfnl_acct_try_del(cur);
361 acct_name = nla_data(tb[NFACCT_NAME]);
363 list_for_each_entry(cur, &net->nfnl_acct_list, head) {
364 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX) != 0)
367 ret = nfnl_acct_try_del(cur);
376 static const struct nla_policy nfnl_acct_policy[NFACCT_MAX+1] = {
377 [NFACCT_NAME] = { .type = NLA_NUL_STRING, .len = NFACCT_NAME_MAX-1 },
378 [NFACCT_BYTES] = { .type = NLA_U64 },
379 [NFACCT_PKTS] = { .type = NLA_U64 },
380 [NFACCT_FLAGS] = { .type = NLA_U32 },
381 [NFACCT_QUOTA] = { .type = NLA_U64 },
382 [NFACCT_FILTER] = {.type = NLA_NESTED },
385 static const struct nfnl_callback nfnl_acct_cb[NFNL_MSG_ACCT_MAX] = {
386 [NFNL_MSG_ACCT_NEW] = { .call = nfnl_acct_new,
387 .attr_count = NFACCT_MAX,
388 .policy = nfnl_acct_policy },
389 [NFNL_MSG_ACCT_GET] = { .call = nfnl_acct_get,
390 .attr_count = NFACCT_MAX,
391 .policy = nfnl_acct_policy },
392 [NFNL_MSG_ACCT_GET_CTRZERO] = { .call = nfnl_acct_get,
393 .attr_count = NFACCT_MAX,
394 .policy = nfnl_acct_policy },
395 [NFNL_MSG_ACCT_DEL] = { .call = nfnl_acct_del,
396 .attr_count = NFACCT_MAX,
397 .policy = nfnl_acct_policy },
400 static const struct nfnetlink_subsystem nfnl_acct_subsys = {
402 .subsys_id = NFNL_SUBSYS_ACCT,
403 .cb_count = NFNL_MSG_ACCT_MAX,
407 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ACCT);
409 struct nf_acct *nfnl_acct_find_get(struct net *net, const char *acct_name)
411 struct nf_acct *cur, *acct = NULL;
414 list_for_each_entry_rcu(cur, &net->nfnl_acct_list, head) {
415 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
418 if (!try_module_get(THIS_MODULE))
421 if (!refcount_inc_not_zero(&cur->refcnt)) {
422 module_put(THIS_MODULE);
433 EXPORT_SYMBOL_GPL(nfnl_acct_find_get);
435 void nfnl_acct_put(struct nf_acct *acct)
437 if (refcount_dec_and_test(&acct->refcnt))
438 kfree_rcu(acct, rcu_head);
440 module_put(THIS_MODULE);
442 EXPORT_SYMBOL_GPL(nfnl_acct_put);
444 void nfnl_acct_update(const struct sk_buff *skb, struct nf_acct *nfacct)
446 atomic64_inc(&nfacct->pkts);
447 atomic64_add(skb->len, &nfacct->bytes);
449 EXPORT_SYMBOL_GPL(nfnl_acct_update);
451 static void nfnl_overquota_report(struct net *net, struct nf_acct *nfacct)
456 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
460 ret = nfnl_acct_fill_info(skb, 0, 0, NFNL_MSG_ACCT_OVERQUOTA, 0,
466 netlink_broadcast(net->nfnl, skb, 0, NFNLGRP_ACCT_QUOTA,
470 int nfnl_acct_overquota(struct net *net, const struct sk_buff *skb,
471 struct nf_acct *nfacct)
475 int ret = NFACCT_UNDERQUOTA;
477 /* no place here if we don't have a quota */
478 if (!(nfacct->flags & NFACCT_F_QUOTA))
479 return NFACCT_NO_QUOTA;
481 quota = (u64 *)nfacct->data;
482 now = (nfacct->flags & NFACCT_F_QUOTA_PKTS) ?
483 atomic64_read(&nfacct->pkts) : atomic64_read(&nfacct->bytes);
488 !test_and_set_bit(NFACCT_OVERQUOTA_BIT, &nfacct->flags)) {
489 nfnl_overquota_report(net, nfacct);
494 EXPORT_SYMBOL_GPL(nfnl_acct_overquota);
496 static int __net_init nfnl_acct_net_init(struct net *net)
498 INIT_LIST_HEAD(&net->nfnl_acct_list);
503 static void __net_exit nfnl_acct_net_exit(struct net *net)
505 struct nf_acct *cur, *tmp;
507 list_for_each_entry_safe(cur, tmp, &net->nfnl_acct_list, head) {
508 list_del_rcu(&cur->head);
510 if (refcount_dec_and_test(&cur->refcnt))
511 kfree_rcu(cur, rcu_head);
515 static struct pernet_operations nfnl_acct_ops = {
516 .init = nfnl_acct_net_init,
517 .exit = nfnl_acct_net_exit,
520 static int __init nfnl_acct_init(void)
524 ret = register_pernet_subsys(&nfnl_acct_ops);
526 pr_err("nfnl_acct_init: failed to register pernet ops\n");
530 pr_info("nfnl_acct: registering with nfnetlink.\n");
531 ret = nfnetlink_subsys_register(&nfnl_acct_subsys);
533 pr_err("nfnl_acct_init: cannot register with nfnetlink.\n");
539 unregister_pernet_subsys(&nfnl_acct_ops);
544 static void __exit nfnl_acct_exit(void)
546 pr_info("nfnl_acct: unregistering from nfnetlink.\n");
547 nfnetlink_subsys_unregister(&nfnl_acct_subsys);
548 unregister_pernet_subsys(&nfnl_acct_ops);
551 module_init(nfnl_acct_init);
552 module_exit(nfnl_acct_exit);