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/netlink.h>
15 #include <linux/rculist.h>
16 #include <linux/slab.h>
17 #include <linux/types.h>
18 #include <linux/errno.h>
19 #include <net/netlink.h>
22 #include <linux/netfilter.h>
23 #include <linux/netfilter/nfnetlink.h>
24 #include <linux/netfilter/nfnetlink_acct.h>
26 MODULE_LICENSE("GPL");
27 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
28 MODULE_DESCRIPTION("nfacct: Extended Netfilter accounting infrastructure");
34 struct list_head head;
36 char name[NFACCT_NAME_MAX];
37 struct rcu_head rcu_head;
41 struct nfacct_filter {
46 #define NFACCT_F_QUOTA (NFACCT_F_QUOTA_PKTS | NFACCT_F_QUOTA_BYTES)
47 #define NFACCT_OVERQUOTA_BIT 2 /* NFACCT_F_OVERQUOTA */
49 static int nfnl_acct_new(struct net *net, struct sock *nfnl,
50 struct sk_buff *skb, const struct nlmsghdr *nlh,
51 const struct nlattr * const tb[])
53 struct nf_acct *nfacct, *matching = NULL;
55 unsigned int size = 0;
61 acct_name = nla_data(tb[NFACCT_NAME]);
62 if (strlen(acct_name) == 0)
65 list_for_each_entry(nfacct, &net->nfnl_acct_list, head) {
66 if (strncmp(nfacct->name, acct_name, NFACCT_NAME_MAX) != 0)
69 if (nlh->nlmsg_flags & NLM_F_EXCL)
77 if (nlh->nlmsg_flags & NLM_F_REPLACE) {
78 /* reset counters if you request a replacement. */
79 atomic64_set(&matching->pkts, 0);
80 atomic64_set(&matching->bytes, 0);
81 smp_mb__before_atomic();
82 /* reset overquota flag if quota is enabled. */
83 if ((matching->flags & NFACCT_F_QUOTA))
84 clear_bit(NFACCT_OVERQUOTA_BIT,
91 if (tb[NFACCT_FLAGS]) {
92 flags = ntohl(nla_get_be32(tb[NFACCT_FLAGS]));
93 if (flags & ~NFACCT_F_QUOTA)
95 if ((flags & NFACCT_F_QUOTA) == NFACCT_F_QUOTA)
97 if (flags & NFACCT_F_OVERQUOTA)
103 nfacct = kzalloc(sizeof(struct nf_acct) + size, GFP_KERNEL);
107 if (flags & NFACCT_F_QUOTA) {
108 u64 *quota = (u64 *)nfacct->data;
110 *quota = be64_to_cpu(nla_get_be64(tb[NFACCT_QUOTA]));
111 nfacct->flags = flags;
114 strncpy(nfacct->name, nla_data(tb[NFACCT_NAME]), NFACCT_NAME_MAX);
116 if (tb[NFACCT_BYTES]) {
117 atomic64_set(&nfacct->bytes,
118 be64_to_cpu(nla_get_be64(tb[NFACCT_BYTES])));
120 if (tb[NFACCT_PKTS]) {
121 atomic64_set(&nfacct->pkts,
122 be64_to_cpu(nla_get_be64(tb[NFACCT_PKTS])));
124 atomic_set(&nfacct->refcnt, 1);
125 list_add_tail_rcu(&nfacct->head, &net->nfnl_acct_list);
130 nfnl_acct_fill_info(struct sk_buff *skb, u32 portid, u32 seq, u32 type,
131 int event, struct nf_acct *acct)
133 struct nlmsghdr *nlh;
134 struct nfgenmsg *nfmsg;
135 unsigned int flags = portid ? NLM_F_MULTI : 0;
139 event |= NFNL_SUBSYS_ACCT << 8;
140 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nfmsg), flags);
144 nfmsg = nlmsg_data(nlh);
145 nfmsg->nfgen_family = AF_UNSPEC;
146 nfmsg->version = NFNETLINK_V0;
149 if (nla_put_string(skb, NFACCT_NAME, acct->name))
150 goto nla_put_failure;
152 old_flags = acct->flags;
153 if (type == NFNL_MSG_ACCT_GET_CTRZERO) {
154 pkts = atomic64_xchg(&acct->pkts, 0);
155 bytes = atomic64_xchg(&acct->bytes, 0);
156 smp_mb__before_atomic();
157 if (acct->flags & NFACCT_F_QUOTA)
158 clear_bit(NFACCT_OVERQUOTA_BIT, &acct->flags);
160 pkts = atomic64_read(&acct->pkts);
161 bytes = atomic64_read(&acct->bytes);
163 if (nla_put_be64(skb, NFACCT_PKTS, cpu_to_be64(pkts)) ||
164 nla_put_be64(skb, NFACCT_BYTES, cpu_to_be64(bytes)) ||
165 nla_put_be32(skb, NFACCT_USE, htonl(atomic_read(&acct->refcnt))))
166 goto nla_put_failure;
167 if (acct->flags & NFACCT_F_QUOTA) {
168 u64 *quota = (u64 *)acct->data;
170 if (nla_put_be32(skb, NFACCT_FLAGS, htonl(old_flags)) ||
171 nla_put_be64(skb, NFACCT_QUOTA, cpu_to_be64(*quota)))
172 goto nla_put_failure;
179 nlmsg_cancel(skb, nlh);
184 nfnl_acct_dump(struct sk_buff *skb, struct netlink_callback *cb)
186 struct net *net = sock_net(skb->sk);
187 struct nf_acct *cur, *last;
188 const struct nfacct_filter *filter = cb->data;
193 last = (struct nf_acct *)cb->args[1];
198 list_for_each_entry_rcu(cur, &net->nfnl_acct_list, head) {
206 if (filter && (cur->flags & filter->mask) != filter->value)
209 if (nfnl_acct_fill_info(skb, NETLINK_CB(cb->skb).portid,
211 NFNL_MSG_TYPE(cb->nlh->nlmsg_type),
212 NFNL_MSG_ACCT_NEW, cur) < 0) {
213 cb->args[1] = (unsigned long)cur;
223 static int nfnl_acct_done(struct netlink_callback *cb)
229 static const struct nla_policy filter_policy[NFACCT_FILTER_MAX + 1] = {
230 [NFACCT_FILTER_MASK] = { .type = NLA_U32 },
231 [NFACCT_FILTER_VALUE] = { .type = NLA_U32 },
234 static struct nfacct_filter *
235 nfacct_filter_alloc(const struct nlattr * const attr)
237 struct nfacct_filter *filter;
238 struct nlattr *tb[NFACCT_FILTER_MAX + 1];
241 err = nla_parse_nested(tb, NFACCT_FILTER_MAX, attr, filter_policy);
245 filter = kzalloc(sizeof(struct nfacct_filter), GFP_KERNEL);
247 return ERR_PTR(-ENOMEM);
249 filter->mask = ntohl(nla_get_be32(tb[NFACCT_FILTER_MASK]));
250 filter->value = ntohl(nla_get_be32(tb[NFACCT_FILTER_VALUE]));
255 static int nfnl_acct_get(struct net *net, struct sock *nfnl,
256 struct sk_buff *skb, const struct nlmsghdr *nlh,
257 const struct nlattr * const tb[])
263 if (nlh->nlmsg_flags & NLM_F_DUMP) {
264 struct netlink_dump_control c = {
265 .dump = nfnl_acct_dump,
266 .done = nfnl_acct_done,
269 if (tb[NFACCT_FILTER]) {
270 struct nfacct_filter *filter;
272 filter = nfacct_filter_alloc(tb[NFACCT_FILTER]);
274 return PTR_ERR(filter);
278 return netlink_dump_start(nfnl, skb, nlh, &c);
281 if (!tb[NFACCT_NAME])
283 acct_name = nla_data(tb[NFACCT_NAME]);
285 list_for_each_entry(cur, &net->nfnl_acct_list, head) {
286 struct sk_buff *skb2;
288 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
291 skb2 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
297 ret = nfnl_acct_fill_info(skb2, NETLINK_CB(skb).portid,
299 NFNL_MSG_TYPE(nlh->nlmsg_type),
300 NFNL_MSG_ACCT_NEW, cur);
305 ret = netlink_unicast(nfnl, skb2, NETLINK_CB(skb).portid,
310 /* this avoids a loop in nfnetlink. */
311 return ret == -EAGAIN ? -ENOBUFS : ret;
316 /* try to delete object, fail if it is still in use. */
317 static int nfnl_acct_try_del(struct nf_acct *cur)
321 /* we want to avoid races with nfnl_acct_find_get. */
322 if (atomic_dec_and_test(&cur->refcnt)) {
323 /* We are protected by nfnl mutex. */
324 list_del_rcu(&cur->head);
325 kfree_rcu(cur, rcu_head);
327 /* still in use, restore reference counter. */
328 atomic_inc(&cur->refcnt);
334 static int nfnl_acct_del(struct net *net, struct sock *nfnl,
335 struct sk_buff *skb, const struct nlmsghdr *nlh,
336 const struct nlattr * const tb[])
342 if (!tb[NFACCT_NAME]) {
343 list_for_each_entry(cur, &net->nfnl_acct_list, head)
344 nfnl_acct_try_del(cur);
348 acct_name = nla_data(tb[NFACCT_NAME]);
350 list_for_each_entry(cur, &net->nfnl_acct_list, head) {
351 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX) != 0)
354 ret = nfnl_acct_try_del(cur);
363 static const struct nla_policy nfnl_acct_policy[NFACCT_MAX+1] = {
364 [NFACCT_NAME] = { .type = NLA_NUL_STRING, .len = NFACCT_NAME_MAX-1 },
365 [NFACCT_BYTES] = { .type = NLA_U64 },
366 [NFACCT_PKTS] = { .type = NLA_U64 },
367 [NFACCT_FLAGS] = { .type = NLA_U32 },
368 [NFACCT_QUOTA] = { .type = NLA_U64 },
369 [NFACCT_FILTER] = {.type = NLA_NESTED },
372 static const struct nfnl_callback nfnl_acct_cb[NFNL_MSG_ACCT_MAX] = {
373 [NFNL_MSG_ACCT_NEW] = { .call = nfnl_acct_new,
374 .attr_count = NFACCT_MAX,
375 .policy = nfnl_acct_policy },
376 [NFNL_MSG_ACCT_GET] = { .call = nfnl_acct_get,
377 .attr_count = NFACCT_MAX,
378 .policy = nfnl_acct_policy },
379 [NFNL_MSG_ACCT_GET_CTRZERO] = { .call = nfnl_acct_get,
380 .attr_count = NFACCT_MAX,
381 .policy = nfnl_acct_policy },
382 [NFNL_MSG_ACCT_DEL] = { .call = nfnl_acct_del,
383 .attr_count = NFACCT_MAX,
384 .policy = nfnl_acct_policy },
387 static const struct nfnetlink_subsystem nfnl_acct_subsys = {
389 .subsys_id = NFNL_SUBSYS_ACCT,
390 .cb_count = NFNL_MSG_ACCT_MAX,
394 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ACCT);
396 struct nf_acct *nfnl_acct_find_get(struct net *net, const char *acct_name)
398 struct nf_acct *cur, *acct = NULL;
401 list_for_each_entry_rcu(cur, &net->nfnl_acct_list, head) {
402 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
405 if (!try_module_get(THIS_MODULE))
408 if (!atomic_inc_not_zero(&cur->refcnt)) {
409 module_put(THIS_MODULE);
420 EXPORT_SYMBOL_GPL(nfnl_acct_find_get);
422 void nfnl_acct_put(struct nf_acct *acct)
424 if (atomic_dec_and_test(&acct->refcnt))
425 kfree_rcu(acct, rcu_head);
427 module_put(THIS_MODULE);
429 EXPORT_SYMBOL_GPL(nfnl_acct_put);
431 void nfnl_acct_update(const struct sk_buff *skb, struct nf_acct *nfacct)
433 atomic64_inc(&nfacct->pkts);
434 atomic64_add(skb->len, &nfacct->bytes);
436 EXPORT_SYMBOL_GPL(nfnl_acct_update);
438 static void nfnl_overquota_report(struct nf_acct *nfacct)
443 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
447 ret = nfnl_acct_fill_info(skb, 0, 0, NFNL_MSG_ACCT_OVERQUOTA, 0,
453 netlink_broadcast(init_net.nfnl, skb, 0, NFNLGRP_ACCT_QUOTA,
457 int nfnl_acct_overquota(const struct sk_buff *skb, struct nf_acct *nfacct)
461 int ret = NFACCT_UNDERQUOTA;
463 /* no place here if we don't have a quota */
464 if (!(nfacct->flags & NFACCT_F_QUOTA))
465 return NFACCT_NO_QUOTA;
467 quota = (u64 *)nfacct->data;
468 now = (nfacct->flags & NFACCT_F_QUOTA_PKTS) ?
469 atomic64_read(&nfacct->pkts) : atomic64_read(&nfacct->bytes);
474 !test_and_set_bit(NFACCT_OVERQUOTA_BIT, &nfacct->flags)) {
475 nfnl_overquota_report(nfacct);
480 EXPORT_SYMBOL_GPL(nfnl_acct_overquota);
482 static int __net_init nfnl_acct_net_init(struct net *net)
484 INIT_LIST_HEAD(&net->nfnl_acct_list);
489 static void __net_exit nfnl_acct_net_exit(struct net *net)
491 struct nf_acct *cur, *tmp;
493 list_for_each_entry_safe(cur, tmp, &net->nfnl_acct_list, head) {
494 list_del_rcu(&cur->head);
496 if (atomic_dec_and_test(&cur->refcnt))
497 kfree_rcu(cur, rcu_head);
501 static struct pernet_operations nfnl_acct_ops = {
502 .init = nfnl_acct_net_init,
503 .exit = nfnl_acct_net_exit,
506 static int __init nfnl_acct_init(void)
510 ret = register_pernet_subsys(&nfnl_acct_ops);
512 pr_err("nfnl_acct_init: failed to register pernet ops\n");
516 pr_info("nfnl_acct: registering with nfnetlink.\n");
517 ret = nfnetlink_subsys_register(&nfnl_acct_subsys);
519 pr_err("nfnl_acct_init: cannot register with nfnetlink.\n");
525 unregister_pernet_subsys(&nfnl_acct_ops);
530 static void __exit nfnl_acct_exit(void)
532 pr_info("nfnl_acct: unregistering from nfnetlink.\n");
533 nfnetlink_subsys_unregister(&nfnl_acct_subsys);
534 unregister_pernet_subsys(&nfnl_acct_ops);
537 module_init(nfnl_acct_init);
538 module_exit(nfnl_acct_exit);