1 /* Netfilter messages via netlink socket. Allows for user space
2 * protocol helpers and general trouble making from userspace.
4 * (C) 2001 by Jay Schulist <jschlst@samba.org>,
5 * (C) 2002-2005 by Harald Welte <laforge@gnumonks.org>
6 * (C) 2005,2007 by Pablo Neira Ayuso <pablo@netfilter.org>
8 * Initial netfilter messages via netlink development funded and
9 * generally made possible by Network Robots, Inc. (www.networkrobots.com)
11 * Further development of this code funded by Astaro AG (http://www.astaro.com)
13 * This software may be used and distributed according to the terms
14 * of the GNU General Public License, incorporated herein by reference.
17 #include <linux/module.h>
18 #include <linux/types.h>
19 #include <linux/socket.h>
20 #include <linux/kernel.h>
21 #include <linux/string.h>
22 #include <linux/sockios.h>
23 #include <linux/net.h>
24 #include <linux/skbuff.h>
25 #include <asm/uaccess.h>
27 #include <linux/init.h>
29 #include <net/netlink.h>
30 #include <linux/netfilter/nfnetlink.h>
32 MODULE_LICENSE("GPL");
33 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
34 MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER);
36 static char __initdata nfversion[] = "0.30";
40 const struct nfnetlink_subsystem __rcu *subsys;
41 } table[NFNL_SUBSYS_COUNT];
43 static const int nfnl_group2type[NFNLGRP_MAX+1] = {
44 [NFNLGRP_CONNTRACK_NEW] = NFNL_SUBSYS_CTNETLINK,
45 [NFNLGRP_CONNTRACK_UPDATE] = NFNL_SUBSYS_CTNETLINK,
46 [NFNLGRP_CONNTRACK_DESTROY] = NFNL_SUBSYS_CTNETLINK,
47 [NFNLGRP_CONNTRACK_EXP_NEW] = NFNL_SUBSYS_CTNETLINK_EXP,
48 [NFNLGRP_CONNTRACK_EXP_UPDATE] = NFNL_SUBSYS_CTNETLINK_EXP,
49 [NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP,
52 void nfnl_lock(__u8 subsys_id)
54 mutex_lock(&table[subsys_id].mutex);
56 EXPORT_SYMBOL_GPL(nfnl_lock);
58 void nfnl_unlock(__u8 subsys_id)
60 mutex_unlock(&table[subsys_id].mutex);
62 EXPORT_SYMBOL_GPL(nfnl_unlock);
64 #ifdef CONFIG_PROVE_LOCKING
65 int lockdep_nfnl_is_held(u8 subsys_id)
67 return lockdep_is_held(&table[subsys_id].mutex);
69 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
72 int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
74 nfnl_lock(n->subsys_id);
75 if (table[n->subsys_id].subsys) {
76 nfnl_unlock(n->subsys_id);
79 rcu_assign_pointer(table[n->subsys_id].subsys, n);
80 nfnl_unlock(n->subsys_id);
84 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
86 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
88 nfnl_lock(n->subsys_id);
89 table[n->subsys_id].subsys = NULL;
90 nfnl_unlock(n->subsys_id);
94 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
96 static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u_int16_t type)
98 u_int8_t subsys_id = NFNL_SUBSYS_ID(type);
100 if (subsys_id >= NFNL_SUBSYS_COUNT)
103 return rcu_dereference(table[subsys_id].subsys);
106 static inline const struct nfnl_callback *
107 nfnetlink_find_client(u_int16_t type, const struct nfnetlink_subsystem *ss)
109 u_int8_t cb_id = NFNL_MSG_TYPE(type);
111 if (cb_id >= ss->cb_count)
114 return &ss->cb[cb_id];
117 int nfnetlink_has_listeners(struct net *net, unsigned int group)
119 return netlink_has_listeners(net->nfnl, group);
121 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
123 struct sk_buff *nfnetlink_alloc_skb(struct net *net, unsigned int size,
124 u32 dst_portid, gfp_t gfp_mask)
126 return netlink_alloc_skb(net->nfnl, size, dst_portid, gfp_mask);
128 EXPORT_SYMBOL_GPL(nfnetlink_alloc_skb);
130 int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
131 unsigned int group, int echo, gfp_t flags)
133 return nlmsg_notify(net->nfnl, skb, portid, group, echo, flags);
135 EXPORT_SYMBOL_GPL(nfnetlink_send);
137 int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
139 return netlink_set_err(net->nfnl, portid, group, error);
141 EXPORT_SYMBOL_GPL(nfnetlink_set_err);
143 int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid,
146 return netlink_unicast(net->nfnl, skb, portid, flags);
148 EXPORT_SYMBOL_GPL(nfnetlink_unicast);
150 /* Process one complete nfnetlink message. */
151 static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
153 struct net *net = sock_net(skb->sk);
154 const struct nfnl_callback *nc;
155 const struct nfnetlink_subsystem *ss;
158 /* All the messages must at least contain nfgenmsg */
159 if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
162 type = nlh->nlmsg_type;
165 ss = nfnetlink_get_subsys(type);
167 #ifdef CONFIG_MODULES
169 request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
171 ss = nfnetlink_get_subsys(type);
180 nc = nfnetlink_find_client(type, ss);
187 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
188 u_int8_t cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
189 struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
190 struct nlattr *attr = (void *)nlh + min_len;
191 int attrlen = nlh->nlmsg_len - min_len;
192 __u8 subsys_id = NFNL_SUBSYS_ID(type);
194 err = nla_parse(cda, ss->cb[cb_id].attr_count,
195 attr, attrlen, ss->cb[cb_id].policy);
202 err = nc->call_rcu(net->nfnl, skb, nlh,
203 (const struct nlattr **)cda);
207 nfnl_lock(subsys_id);
208 if (rcu_dereference_protected(table[subsys_id].subsys,
209 lockdep_is_held(&table[subsys_id].mutex)) != ss ||
210 nfnetlink_find_client(type, ss) != nc)
213 err = nc->call(net->nfnl, skb, nlh,
214 (const struct nlattr **)cda);
217 nfnl_unlock(subsys_id);
225 static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
228 struct sk_buff *nskb, *oskb = skb;
229 struct net *net = sock_net(skb->sk);
230 const struct nfnetlink_subsystem *ss;
231 const struct nfnl_callback *nc;
232 bool success = true, done = false;
235 if (subsys_id >= NFNL_SUBSYS_COUNT)
236 return netlink_ack(skb, nlh, -EINVAL);
238 nskb = netlink_skb_clone(oskb, GFP_KERNEL);
240 return netlink_ack(oskb, nlh, -ENOMEM);
245 nfnl_lock(subsys_id);
246 ss = rcu_dereference_protected(table[subsys_id].subsys,
247 lockdep_is_held(&table[subsys_id].mutex));
249 #ifdef CONFIG_MODULES
250 nfnl_unlock(subsys_id);
251 request_module("nfnetlink-subsys-%d", subsys_id);
252 nfnl_lock(subsys_id);
253 ss = rcu_dereference_protected(table[subsys_id].subsys,
254 lockdep_is_held(&table[subsys_id].mutex));
258 nfnl_unlock(subsys_id);
260 return netlink_ack(skb, nlh, -EOPNOTSUPP);
264 if (!ss->commit || !ss->abort) {
265 nfnl_unlock(subsys_id);
267 return netlink_ack(skb, nlh, -EOPNOTSUPP);
270 while (skb->len >= nlmsg_total_size(0)) {
273 nlh = nlmsg_hdr(skb);
276 if (nlh->nlmsg_len < NLMSG_HDRLEN) {
281 /* Only requests are handled by the kernel */
282 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
287 type = nlh->nlmsg_type;
288 if (type == NFNL_MSG_BATCH_BEGIN) {
289 /* Malformed: Batch begin twice */
292 } else if (type == NFNL_MSG_BATCH_END) {
295 } else if (type < NLMSG_MIN_TYPE) {
300 /* We only accept a batch with messages for the same
303 if (NFNL_SUBSYS_ID(type) != subsys_id) {
308 nc = nfnetlink_find_client(type, ss);
315 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
316 u_int8_t cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
317 struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
318 struct nlattr *attr = (void *)nlh + min_len;
319 int attrlen = nlh->nlmsg_len - min_len;
321 err = nla_parse(cda, ss->cb[cb_id].attr_count,
322 attr, attrlen, ss->cb[cb_id].policy);
326 if (nc->call_batch) {
327 err = nc->call_batch(net->nfnl, skb, nlh,
328 (const struct nlattr **)cda);
331 /* The lock was released to autoload some module, we
332 * have to abort and start from scratch using the
335 if (err == -EAGAIN) {
337 nfnl_unlock(subsys_id);
343 if (nlh->nlmsg_flags & NLM_F_ACK || err) {
344 /* We don't stop processing the batch on errors, thus,
345 * userspace gets all the errors that the batch
348 netlink_ack(skb, nlh, err);
353 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
354 if (msglen > skb->len)
356 skb_pull(skb, msglen);
364 nfnl_unlock(subsys_id);
368 static void nfnetlink_rcv(struct sk_buff *skb)
370 struct nlmsghdr *nlh = nlmsg_hdr(skb);
371 struct net *net = sock_net(skb->sk);
374 if (nlh->nlmsg_len < NLMSG_HDRLEN ||
375 skb->len < nlh->nlmsg_len)
378 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) {
379 netlink_ack(skb, nlh, -EPERM);
383 if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN) {
384 struct nfgenmsg *nfgenmsg;
386 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
387 if (msglen > skb->len)
390 if (nlh->nlmsg_len < NLMSG_HDRLEN ||
391 skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
394 nfgenmsg = nlmsg_data(nlh);
395 skb_pull(skb, msglen);
396 nfnetlink_rcv_batch(skb, nlh, nfgenmsg->res_id);
398 netlink_rcv_skb(skb, &nfnetlink_rcv_msg);
402 #ifdef CONFIG_MODULES
403 static int nfnetlink_bind(int group)
405 const struct nfnetlink_subsystem *ss;
406 int type = nfnl_group2type[group];
409 ss = nfnetlink_get_subsys(type);
412 request_module("nfnetlink-subsys-%d", type);
417 static int __net_init nfnetlink_net_init(struct net *net)
420 struct netlink_kernel_cfg cfg = {
421 .groups = NFNLGRP_MAX,
422 .input = nfnetlink_rcv,
423 #ifdef CONFIG_MODULES
424 .bind = nfnetlink_bind,
428 nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
431 net->nfnl_stash = nfnl;
432 rcu_assign_pointer(net->nfnl, nfnl);
436 static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
440 list_for_each_entry(net, net_exit_list, exit_list)
441 RCU_INIT_POINTER(net->nfnl, NULL);
443 list_for_each_entry(net, net_exit_list, exit_list)
444 netlink_kernel_release(net->nfnl_stash);
447 static struct pernet_operations nfnetlink_net_ops = {
448 .init = nfnetlink_net_init,
449 .exit_batch = nfnetlink_net_exit_batch,
452 static int __init nfnetlink_init(void)
456 for (i=0; i<NFNL_SUBSYS_COUNT; i++)
457 mutex_init(&table[i].mutex);
459 pr_info("Netfilter messages via NETLINK v%s.\n", nfversion);
460 return register_pernet_subsys(&nfnetlink_net_ops);
463 static void __exit nfnetlink_exit(void)
465 pr_info("Removing netfilter NETLINK layer.\n");
466 unregister_pernet_subsys(&nfnetlink_net_ops);
468 module_init(nfnetlink_init);
469 module_exit(nfnetlink_exit);