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 #define nfnl_dereference_protected(id) \
37 rcu_dereference_protected(table[(id)].subsys, \
38 lockdep_nfnl_is_held((id)))
40 static char __initdata nfversion[] = "0.30";
44 const struct nfnetlink_subsystem __rcu *subsys;
45 } table[NFNL_SUBSYS_COUNT];
47 static const int nfnl_group2type[NFNLGRP_MAX+1] = {
48 [NFNLGRP_CONNTRACK_NEW] = NFNL_SUBSYS_CTNETLINK,
49 [NFNLGRP_CONNTRACK_UPDATE] = NFNL_SUBSYS_CTNETLINK,
50 [NFNLGRP_CONNTRACK_DESTROY] = NFNL_SUBSYS_CTNETLINK,
51 [NFNLGRP_CONNTRACK_EXP_NEW] = NFNL_SUBSYS_CTNETLINK_EXP,
52 [NFNLGRP_CONNTRACK_EXP_UPDATE] = NFNL_SUBSYS_CTNETLINK_EXP,
53 [NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP,
54 [NFNLGRP_NFTABLES] = NFNL_SUBSYS_NFTABLES,
55 [NFNLGRP_ACCT_QUOTA] = NFNL_SUBSYS_ACCT,
56 [NFNLGRP_NFTRACE] = NFNL_SUBSYS_NFTABLES,
59 void nfnl_lock(__u8 subsys_id)
61 mutex_lock(&table[subsys_id].mutex);
63 EXPORT_SYMBOL_GPL(nfnl_lock);
65 void nfnl_unlock(__u8 subsys_id)
67 mutex_unlock(&table[subsys_id].mutex);
69 EXPORT_SYMBOL_GPL(nfnl_unlock);
71 #ifdef CONFIG_PROVE_LOCKING
72 bool lockdep_nfnl_is_held(u8 subsys_id)
74 return lockdep_is_held(&table[subsys_id].mutex);
76 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
79 int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
81 nfnl_lock(n->subsys_id);
82 if (table[n->subsys_id].subsys) {
83 nfnl_unlock(n->subsys_id);
86 rcu_assign_pointer(table[n->subsys_id].subsys, n);
87 nfnl_unlock(n->subsys_id);
91 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
93 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
95 nfnl_lock(n->subsys_id);
96 table[n->subsys_id].subsys = NULL;
97 nfnl_unlock(n->subsys_id);
101 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
103 static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u_int16_t type)
105 u_int8_t subsys_id = NFNL_SUBSYS_ID(type);
107 if (subsys_id >= NFNL_SUBSYS_COUNT)
110 return rcu_dereference(table[subsys_id].subsys);
113 static inline const struct nfnl_callback *
114 nfnetlink_find_client(u_int16_t type, const struct nfnetlink_subsystem *ss)
116 u_int8_t cb_id = NFNL_MSG_TYPE(type);
118 if (cb_id >= ss->cb_count)
121 return &ss->cb[cb_id];
124 int nfnetlink_has_listeners(struct net *net, unsigned int group)
126 return netlink_has_listeners(net->nfnl, group);
128 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
130 struct sk_buff *nfnetlink_alloc_skb(struct net *net, unsigned int size,
131 u32 dst_portid, gfp_t gfp_mask)
133 return netlink_alloc_skb(net->nfnl, size, dst_portid, gfp_mask);
135 EXPORT_SYMBOL_GPL(nfnetlink_alloc_skb);
137 int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
138 unsigned int group, int echo, gfp_t flags)
140 return nlmsg_notify(net->nfnl, skb, portid, group, echo, flags);
142 EXPORT_SYMBOL_GPL(nfnetlink_send);
144 int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
146 return netlink_set_err(net->nfnl, portid, group, error);
148 EXPORT_SYMBOL_GPL(nfnetlink_set_err);
150 int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid,
153 return netlink_unicast(net->nfnl, skb, portid, flags);
155 EXPORT_SYMBOL_GPL(nfnetlink_unicast);
157 /* Process one complete nfnetlink message. */
158 static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
160 struct net *net = sock_net(skb->sk);
161 const struct nfnl_callback *nc;
162 const struct nfnetlink_subsystem *ss;
165 /* All the messages must at least contain nfgenmsg */
166 if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
169 type = nlh->nlmsg_type;
172 ss = nfnetlink_get_subsys(type);
174 #ifdef CONFIG_MODULES
176 request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
178 ss = nfnetlink_get_subsys(type);
187 nc = nfnetlink_find_client(type, ss);
194 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
195 u_int8_t cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
196 struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
197 struct nlattr *attr = (void *)nlh + min_len;
198 int attrlen = nlh->nlmsg_len - min_len;
199 __u8 subsys_id = NFNL_SUBSYS_ID(type);
201 err = nla_parse(cda, ss->cb[cb_id].attr_count,
202 attr, attrlen, ss->cb[cb_id].policy);
209 err = nc->call_rcu(net, net->nfnl, skb, nlh,
210 (const struct nlattr **)cda);
214 nfnl_lock(subsys_id);
215 if (nfnl_dereference_protected(subsys_id) != ss ||
216 nfnetlink_find_client(type, ss) != nc)
219 err = nc->call(net, net->nfnl, skb, nlh,
220 (const struct nlattr **)cda);
223 nfnl_unlock(subsys_id);
232 struct list_head head;
233 struct nlmsghdr *nlh;
237 static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err)
239 struct nfnl_err *nfnl_err;
241 nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL);
242 if (nfnl_err == NULL)
247 list_add_tail(&nfnl_err->head, list);
252 static void nfnl_err_del(struct nfnl_err *nfnl_err)
254 list_del(&nfnl_err->head);
258 static void nfnl_err_reset(struct list_head *err_list)
260 struct nfnl_err *nfnl_err, *next;
262 list_for_each_entry_safe(nfnl_err, next, err_list, head)
263 nfnl_err_del(nfnl_err);
266 static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
268 struct nfnl_err *nfnl_err, *next;
270 list_for_each_entry_safe(nfnl_err, next, err_list, head) {
271 netlink_ack(skb, nfnl_err->nlh, nfnl_err->err);
272 nfnl_err_del(nfnl_err);
277 NFNL_BATCH_FAILURE = (1 << 0),
278 NFNL_BATCH_DONE = (1 << 1),
279 NFNL_BATCH_REPLAY = (1 << 2),
282 static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
285 struct sk_buff *oskb = skb;
286 struct net *net = sock_net(skb->sk);
287 const struct nfnetlink_subsystem *ss;
288 const struct nfnl_callback *nc;
289 static LIST_HEAD(err_list);
293 if (subsys_id >= NFNL_SUBSYS_COUNT)
294 return netlink_ack(skb, nlh, -EINVAL);
298 skb = netlink_skb_clone(oskb, GFP_KERNEL);
300 return netlink_ack(oskb, nlh, -ENOMEM);
302 nfnl_lock(subsys_id);
303 ss = nfnl_dereference_protected(subsys_id);
305 #ifdef CONFIG_MODULES
306 nfnl_unlock(subsys_id);
307 request_module("nfnetlink-subsys-%d", subsys_id);
308 nfnl_lock(subsys_id);
309 ss = nfnl_dereference_protected(subsys_id);
313 nfnl_unlock(subsys_id);
314 netlink_ack(oskb, nlh, -EOPNOTSUPP);
315 return kfree_skb(skb);
319 if (!ss->commit || !ss->abort) {
320 nfnl_unlock(subsys_id);
321 netlink_ack(oskb, nlh, -EOPNOTSUPP);
322 return kfree_skb(skb);
325 while (skb->len >= nlmsg_total_size(0)) {
328 nlh = nlmsg_hdr(skb);
331 if (nlh->nlmsg_len < NLMSG_HDRLEN ||
332 skb->len < nlh->nlmsg_len ||
333 nlmsg_len(nlh) < sizeof(struct nfgenmsg)) {
334 nfnl_err_reset(&err_list);
335 status |= NFNL_BATCH_FAILURE;
339 /* Only requests are handled by the kernel */
340 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
345 type = nlh->nlmsg_type;
346 if (type == NFNL_MSG_BATCH_BEGIN) {
347 /* Malformed: Batch begin twice */
348 nfnl_err_reset(&err_list);
349 status |= NFNL_BATCH_FAILURE;
351 } else if (type == NFNL_MSG_BATCH_END) {
352 status |= NFNL_BATCH_DONE;
354 } else if (type < NLMSG_MIN_TYPE) {
359 /* We only accept a batch with messages for the same
362 if (NFNL_SUBSYS_ID(type) != subsys_id) {
367 nc = nfnetlink_find_client(type, ss);
374 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
375 u_int8_t cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
376 struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
377 struct nlattr *attr = (void *)nlh + min_len;
378 int attrlen = nlh->nlmsg_len - min_len;
380 err = nla_parse(cda, ss->cb[cb_id].attr_count,
381 attr, attrlen, ss->cb[cb_id].policy);
385 if (nc->call_batch) {
386 err = nc->call_batch(net, net->nfnl, skb, nlh,
387 (const struct nlattr **)cda);
390 /* The lock was released to autoload some module, we
391 * have to abort and start from scratch using the
394 if (err == -EAGAIN) {
395 status |= NFNL_BATCH_REPLAY;
400 if (nlh->nlmsg_flags & NLM_F_ACK || err) {
401 /* Errors are delivered once the full batch has been
402 * processed, this avoids that the same error is
403 * reported several times when replaying the batch.
405 if (nfnl_err_add(&err_list, nlh, err) < 0) {
406 /* We failed to enqueue an error, reset the
407 * list of errors and send OOM to userspace
408 * pointing to the batch header.
410 nfnl_err_reset(&err_list);
411 netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM);
412 status |= NFNL_BATCH_FAILURE;
415 /* We don't stop processing the batch on errors, thus,
416 * userspace gets all the errors that the batch
420 status |= NFNL_BATCH_FAILURE;
423 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
424 if (msglen > skb->len)
426 skb_pull(skb, msglen);
429 if (status & NFNL_BATCH_REPLAY) {
430 ss->abort(net, oskb);
431 nfnl_err_reset(&err_list);
432 nfnl_unlock(subsys_id);
435 } else if (status == NFNL_BATCH_DONE) {
436 ss->commit(net, oskb);
438 ss->abort(net, oskb);
441 nfnl_err_deliver(&err_list, oskb);
442 nfnl_unlock(subsys_id);
446 static void nfnetlink_rcv(struct sk_buff *skb)
448 struct nlmsghdr *nlh = nlmsg_hdr(skb);
452 if (nlh->nlmsg_len < NLMSG_HDRLEN ||
453 skb->len < nlh->nlmsg_len)
456 if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
457 netlink_ack(skb, nlh, -EPERM);
461 if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN) {
462 struct nfgenmsg *nfgenmsg;
464 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
465 if (msglen > skb->len)
468 if (nlh->nlmsg_len < NLMSG_HDRLEN ||
469 skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
472 nfgenmsg = nlmsg_data(nlh);
473 skb_pull(skb, msglen);
474 /* Work around old nft using host byte order */
475 if (nfgenmsg->res_id == NFNL_SUBSYS_NFTABLES)
476 res_id = NFNL_SUBSYS_NFTABLES;
478 res_id = ntohs(nfgenmsg->res_id);
479 nfnetlink_rcv_batch(skb, nlh, res_id);
481 netlink_rcv_skb(skb, &nfnetlink_rcv_msg);
485 #ifdef CONFIG_MODULES
486 static int nfnetlink_bind(struct net *net, int group)
488 const struct nfnetlink_subsystem *ss;
491 if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
494 type = nfnl_group2type[group];
497 ss = nfnetlink_get_subsys(type << 8);
500 request_module("nfnetlink-subsys-%d", type);
505 static int __net_init nfnetlink_net_init(struct net *net)
508 struct netlink_kernel_cfg cfg = {
509 .groups = NFNLGRP_MAX,
510 .input = nfnetlink_rcv,
511 #ifdef CONFIG_MODULES
512 .bind = nfnetlink_bind,
516 nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
519 net->nfnl_stash = nfnl;
520 rcu_assign_pointer(net->nfnl, nfnl);
524 static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
528 list_for_each_entry(net, net_exit_list, exit_list)
529 RCU_INIT_POINTER(net->nfnl, NULL);
531 list_for_each_entry(net, net_exit_list, exit_list)
532 netlink_kernel_release(net->nfnl_stash);
535 static struct pernet_operations nfnetlink_net_ops = {
536 .init = nfnetlink_net_init,
537 .exit_batch = nfnetlink_net_exit_batch,
540 static int __init nfnetlink_init(void)
544 for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
545 BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
547 for (i=0; i<NFNL_SUBSYS_COUNT; i++)
548 mutex_init(&table[i].mutex);
550 pr_info("Netfilter messages via NETLINK v%s.\n", nfversion);
551 return register_pernet_subsys(&nfnetlink_net_ops);
554 static void __exit nfnetlink_exit(void)
556 pr_info("Removing netfilter NETLINK layer.\n");
557 unregister_pernet_subsys(&nfnetlink_net_ops);
559 module_init(nfnetlink_init);
560 module_exit(nfnetlink_exit);