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) 2005-2017 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 <linux/uaccess.h>
27 #include <linux/init.h>
28 #include <linux/sched/signal.h>
30 #include <net/netlink.h>
31 #include <linux/netfilter/nfnetlink.h>
33 MODULE_LICENSE("GPL");
34 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
35 MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER);
37 #define nfnl_dereference_protected(id) \
38 rcu_dereference_protected(table[(id)].subsys, \
39 lockdep_nfnl_is_held((id)))
41 #define NFNL_MAX_ATTR_COUNT 32
45 const struct nfnetlink_subsystem __rcu *subsys;
46 } table[NFNL_SUBSYS_COUNT];
48 static const int nfnl_group2type[NFNLGRP_MAX+1] = {
49 [NFNLGRP_CONNTRACK_NEW] = NFNL_SUBSYS_CTNETLINK,
50 [NFNLGRP_CONNTRACK_UPDATE] = NFNL_SUBSYS_CTNETLINK,
51 [NFNLGRP_CONNTRACK_DESTROY] = NFNL_SUBSYS_CTNETLINK,
52 [NFNLGRP_CONNTRACK_EXP_NEW] = NFNL_SUBSYS_CTNETLINK_EXP,
53 [NFNLGRP_CONNTRACK_EXP_UPDATE] = NFNL_SUBSYS_CTNETLINK_EXP,
54 [NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP,
55 [NFNLGRP_NFTABLES] = NFNL_SUBSYS_NFTABLES,
56 [NFNLGRP_ACCT_QUOTA] = NFNL_SUBSYS_ACCT,
57 [NFNLGRP_NFTRACE] = NFNL_SUBSYS_NFTABLES,
60 void nfnl_lock(__u8 subsys_id)
62 mutex_lock(&table[subsys_id].mutex);
64 EXPORT_SYMBOL_GPL(nfnl_lock);
66 void nfnl_unlock(__u8 subsys_id)
68 mutex_unlock(&table[subsys_id].mutex);
70 EXPORT_SYMBOL_GPL(nfnl_unlock);
72 #ifdef CONFIG_PROVE_LOCKING
73 bool lockdep_nfnl_is_held(u8 subsys_id)
75 return lockdep_is_held(&table[subsys_id].mutex);
77 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
80 int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
84 /* Sanity-check attr_count size to avoid stack buffer overflow. */
85 for (cb_id = 0; cb_id < n->cb_count; cb_id++)
86 if (WARN_ON(n->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT))
89 nfnl_lock(n->subsys_id);
90 if (table[n->subsys_id].subsys) {
91 nfnl_unlock(n->subsys_id);
94 rcu_assign_pointer(table[n->subsys_id].subsys, n);
95 nfnl_unlock(n->subsys_id);
99 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
101 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
103 nfnl_lock(n->subsys_id);
104 table[n->subsys_id].subsys = NULL;
105 nfnl_unlock(n->subsys_id);
109 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
111 static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u16 type)
113 u8 subsys_id = NFNL_SUBSYS_ID(type);
115 if (subsys_id >= NFNL_SUBSYS_COUNT)
118 return rcu_dereference(table[subsys_id].subsys);
121 static inline const struct nfnl_callback *
122 nfnetlink_find_client(u16 type, const struct nfnetlink_subsystem *ss)
124 u8 cb_id = NFNL_MSG_TYPE(type);
126 if (cb_id >= ss->cb_count)
129 return &ss->cb[cb_id];
132 int nfnetlink_has_listeners(struct net *net, unsigned int group)
134 return netlink_has_listeners(net->nfnl, group);
136 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
138 int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
139 unsigned int group, int echo, gfp_t flags)
141 return nlmsg_notify(net->nfnl, skb, portid, group, echo, flags);
143 EXPORT_SYMBOL_GPL(nfnetlink_send);
145 int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
147 return netlink_set_err(net->nfnl, portid, group, error);
149 EXPORT_SYMBOL_GPL(nfnetlink_set_err);
151 int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid)
155 err = nlmsg_unicast(net->nfnl, skb, portid);
161 EXPORT_SYMBOL_GPL(nfnetlink_unicast);
163 /* Process one complete nfnetlink message. */
164 static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
165 struct netlink_ext_ack *extack)
167 struct net *net = sock_net(skb->sk);
168 const struct nfnl_callback *nc;
169 const struct nfnetlink_subsystem *ss;
172 /* All the messages must at least contain nfgenmsg */
173 if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
176 type = nlh->nlmsg_type;
179 ss = nfnetlink_get_subsys(type);
181 #ifdef CONFIG_MODULES
183 request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
185 ss = nfnetlink_get_subsys(type);
194 nc = nfnetlink_find_client(type, ss);
201 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
202 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
203 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
204 struct nlattr *attr = (void *)nlh + min_len;
205 int attrlen = nlh->nlmsg_len - min_len;
206 __u8 subsys_id = NFNL_SUBSYS_ID(type);
208 /* Sanity-check NFNL_MAX_ATTR_COUNT */
209 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
214 err = nla_parse(cda, ss->cb[cb_id].attr_count, attr, attrlen,
215 ss->cb[cb_id].policy, extack);
222 err = nc->call_rcu(net, net->nfnl, skb, nlh,
223 (const struct nlattr **)cda,
228 nfnl_lock(subsys_id);
229 if (nfnl_dereference_protected(subsys_id) != ss ||
230 nfnetlink_find_client(type, ss) != nc)
233 err = nc->call(net, net->nfnl, skb, nlh,
234 (const struct nlattr **)cda,
238 nfnl_unlock(subsys_id);
247 struct list_head head;
248 struct nlmsghdr *nlh;
250 struct netlink_ext_ack extack;
253 static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err,
254 const struct netlink_ext_ack *extack)
256 struct nfnl_err *nfnl_err;
258 nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL);
259 if (nfnl_err == NULL)
264 nfnl_err->extack = *extack;
265 list_add_tail(&nfnl_err->head, list);
270 static void nfnl_err_del(struct nfnl_err *nfnl_err)
272 list_del(&nfnl_err->head);
276 static void nfnl_err_reset(struct list_head *err_list)
278 struct nfnl_err *nfnl_err, *next;
280 list_for_each_entry_safe(nfnl_err, next, err_list, head)
281 nfnl_err_del(nfnl_err);
284 static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
286 struct nfnl_err *nfnl_err, *next;
288 list_for_each_entry_safe(nfnl_err, next, err_list, head) {
289 netlink_ack(skb, nfnl_err->nlh, nfnl_err->err,
291 nfnl_err_del(nfnl_err);
296 NFNL_BATCH_FAILURE = (1 << 0),
297 NFNL_BATCH_DONE = (1 << 1),
298 NFNL_BATCH_REPLAY = (1 << 2),
301 static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
302 u16 subsys_id, u32 genid)
304 struct sk_buff *oskb = skb;
305 struct net *net = sock_net(skb->sk);
306 const struct nfnetlink_subsystem *ss;
307 const struct nfnl_callback *nc;
308 struct netlink_ext_ack extack;
313 if (subsys_id >= NFNL_SUBSYS_COUNT)
314 return netlink_ack(skb, nlh, -EINVAL, NULL);
318 skb = netlink_skb_clone(oskb, GFP_KERNEL);
320 return netlink_ack(oskb, nlh, -ENOMEM, NULL);
322 nfnl_lock(subsys_id);
323 ss = nfnl_dereference_protected(subsys_id);
325 #ifdef CONFIG_MODULES
326 nfnl_unlock(subsys_id);
327 request_module("nfnetlink-subsys-%d", subsys_id);
328 nfnl_lock(subsys_id);
329 ss = nfnl_dereference_protected(subsys_id);
333 nfnl_unlock(subsys_id);
334 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
335 return kfree_skb(skb);
339 if (!ss->valid_genid || !ss->commit || !ss->abort) {
340 nfnl_unlock(subsys_id);
341 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
342 return kfree_skb(skb);
345 if (!try_module_get(ss->owner)) {
346 nfnl_unlock(subsys_id);
347 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
348 return kfree_skb(skb);
351 if (!ss->valid_genid(net, genid)) {
352 module_put(ss->owner);
353 nfnl_unlock(subsys_id);
354 netlink_ack(oskb, nlh, -ERESTART, NULL);
355 return kfree_skb(skb);
358 nfnl_unlock(subsys_id);
360 while (skb->len >= nlmsg_total_size(0)) {
363 if (fatal_signal_pending(current)) {
364 nfnl_err_reset(&err_list);
366 status = NFNL_BATCH_FAILURE;
370 memset(&extack, 0, sizeof(extack));
371 nlh = nlmsg_hdr(skb);
374 if (nlh->nlmsg_len < NLMSG_HDRLEN ||
375 skb->len < nlh->nlmsg_len ||
376 nlmsg_len(nlh) < sizeof(struct nfgenmsg)) {
377 nfnl_err_reset(&err_list);
378 status |= NFNL_BATCH_FAILURE;
382 /* Only requests are handled by the kernel */
383 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
388 type = nlh->nlmsg_type;
389 if (type == NFNL_MSG_BATCH_BEGIN) {
390 /* Malformed: Batch begin twice */
391 nfnl_err_reset(&err_list);
392 status |= NFNL_BATCH_FAILURE;
394 } else if (type == NFNL_MSG_BATCH_END) {
395 status |= NFNL_BATCH_DONE;
397 } else if (type < NLMSG_MIN_TYPE) {
402 /* We only accept a batch with messages for the same
405 if (NFNL_SUBSYS_ID(type) != subsys_id) {
410 nc = nfnetlink_find_client(type, ss);
417 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
418 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
419 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
420 struct nlattr *attr = (void *)nlh + min_len;
421 int attrlen = nlh->nlmsg_len - min_len;
423 /* Sanity-check NFTA_MAX_ATTR */
424 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
429 err = nla_parse(cda, ss->cb[cb_id].attr_count, attr,
430 attrlen, ss->cb[cb_id].policy, NULL);
434 if (nc->call_batch) {
435 err = nc->call_batch(net, net->nfnl, skb, nlh,
436 (const struct nlattr **)cda,
440 /* The lock was released to autoload some module, we
441 * have to abort and start from scratch using the
444 if (err == -EAGAIN) {
445 status |= NFNL_BATCH_REPLAY;
450 if (nlh->nlmsg_flags & NLM_F_ACK || err) {
451 /* Errors are delivered once the full batch has been
452 * processed, this avoids that the same error is
453 * reported several times when replaying the batch.
455 if (nfnl_err_add(&err_list, nlh, err, &extack) < 0) {
456 /* We failed to enqueue an error, reset the
457 * list of errors and send OOM to userspace
458 * pointing to the batch header.
460 nfnl_err_reset(&err_list);
461 netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM,
463 status |= NFNL_BATCH_FAILURE;
466 /* We don't stop processing the batch on errors, thus,
467 * userspace gets all the errors that the batch
471 status |= NFNL_BATCH_FAILURE;
474 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
475 if (msglen > skb->len)
477 skb_pull(skb, msglen);
480 if (status & NFNL_BATCH_REPLAY) {
481 ss->abort(net, oskb);
482 nfnl_err_reset(&err_list);
484 module_put(ss->owner);
486 } else if (status == NFNL_BATCH_DONE) {
487 err = ss->commit(net, oskb);
488 if (err == -EAGAIN) {
489 status |= NFNL_BATCH_REPLAY;
492 ss->abort(net, oskb);
493 netlink_ack(oskb, nlmsg_hdr(oskb), err, NULL);
496 ss->abort(net, oskb);
501 nfnl_err_deliver(&err_list, oskb);
503 module_put(ss->owner);
506 static const struct nla_policy nfnl_batch_policy[NFNL_BATCH_MAX + 1] = {
507 [NFNL_BATCH_GENID] = { .type = NLA_U32 },
510 static void nfnetlink_rcv_skb_batch(struct sk_buff *skb, struct nlmsghdr *nlh)
512 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
513 struct nlattr *attr = (void *)nlh + min_len;
514 struct nlattr *cda[NFNL_BATCH_MAX + 1];
515 int attrlen = nlh->nlmsg_len - min_len;
516 struct nfgenmsg *nfgenmsg;
521 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
522 if (msglen > skb->len)
525 if (skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
528 err = nla_parse(cda, NFNL_BATCH_MAX, attr, attrlen, nfnl_batch_policy,
531 netlink_ack(skb, nlh, err, NULL);
534 if (cda[NFNL_BATCH_GENID])
535 gen_id = ntohl(nla_get_be32(cda[NFNL_BATCH_GENID]));
537 nfgenmsg = nlmsg_data(nlh);
538 skb_pull(skb, msglen);
539 /* Work around old nft using host byte order */
540 if (nfgenmsg->res_id == NFNL_SUBSYS_NFTABLES)
541 res_id = NFNL_SUBSYS_NFTABLES;
543 res_id = ntohs(nfgenmsg->res_id);
545 nfnetlink_rcv_batch(skb, nlh, res_id, gen_id);
548 static void nfnetlink_rcv(struct sk_buff *skb)
550 struct nlmsghdr *nlh = nlmsg_hdr(skb);
552 if (skb->len < NLMSG_HDRLEN ||
553 nlh->nlmsg_len < NLMSG_HDRLEN ||
554 skb->len < nlh->nlmsg_len)
557 if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
558 netlink_ack(skb, nlh, -EPERM, NULL);
562 if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN)
563 nfnetlink_rcv_skb_batch(skb, nlh);
565 netlink_rcv_skb(skb, nfnetlink_rcv_msg);
568 #ifdef CONFIG_MODULES
569 static int nfnetlink_bind(struct net *net, int group)
571 const struct nfnetlink_subsystem *ss;
574 if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
577 type = nfnl_group2type[group];
580 ss = nfnetlink_get_subsys(type << 8);
583 request_module_nowait("nfnetlink-subsys-%d", type);
588 static int __net_init nfnetlink_net_init(struct net *net)
591 struct netlink_kernel_cfg cfg = {
592 .groups = NFNLGRP_MAX,
593 .input = nfnetlink_rcv,
594 #ifdef CONFIG_MODULES
595 .bind = nfnetlink_bind,
599 nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
602 net->nfnl_stash = nfnl;
603 rcu_assign_pointer(net->nfnl, nfnl);
607 static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
611 list_for_each_entry(net, net_exit_list, exit_list)
612 RCU_INIT_POINTER(net->nfnl, NULL);
614 list_for_each_entry(net, net_exit_list, exit_list)
615 netlink_kernel_release(net->nfnl_stash);
618 static struct pernet_operations nfnetlink_net_ops = {
619 .init = nfnetlink_net_init,
620 .exit_batch = nfnetlink_net_exit_batch,
623 static int __init nfnetlink_init(void)
627 for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
628 BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
630 for (i=0; i<NFNL_SUBSYS_COUNT; i++)
631 mutex_init(&table[i].mutex);
633 return register_pernet_subsys(&nfnetlink_net_ops);
636 static void __exit nfnetlink_exit(void)
638 unregister_pernet_subsys(&nfnetlink_net_ops);
640 module_init(nfnetlink_init);
641 module_exit(nfnetlink_exit);