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 <net/netns/generic.h>
32 #include <linux/netfilter.h>
33 #include <linux/netfilter/nfnetlink.h>
35 MODULE_LICENSE("GPL");
36 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
37 MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER);
38 MODULE_DESCRIPTION("Netfilter messages via netlink socket");
40 #define nfnl_dereference_protected(id) \
41 rcu_dereference_protected(table[(id)].subsys, \
42 lockdep_nfnl_is_held((id)))
44 #define NFNL_MAX_ATTR_COUNT 32
46 static unsigned int nfnetlink_pernet_id __read_mostly;
48 #ifdef CONFIG_NF_CONNTRACK_EVENTS
49 static DEFINE_SPINLOCK(nfnl_grp_active_lock);
58 const struct nfnetlink_subsystem __rcu *subsys;
59 } table[NFNL_SUBSYS_COUNT];
61 static struct lock_class_key nfnl_lockdep_keys[NFNL_SUBSYS_COUNT];
63 static const char *const nfnl_lockdep_names[NFNL_SUBSYS_COUNT] = {
64 [NFNL_SUBSYS_NONE] = "nfnl_subsys_none",
65 [NFNL_SUBSYS_CTNETLINK] = "nfnl_subsys_ctnetlink",
66 [NFNL_SUBSYS_CTNETLINK_EXP] = "nfnl_subsys_ctnetlink_exp",
67 [NFNL_SUBSYS_QUEUE] = "nfnl_subsys_queue",
68 [NFNL_SUBSYS_ULOG] = "nfnl_subsys_ulog",
69 [NFNL_SUBSYS_OSF] = "nfnl_subsys_osf",
70 [NFNL_SUBSYS_IPSET] = "nfnl_subsys_ipset",
71 [NFNL_SUBSYS_ACCT] = "nfnl_subsys_acct",
72 [NFNL_SUBSYS_CTNETLINK_TIMEOUT] = "nfnl_subsys_cttimeout",
73 [NFNL_SUBSYS_CTHELPER] = "nfnl_subsys_cthelper",
74 [NFNL_SUBSYS_NFTABLES] = "nfnl_subsys_nftables",
75 [NFNL_SUBSYS_NFT_COMPAT] = "nfnl_subsys_nftcompat",
76 [NFNL_SUBSYS_HOOK] = "nfnl_subsys_hook",
79 static const int nfnl_group2type[NFNLGRP_MAX+1] = {
80 [NFNLGRP_CONNTRACK_NEW] = NFNL_SUBSYS_CTNETLINK,
81 [NFNLGRP_CONNTRACK_UPDATE] = NFNL_SUBSYS_CTNETLINK,
82 [NFNLGRP_CONNTRACK_DESTROY] = NFNL_SUBSYS_CTNETLINK,
83 [NFNLGRP_CONNTRACK_EXP_NEW] = NFNL_SUBSYS_CTNETLINK_EXP,
84 [NFNLGRP_CONNTRACK_EXP_UPDATE] = NFNL_SUBSYS_CTNETLINK_EXP,
85 [NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP,
86 [NFNLGRP_NFTABLES] = NFNL_SUBSYS_NFTABLES,
87 [NFNLGRP_ACCT_QUOTA] = NFNL_SUBSYS_ACCT,
88 [NFNLGRP_NFTRACE] = NFNL_SUBSYS_NFTABLES,
91 static struct nfnl_net *nfnl_pernet(struct net *net)
93 return net_generic(net, nfnetlink_pernet_id);
96 void nfnl_lock(__u8 subsys_id)
98 mutex_lock(&table[subsys_id].mutex);
100 EXPORT_SYMBOL_GPL(nfnl_lock);
102 void nfnl_unlock(__u8 subsys_id)
104 mutex_unlock(&table[subsys_id].mutex);
106 EXPORT_SYMBOL_GPL(nfnl_unlock);
108 #ifdef CONFIG_PROVE_LOCKING
109 bool lockdep_nfnl_is_held(u8 subsys_id)
111 return lockdep_is_held(&table[subsys_id].mutex);
113 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
116 int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
120 /* Sanity-check attr_count size to avoid stack buffer overflow. */
121 for (cb_id = 0; cb_id < n->cb_count; cb_id++)
122 if (WARN_ON(n->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT))
125 nfnl_lock(n->subsys_id);
126 if (table[n->subsys_id].subsys) {
127 nfnl_unlock(n->subsys_id);
130 rcu_assign_pointer(table[n->subsys_id].subsys, n);
131 nfnl_unlock(n->subsys_id);
135 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
137 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
139 nfnl_lock(n->subsys_id);
140 table[n->subsys_id].subsys = NULL;
141 nfnl_unlock(n->subsys_id);
145 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
147 static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u16 type)
149 u8 subsys_id = NFNL_SUBSYS_ID(type);
151 if (subsys_id >= NFNL_SUBSYS_COUNT)
154 return rcu_dereference(table[subsys_id].subsys);
157 static inline const struct nfnl_callback *
158 nfnetlink_find_client(u16 type, const struct nfnetlink_subsystem *ss)
160 u8 cb_id = NFNL_MSG_TYPE(type);
162 if (cb_id >= ss->cb_count)
165 return &ss->cb[cb_id];
168 int nfnetlink_has_listeners(struct net *net, unsigned int group)
170 struct nfnl_net *nfnlnet = nfnl_pernet(net);
172 return netlink_has_listeners(nfnlnet->nfnl, group);
174 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
176 int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
177 unsigned int group, int echo, gfp_t flags)
179 struct nfnl_net *nfnlnet = nfnl_pernet(net);
181 return nlmsg_notify(nfnlnet->nfnl, skb, portid, group, echo, flags);
183 EXPORT_SYMBOL_GPL(nfnetlink_send);
185 int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
187 struct nfnl_net *nfnlnet = nfnl_pernet(net);
189 return netlink_set_err(nfnlnet->nfnl, portid, group, error);
191 EXPORT_SYMBOL_GPL(nfnetlink_set_err);
193 int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid)
195 struct nfnl_net *nfnlnet = nfnl_pernet(net);
198 err = nlmsg_unicast(nfnlnet->nfnl, skb, portid);
204 EXPORT_SYMBOL_GPL(nfnetlink_unicast);
206 void nfnetlink_broadcast(struct net *net, struct sk_buff *skb, __u32 portid,
207 __u32 group, gfp_t allocation)
209 struct nfnl_net *nfnlnet = nfnl_pernet(net);
211 netlink_broadcast(nfnlnet->nfnl, skb, portid, group, allocation);
213 EXPORT_SYMBOL_GPL(nfnetlink_broadcast);
215 /* Process one complete nfnetlink message. */
216 static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
217 struct netlink_ext_ack *extack)
219 struct net *net = sock_net(skb->sk);
220 const struct nfnl_callback *nc;
221 const struct nfnetlink_subsystem *ss;
224 /* All the messages must at least contain nfgenmsg */
225 if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
228 type = nlh->nlmsg_type;
232 ss = nfnetlink_get_subsys(type);
234 #ifdef CONFIG_MODULES
236 request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
238 ss = nfnetlink_get_subsys(type);
247 nc = nfnetlink_find_client(type, ss);
254 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
255 struct nfnl_net *nfnlnet = nfnl_pernet(net);
256 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
257 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
258 struct nlattr *attr = (void *)nlh + min_len;
259 int attrlen = nlh->nlmsg_len - min_len;
260 __u8 subsys_id = NFNL_SUBSYS_ID(type);
261 struct nfnl_info info = {
265 .nfmsg = nlmsg_data(nlh),
269 /* Sanity-check NFNL_MAX_ATTR_COUNT */
270 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
275 err = nla_parse_deprecated(cda, ss->cb[cb_id].attr_count,
277 ss->cb[cb_id].policy, extack);
290 err = nc->call(skb, &info, (const struct nlattr **)cda);
295 nfnl_lock(subsys_id);
296 if (nfnl_dereference_protected(subsys_id) != ss ||
297 nfnetlink_find_client(type, ss) != nc) {
298 nfnl_unlock(subsys_id);
302 err = nc->call(skb, &info, (const struct nlattr **)cda);
303 nfnl_unlock(subsys_id);
317 struct list_head head;
318 struct nlmsghdr *nlh;
320 struct netlink_ext_ack extack;
323 static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err,
324 const struct netlink_ext_ack *extack)
326 struct nfnl_err *nfnl_err;
328 nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL);
329 if (nfnl_err == NULL)
334 nfnl_err->extack = *extack;
335 list_add_tail(&nfnl_err->head, list);
340 static void nfnl_err_del(struct nfnl_err *nfnl_err)
342 list_del(&nfnl_err->head);
346 static void nfnl_err_reset(struct list_head *err_list)
348 struct nfnl_err *nfnl_err, *next;
350 list_for_each_entry_safe(nfnl_err, next, err_list, head)
351 nfnl_err_del(nfnl_err);
354 static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
356 struct nfnl_err *nfnl_err, *next;
358 list_for_each_entry_safe(nfnl_err, next, err_list, head) {
359 netlink_ack(skb, nfnl_err->nlh, nfnl_err->err,
361 nfnl_err_del(nfnl_err);
366 NFNL_BATCH_FAILURE = (1 << 0),
367 NFNL_BATCH_DONE = (1 << 1),
368 NFNL_BATCH_REPLAY = (1 << 2),
371 static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
372 u16 subsys_id, u32 genid)
374 struct sk_buff *oskb = skb;
375 struct net *net = sock_net(skb->sk);
376 const struct nfnetlink_subsystem *ss;
377 const struct nfnl_callback *nc;
378 struct netlink_ext_ack extack;
383 if (subsys_id >= NFNL_SUBSYS_COUNT)
384 return netlink_ack(skb, nlh, -EINVAL, NULL);
388 skb = netlink_skb_clone(oskb, GFP_KERNEL);
390 return netlink_ack(oskb, nlh, -ENOMEM, NULL);
392 nfnl_lock(subsys_id);
393 ss = nfnl_dereference_protected(subsys_id);
395 #ifdef CONFIG_MODULES
396 nfnl_unlock(subsys_id);
397 request_module("nfnetlink-subsys-%d", subsys_id);
398 nfnl_lock(subsys_id);
399 ss = nfnl_dereference_protected(subsys_id);
403 nfnl_unlock(subsys_id);
404 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
405 return kfree_skb(skb);
409 if (!ss->valid_genid || !ss->commit || !ss->abort) {
410 nfnl_unlock(subsys_id);
411 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
412 return kfree_skb(skb);
415 if (!try_module_get(ss->owner)) {
416 nfnl_unlock(subsys_id);
417 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
418 return kfree_skb(skb);
421 if (!ss->valid_genid(net, genid)) {
422 module_put(ss->owner);
423 nfnl_unlock(subsys_id);
424 netlink_ack(oskb, nlh, -ERESTART, NULL);
425 return kfree_skb(skb);
428 nfnl_unlock(subsys_id);
430 while (skb->len >= nlmsg_total_size(0)) {
433 if (fatal_signal_pending(current)) {
434 nfnl_err_reset(&err_list);
436 status = NFNL_BATCH_FAILURE;
440 memset(&extack, 0, sizeof(extack));
441 nlh = nlmsg_hdr(skb);
444 if (nlh->nlmsg_len < NLMSG_HDRLEN ||
445 skb->len < nlh->nlmsg_len ||
446 nlmsg_len(nlh) < sizeof(struct nfgenmsg)) {
447 nfnl_err_reset(&err_list);
448 status |= NFNL_BATCH_FAILURE;
452 /* Only requests are handled by the kernel */
453 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
458 type = nlh->nlmsg_type;
459 if (type == NFNL_MSG_BATCH_BEGIN) {
460 /* Malformed: Batch begin twice */
461 nfnl_err_reset(&err_list);
462 status |= NFNL_BATCH_FAILURE;
464 } else if (type == NFNL_MSG_BATCH_END) {
465 status |= NFNL_BATCH_DONE;
467 } else if (type < NLMSG_MIN_TYPE) {
472 /* We only accept a batch with messages for the same
475 if (NFNL_SUBSYS_ID(type) != subsys_id) {
480 nc = nfnetlink_find_client(type, ss);
486 if (nc->type != NFNL_CB_BATCH) {
492 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
493 struct nfnl_net *nfnlnet = nfnl_pernet(net);
494 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
495 struct nlattr *attr = (void *)nlh + min_len;
496 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
497 int attrlen = nlh->nlmsg_len - min_len;
498 struct nfnl_info info = {
502 .nfmsg = nlmsg_data(nlh),
506 /* Sanity-check NFTA_MAX_ATTR */
507 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
512 err = nla_parse_deprecated(cda,
513 ss->cb[cb_id].attr_count,
515 ss->cb[cb_id].policy, NULL);
519 err = nc->call(skb, &info, (const struct nlattr **)cda);
521 /* The lock was released to autoload some module, we
522 * have to abort and start from scratch using the
525 if (err == -EAGAIN) {
526 status |= NFNL_BATCH_REPLAY;
531 if (nlh->nlmsg_flags & NLM_F_ACK || err) {
532 /* Errors are delivered once the full batch has been
533 * processed, this avoids that the same error is
534 * reported several times when replaying the batch.
536 if (err == -ENOMEM ||
537 nfnl_err_add(&err_list, nlh, err, &extack) < 0) {
538 /* We failed to enqueue an error, reset the
539 * list of errors and send OOM to userspace
540 * pointing to the batch header.
542 nfnl_err_reset(&err_list);
543 netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM,
545 status |= NFNL_BATCH_FAILURE;
548 /* We don't stop processing the batch on errors, thus,
549 * userspace gets all the errors that the batch
553 status |= NFNL_BATCH_FAILURE;
556 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
557 if (msglen > skb->len)
559 skb_pull(skb, msglen);
562 if (status & NFNL_BATCH_REPLAY) {
563 ss->abort(net, oskb, NFNL_ABORT_AUTOLOAD);
564 nfnl_err_reset(&err_list);
566 module_put(ss->owner);
568 } else if (status == NFNL_BATCH_DONE) {
569 err = ss->commit(net, oskb);
570 if (err == -EAGAIN) {
571 status |= NFNL_BATCH_REPLAY;
574 ss->abort(net, oskb, NFNL_ABORT_NONE);
575 netlink_ack(oskb, nlmsg_hdr(oskb), err, NULL);
578 enum nfnl_abort_action abort_action;
580 if (status & NFNL_BATCH_FAILURE)
581 abort_action = NFNL_ABORT_NONE;
583 abort_action = NFNL_ABORT_VALIDATE;
585 err = ss->abort(net, oskb, abort_action);
586 if (err == -EAGAIN) {
587 nfnl_err_reset(&err_list);
589 module_put(ss->owner);
590 status |= NFNL_BATCH_FAILURE;
595 nfnl_err_deliver(&err_list, oskb);
597 module_put(ss->owner);
600 static const struct nla_policy nfnl_batch_policy[NFNL_BATCH_MAX + 1] = {
601 [NFNL_BATCH_GENID] = { .type = NLA_U32 },
604 static void nfnetlink_rcv_skb_batch(struct sk_buff *skb, struct nlmsghdr *nlh)
606 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
607 struct nlattr *attr = (void *)nlh + min_len;
608 struct nlattr *cda[NFNL_BATCH_MAX + 1];
609 int attrlen = nlh->nlmsg_len - min_len;
610 struct nfgenmsg *nfgenmsg;
615 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
616 if (msglen > skb->len)
619 if (skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
622 err = nla_parse_deprecated(cda, NFNL_BATCH_MAX, attr, attrlen,
623 nfnl_batch_policy, NULL);
625 netlink_ack(skb, nlh, err, NULL);
628 if (cda[NFNL_BATCH_GENID])
629 gen_id = ntohl(nla_get_be32(cda[NFNL_BATCH_GENID]));
631 nfgenmsg = nlmsg_data(nlh);
632 skb_pull(skb, msglen);
633 /* Work around old nft using host byte order */
634 if (nfgenmsg->res_id == (__force __be16)NFNL_SUBSYS_NFTABLES)
635 res_id = NFNL_SUBSYS_NFTABLES;
637 res_id = ntohs(nfgenmsg->res_id);
639 nfnetlink_rcv_batch(skb, nlh, res_id, gen_id);
642 static void nfnetlink_rcv(struct sk_buff *skb)
644 struct nlmsghdr *nlh = nlmsg_hdr(skb);
646 if (skb->len < NLMSG_HDRLEN ||
647 nlh->nlmsg_len < NLMSG_HDRLEN ||
648 skb->len < nlh->nlmsg_len)
651 if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
652 netlink_ack(skb, nlh, -EPERM, NULL);
656 if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN)
657 nfnetlink_rcv_skb_batch(skb, nlh);
659 netlink_rcv_skb(skb, nfnetlink_rcv_msg);
662 static void nfnetlink_bind_event(struct net *net, unsigned int group)
664 #ifdef CONFIG_NF_CONNTRACK_EVENTS
668 /* All NFNLGRP_CONNTRACK_* group bits fit into u8.
669 * The other groups are not relevant and can be ignored.
674 type = nfnl_group2type[group];
677 case NFNL_SUBSYS_CTNETLINK:
679 case NFNL_SUBSYS_CTNETLINK_EXP:
685 group_bit = (1 << group);
687 spin_lock(&nfnl_grp_active_lock);
688 v = READ_ONCE(nf_ctnetlink_has_listener);
689 if ((v & group_bit) == 0) {
692 /* read concurrently without nfnl_grp_active_lock held. */
693 WRITE_ONCE(nf_ctnetlink_has_listener, v);
696 spin_unlock(&nfnl_grp_active_lock);
700 static int nfnetlink_bind(struct net *net, int group)
702 const struct nfnetlink_subsystem *ss;
705 if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
708 type = nfnl_group2type[group];
711 ss = nfnetlink_get_subsys(type << 8);
714 request_module_nowait("nfnetlink-subsys-%d", type);
716 nfnetlink_bind_event(net, group);
720 static void nfnetlink_unbind(struct net *net, int group)
722 #ifdef CONFIG_NF_CONNTRACK_EVENTS
725 if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
728 type = nfnl_group2type[group];
731 case NFNL_SUBSYS_CTNETLINK:
733 case NFNL_SUBSYS_CTNETLINK_EXP:
739 /* ctnetlink_has_listener is u8 */
743 group_bit = (1 << group);
745 spin_lock(&nfnl_grp_active_lock);
746 if (!nfnetlink_has_listeners(net, group)) {
747 u8 v = READ_ONCE(nf_ctnetlink_has_listener);
751 /* read concurrently without nfnl_grp_active_lock held. */
752 WRITE_ONCE(nf_ctnetlink_has_listener, v);
754 spin_unlock(&nfnl_grp_active_lock);
758 static int __net_init nfnetlink_net_init(struct net *net)
760 struct nfnl_net *nfnlnet = nfnl_pernet(net);
761 struct netlink_kernel_cfg cfg = {
762 .groups = NFNLGRP_MAX,
763 .input = nfnetlink_rcv,
764 .bind = nfnetlink_bind,
765 .unbind = nfnetlink_unbind,
768 nfnlnet->nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
774 static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
776 struct nfnl_net *nfnlnet;
779 list_for_each_entry(net, net_exit_list, exit_list) {
780 nfnlnet = nfnl_pernet(net);
782 netlink_kernel_release(nfnlnet->nfnl);
786 static struct pernet_operations nfnetlink_net_ops = {
787 .init = nfnetlink_net_init,
788 .exit_batch = nfnetlink_net_exit_batch,
789 .id = &nfnetlink_pernet_id,
790 .size = sizeof(struct nfnl_net),
793 static int __init nfnetlink_init(void)
797 for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
798 BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
800 for (i=0; i<NFNL_SUBSYS_COUNT; i++)
801 __mutex_init(&table[i].mutex, nfnl_lockdep_names[i], &nfnl_lockdep_keys[i]);
803 return register_pernet_subsys(&nfnetlink_net_ops);
806 static void __exit nfnetlink_exit(void)
808 unregister_pernet_subsys(&nfnetlink_net_ops);
810 module_init(nfnetlink_init);
811 module_exit(nfnetlink_exit);