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/nfnetlink.h>
34 MODULE_LICENSE("GPL");
35 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
36 MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER);
37 MODULE_DESCRIPTION("Netfilter messages via netlink socket");
39 #define nfnl_dereference_protected(id) \
40 rcu_dereference_protected(table[(id)].subsys, \
41 lockdep_nfnl_is_held((id)))
43 #define NFNL_MAX_ATTR_COUNT 32
45 static unsigned int nfnetlink_pernet_id __read_mostly;
47 #ifdef CONFIG_NF_CONNTRACK_EVENTS
48 static DEFINE_SPINLOCK(nfnl_grp_active_lock);
57 const struct nfnetlink_subsystem __rcu *subsys;
58 } table[NFNL_SUBSYS_COUNT];
60 static struct lock_class_key nfnl_lockdep_keys[NFNL_SUBSYS_COUNT];
62 static const char *const nfnl_lockdep_names[NFNL_SUBSYS_COUNT] = {
63 [NFNL_SUBSYS_NONE] = "nfnl_subsys_none",
64 [NFNL_SUBSYS_CTNETLINK] = "nfnl_subsys_ctnetlink",
65 [NFNL_SUBSYS_CTNETLINK_EXP] = "nfnl_subsys_ctnetlink_exp",
66 [NFNL_SUBSYS_QUEUE] = "nfnl_subsys_queue",
67 [NFNL_SUBSYS_ULOG] = "nfnl_subsys_ulog",
68 [NFNL_SUBSYS_OSF] = "nfnl_subsys_osf",
69 [NFNL_SUBSYS_IPSET] = "nfnl_subsys_ipset",
70 [NFNL_SUBSYS_ACCT] = "nfnl_subsys_acct",
71 [NFNL_SUBSYS_CTNETLINK_TIMEOUT] = "nfnl_subsys_cttimeout",
72 [NFNL_SUBSYS_CTHELPER] = "nfnl_subsys_cthelper",
73 [NFNL_SUBSYS_NFTABLES] = "nfnl_subsys_nftables",
74 [NFNL_SUBSYS_NFT_COMPAT] = "nfnl_subsys_nftcompat",
75 [NFNL_SUBSYS_HOOK] = "nfnl_subsys_hook",
78 static const int nfnl_group2type[NFNLGRP_MAX+1] = {
79 [NFNLGRP_CONNTRACK_NEW] = NFNL_SUBSYS_CTNETLINK,
80 [NFNLGRP_CONNTRACK_UPDATE] = NFNL_SUBSYS_CTNETLINK,
81 [NFNLGRP_CONNTRACK_DESTROY] = NFNL_SUBSYS_CTNETLINK,
82 [NFNLGRP_CONNTRACK_EXP_NEW] = NFNL_SUBSYS_CTNETLINK_EXP,
83 [NFNLGRP_CONNTRACK_EXP_UPDATE] = NFNL_SUBSYS_CTNETLINK_EXP,
84 [NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP,
85 [NFNLGRP_NFTABLES] = NFNL_SUBSYS_NFTABLES,
86 [NFNLGRP_ACCT_QUOTA] = NFNL_SUBSYS_ACCT,
87 [NFNLGRP_NFTRACE] = NFNL_SUBSYS_NFTABLES,
90 static struct nfnl_net *nfnl_pernet(struct net *net)
92 return net_generic(net, nfnetlink_pernet_id);
95 void nfnl_lock(__u8 subsys_id)
97 mutex_lock(&table[subsys_id].mutex);
99 EXPORT_SYMBOL_GPL(nfnl_lock);
101 void nfnl_unlock(__u8 subsys_id)
103 mutex_unlock(&table[subsys_id].mutex);
105 EXPORT_SYMBOL_GPL(nfnl_unlock);
107 #ifdef CONFIG_PROVE_LOCKING
108 bool lockdep_nfnl_is_held(u8 subsys_id)
110 return lockdep_is_held(&table[subsys_id].mutex);
112 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
115 int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
119 /* Sanity-check attr_count size to avoid stack buffer overflow. */
120 for (cb_id = 0; cb_id < n->cb_count; cb_id++)
121 if (WARN_ON(n->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT))
124 nfnl_lock(n->subsys_id);
125 if (table[n->subsys_id].subsys) {
126 nfnl_unlock(n->subsys_id);
129 rcu_assign_pointer(table[n->subsys_id].subsys, n);
130 nfnl_unlock(n->subsys_id);
134 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
136 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
138 nfnl_lock(n->subsys_id);
139 table[n->subsys_id].subsys = NULL;
140 nfnl_unlock(n->subsys_id);
144 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
146 static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u16 type)
148 u8 subsys_id = NFNL_SUBSYS_ID(type);
150 if (subsys_id >= NFNL_SUBSYS_COUNT)
153 return rcu_dereference(table[subsys_id].subsys);
156 static inline const struct nfnl_callback *
157 nfnetlink_find_client(u16 type, const struct nfnetlink_subsystem *ss)
159 u8 cb_id = NFNL_MSG_TYPE(type);
161 if (cb_id >= ss->cb_count)
164 return &ss->cb[cb_id];
167 int nfnetlink_has_listeners(struct net *net, unsigned int group)
169 struct nfnl_net *nfnlnet = nfnl_pernet(net);
171 return netlink_has_listeners(nfnlnet->nfnl, group);
173 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
175 int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
176 unsigned int group, int echo, gfp_t flags)
178 struct nfnl_net *nfnlnet = nfnl_pernet(net);
180 return nlmsg_notify(nfnlnet->nfnl, skb, portid, group, echo, flags);
182 EXPORT_SYMBOL_GPL(nfnetlink_send);
184 int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
186 struct nfnl_net *nfnlnet = nfnl_pernet(net);
188 return netlink_set_err(nfnlnet->nfnl, portid, group, error);
190 EXPORT_SYMBOL_GPL(nfnetlink_set_err);
192 int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid)
194 struct nfnl_net *nfnlnet = nfnl_pernet(net);
197 err = nlmsg_unicast(nfnlnet->nfnl, skb, portid);
203 EXPORT_SYMBOL_GPL(nfnetlink_unicast);
205 void nfnetlink_broadcast(struct net *net, struct sk_buff *skb, __u32 portid,
206 __u32 group, gfp_t allocation)
208 struct nfnl_net *nfnlnet = nfnl_pernet(net);
210 netlink_broadcast(nfnlnet->nfnl, skb, portid, group, allocation);
212 EXPORT_SYMBOL_GPL(nfnetlink_broadcast);
214 /* Process one complete nfnetlink message. */
215 static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
216 struct netlink_ext_ack *extack)
218 struct net *net = sock_net(skb->sk);
219 const struct nfnl_callback *nc;
220 const struct nfnetlink_subsystem *ss;
223 /* All the messages must at least contain nfgenmsg */
224 if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
227 type = nlh->nlmsg_type;
231 ss = nfnetlink_get_subsys(type);
233 #ifdef CONFIG_MODULES
235 request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
237 ss = nfnetlink_get_subsys(type);
246 nc = nfnetlink_find_client(type, ss);
253 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
254 struct nfnl_net *nfnlnet = nfnl_pernet(net);
255 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
256 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
257 struct nlattr *attr = (void *)nlh + min_len;
258 int attrlen = nlh->nlmsg_len - min_len;
259 __u8 subsys_id = NFNL_SUBSYS_ID(type);
260 struct nfnl_info info = {
264 .nfmsg = nlmsg_data(nlh),
268 /* Sanity-check NFNL_MAX_ATTR_COUNT */
269 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
274 err = nla_parse_deprecated(cda, ss->cb[cb_id].attr_count,
276 ss->cb[cb_id].policy, extack);
289 err = nc->call(skb, &info, (const struct nlattr **)cda);
294 nfnl_lock(subsys_id);
295 if (nfnl_dereference_protected(subsys_id) != ss ||
296 nfnetlink_find_client(type, ss) != nc) {
297 nfnl_unlock(subsys_id);
301 err = nc->call(skb, &info, (const struct nlattr **)cda);
302 nfnl_unlock(subsys_id);
316 struct list_head head;
317 struct nlmsghdr *nlh;
319 struct netlink_ext_ack extack;
322 static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err,
323 const struct netlink_ext_ack *extack)
325 struct nfnl_err *nfnl_err;
327 nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL);
328 if (nfnl_err == NULL)
333 nfnl_err->extack = *extack;
334 list_add_tail(&nfnl_err->head, list);
339 static void nfnl_err_del(struct nfnl_err *nfnl_err)
341 list_del(&nfnl_err->head);
345 static void nfnl_err_reset(struct list_head *err_list)
347 struct nfnl_err *nfnl_err, *next;
349 list_for_each_entry_safe(nfnl_err, next, err_list, head)
350 nfnl_err_del(nfnl_err);
353 static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
355 struct nfnl_err *nfnl_err, *next;
357 list_for_each_entry_safe(nfnl_err, next, err_list, head) {
358 netlink_ack(skb, nfnl_err->nlh, nfnl_err->err,
360 nfnl_err_del(nfnl_err);
365 NFNL_BATCH_FAILURE = (1 << 0),
366 NFNL_BATCH_DONE = (1 << 1),
367 NFNL_BATCH_REPLAY = (1 << 2),
370 static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
371 u16 subsys_id, u32 genid)
373 struct sk_buff *oskb = skb;
374 struct net *net = sock_net(skb->sk);
375 const struct nfnetlink_subsystem *ss;
376 const struct nfnl_callback *nc;
377 struct netlink_ext_ack extack;
382 if (subsys_id >= NFNL_SUBSYS_COUNT)
383 return netlink_ack(skb, nlh, -EINVAL, NULL);
387 skb = netlink_skb_clone(oskb, GFP_KERNEL);
389 return netlink_ack(oskb, nlh, -ENOMEM, NULL);
391 nfnl_lock(subsys_id);
392 ss = nfnl_dereference_protected(subsys_id);
394 #ifdef CONFIG_MODULES
395 nfnl_unlock(subsys_id);
396 request_module("nfnetlink-subsys-%d", subsys_id);
397 nfnl_lock(subsys_id);
398 ss = nfnl_dereference_protected(subsys_id);
402 nfnl_unlock(subsys_id);
403 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
404 return kfree_skb(skb);
408 if (!ss->valid_genid || !ss->commit || !ss->abort) {
409 nfnl_unlock(subsys_id);
410 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
411 return kfree_skb(skb);
414 if (!try_module_get(ss->owner)) {
415 nfnl_unlock(subsys_id);
416 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
417 return kfree_skb(skb);
420 if (!ss->valid_genid(net, genid)) {
421 module_put(ss->owner);
422 nfnl_unlock(subsys_id);
423 netlink_ack(oskb, nlh, -ERESTART, NULL);
424 return kfree_skb(skb);
427 nfnl_unlock(subsys_id);
429 while (skb->len >= nlmsg_total_size(0)) {
432 if (fatal_signal_pending(current)) {
433 nfnl_err_reset(&err_list);
435 status = NFNL_BATCH_FAILURE;
439 memset(&extack, 0, sizeof(extack));
440 nlh = nlmsg_hdr(skb);
443 if (nlh->nlmsg_len < NLMSG_HDRLEN ||
444 skb->len < nlh->nlmsg_len ||
445 nlmsg_len(nlh) < sizeof(struct nfgenmsg)) {
446 nfnl_err_reset(&err_list);
447 status |= NFNL_BATCH_FAILURE;
451 /* Only requests are handled by the kernel */
452 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
457 type = nlh->nlmsg_type;
458 if (type == NFNL_MSG_BATCH_BEGIN) {
459 /* Malformed: Batch begin twice */
460 nfnl_err_reset(&err_list);
461 status |= NFNL_BATCH_FAILURE;
463 } else if (type == NFNL_MSG_BATCH_END) {
464 status |= NFNL_BATCH_DONE;
466 } else if (type < NLMSG_MIN_TYPE) {
471 /* We only accept a batch with messages for the same
474 if (NFNL_SUBSYS_ID(type) != subsys_id) {
479 nc = nfnetlink_find_client(type, ss);
485 if (nc->type != NFNL_CB_BATCH) {
491 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
492 struct nfnl_net *nfnlnet = nfnl_pernet(net);
493 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
494 struct nlattr *attr = (void *)nlh + min_len;
495 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
496 int attrlen = nlh->nlmsg_len - min_len;
497 struct nfnl_info info = {
501 .nfmsg = nlmsg_data(nlh),
505 /* Sanity-check NFTA_MAX_ATTR */
506 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
511 err = nla_parse_deprecated(cda,
512 ss->cb[cb_id].attr_count,
514 ss->cb[cb_id].policy, NULL);
518 err = nc->call(skb, &info, (const struct nlattr **)cda);
520 /* The lock was released to autoload some module, we
521 * have to abort and start from scratch using the
524 if (err == -EAGAIN) {
525 status |= NFNL_BATCH_REPLAY;
530 if (nlh->nlmsg_flags & NLM_F_ACK || err) {
531 /* Errors are delivered once the full batch has been
532 * processed, this avoids that the same error is
533 * reported several times when replaying the batch.
535 if (nfnl_err_add(&err_list, nlh, err, &extack) < 0) {
536 /* We failed to enqueue an error, reset the
537 * list of errors and send OOM to userspace
538 * pointing to the batch header.
540 nfnl_err_reset(&err_list);
541 netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM,
543 status |= NFNL_BATCH_FAILURE;
546 /* We don't stop processing the batch on errors, thus,
547 * userspace gets all the errors that the batch
551 status |= NFNL_BATCH_FAILURE;
554 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
555 if (msglen > skb->len)
557 skb_pull(skb, msglen);
560 if (status & NFNL_BATCH_REPLAY) {
561 ss->abort(net, oskb, NFNL_ABORT_AUTOLOAD);
562 nfnl_err_reset(&err_list);
564 module_put(ss->owner);
566 } else if (status == NFNL_BATCH_DONE) {
567 err = ss->commit(net, oskb);
568 if (err == -EAGAIN) {
569 status |= NFNL_BATCH_REPLAY;
572 ss->abort(net, oskb, NFNL_ABORT_NONE);
573 netlink_ack(oskb, nlmsg_hdr(oskb), err, NULL);
576 enum nfnl_abort_action abort_action;
578 if (status & NFNL_BATCH_FAILURE)
579 abort_action = NFNL_ABORT_NONE;
581 abort_action = NFNL_ABORT_VALIDATE;
583 err = ss->abort(net, oskb, abort_action);
584 if (err == -EAGAIN) {
585 nfnl_err_reset(&err_list);
587 module_put(ss->owner);
588 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(net->ct.ctnetlink_has_listener);
689 if ((v & group_bit) == 0) {
692 /* read concurrently without nfnl_grp_active_lock held. */
693 WRITE_ONCE(net->ct.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(net->ct.ctnetlink_has_listener);
751 /* read concurrently without nfnl_grp_active_lock held. */
752 WRITE_ONCE(net->ct.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);