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;
53 const struct nfnetlink_subsystem __rcu *subsys;
54 } table[NFNL_SUBSYS_COUNT];
56 static struct lock_class_key nfnl_lockdep_keys[NFNL_SUBSYS_COUNT];
58 static const char *const nfnl_lockdep_names[NFNL_SUBSYS_COUNT] = {
59 [NFNL_SUBSYS_NONE] = "nfnl_subsys_none",
60 [NFNL_SUBSYS_CTNETLINK] = "nfnl_subsys_ctnetlink",
61 [NFNL_SUBSYS_CTNETLINK_EXP] = "nfnl_subsys_ctnetlink_exp",
62 [NFNL_SUBSYS_QUEUE] = "nfnl_subsys_queue",
63 [NFNL_SUBSYS_ULOG] = "nfnl_subsys_ulog",
64 [NFNL_SUBSYS_OSF] = "nfnl_subsys_osf",
65 [NFNL_SUBSYS_IPSET] = "nfnl_subsys_ipset",
66 [NFNL_SUBSYS_ACCT] = "nfnl_subsys_acct",
67 [NFNL_SUBSYS_CTNETLINK_TIMEOUT] = "nfnl_subsys_cttimeout",
68 [NFNL_SUBSYS_CTHELPER] = "nfnl_subsys_cthelper",
69 [NFNL_SUBSYS_NFTABLES] = "nfnl_subsys_nftables",
70 [NFNL_SUBSYS_NFT_COMPAT] = "nfnl_subsys_nftcompat",
73 static const int nfnl_group2type[NFNLGRP_MAX+1] = {
74 [NFNLGRP_CONNTRACK_NEW] = NFNL_SUBSYS_CTNETLINK,
75 [NFNLGRP_CONNTRACK_UPDATE] = NFNL_SUBSYS_CTNETLINK,
76 [NFNLGRP_CONNTRACK_DESTROY] = NFNL_SUBSYS_CTNETLINK,
77 [NFNLGRP_CONNTRACK_EXP_NEW] = NFNL_SUBSYS_CTNETLINK_EXP,
78 [NFNLGRP_CONNTRACK_EXP_UPDATE] = NFNL_SUBSYS_CTNETLINK_EXP,
79 [NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP,
80 [NFNLGRP_NFTABLES] = NFNL_SUBSYS_NFTABLES,
81 [NFNLGRP_ACCT_QUOTA] = NFNL_SUBSYS_ACCT,
82 [NFNLGRP_NFTRACE] = NFNL_SUBSYS_NFTABLES,
85 static struct nfnl_net *nfnl_pernet(struct net *net)
87 return net_generic(net, nfnetlink_pernet_id);
90 void nfnl_lock(__u8 subsys_id)
92 mutex_lock(&table[subsys_id].mutex);
94 EXPORT_SYMBOL_GPL(nfnl_lock);
96 void nfnl_unlock(__u8 subsys_id)
98 mutex_unlock(&table[subsys_id].mutex);
100 EXPORT_SYMBOL_GPL(nfnl_unlock);
102 #ifdef CONFIG_PROVE_LOCKING
103 bool lockdep_nfnl_is_held(u8 subsys_id)
105 return lockdep_is_held(&table[subsys_id].mutex);
107 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
110 int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
114 /* Sanity-check attr_count size to avoid stack buffer overflow. */
115 for (cb_id = 0; cb_id < n->cb_count; cb_id++)
116 if (WARN_ON(n->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT))
119 nfnl_lock(n->subsys_id);
120 if (table[n->subsys_id].subsys) {
121 nfnl_unlock(n->subsys_id);
124 rcu_assign_pointer(table[n->subsys_id].subsys, n);
125 nfnl_unlock(n->subsys_id);
129 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
131 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
133 nfnl_lock(n->subsys_id);
134 table[n->subsys_id].subsys = NULL;
135 nfnl_unlock(n->subsys_id);
139 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
141 static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u16 type)
143 u8 subsys_id = NFNL_SUBSYS_ID(type);
145 if (subsys_id >= NFNL_SUBSYS_COUNT)
148 return rcu_dereference(table[subsys_id].subsys);
151 static inline const struct nfnl_callback *
152 nfnetlink_find_client(u16 type, const struct nfnetlink_subsystem *ss)
154 u8 cb_id = NFNL_MSG_TYPE(type);
156 if (cb_id >= ss->cb_count)
159 return &ss->cb[cb_id];
162 int nfnetlink_has_listeners(struct net *net, unsigned int group)
164 struct nfnl_net *nfnlnet = nfnl_pernet(net);
166 return netlink_has_listeners(nfnlnet->nfnl, group);
168 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
170 int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
171 unsigned int group, int echo, gfp_t flags)
173 struct nfnl_net *nfnlnet = nfnl_pernet(net);
175 return nlmsg_notify(nfnlnet->nfnl, skb, portid, group, echo, flags);
177 EXPORT_SYMBOL_GPL(nfnetlink_send);
179 int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
181 struct nfnl_net *nfnlnet = nfnl_pernet(net);
183 return netlink_set_err(nfnlnet->nfnl, portid, group, error);
185 EXPORT_SYMBOL_GPL(nfnetlink_set_err);
187 int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid)
189 struct nfnl_net *nfnlnet = nfnl_pernet(net);
192 err = nlmsg_unicast(nfnlnet->nfnl, skb, portid);
198 EXPORT_SYMBOL_GPL(nfnetlink_unicast);
200 void nfnetlink_broadcast(struct net *net, struct sk_buff *skb, __u32 portid,
201 __u32 group, gfp_t allocation)
203 struct nfnl_net *nfnlnet = nfnl_pernet(net);
205 netlink_broadcast(nfnlnet->nfnl, skb, portid, group, allocation);
207 EXPORT_SYMBOL_GPL(nfnetlink_broadcast);
209 /* Process one complete nfnetlink message. */
210 static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
211 struct netlink_ext_ack *extack)
213 struct net *net = sock_net(skb->sk);
214 const struct nfnl_callback *nc;
215 const struct nfnetlink_subsystem *ss;
218 /* All the messages must at least contain nfgenmsg */
219 if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
222 type = nlh->nlmsg_type;
226 ss = nfnetlink_get_subsys(type);
228 #ifdef CONFIG_MODULES
230 request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
232 ss = nfnetlink_get_subsys(type);
241 nc = nfnetlink_find_client(type, ss);
248 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
249 struct nfnl_net *nfnlnet = nfnl_pernet(net);
250 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
251 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
252 struct nlattr *attr = (void *)nlh + min_len;
253 int attrlen = nlh->nlmsg_len - min_len;
254 __u8 subsys_id = NFNL_SUBSYS_ID(type);
255 struct nfnl_info info = {
262 /* Sanity-check NFNL_MAX_ATTR_COUNT */
263 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
268 err = nla_parse_deprecated(cda, ss->cb[cb_id].attr_count,
270 ss->cb[cb_id].policy, extack);
283 err = nc->call(skb, &info, (const struct nlattr **)cda);
288 nfnl_lock(subsys_id);
289 if (nfnl_dereference_protected(subsys_id) != ss ||
290 nfnetlink_find_client(type, ss) != nc) {
294 err = nc->call(skb, &info, (const struct nlattr **)cda);
295 nfnl_unlock(subsys_id);
309 struct list_head head;
310 struct nlmsghdr *nlh;
312 struct netlink_ext_ack extack;
315 static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err,
316 const struct netlink_ext_ack *extack)
318 struct nfnl_err *nfnl_err;
320 nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL);
321 if (nfnl_err == NULL)
326 nfnl_err->extack = *extack;
327 list_add_tail(&nfnl_err->head, list);
332 static void nfnl_err_del(struct nfnl_err *nfnl_err)
334 list_del(&nfnl_err->head);
338 static void nfnl_err_reset(struct list_head *err_list)
340 struct nfnl_err *nfnl_err, *next;
342 list_for_each_entry_safe(nfnl_err, next, err_list, head)
343 nfnl_err_del(nfnl_err);
346 static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
348 struct nfnl_err *nfnl_err, *next;
350 list_for_each_entry_safe(nfnl_err, next, err_list, head) {
351 netlink_ack(skb, nfnl_err->nlh, nfnl_err->err,
353 nfnl_err_del(nfnl_err);
358 NFNL_BATCH_FAILURE = (1 << 0),
359 NFNL_BATCH_DONE = (1 << 1),
360 NFNL_BATCH_REPLAY = (1 << 2),
363 static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
364 u16 subsys_id, u32 genid)
366 struct sk_buff *oskb = skb;
367 struct net *net = sock_net(skb->sk);
368 const struct nfnetlink_subsystem *ss;
369 const struct nfnl_callback *nc;
370 struct netlink_ext_ack extack;
375 if (subsys_id >= NFNL_SUBSYS_COUNT)
376 return netlink_ack(skb, nlh, -EINVAL, NULL);
380 skb = netlink_skb_clone(oskb, GFP_KERNEL);
382 return netlink_ack(oskb, nlh, -ENOMEM, NULL);
384 nfnl_lock(subsys_id);
385 ss = nfnl_dereference_protected(subsys_id);
387 #ifdef CONFIG_MODULES
388 nfnl_unlock(subsys_id);
389 request_module("nfnetlink-subsys-%d", subsys_id);
390 nfnl_lock(subsys_id);
391 ss = nfnl_dereference_protected(subsys_id);
395 nfnl_unlock(subsys_id);
396 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
397 return kfree_skb(skb);
401 if (!ss->valid_genid || !ss->commit || !ss->abort) {
402 nfnl_unlock(subsys_id);
403 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
404 return kfree_skb(skb);
407 if (!try_module_get(ss->owner)) {
408 nfnl_unlock(subsys_id);
409 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
410 return kfree_skb(skb);
413 if (!ss->valid_genid(net, genid)) {
414 module_put(ss->owner);
415 nfnl_unlock(subsys_id);
416 netlink_ack(oskb, nlh, -ERESTART, NULL);
417 return kfree_skb(skb);
420 nfnl_unlock(subsys_id);
422 while (skb->len >= nlmsg_total_size(0)) {
425 if (fatal_signal_pending(current)) {
426 nfnl_err_reset(&err_list);
428 status = NFNL_BATCH_FAILURE;
432 memset(&extack, 0, sizeof(extack));
433 nlh = nlmsg_hdr(skb);
436 if (nlh->nlmsg_len < NLMSG_HDRLEN ||
437 skb->len < nlh->nlmsg_len ||
438 nlmsg_len(nlh) < sizeof(struct nfgenmsg)) {
439 nfnl_err_reset(&err_list);
440 status |= NFNL_BATCH_FAILURE;
444 /* Only requests are handled by the kernel */
445 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
450 type = nlh->nlmsg_type;
451 if (type == NFNL_MSG_BATCH_BEGIN) {
452 /* Malformed: Batch begin twice */
453 nfnl_err_reset(&err_list);
454 status |= NFNL_BATCH_FAILURE;
456 } else if (type == NFNL_MSG_BATCH_END) {
457 status |= NFNL_BATCH_DONE;
459 } else if (type < NLMSG_MIN_TYPE) {
464 /* We only accept a batch with messages for the same
467 if (NFNL_SUBSYS_ID(type) != subsys_id) {
472 nc = nfnetlink_find_client(type, ss);
478 if (nc->type != NFNL_CB_BATCH) {
484 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
485 struct nfnl_net *nfnlnet = nfnl_pernet(net);
486 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
487 struct nlattr *attr = (void *)nlh + min_len;
488 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
489 int attrlen = nlh->nlmsg_len - min_len;
490 struct nfnl_info info = {
497 /* Sanity-check NFTA_MAX_ATTR */
498 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
503 err = nla_parse_deprecated(cda,
504 ss->cb[cb_id].attr_count,
506 ss->cb[cb_id].policy, NULL);
510 err = nc->call(skb, &info, (const struct nlattr **)cda);
512 /* The lock was released to autoload some module, we
513 * have to abort and start from scratch using the
516 if (err == -EAGAIN) {
517 status |= NFNL_BATCH_REPLAY;
522 if (nlh->nlmsg_flags & NLM_F_ACK || err) {
523 /* Errors are delivered once the full batch has been
524 * processed, this avoids that the same error is
525 * reported several times when replaying the batch.
527 if (nfnl_err_add(&err_list, nlh, err, &extack) < 0) {
528 /* We failed to enqueue an error, reset the
529 * list of errors and send OOM to userspace
530 * pointing to the batch header.
532 nfnl_err_reset(&err_list);
533 netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM,
535 status |= NFNL_BATCH_FAILURE;
538 /* We don't stop processing the batch on errors, thus,
539 * userspace gets all the errors that the batch
543 status |= NFNL_BATCH_FAILURE;
546 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
547 if (msglen > skb->len)
549 skb_pull(skb, msglen);
552 if (status & NFNL_BATCH_REPLAY) {
553 ss->abort(net, oskb, NFNL_ABORT_AUTOLOAD);
554 nfnl_err_reset(&err_list);
556 module_put(ss->owner);
558 } else if (status == NFNL_BATCH_DONE) {
559 err = ss->commit(net, oskb);
560 if (err == -EAGAIN) {
561 status |= NFNL_BATCH_REPLAY;
564 ss->abort(net, oskb, NFNL_ABORT_NONE);
565 netlink_ack(oskb, nlmsg_hdr(oskb), err, NULL);
568 enum nfnl_abort_action abort_action;
570 if (status & NFNL_BATCH_FAILURE)
571 abort_action = NFNL_ABORT_NONE;
573 abort_action = NFNL_ABORT_VALIDATE;
575 err = ss->abort(net, oskb, abort_action);
576 if (err == -EAGAIN) {
577 nfnl_err_reset(&err_list);
579 module_put(ss->owner);
580 status |= NFNL_BATCH_FAILURE;
587 nfnl_err_deliver(&err_list, oskb);
589 module_put(ss->owner);
592 static const struct nla_policy nfnl_batch_policy[NFNL_BATCH_MAX + 1] = {
593 [NFNL_BATCH_GENID] = { .type = NLA_U32 },
596 static void nfnetlink_rcv_skb_batch(struct sk_buff *skb, struct nlmsghdr *nlh)
598 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
599 struct nlattr *attr = (void *)nlh + min_len;
600 struct nlattr *cda[NFNL_BATCH_MAX + 1];
601 int attrlen = nlh->nlmsg_len - min_len;
602 struct nfgenmsg *nfgenmsg;
607 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
608 if (msglen > skb->len)
611 if (skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
614 err = nla_parse_deprecated(cda, NFNL_BATCH_MAX, attr, attrlen,
615 nfnl_batch_policy, NULL);
617 netlink_ack(skb, nlh, err, NULL);
620 if (cda[NFNL_BATCH_GENID])
621 gen_id = ntohl(nla_get_be32(cda[NFNL_BATCH_GENID]));
623 nfgenmsg = nlmsg_data(nlh);
624 skb_pull(skb, msglen);
625 /* Work around old nft using host byte order */
626 if (nfgenmsg->res_id == NFNL_SUBSYS_NFTABLES)
627 res_id = NFNL_SUBSYS_NFTABLES;
629 res_id = ntohs(nfgenmsg->res_id);
631 nfnetlink_rcv_batch(skb, nlh, res_id, gen_id);
634 static void nfnetlink_rcv(struct sk_buff *skb)
636 struct nlmsghdr *nlh = nlmsg_hdr(skb);
638 if (skb->len < NLMSG_HDRLEN ||
639 nlh->nlmsg_len < NLMSG_HDRLEN ||
640 skb->len < nlh->nlmsg_len)
643 if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
644 netlink_ack(skb, nlh, -EPERM, NULL);
648 if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN)
649 nfnetlink_rcv_skb_batch(skb, nlh);
651 netlink_rcv_skb(skb, nfnetlink_rcv_msg);
654 #ifdef CONFIG_MODULES
655 static int nfnetlink_bind(struct net *net, int group)
657 const struct nfnetlink_subsystem *ss;
660 if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
663 type = nfnl_group2type[group];
666 ss = nfnetlink_get_subsys(type << 8);
669 request_module_nowait("nfnetlink-subsys-%d", type);
674 static int __net_init nfnetlink_net_init(struct net *net)
676 struct nfnl_net *nfnlnet = nfnl_pernet(net);
677 struct netlink_kernel_cfg cfg = {
678 .groups = NFNLGRP_MAX,
679 .input = nfnetlink_rcv,
680 #ifdef CONFIG_MODULES
681 .bind = nfnetlink_bind,
685 nfnlnet->nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
691 static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
693 struct nfnl_net *nfnlnet;
696 list_for_each_entry(net, net_exit_list, exit_list) {
697 nfnlnet = nfnl_pernet(net);
699 netlink_kernel_release(nfnlnet->nfnl);
703 static struct pernet_operations nfnetlink_net_ops = {
704 .init = nfnetlink_net_init,
705 .exit_batch = nfnetlink_net_exit_batch,
706 .id = &nfnetlink_pernet_id,
707 .size = sizeof(struct nfnl_net),
710 static int __init nfnetlink_init(void)
714 for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
715 BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
717 for (i=0; i<NFNL_SUBSYS_COUNT; i++)
718 __mutex_init(&table[i].mutex, nfnl_lockdep_names[i], &nfnl_lockdep_keys[i]);
720 return register_pernet_subsys(&nfnetlink_net_ops);
723 static void __exit nfnetlink_exit(void)
725 unregister_pernet_subsys(&nfnetlink_net_ops);
727 module_init(nfnetlink_init);
728 module_exit(nfnetlink_exit);