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",
71 [NFNL_SUBSYS_HOOK] = "nfnl_subsys_hook",
74 static const int nfnl_group2type[NFNLGRP_MAX+1] = {
75 [NFNLGRP_CONNTRACK_NEW] = NFNL_SUBSYS_CTNETLINK,
76 [NFNLGRP_CONNTRACK_UPDATE] = NFNL_SUBSYS_CTNETLINK,
77 [NFNLGRP_CONNTRACK_DESTROY] = NFNL_SUBSYS_CTNETLINK,
78 [NFNLGRP_CONNTRACK_EXP_NEW] = NFNL_SUBSYS_CTNETLINK_EXP,
79 [NFNLGRP_CONNTRACK_EXP_UPDATE] = NFNL_SUBSYS_CTNETLINK_EXP,
80 [NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP,
81 [NFNLGRP_NFTABLES] = NFNL_SUBSYS_NFTABLES,
82 [NFNLGRP_ACCT_QUOTA] = NFNL_SUBSYS_ACCT,
83 [NFNLGRP_NFTRACE] = NFNL_SUBSYS_NFTABLES,
86 static struct nfnl_net *nfnl_pernet(struct net *net)
88 return net_generic(net, nfnetlink_pernet_id);
91 void nfnl_lock(__u8 subsys_id)
93 mutex_lock(&table[subsys_id].mutex);
95 EXPORT_SYMBOL_GPL(nfnl_lock);
97 void nfnl_unlock(__u8 subsys_id)
99 mutex_unlock(&table[subsys_id].mutex);
101 EXPORT_SYMBOL_GPL(nfnl_unlock);
103 #ifdef CONFIG_PROVE_LOCKING
104 bool lockdep_nfnl_is_held(u8 subsys_id)
106 return lockdep_is_held(&table[subsys_id].mutex);
108 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
111 int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
115 /* Sanity-check attr_count size to avoid stack buffer overflow. */
116 for (cb_id = 0; cb_id < n->cb_count; cb_id++)
117 if (WARN_ON(n->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT))
120 nfnl_lock(n->subsys_id);
121 if (table[n->subsys_id].subsys) {
122 nfnl_unlock(n->subsys_id);
125 rcu_assign_pointer(table[n->subsys_id].subsys, n);
126 nfnl_unlock(n->subsys_id);
130 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
132 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
134 nfnl_lock(n->subsys_id);
135 table[n->subsys_id].subsys = NULL;
136 nfnl_unlock(n->subsys_id);
140 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
142 static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u16 type)
144 u8 subsys_id = NFNL_SUBSYS_ID(type);
146 if (subsys_id >= NFNL_SUBSYS_COUNT)
149 return rcu_dereference(table[subsys_id].subsys);
152 static inline const struct nfnl_callback *
153 nfnetlink_find_client(u16 type, const struct nfnetlink_subsystem *ss)
155 u8 cb_id = NFNL_MSG_TYPE(type);
157 if (cb_id >= ss->cb_count)
160 return &ss->cb[cb_id];
163 int nfnetlink_has_listeners(struct net *net, unsigned int group)
165 struct nfnl_net *nfnlnet = nfnl_pernet(net);
167 return netlink_has_listeners(nfnlnet->nfnl, group);
169 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
171 int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
172 unsigned int group, int echo, gfp_t flags)
174 struct nfnl_net *nfnlnet = nfnl_pernet(net);
176 return nlmsg_notify(nfnlnet->nfnl, skb, portid, group, echo, flags);
178 EXPORT_SYMBOL_GPL(nfnetlink_send);
180 int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
182 struct nfnl_net *nfnlnet = nfnl_pernet(net);
184 return netlink_set_err(nfnlnet->nfnl, portid, group, error);
186 EXPORT_SYMBOL_GPL(nfnetlink_set_err);
188 int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid)
190 struct nfnl_net *nfnlnet = nfnl_pernet(net);
193 err = nlmsg_unicast(nfnlnet->nfnl, skb, portid);
199 EXPORT_SYMBOL_GPL(nfnetlink_unicast);
201 void nfnetlink_broadcast(struct net *net, struct sk_buff *skb, __u32 portid,
202 __u32 group, gfp_t allocation)
204 struct nfnl_net *nfnlnet = nfnl_pernet(net);
206 netlink_broadcast(nfnlnet->nfnl, skb, portid, group, allocation);
208 EXPORT_SYMBOL_GPL(nfnetlink_broadcast);
210 /* Process one complete nfnetlink message. */
211 static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
212 struct netlink_ext_ack *extack)
214 struct net *net = sock_net(skb->sk);
215 const struct nfnl_callback *nc;
216 const struct nfnetlink_subsystem *ss;
219 /* All the messages must at least contain nfgenmsg */
220 if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
223 type = nlh->nlmsg_type;
227 ss = nfnetlink_get_subsys(type);
229 #ifdef CONFIG_MODULES
231 request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
233 ss = nfnetlink_get_subsys(type);
242 nc = nfnetlink_find_client(type, ss);
249 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
250 struct nfnl_net *nfnlnet = nfnl_pernet(net);
251 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
252 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
253 struct nlattr *attr = (void *)nlh + min_len;
254 int attrlen = nlh->nlmsg_len - min_len;
255 __u8 subsys_id = NFNL_SUBSYS_ID(type);
256 struct nfnl_info info = {
260 .nfmsg = nlmsg_data(nlh),
264 /* Sanity-check NFNL_MAX_ATTR_COUNT */
265 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
270 err = nla_parse_deprecated(cda, ss->cb[cb_id].attr_count,
272 ss->cb[cb_id].policy, extack);
285 err = nc->call(skb, &info, (const struct nlattr **)cda);
290 nfnl_lock(subsys_id);
291 if (nfnl_dereference_protected(subsys_id) != ss ||
292 nfnetlink_find_client(type, ss) != nc) {
293 nfnl_unlock(subsys_id);
297 err = nc->call(skb, &info, (const struct nlattr **)cda);
298 nfnl_unlock(subsys_id);
312 struct list_head head;
313 struct nlmsghdr *nlh;
315 struct netlink_ext_ack extack;
318 static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err,
319 const struct netlink_ext_ack *extack)
321 struct nfnl_err *nfnl_err;
323 nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL);
324 if (nfnl_err == NULL)
329 nfnl_err->extack = *extack;
330 list_add_tail(&nfnl_err->head, list);
335 static void nfnl_err_del(struct nfnl_err *nfnl_err)
337 list_del(&nfnl_err->head);
341 static void nfnl_err_reset(struct list_head *err_list)
343 struct nfnl_err *nfnl_err, *next;
345 list_for_each_entry_safe(nfnl_err, next, err_list, head)
346 nfnl_err_del(nfnl_err);
349 static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
351 struct nfnl_err *nfnl_err, *next;
353 list_for_each_entry_safe(nfnl_err, next, err_list, head) {
354 netlink_ack(skb, nfnl_err->nlh, nfnl_err->err,
356 nfnl_err_del(nfnl_err);
361 NFNL_BATCH_FAILURE = (1 << 0),
362 NFNL_BATCH_DONE = (1 << 1),
363 NFNL_BATCH_REPLAY = (1 << 2),
366 static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
367 u16 subsys_id, u32 genid)
369 struct sk_buff *oskb = skb;
370 struct net *net = sock_net(skb->sk);
371 const struct nfnetlink_subsystem *ss;
372 const struct nfnl_callback *nc;
373 struct netlink_ext_ack extack;
378 if (subsys_id >= NFNL_SUBSYS_COUNT)
379 return netlink_ack(skb, nlh, -EINVAL, NULL);
383 skb = netlink_skb_clone(oskb, GFP_KERNEL);
385 return netlink_ack(oskb, nlh, -ENOMEM, NULL);
387 nfnl_lock(subsys_id);
388 ss = nfnl_dereference_protected(subsys_id);
390 #ifdef CONFIG_MODULES
391 nfnl_unlock(subsys_id);
392 request_module("nfnetlink-subsys-%d", subsys_id);
393 nfnl_lock(subsys_id);
394 ss = nfnl_dereference_protected(subsys_id);
398 nfnl_unlock(subsys_id);
399 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
400 return kfree_skb(skb);
404 if (!ss->valid_genid || !ss->commit || !ss->abort) {
405 nfnl_unlock(subsys_id);
406 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
407 return kfree_skb(skb);
410 if (!try_module_get(ss->owner)) {
411 nfnl_unlock(subsys_id);
412 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
413 return kfree_skb(skb);
416 if (!ss->valid_genid(net, genid)) {
417 module_put(ss->owner);
418 nfnl_unlock(subsys_id);
419 netlink_ack(oskb, nlh, -ERESTART, NULL);
420 return kfree_skb(skb);
423 nfnl_unlock(subsys_id);
425 while (skb->len >= nlmsg_total_size(0)) {
428 if (fatal_signal_pending(current)) {
429 nfnl_err_reset(&err_list);
431 status = NFNL_BATCH_FAILURE;
435 memset(&extack, 0, sizeof(extack));
436 nlh = nlmsg_hdr(skb);
439 if (nlh->nlmsg_len < NLMSG_HDRLEN ||
440 skb->len < nlh->nlmsg_len ||
441 nlmsg_len(nlh) < sizeof(struct nfgenmsg)) {
442 nfnl_err_reset(&err_list);
443 status |= NFNL_BATCH_FAILURE;
447 /* Only requests are handled by the kernel */
448 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
453 type = nlh->nlmsg_type;
454 if (type == NFNL_MSG_BATCH_BEGIN) {
455 /* Malformed: Batch begin twice */
456 nfnl_err_reset(&err_list);
457 status |= NFNL_BATCH_FAILURE;
459 } else if (type == NFNL_MSG_BATCH_END) {
460 status |= NFNL_BATCH_DONE;
462 } else if (type < NLMSG_MIN_TYPE) {
467 /* We only accept a batch with messages for the same
470 if (NFNL_SUBSYS_ID(type) != subsys_id) {
475 nc = nfnetlink_find_client(type, ss);
481 if (nc->type != NFNL_CB_BATCH) {
487 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
488 struct nfnl_net *nfnlnet = nfnl_pernet(net);
489 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
490 struct nlattr *attr = (void *)nlh + min_len;
491 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
492 int attrlen = nlh->nlmsg_len - min_len;
493 struct nfnl_info info = {
497 .nfmsg = nlmsg_data(nlh),
501 /* Sanity-check NFTA_MAX_ATTR */
502 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
507 err = nla_parse_deprecated(cda,
508 ss->cb[cb_id].attr_count,
510 ss->cb[cb_id].policy, NULL);
514 err = nc->call(skb, &info, (const struct nlattr **)cda);
516 /* The lock was released to autoload some module, we
517 * have to abort and start from scratch using the
520 if (err == -EAGAIN) {
521 status |= NFNL_BATCH_REPLAY;
526 if (nlh->nlmsg_flags & NLM_F_ACK || err) {
527 /* Errors are delivered once the full batch has been
528 * processed, this avoids that the same error is
529 * reported several times when replaying the batch.
531 if (nfnl_err_add(&err_list, nlh, err, &extack) < 0) {
532 /* We failed to enqueue an error, reset the
533 * list of errors and send OOM to userspace
534 * pointing to the batch header.
536 nfnl_err_reset(&err_list);
537 netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM,
539 status |= NFNL_BATCH_FAILURE;
542 /* We don't stop processing the batch on errors, thus,
543 * userspace gets all the errors that the batch
547 status |= NFNL_BATCH_FAILURE;
550 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
551 if (msglen > skb->len)
553 skb_pull(skb, msglen);
556 if (status & NFNL_BATCH_REPLAY) {
557 ss->abort(net, oskb, NFNL_ABORT_AUTOLOAD);
558 nfnl_err_reset(&err_list);
560 module_put(ss->owner);
562 } else if (status == NFNL_BATCH_DONE) {
563 err = ss->commit(net, oskb);
564 if (err == -EAGAIN) {
565 status |= NFNL_BATCH_REPLAY;
568 ss->abort(net, oskb, NFNL_ABORT_NONE);
569 netlink_ack(oskb, nlmsg_hdr(oskb), err, NULL);
572 enum nfnl_abort_action abort_action;
574 if (status & NFNL_BATCH_FAILURE)
575 abort_action = NFNL_ABORT_NONE;
577 abort_action = NFNL_ABORT_VALIDATE;
579 err = ss->abort(net, oskb, abort_action);
580 if (err == -EAGAIN) {
581 nfnl_err_reset(&err_list);
583 module_put(ss->owner);
584 status |= NFNL_BATCH_FAILURE;
591 nfnl_err_deliver(&err_list, oskb);
593 module_put(ss->owner);
596 static const struct nla_policy nfnl_batch_policy[NFNL_BATCH_MAX + 1] = {
597 [NFNL_BATCH_GENID] = { .type = NLA_U32 },
600 static void nfnetlink_rcv_skb_batch(struct sk_buff *skb, struct nlmsghdr *nlh)
602 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
603 struct nlattr *attr = (void *)nlh + min_len;
604 struct nlattr *cda[NFNL_BATCH_MAX + 1];
605 int attrlen = nlh->nlmsg_len - min_len;
606 struct nfgenmsg *nfgenmsg;
611 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
612 if (msglen > skb->len)
615 if (skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
618 err = nla_parse_deprecated(cda, NFNL_BATCH_MAX, attr, attrlen,
619 nfnl_batch_policy, NULL);
621 netlink_ack(skb, nlh, err, NULL);
624 if (cda[NFNL_BATCH_GENID])
625 gen_id = ntohl(nla_get_be32(cda[NFNL_BATCH_GENID]));
627 nfgenmsg = nlmsg_data(nlh);
628 skb_pull(skb, msglen);
629 /* Work around old nft using host byte order */
630 if (nfgenmsg->res_id == NFNL_SUBSYS_NFTABLES)
631 res_id = NFNL_SUBSYS_NFTABLES;
633 res_id = ntohs(nfgenmsg->res_id);
635 nfnetlink_rcv_batch(skb, nlh, res_id, gen_id);
638 static void nfnetlink_rcv(struct sk_buff *skb)
640 struct nlmsghdr *nlh = nlmsg_hdr(skb);
642 if (skb->len < NLMSG_HDRLEN ||
643 nlh->nlmsg_len < NLMSG_HDRLEN ||
644 skb->len < nlh->nlmsg_len)
647 if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
648 netlink_ack(skb, nlh, -EPERM, NULL);
652 if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN)
653 nfnetlink_rcv_skb_batch(skb, nlh);
655 netlink_rcv_skb(skb, nfnetlink_rcv_msg);
658 #ifdef CONFIG_MODULES
659 static int nfnetlink_bind(struct net *net, int group)
661 const struct nfnetlink_subsystem *ss;
664 if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
667 type = nfnl_group2type[group];
670 ss = nfnetlink_get_subsys(type << 8);
673 request_module_nowait("nfnetlink-subsys-%d", type);
678 static int __net_init nfnetlink_net_init(struct net *net)
680 struct nfnl_net *nfnlnet = nfnl_pernet(net);
681 struct netlink_kernel_cfg cfg = {
682 .groups = NFNLGRP_MAX,
683 .input = nfnetlink_rcv,
684 #ifdef CONFIG_MODULES
685 .bind = nfnetlink_bind,
689 nfnlnet->nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
695 static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
697 struct nfnl_net *nfnlnet;
700 list_for_each_entry(net, net_exit_list, exit_list) {
701 nfnlnet = nfnl_pernet(net);
703 netlink_kernel_release(nfnlnet->nfnl);
707 static struct pernet_operations nfnetlink_net_ops = {
708 .init = nfnetlink_net_init,
709 .exit_batch = nfnetlink_net_exit_batch,
710 .id = &nfnetlink_pernet_id,
711 .size = sizeof(struct nfnl_net),
714 static int __init nfnetlink_init(void)
718 for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
719 BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
721 for (i=0; i<NFNL_SUBSYS_COUNT; i++)
722 __mutex_init(&table[i].mutex, nfnl_lockdep_names[i], &nfnl_lockdep_keys[i]);
724 return register_pernet_subsys(&nfnetlink_net_ops);
727 static void __exit nfnetlink_exit(void)
729 unregister_pernet_subsys(&nfnetlink_net_ops);
731 module_init(nfnetlink_init);
732 module_exit(nfnetlink_exit);