2 * NETLINK Kernel-user communication protocol.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6 * Patrick McHardy <kaber@trash.net>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
13 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
14 * added netlink_proto_exit
15 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
16 * use nlk_sk, as sk->protinfo is on a diet 8)
17 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
18 * - inc module use count of module that owns
19 * the kernel socket in case userspace opens
20 * socket of same protocol
21 * - remove all module support, since netlink is
22 * mandatory if CONFIG_NET=y these days
25 #include <linux/module.h>
27 #include <linux/capability.h>
28 #include <linux/kernel.h>
29 #include <linux/init.h>
30 #include <linux/signal.h>
31 #include <linux/sched.h>
32 #include <linux/errno.h>
33 #include <linux/string.h>
34 #include <linux/stat.h>
35 #include <linux/socket.h>
37 #include <linux/fcntl.h>
38 #include <linux/termios.h>
39 #include <linux/sockios.h>
40 #include <linux/net.h>
42 #include <linux/slab.h>
43 #include <linux/uaccess.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/rtnetlink.h>
47 #include <linux/proc_fs.h>
48 #include <linux/seq_file.h>
49 #include <linux/notifier.h>
50 #include <linux/security.h>
51 #include <linux/jhash.h>
52 #include <linux/jiffies.h>
53 #include <linux/random.h>
54 #include <linux/bitops.h>
56 #include <linux/types.h>
57 #include <linux/audit.h>
58 #include <linux/mutex.h>
59 #include <linux/vmalloc.h>
60 #include <linux/if_arp.h>
61 #include <linux/rhashtable.h>
62 #include <asm/cacheflush.h>
63 #include <linux/hash.h>
64 #include <linux/genetlink.h>
65 #include <linux/net_namespace.h>
67 #include <net/net_namespace.h>
68 #include <net/netns/generic.h>
71 #include <net/netlink.h>
73 #include "af_netlink.h"
77 unsigned long masks[0];
81 #define NETLINK_S_CONGESTED 0x0
83 static inline int netlink_is_kernel(struct sock *sk)
85 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
88 struct netlink_table *nl_table __read_mostly;
89 EXPORT_SYMBOL_GPL(nl_table);
91 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
93 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
95 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
98 "nlk_cb_mutex-USERSOCK",
99 "nlk_cb_mutex-FIREWALL",
100 "nlk_cb_mutex-SOCK_DIAG",
101 "nlk_cb_mutex-NFLOG",
103 "nlk_cb_mutex-SELINUX",
104 "nlk_cb_mutex-ISCSI",
105 "nlk_cb_mutex-AUDIT",
106 "nlk_cb_mutex-FIB_LOOKUP",
107 "nlk_cb_mutex-CONNECTOR",
108 "nlk_cb_mutex-NETFILTER",
109 "nlk_cb_mutex-IP6_FW",
110 "nlk_cb_mutex-DNRTMSG",
111 "nlk_cb_mutex-KOBJECT_UEVENT",
112 "nlk_cb_mutex-GENERIC",
114 "nlk_cb_mutex-SCSITRANSPORT",
115 "nlk_cb_mutex-ECRYPTFS",
117 "nlk_cb_mutex-CRYPTO",
128 "nlk_cb_mutex-MAX_LINKS"
131 static int netlink_dump(struct sock *sk);
133 /* nl_table locking explained:
134 * Lookup and traversal are protected with an RCU read-side lock. Insertion
135 * and removal are protected with per bucket lock while using RCU list
136 * modification primitives and may run in parallel to RCU protected lookups.
137 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
138 * been acquired * either during or after the socket has been removed from
139 * the list and after an RCU grace period.
141 DEFINE_RWLOCK(nl_table_lock);
142 EXPORT_SYMBOL_GPL(nl_table_lock);
143 static atomic_t nl_table_users = ATOMIC_INIT(0);
145 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
147 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
150 static const struct rhashtable_params netlink_rhashtable_params;
152 static inline u32 netlink_group_mask(u32 group)
154 return group ? 1 << (group - 1) : 0;
157 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
160 unsigned int len = skb_end_offset(skb);
163 new = alloc_skb(len, gfp_mask);
167 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
168 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
169 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
171 skb_put_data(new, skb->data, len);
175 static unsigned int netlink_tap_net_id;
177 struct netlink_tap_net {
178 struct list_head netlink_tap_all;
179 struct mutex netlink_tap_lock;
182 int netlink_add_tap(struct netlink_tap *nt)
184 struct net *net = dev_net(nt->dev);
185 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
187 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
190 mutex_lock(&nn->netlink_tap_lock);
191 list_add_rcu(&nt->list, &nn->netlink_tap_all);
192 mutex_unlock(&nn->netlink_tap_lock);
194 __module_get(nt->module);
198 EXPORT_SYMBOL_GPL(netlink_add_tap);
200 static int __netlink_remove_tap(struct netlink_tap *nt)
202 struct net *net = dev_net(nt->dev);
203 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
205 struct netlink_tap *tmp;
207 mutex_lock(&nn->netlink_tap_lock);
209 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
211 list_del_rcu(&nt->list);
217 pr_warn("__netlink_remove_tap: %p not found\n", nt);
219 mutex_unlock(&nn->netlink_tap_lock);
222 module_put(nt->module);
224 return found ? 0 : -ENODEV;
227 int netlink_remove_tap(struct netlink_tap *nt)
231 ret = __netlink_remove_tap(nt);
236 EXPORT_SYMBOL_GPL(netlink_remove_tap);
238 static __net_init int netlink_tap_init_net(struct net *net)
240 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
242 INIT_LIST_HEAD(&nn->netlink_tap_all);
243 mutex_init(&nn->netlink_tap_lock);
247 static void __net_exit netlink_tap_exit_net(struct net *net)
251 static struct pernet_operations netlink_tap_net_ops = {
252 .init = netlink_tap_init_net,
253 .exit = netlink_tap_exit_net,
254 .id = &netlink_tap_net_id,
255 .size = sizeof(struct netlink_tap_net),
258 static bool netlink_filter_tap(const struct sk_buff *skb)
260 struct sock *sk = skb->sk;
262 /* We take the more conservative approach and
263 * whitelist socket protocols that may pass.
265 switch (sk->sk_protocol) {
267 case NETLINK_USERSOCK:
268 case NETLINK_SOCK_DIAG:
271 case NETLINK_FIB_LOOKUP:
272 case NETLINK_NETFILTER:
273 case NETLINK_GENERIC:
280 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
281 struct net_device *dev)
283 struct sk_buff *nskb;
284 struct sock *sk = skb->sk;
287 if (!net_eq(dev_net(dev), sock_net(sk)))
292 if (is_vmalloc_addr(skb->head))
293 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
295 nskb = skb_clone(skb, GFP_ATOMIC);
298 nskb->protocol = htons((u16) sk->sk_protocol);
299 nskb->pkt_type = netlink_is_kernel(sk) ?
300 PACKET_KERNEL : PACKET_USER;
301 skb_reset_network_header(nskb);
302 ret = dev_queue_xmit(nskb);
303 if (unlikely(ret > 0))
304 ret = net_xmit_errno(ret);
311 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
314 struct netlink_tap *tmp;
316 if (!netlink_filter_tap(skb))
319 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
320 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
326 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
328 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
332 if (unlikely(!list_empty(&nn->netlink_tap_all)))
333 __netlink_deliver_tap(skb, nn);
338 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
341 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
342 netlink_deliver_tap(sock_net(dst), skb);
345 static void netlink_overrun(struct sock *sk)
347 struct netlink_sock *nlk = nlk_sk(sk);
349 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
350 if (!test_and_set_bit(NETLINK_S_CONGESTED,
351 &nlk_sk(sk)->state)) {
352 sk->sk_err = ENOBUFS;
353 sk->sk_error_report(sk);
356 atomic_inc(&sk->sk_drops);
359 static void netlink_rcv_wake(struct sock *sk)
361 struct netlink_sock *nlk = nlk_sk(sk);
363 if (skb_queue_empty(&sk->sk_receive_queue))
364 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
365 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
366 wake_up_interruptible(&nlk->wait);
369 static void netlink_skb_destructor(struct sk_buff *skb)
371 if (is_vmalloc_addr(skb->head)) {
373 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
382 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
384 WARN_ON(skb->sk != NULL);
386 skb->destructor = netlink_skb_destructor;
387 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
388 sk_mem_charge(sk, skb->truesize);
391 static void netlink_sock_destruct(struct sock *sk)
393 struct netlink_sock *nlk = nlk_sk(sk);
395 if (nlk->cb_running) {
397 nlk->cb.done(&nlk->cb);
398 module_put(nlk->cb.module);
399 kfree_skb(nlk->cb.skb);
402 skb_queue_purge(&sk->sk_receive_queue);
404 if (!sock_flag(sk, SOCK_DEAD)) {
405 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
409 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
410 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
411 WARN_ON(nlk_sk(sk)->groups);
414 static void netlink_sock_destruct_work(struct work_struct *work)
416 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
422 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
423 * SMP. Look, when several writers sleep and reader wakes them up, all but one
424 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
425 * this, _but_ remember, it adds useless work on UP machines.
428 void netlink_table_grab(void)
429 __acquires(nl_table_lock)
433 write_lock_irq(&nl_table_lock);
435 if (atomic_read(&nl_table_users)) {
436 DECLARE_WAITQUEUE(wait, current);
438 add_wait_queue_exclusive(&nl_table_wait, &wait);
440 set_current_state(TASK_UNINTERRUPTIBLE);
441 if (atomic_read(&nl_table_users) == 0)
443 write_unlock_irq(&nl_table_lock);
445 write_lock_irq(&nl_table_lock);
448 __set_current_state(TASK_RUNNING);
449 remove_wait_queue(&nl_table_wait, &wait);
453 void netlink_table_ungrab(void)
454 __releases(nl_table_lock)
456 write_unlock_irq(&nl_table_lock);
457 wake_up(&nl_table_wait);
461 netlink_lock_table(void)
463 /* read_lock() synchronizes us to netlink_table_grab */
465 read_lock(&nl_table_lock);
466 atomic_inc(&nl_table_users);
467 read_unlock(&nl_table_lock);
471 netlink_unlock_table(void)
473 if (atomic_dec_and_test(&nl_table_users))
474 wake_up(&nl_table_wait);
477 struct netlink_compare_arg
483 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
484 #define netlink_compare_arg_len \
485 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
487 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
490 const struct netlink_compare_arg *x = arg->key;
491 const struct netlink_sock *nlk = ptr;
493 return nlk->portid != x->portid ||
494 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
497 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
498 struct net *net, u32 portid)
500 memset(arg, 0, sizeof(*arg));
501 write_pnet(&arg->pnet, net);
502 arg->portid = portid;
505 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
508 struct netlink_compare_arg arg;
510 netlink_compare_arg_init(&arg, net, portid);
511 return rhashtable_lookup_fast(&table->hash, &arg,
512 netlink_rhashtable_params);
515 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
517 struct netlink_compare_arg arg;
519 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
520 return rhashtable_lookup_insert_key(&table->hash, &arg,
522 netlink_rhashtable_params);
525 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
527 struct netlink_table *table = &nl_table[protocol];
531 sk = __netlink_lookup(table, portid, net);
539 static const struct proto_ops netlink_ops;
542 netlink_update_listeners(struct sock *sk)
544 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
547 struct listeners *listeners;
549 listeners = nl_deref_protected(tbl->listeners);
553 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
555 sk_for_each_bound(sk, &tbl->mc_list) {
556 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
557 mask |= nlk_sk(sk)->groups[i];
559 listeners->masks[i] = mask;
561 /* this function is only called with the netlink table "grabbed", which
562 * makes sure updates are visible before bind or setsockopt return. */
565 static int netlink_insert(struct sock *sk, u32 portid)
567 struct netlink_table *table = &nl_table[sk->sk_protocol];
572 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
573 if (nlk_sk(sk)->bound)
577 if (BITS_PER_LONG > 32 &&
578 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
581 nlk_sk(sk)->portid = portid;
584 err = __netlink_insert(table, sk);
586 /* In case the hashtable backend returns with -EBUSY
587 * from here, it must not escape to the caller.
589 if (unlikely(err == -EBUSY))
597 /* We need to ensure that the socket is hashed and visible. */
599 nlk_sk(sk)->bound = portid;
606 static void netlink_remove(struct sock *sk)
608 struct netlink_table *table;
610 table = &nl_table[sk->sk_protocol];
611 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
612 netlink_rhashtable_params)) {
613 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
617 netlink_table_grab();
618 if (nlk_sk(sk)->subscriptions) {
619 __sk_del_bind_node(sk);
620 netlink_update_listeners(sk);
622 if (sk->sk_protocol == NETLINK_GENERIC)
623 atomic_inc(&genl_sk_destructing_cnt);
624 netlink_table_ungrab();
627 static struct proto netlink_proto = {
629 .owner = THIS_MODULE,
630 .obj_size = sizeof(struct netlink_sock),
633 static int __netlink_create(struct net *net, struct socket *sock,
634 struct mutex *cb_mutex, int protocol,
638 struct netlink_sock *nlk;
640 sock->ops = &netlink_ops;
642 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
646 sock_init_data(sock, sk);
650 nlk->cb_mutex = cb_mutex;
652 nlk->cb_mutex = &nlk->cb_def_mutex;
653 mutex_init(nlk->cb_mutex);
654 lockdep_set_class_and_name(nlk->cb_mutex,
655 nlk_cb_mutex_keys + protocol,
656 nlk_cb_mutex_key_strings[protocol]);
658 init_waitqueue_head(&nlk->wait);
660 sk->sk_destruct = netlink_sock_destruct;
661 sk->sk_protocol = protocol;
665 static int netlink_create(struct net *net, struct socket *sock, int protocol,
668 struct module *module = NULL;
669 struct mutex *cb_mutex;
670 struct netlink_sock *nlk;
671 int (*bind)(struct net *net, int group);
672 void (*unbind)(struct net *net, int group);
675 sock->state = SS_UNCONNECTED;
677 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
678 return -ESOCKTNOSUPPORT;
680 if (protocol < 0 || protocol >= MAX_LINKS)
681 return -EPROTONOSUPPORT;
683 netlink_lock_table();
684 #ifdef CONFIG_MODULES
685 if (!nl_table[protocol].registered) {
686 netlink_unlock_table();
687 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
688 netlink_lock_table();
691 if (nl_table[protocol].registered &&
692 try_module_get(nl_table[protocol].module))
693 module = nl_table[protocol].module;
695 err = -EPROTONOSUPPORT;
696 cb_mutex = nl_table[protocol].cb_mutex;
697 bind = nl_table[protocol].bind;
698 unbind = nl_table[protocol].unbind;
699 netlink_unlock_table();
704 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
709 sock_prot_inuse_add(net, &netlink_proto, 1);
712 nlk = nlk_sk(sock->sk);
713 nlk->module = module;
714 nlk->netlink_bind = bind;
715 nlk->netlink_unbind = unbind;
724 static void deferred_put_nlk_sk(struct rcu_head *head)
726 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
727 struct sock *sk = &nlk->sk;
732 if (!refcount_dec_and_test(&sk->sk_refcnt))
735 if (nlk->cb_running && nlk->cb.done) {
736 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
737 schedule_work(&nlk->work);
744 static int netlink_release(struct socket *sock)
746 struct sock *sk = sock->sk;
747 struct netlink_sock *nlk;
757 * OK. Socket is unlinked, any packets that arrive now
761 /* must not acquire netlink_table_lock in any way again before unbind
762 * and notifying genetlink is done as otherwise it might deadlock
764 if (nlk->netlink_unbind) {
767 for (i = 0; i < nlk->ngroups; i++)
768 if (test_bit(i, nlk->groups))
769 nlk->netlink_unbind(sock_net(sk), i + 1);
771 if (sk->sk_protocol == NETLINK_GENERIC &&
772 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
773 wake_up(&genl_sk_destructing_waitq);
776 wake_up_interruptible_all(&nlk->wait);
778 skb_queue_purge(&sk->sk_write_queue);
780 if (nlk->portid && nlk->bound) {
781 struct netlink_notify n = {
783 .protocol = sk->sk_protocol,
784 .portid = nlk->portid,
786 blocking_notifier_call_chain(&netlink_chain,
787 NETLINK_URELEASE, &n);
790 module_put(nlk->module);
792 if (netlink_is_kernel(sk)) {
793 netlink_table_grab();
794 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
795 if (--nl_table[sk->sk_protocol].registered == 0) {
796 struct listeners *old;
798 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
799 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
801 nl_table[sk->sk_protocol].module = NULL;
802 nl_table[sk->sk_protocol].bind = NULL;
803 nl_table[sk->sk_protocol].unbind = NULL;
804 nl_table[sk->sk_protocol].flags = 0;
805 nl_table[sk->sk_protocol].registered = 0;
807 netlink_table_ungrab();
811 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
813 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
817 static int netlink_autobind(struct socket *sock)
819 struct sock *sk = sock->sk;
820 struct net *net = sock_net(sk);
821 struct netlink_table *table = &nl_table[sk->sk_protocol];
822 s32 portid = task_tgid_vnr(current);
830 ok = !__netlink_lookup(table, portid, net);
833 /* Bind collision, search negative portid values. */
835 /* rover will be in range [S32_MIN, -4097] */
836 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
837 else if (rover >= -4096)
843 err = netlink_insert(sk, portid);
844 if (err == -EADDRINUSE)
847 /* If 2 threads race to autobind, that is fine. */
855 * __netlink_ns_capable - General netlink message capability test
856 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
857 * @user_ns: The user namespace of the capability to use
858 * @cap: The capability to use
860 * Test to see if the opener of the socket we received the message
861 * from had when the netlink socket was created and the sender of the
862 * message has has the capability @cap in the user namespace @user_ns.
864 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
865 struct user_namespace *user_ns, int cap)
867 return ((nsp->flags & NETLINK_SKB_DST) ||
868 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
869 ns_capable(user_ns, cap);
871 EXPORT_SYMBOL(__netlink_ns_capable);
874 * netlink_ns_capable - General netlink message capability test
875 * @skb: socket buffer holding a netlink command from userspace
876 * @user_ns: The user namespace of the capability to use
877 * @cap: The capability to use
879 * Test to see if the opener of the socket we received the message
880 * from had when the netlink socket was created and the sender of the
881 * message has has the capability @cap in the user namespace @user_ns.
883 bool netlink_ns_capable(const struct sk_buff *skb,
884 struct user_namespace *user_ns, int cap)
886 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
888 EXPORT_SYMBOL(netlink_ns_capable);
891 * netlink_capable - Netlink global message capability test
892 * @skb: socket buffer holding a netlink command from userspace
893 * @cap: The capability to use
895 * Test to see if the opener of the socket we received the message
896 * from had when the netlink socket was created and the sender of the
897 * message has has the capability @cap in all user namespaces.
899 bool netlink_capable(const struct sk_buff *skb, int cap)
901 return netlink_ns_capable(skb, &init_user_ns, cap);
903 EXPORT_SYMBOL(netlink_capable);
906 * netlink_net_capable - Netlink network namespace message capability test
907 * @skb: socket buffer holding a netlink command from userspace
908 * @cap: The capability to use
910 * Test to see if the opener of the socket we received the message
911 * from had when the netlink socket was created and the sender of the
912 * message has has the capability @cap over the network namespace of
913 * the socket we received the message from.
915 bool netlink_net_capable(const struct sk_buff *skb, int cap)
917 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
919 EXPORT_SYMBOL(netlink_net_capable);
921 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
923 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
924 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
928 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
930 struct netlink_sock *nlk = nlk_sk(sk);
932 if (nlk->subscriptions && !subscriptions)
933 __sk_del_bind_node(sk);
934 else if (!nlk->subscriptions && subscriptions)
935 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
936 nlk->subscriptions = subscriptions;
939 static int netlink_realloc_groups(struct sock *sk)
941 struct netlink_sock *nlk = nlk_sk(sk);
943 unsigned long *new_groups;
946 netlink_table_grab();
948 groups = nl_table[sk->sk_protocol].groups;
949 if (!nl_table[sk->sk_protocol].registered) {
954 if (nlk->ngroups >= groups)
957 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
958 if (new_groups == NULL) {
962 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
963 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
965 nlk->groups = new_groups;
966 nlk->ngroups = groups;
968 netlink_table_ungrab();
972 static void netlink_undo_bind(int group, long unsigned int groups,
975 struct netlink_sock *nlk = nlk_sk(sk);
978 if (!nlk->netlink_unbind)
981 for (undo = 0; undo < group; undo++)
982 if (test_bit(undo, &groups))
983 nlk->netlink_unbind(sock_net(sk), undo + 1);
986 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
989 struct sock *sk = sock->sk;
990 struct net *net = sock_net(sk);
991 struct netlink_sock *nlk = nlk_sk(sk);
992 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
994 long unsigned int groups = nladdr->nl_groups;
997 if (addr_len < sizeof(struct sockaddr_nl))
1000 if (nladdr->nl_family != AF_NETLINK)
1003 /* Only superuser is allowed to listen multicasts */
1005 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1007 err = netlink_realloc_groups(sk);
1014 /* Ensure nlk->portid is up-to-date. */
1017 if (nladdr->nl_pid != nlk->portid)
1021 netlink_lock_table();
1022 if (nlk->netlink_bind && groups) {
1025 for (group = 0; group < nlk->ngroups; group++) {
1026 if (!test_bit(group, &groups))
1028 err = nlk->netlink_bind(net, group + 1);
1031 netlink_undo_bind(group, groups, sk);
1036 /* No need for barriers here as we return to user-space without
1037 * using any of the bound attributes.
1040 err = nladdr->nl_pid ?
1041 netlink_insert(sk, nladdr->nl_pid) :
1042 netlink_autobind(sock);
1044 netlink_undo_bind(nlk->ngroups, groups, sk);
1049 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1051 netlink_unlock_table();
1053 netlink_table_grab();
1054 netlink_update_subscriptions(sk, nlk->subscriptions +
1056 hweight32(nlk->groups[0]));
1057 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1058 netlink_update_listeners(sk);
1059 netlink_table_ungrab();
1064 netlink_unlock_table();
1068 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1069 int alen, int flags)
1072 struct sock *sk = sock->sk;
1073 struct netlink_sock *nlk = nlk_sk(sk);
1074 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1076 if (alen < sizeof(addr->sa_family))
1079 if (addr->sa_family == AF_UNSPEC) {
1080 sk->sk_state = NETLINK_UNCONNECTED;
1081 nlk->dst_portid = 0;
1085 if (addr->sa_family != AF_NETLINK)
1088 if (alen < sizeof(struct sockaddr_nl))
1091 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1092 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1095 /* No need for barriers here as we return to user-space without
1096 * using any of the bound attributes.
1099 err = netlink_autobind(sock);
1102 sk->sk_state = NETLINK_CONNECTED;
1103 nlk->dst_portid = nladdr->nl_pid;
1104 nlk->dst_group = ffs(nladdr->nl_groups);
1110 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1111 int *addr_len, int peer)
1113 struct sock *sk = sock->sk;
1114 struct netlink_sock *nlk = nlk_sk(sk);
1115 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1117 nladdr->nl_family = AF_NETLINK;
1119 *addr_len = sizeof(*nladdr);
1122 nladdr->nl_pid = nlk->dst_portid;
1123 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1125 nladdr->nl_pid = nlk->portid;
1126 netlink_lock_table();
1127 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1128 netlink_unlock_table();
1133 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1136 /* try to hand this ioctl down to the NIC drivers.
1138 return -ENOIOCTLCMD;
1141 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1144 struct netlink_sock *nlk;
1146 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1148 return ERR_PTR(-ECONNREFUSED);
1150 /* Don't bother queuing skb if kernel socket has no input function */
1152 if (sock->sk_state == NETLINK_CONNECTED &&
1153 nlk->dst_portid != nlk_sk(ssk)->portid) {
1155 return ERR_PTR(-ECONNREFUSED);
1160 struct sock *netlink_getsockbyfilp(struct file *filp)
1162 struct inode *inode = file_inode(filp);
1165 if (!S_ISSOCK(inode->i_mode))
1166 return ERR_PTR(-ENOTSOCK);
1168 sock = SOCKET_I(inode)->sk;
1169 if (sock->sk_family != AF_NETLINK)
1170 return ERR_PTR(-EINVAL);
1176 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1179 struct sk_buff *skb;
1182 if (size <= NLMSG_GOODSIZE || broadcast)
1183 return alloc_skb(size, GFP_KERNEL);
1185 size = SKB_DATA_ALIGN(size) +
1186 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1188 data = vmalloc(size);
1192 skb = __build_skb(data, size);
1196 skb->destructor = netlink_skb_destructor;
1202 * Attach a skb to a netlink socket.
1203 * The caller must hold a reference to the destination socket. On error, the
1204 * reference is dropped. The skb is not send to the destination, just all
1205 * all error checks are performed and memory in the queue is reserved.
1207 * < 0: error. skb freed, reference to sock dropped.
1209 * 1: repeat lookup - reference dropped while waiting for socket memory.
1211 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1212 long *timeo, struct sock *ssk)
1214 struct netlink_sock *nlk;
1218 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1219 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1220 DECLARE_WAITQUEUE(wait, current);
1222 if (!ssk || netlink_is_kernel(ssk))
1223 netlink_overrun(sk);
1229 __set_current_state(TASK_INTERRUPTIBLE);
1230 add_wait_queue(&nlk->wait, &wait);
1232 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1233 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1234 !sock_flag(sk, SOCK_DEAD))
1235 *timeo = schedule_timeout(*timeo);
1237 __set_current_state(TASK_RUNNING);
1238 remove_wait_queue(&nlk->wait, &wait);
1241 if (signal_pending(current)) {
1243 return sock_intr_errno(*timeo);
1247 netlink_skb_set_owner_r(skb, sk);
1251 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1255 netlink_deliver_tap(sock_net(sk), skb);
1257 skb_queue_tail(&sk->sk_receive_queue, skb);
1258 sk->sk_data_ready(sk);
1262 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1264 int len = __netlink_sendskb(sk, skb);
1270 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1276 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1280 WARN_ON(skb->sk != NULL);
1281 delta = skb->end - skb->tail;
1282 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1285 if (skb_shared(skb)) {
1286 struct sk_buff *nskb = skb_clone(skb, allocation);
1293 pskb_expand_head(skb, 0, -delta,
1294 (allocation & ~__GFP_DIRECT_RECLAIM) |
1295 __GFP_NOWARN | __GFP_NORETRY);
1299 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1303 struct netlink_sock *nlk = nlk_sk(sk);
1305 ret = -ECONNREFUSED;
1306 if (nlk->netlink_rcv != NULL) {
1308 netlink_skb_set_owner_r(skb, sk);
1309 NETLINK_CB(skb).sk = ssk;
1310 netlink_deliver_tap_kernel(sk, ssk, skb);
1311 nlk->netlink_rcv(skb);
1320 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1321 u32 portid, int nonblock)
1327 skb = netlink_trim(skb, gfp_any());
1329 timeo = sock_sndtimeo(ssk, nonblock);
1331 sk = netlink_getsockbyportid(ssk, portid);
1336 if (netlink_is_kernel(sk))
1337 return netlink_unicast_kernel(sk, skb, ssk);
1339 if (sk_filter(sk, skb)) {
1346 err = netlink_attachskb(sk, skb, &timeo, ssk);
1352 return netlink_sendskb(sk, skb);
1354 EXPORT_SYMBOL(netlink_unicast);
1356 int netlink_has_listeners(struct sock *sk, unsigned int group)
1359 struct listeners *listeners;
1361 BUG_ON(!netlink_is_kernel(sk));
1364 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1366 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1367 res = test_bit(group - 1, listeners->masks);
1373 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1375 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1377 struct netlink_sock *nlk = nlk_sk(sk);
1379 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1380 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1381 netlink_skb_set_owner_r(skb, sk);
1382 __netlink_sendskb(sk, skb);
1383 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1388 struct netlink_broadcast_data {
1389 struct sock *exclude_sk;
1394 int delivery_failure;
1398 struct sk_buff *skb, *skb2;
1399 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1403 static void do_one_broadcast(struct sock *sk,
1404 struct netlink_broadcast_data *p)
1406 struct netlink_sock *nlk = nlk_sk(sk);
1409 if (p->exclude_sk == sk)
1412 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1413 !test_bit(p->group - 1, nlk->groups))
1416 if (!net_eq(sock_net(sk), p->net)) {
1417 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1420 if (!peernet_has_id(sock_net(sk), p->net))
1423 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1429 netlink_overrun(sk);
1434 if (p->skb2 == NULL) {
1435 if (skb_shared(p->skb)) {
1436 p->skb2 = skb_clone(p->skb, p->allocation);
1438 p->skb2 = skb_get(p->skb);
1440 * skb ownership may have been set when
1441 * delivered to a previous socket.
1443 skb_orphan(p->skb2);
1446 if (p->skb2 == NULL) {
1447 netlink_overrun(sk);
1448 /* Clone failed. Notify ALL listeners. */
1450 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1451 p->delivery_failure = 1;
1454 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1459 if (sk_filter(sk, p->skb2)) {
1464 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1465 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1466 NETLINK_CB(p->skb2).nsid_is_set = true;
1467 val = netlink_broadcast_deliver(sk, p->skb2);
1469 netlink_overrun(sk);
1470 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1471 p->delivery_failure = 1;
1473 p->congested |= val;
1481 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1482 u32 group, gfp_t allocation,
1483 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1486 struct net *net = sock_net(ssk);
1487 struct netlink_broadcast_data info;
1490 skb = netlink_trim(skb, allocation);
1492 info.exclude_sk = ssk;
1494 info.portid = portid;
1497 info.delivery_failure = 0;
1500 info.allocation = allocation;
1503 info.tx_filter = filter;
1504 info.tx_data = filter_data;
1506 /* While we sleep in clone, do not allow to change socket list */
1508 netlink_lock_table();
1510 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1511 do_one_broadcast(sk, &info);
1515 netlink_unlock_table();
1517 if (info.delivery_failure) {
1518 kfree_skb(info.skb2);
1521 consume_skb(info.skb2);
1523 if (info.delivered) {
1524 if (info.congested && gfpflags_allow_blocking(allocation))
1530 EXPORT_SYMBOL(netlink_broadcast_filtered);
1532 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1533 u32 group, gfp_t allocation)
1535 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1538 EXPORT_SYMBOL(netlink_broadcast);
1540 struct netlink_set_err_data {
1541 struct sock *exclude_sk;
1547 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1549 struct netlink_sock *nlk = nlk_sk(sk);
1552 if (sk == p->exclude_sk)
1555 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1558 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1559 !test_bit(p->group - 1, nlk->groups))
1562 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1567 sk->sk_err = p->code;
1568 sk->sk_error_report(sk);
1574 * netlink_set_err - report error to broadcast listeners
1575 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1576 * @portid: the PORTID of a process that we want to skip (if any)
1577 * @group: the broadcast group that will notice the error
1578 * @code: error code, must be negative (as usual in kernelspace)
1580 * This function returns the number of broadcast listeners that have set the
1581 * NETLINK_NO_ENOBUFS socket option.
1583 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1585 struct netlink_set_err_data info;
1589 info.exclude_sk = ssk;
1590 info.portid = portid;
1592 /* sk->sk_err wants a positive error value */
1595 read_lock(&nl_table_lock);
1597 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1598 ret += do_one_set_err(sk, &info);
1600 read_unlock(&nl_table_lock);
1603 EXPORT_SYMBOL(netlink_set_err);
1605 /* must be called with netlink table grabbed */
1606 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1610 int old, new = !!is_new, subscriptions;
1612 old = test_bit(group - 1, nlk->groups);
1613 subscriptions = nlk->subscriptions - old + new;
1615 __set_bit(group - 1, nlk->groups);
1617 __clear_bit(group - 1, nlk->groups);
1618 netlink_update_subscriptions(&nlk->sk, subscriptions);
1619 netlink_update_listeners(&nlk->sk);
1622 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1623 char __user *optval, unsigned int optlen)
1625 struct sock *sk = sock->sk;
1626 struct netlink_sock *nlk = nlk_sk(sk);
1627 unsigned int val = 0;
1630 if (level != SOL_NETLINK)
1631 return -ENOPROTOOPT;
1633 if (optlen >= sizeof(int) &&
1634 get_user(val, (unsigned int __user *)optval))
1638 case NETLINK_PKTINFO:
1640 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1642 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1645 case NETLINK_ADD_MEMBERSHIP:
1646 case NETLINK_DROP_MEMBERSHIP: {
1647 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1649 err = netlink_realloc_groups(sk);
1652 if (!val || val - 1 >= nlk->ngroups)
1654 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1655 err = nlk->netlink_bind(sock_net(sk), val);
1659 netlink_table_grab();
1660 netlink_update_socket_mc(nlk, val,
1661 optname == NETLINK_ADD_MEMBERSHIP);
1662 netlink_table_ungrab();
1663 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1664 nlk->netlink_unbind(sock_net(sk), val);
1669 case NETLINK_BROADCAST_ERROR:
1671 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1673 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1676 case NETLINK_NO_ENOBUFS:
1678 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1679 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1680 wake_up_interruptible(&nlk->wait);
1682 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1686 case NETLINK_LISTEN_ALL_NSID:
1687 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1691 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1693 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1696 case NETLINK_CAP_ACK:
1698 nlk->flags |= NETLINK_F_CAP_ACK;
1700 nlk->flags &= ~NETLINK_F_CAP_ACK;
1703 case NETLINK_EXT_ACK:
1705 nlk->flags |= NETLINK_F_EXT_ACK;
1707 nlk->flags &= ~NETLINK_F_EXT_ACK;
1716 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1717 char __user *optval, int __user *optlen)
1719 struct sock *sk = sock->sk;
1720 struct netlink_sock *nlk = nlk_sk(sk);
1723 if (level != SOL_NETLINK)
1724 return -ENOPROTOOPT;
1726 if (get_user(len, optlen))
1732 case NETLINK_PKTINFO:
1733 if (len < sizeof(int))
1736 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1737 if (put_user(len, optlen) ||
1738 put_user(val, optval))
1742 case NETLINK_BROADCAST_ERROR:
1743 if (len < sizeof(int))
1746 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1747 if (put_user(len, optlen) ||
1748 put_user(val, optval))
1752 case NETLINK_NO_ENOBUFS:
1753 if (len < sizeof(int))
1756 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1757 if (put_user(len, optlen) ||
1758 put_user(val, optval))
1762 case NETLINK_LIST_MEMBERSHIPS: {
1763 int pos, idx, shift;
1766 netlink_lock_table();
1767 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1768 if (len - pos < sizeof(u32))
1771 idx = pos / sizeof(unsigned long);
1772 shift = (pos % sizeof(unsigned long)) * 8;
1773 if (put_user((u32)(nlk->groups[idx] >> shift),
1774 (u32 __user *)(optval + pos))) {
1779 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1781 netlink_unlock_table();
1784 case NETLINK_CAP_ACK:
1785 if (len < sizeof(int))
1788 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1789 if (put_user(len, optlen) ||
1790 put_user(val, optval))
1794 case NETLINK_EXT_ACK:
1795 if (len < sizeof(int))
1798 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1799 if (put_user(len, optlen) || put_user(val, optval))
1809 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1811 struct nl_pktinfo info;
1813 info.group = NETLINK_CB(skb).dst_group;
1814 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1817 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1818 struct sk_buff *skb)
1820 if (!NETLINK_CB(skb).nsid_is_set)
1823 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1824 &NETLINK_CB(skb).nsid);
1827 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1829 struct sock *sk = sock->sk;
1830 struct netlink_sock *nlk = nlk_sk(sk);
1831 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1834 struct sk_buff *skb;
1836 struct scm_cookie scm;
1837 u32 netlink_skb_flags = 0;
1839 if (msg->msg_flags&MSG_OOB)
1842 err = scm_send(sock, msg, &scm, true);
1846 if (msg->msg_namelen) {
1848 if (addr->nl_family != AF_NETLINK)
1850 dst_portid = addr->nl_pid;
1851 dst_group = ffs(addr->nl_groups);
1853 if ((dst_group || dst_portid) &&
1854 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1856 netlink_skb_flags |= NETLINK_SKB_DST;
1858 dst_portid = nlk->dst_portid;
1859 dst_group = nlk->dst_group;
1863 err = netlink_autobind(sock);
1867 /* Ensure nlk is hashed and visible. */
1872 if (len > sk->sk_sndbuf - 32)
1875 skb = netlink_alloc_large_skb(len, dst_group);
1879 NETLINK_CB(skb).portid = nlk->portid;
1880 NETLINK_CB(skb).dst_group = dst_group;
1881 NETLINK_CB(skb).creds = scm.creds;
1882 NETLINK_CB(skb).flags = netlink_skb_flags;
1885 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1890 err = security_netlink_send(sk, skb);
1897 refcount_inc(&skb->users);
1898 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1900 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1907 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1910 struct scm_cookie scm;
1911 struct sock *sk = sock->sk;
1912 struct netlink_sock *nlk = nlk_sk(sk);
1913 int noblock = flags&MSG_DONTWAIT;
1915 struct sk_buff *skb, *data_skb;
1923 skb = skb_recv_datagram(sk, flags, noblock, &err);
1929 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1930 if (unlikely(skb_shinfo(skb)->frag_list)) {
1932 * If this skb has a frag_list, then here that means that we
1933 * will have to use the frag_list skb's data for compat tasks
1934 * and the regular skb's data for normal (non-compat) tasks.
1936 * If we need to send the compat skb, assign it to the
1937 * 'data_skb' variable so that it will be used below for data
1938 * copying. We keep 'skb' for everything else, including
1939 * freeing both later.
1941 if (flags & MSG_CMSG_COMPAT)
1942 data_skb = skb_shinfo(skb)->frag_list;
1946 /* Record the max length of recvmsg() calls for future allocations */
1947 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1948 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1949 SKB_WITH_OVERHEAD(32768));
1951 copied = data_skb->len;
1953 msg->msg_flags |= MSG_TRUNC;
1957 skb_reset_transport_header(data_skb);
1958 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1960 if (msg->msg_name) {
1961 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1962 addr->nl_family = AF_NETLINK;
1964 addr->nl_pid = NETLINK_CB(skb).portid;
1965 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1966 msg->msg_namelen = sizeof(*addr);
1969 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1970 netlink_cmsg_recv_pktinfo(msg, skb);
1971 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1972 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1974 memset(&scm, 0, sizeof(scm));
1975 scm.creds = *NETLINK_CREDS(skb);
1976 if (flags & MSG_TRUNC)
1977 copied = data_skb->len;
1979 skb_free_datagram(sk, skb);
1981 if (nlk->cb_running &&
1982 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1983 ret = netlink_dump(sk);
1986 sk->sk_error_report(sk);
1990 scm_recv(sock, msg, &scm, flags);
1992 netlink_rcv_wake(sk);
1993 return err ? : copied;
1996 static void netlink_data_ready(struct sock *sk)
2002 * We export these functions to other modules. They provide a
2003 * complete set of kernel non-blocking support for message
2008 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2009 struct netlink_kernel_cfg *cfg)
2011 struct socket *sock;
2013 struct netlink_sock *nlk;
2014 struct listeners *listeners = NULL;
2015 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2016 unsigned int groups;
2020 if (unit < 0 || unit >= MAX_LINKS)
2023 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2026 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2027 goto out_sock_release_nosk;
2031 if (!cfg || cfg->groups < 32)
2034 groups = cfg->groups;
2036 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2038 goto out_sock_release;
2040 sk->sk_data_ready = netlink_data_ready;
2041 if (cfg && cfg->input)
2042 nlk_sk(sk)->netlink_rcv = cfg->input;
2044 if (netlink_insert(sk, 0))
2045 goto out_sock_release;
2048 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2050 netlink_table_grab();
2051 if (!nl_table[unit].registered) {
2052 nl_table[unit].groups = groups;
2053 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2054 nl_table[unit].cb_mutex = cb_mutex;
2055 nl_table[unit].module = module;
2057 nl_table[unit].bind = cfg->bind;
2058 nl_table[unit].unbind = cfg->unbind;
2059 nl_table[unit].flags = cfg->flags;
2061 nl_table[unit].compare = cfg->compare;
2063 nl_table[unit].registered = 1;
2066 nl_table[unit].registered++;
2068 netlink_table_ungrab();
2073 netlink_kernel_release(sk);
2076 out_sock_release_nosk:
2080 EXPORT_SYMBOL(__netlink_kernel_create);
2083 netlink_kernel_release(struct sock *sk)
2085 if (sk == NULL || sk->sk_socket == NULL)
2088 sock_release(sk->sk_socket);
2090 EXPORT_SYMBOL(netlink_kernel_release);
2092 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2094 struct listeners *new, *old;
2095 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2100 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2101 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2104 old = nl_deref_protected(tbl->listeners);
2105 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2106 rcu_assign_pointer(tbl->listeners, new);
2108 kfree_rcu(old, rcu);
2110 tbl->groups = groups;
2116 * netlink_change_ngroups - change number of multicast groups
2118 * This changes the number of multicast groups that are available
2119 * on a certain netlink family. Note that it is not possible to
2120 * change the number of groups to below 32. Also note that it does
2121 * not implicitly call netlink_clear_multicast_users() when the
2122 * number of groups is reduced.
2124 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2125 * @groups: The new number of groups.
2127 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2131 netlink_table_grab();
2132 err = __netlink_change_ngroups(sk, groups);
2133 netlink_table_ungrab();
2138 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2141 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2143 sk_for_each_bound(sk, &tbl->mc_list)
2144 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2148 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2150 struct nlmsghdr *nlh;
2151 int size = nlmsg_msg_size(len);
2153 nlh = skb_put(skb, NLMSG_ALIGN(size));
2154 nlh->nlmsg_type = type;
2155 nlh->nlmsg_len = size;
2156 nlh->nlmsg_flags = flags;
2157 nlh->nlmsg_pid = portid;
2158 nlh->nlmsg_seq = seq;
2159 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2160 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2163 EXPORT_SYMBOL(__nlmsg_put);
2166 * It looks a bit ugly.
2167 * It would be better to create kernel thread.
2170 static int netlink_dump(struct sock *sk)
2172 struct netlink_sock *nlk = nlk_sk(sk);
2173 struct netlink_callback *cb;
2174 struct sk_buff *skb = NULL;
2175 struct nlmsghdr *nlh;
2176 struct module *module;
2181 mutex_lock(nlk->cb_mutex);
2182 if (!nlk->cb_running) {
2187 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2190 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2191 * required, but it makes sense to _attempt_ a 16K bytes allocation
2192 * to reduce number of system calls on dump operations, if user
2193 * ever provided a big enough buffer.
2196 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2198 if (alloc_min_size < nlk->max_recvmsg_len) {
2199 alloc_size = nlk->max_recvmsg_len;
2200 skb = alloc_skb(alloc_size,
2201 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2202 __GFP_NOWARN | __GFP_NORETRY);
2205 alloc_size = alloc_min_size;
2206 skb = alloc_skb(alloc_size, GFP_KERNEL);
2211 /* Trim skb to allocated size. User is expected to provide buffer as
2212 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2213 * netlink_recvmsg())). dump will pack as many smaller messages as
2214 * could fit within the allocated skb. skb is typically allocated
2215 * with larger space than required (could be as much as near 2x the
2216 * requested size with align to next power of 2 approach). Allowing
2217 * dump to use the excess space makes it difficult for a user to have a
2218 * reasonable static buffer based on the expected largest dump of a
2219 * single netdev. The outcome is MSG_TRUNC error.
2221 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2222 netlink_skb_set_owner_r(skb, sk);
2224 if (nlk->dump_done_errno > 0)
2225 nlk->dump_done_errno = cb->dump(skb, cb);
2227 if (nlk->dump_done_errno > 0 ||
2228 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2229 mutex_unlock(nlk->cb_mutex);
2231 if (sk_filter(sk, skb))
2234 __netlink_sendskb(sk, skb);
2238 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE,
2239 sizeof(nlk->dump_done_errno), NLM_F_MULTI);
2243 nl_dump_check_consistent(cb, nlh);
2245 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno,
2246 sizeof(nlk->dump_done_errno));
2248 if (sk_filter(sk, skb))
2251 __netlink_sendskb(sk, skb);
2256 nlk->cb_running = false;
2257 module = cb->module;
2259 mutex_unlock(nlk->cb_mutex);
2265 mutex_unlock(nlk->cb_mutex);
2270 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2271 const struct nlmsghdr *nlh,
2272 struct netlink_dump_control *control)
2274 struct netlink_callback *cb;
2276 struct netlink_sock *nlk;
2279 refcount_inc(&skb->users);
2281 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2283 ret = -ECONNREFUSED;
2288 mutex_lock(nlk->cb_mutex);
2289 /* A dump is in progress... */
2290 if (nlk->cb_running) {
2294 /* add reference of module which cb->dump belongs to */
2295 if (!try_module_get(control->module)) {
2296 ret = -EPROTONOSUPPORT;
2301 memset(cb, 0, sizeof(*cb));
2302 cb->start = control->start;
2303 cb->dump = control->dump;
2304 cb->done = control->done;
2306 cb->data = control->data;
2307 cb->module = control->module;
2308 cb->min_dump_alloc = control->min_dump_alloc;
2312 ret = cb->start(cb);
2317 nlk->cb_running = true;
2318 nlk->dump_done_errno = INT_MAX;
2320 mutex_unlock(nlk->cb_mutex);
2322 ret = netlink_dump(sk);
2329 /* We successfully started a dump, by returning -EINTR we
2330 * signal not to send ACK even if it was requested.
2335 module_put(control->module);
2338 mutex_unlock(nlk->cb_mutex);
2343 EXPORT_SYMBOL(__netlink_dump_start);
2345 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2346 const struct netlink_ext_ack *extack)
2348 struct sk_buff *skb;
2349 struct nlmsghdr *rep;
2350 struct nlmsgerr *errmsg;
2351 size_t payload = sizeof(*errmsg);
2353 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2354 unsigned int flags = 0;
2355 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2357 /* Error messages get the original request appened, unless the user
2358 * requests to cap the error message, and get extra error data if
2361 if (nlk_has_extack && extack && extack->_msg)
2362 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2365 if (!(nlk->flags & NETLINK_F_CAP_ACK))
2366 payload += nlmsg_len(nlh);
2368 flags |= NLM_F_CAPPED;
2369 if (nlk_has_extack && extack && extack->bad_attr)
2370 tlvlen += nla_total_size(sizeof(u32));
2372 flags |= NLM_F_CAPPED;
2374 if (nlk_has_extack && extack && extack->cookie_len)
2375 tlvlen += nla_total_size(extack->cookie_len);
2379 flags |= NLM_F_ACK_TLVS;
2381 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2383 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2384 NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk);
2388 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2389 NLMSG_ERROR, payload, flags);
2390 errmsg = nlmsg_data(rep);
2391 errmsg->error = err;
2392 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2394 if (nlk_has_extack && extack) {
2396 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2400 if (extack->bad_attr &&
2401 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2402 (u8 *)extack->bad_attr >= in_skb->data +
2404 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2405 (u8 *)extack->bad_attr -
2408 if (extack->cookie_len)
2409 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2415 nlmsg_end(skb, rep);
2417 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2419 EXPORT_SYMBOL(netlink_ack);
2421 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2423 struct netlink_ext_ack *))
2425 struct netlink_ext_ack extack;
2426 struct nlmsghdr *nlh;
2429 while (skb->len >= nlmsg_total_size(0)) {
2432 memset(&extack, 0, sizeof(extack));
2433 nlh = nlmsg_hdr(skb);
2436 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2439 /* Only requests are handled by the kernel */
2440 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2443 /* Skip control messages */
2444 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2447 err = cb(skb, nlh, &extack);
2452 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2453 netlink_ack(skb, nlh, err, &extack);
2456 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2457 if (msglen > skb->len)
2459 skb_pull(skb, msglen);
2464 EXPORT_SYMBOL(netlink_rcv_skb);
2467 * nlmsg_notify - send a notification netlink message
2468 * @sk: netlink socket to use
2469 * @skb: notification message
2470 * @portid: destination netlink portid for reports or 0
2471 * @group: destination multicast group or 0
2472 * @report: 1 to report back, 0 to disable
2473 * @flags: allocation flags
2475 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2476 unsigned int group, int report, gfp_t flags)
2481 int exclude_portid = 0;
2484 refcount_inc(&skb->users);
2485 exclude_portid = portid;
2488 /* errors reported via destination sk->sk_err, but propagate
2489 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2490 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2496 err2 = nlmsg_unicast(sk, skb, portid);
2497 if (!err || err == -ESRCH)
2503 EXPORT_SYMBOL(nlmsg_notify);
2505 #ifdef CONFIG_PROC_FS
2506 struct nl_seq_iter {
2507 struct seq_net_private p;
2508 struct rhashtable_iter hti;
2512 static int netlink_walk_start(struct nl_seq_iter *iter)
2516 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti,
2519 iter->link = MAX_LINKS;
2523 rhashtable_walk_start(&iter->hti);
2528 static void netlink_walk_stop(struct nl_seq_iter *iter)
2530 rhashtable_walk_stop(&iter->hti);
2531 rhashtable_walk_exit(&iter->hti);
2534 static void *__netlink_seq_next(struct seq_file *seq)
2536 struct nl_seq_iter *iter = seq->private;
2537 struct netlink_sock *nlk;
2543 nlk = rhashtable_walk_next(&iter->hti);
2546 if (PTR_ERR(nlk) == -EAGAIN)
2555 netlink_walk_stop(iter);
2556 if (++iter->link >= MAX_LINKS)
2559 err = netlink_walk_start(iter);
2561 return ERR_PTR(err);
2563 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2568 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2570 struct nl_seq_iter *iter = seq->private;
2571 void *obj = SEQ_START_TOKEN;
2577 err = netlink_walk_start(iter);
2579 return ERR_PTR(err);
2581 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2582 obj = __netlink_seq_next(seq);
2587 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2590 return __netlink_seq_next(seq);
2593 static void netlink_seq_stop(struct seq_file *seq, void *v)
2595 struct nl_seq_iter *iter = seq->private;
2597 if (iter->link >= MAX_LINKS)
2600 netlink_walk_stop(iter);
2604 static int netlink_seq_show(struct seq_file *seq, void *v)
2606 if (v == SEQ_START_TOKEN) {
2608 "sk Eth Pid Groups "
2609 "Rmem Wmem Dump Locks Drops Inode\n");
2612 struct netlink_sock *nlk = nlk_sk(s);
2614 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2618 nlk->groups ? (u32)nlk->groups[0] : 0,
2619 sk_rmem_alloc_get(s),
2620 sk_wmem_alloc_get(s),
2622 refcount_read(&s->sk_refcnt),
2623 atomic_read(&s->sk_drops),
2631 static const struct seq_operations netlink_seq_ops = {
2632 .start = netlink_seq_start,
2633 .next = netlink_seq_next,
2634 .stop = netlink_seq_stop,
2635 .show = netlink_seq_show,
2639 static int netlink_seq_open(struct inode *inode, struct file *file)
2641 return seq_open_net(inode, file, &netlink_seq_ops,
2642 sizeof(struct nl_seq_iter));
2645 static const struct file_operations netlink_seq_fops = {
2646 .open = netlink_seq_open,
2648 .llseek = seq_lseek,
2649 .release = seq_release_net,
2654 int netlink_register_notifier(struct notifier_block *nb)
2656 return blocking_notifier_chain_register(&netlink_chain, nb);
2658 EXPORT_SYMBOL(netlink_register_notifier);
2660 int netlink_unregister_notifier(struct notifier_block *nb)
2662 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2664 EXPORT_SYMBOL(netlink_unregister_notifier);
2666 static const struct proto_ops netlink_ops = {
2667 .family = PF_NETLINK,
2668 .owner = THIS_MODULE,
2669 .release = netlink_release,
2670 .bind = netlink_bind,
2671 .connect = netlink_connect,
2672 .socketpair = sock_no_socketpair,
2673 .accept = sock_no_accept,
2674 .getname = netlink_getname,
2675 .poll = datagram_poll,
2676 .ioctl = netlink_ioctl,
2677 .listen = sock_no_listen,
2678 .shutdown = sock_no_shutdown,
2679 .setsockopt = netlink_setsockopt,
2680 .getsockopt = netlink_getsockopt,
2681 .sendmsg = netlink_sendmsg,
2682 .recvmsg = netlink_recvmsg,
2683 .mmap = sock_no_mmap,
2684 .sendpage = sock_no_sendpage,
2687 static const struct net_proto_family netlink_family_ops = {
2688 .family = PF_NETLINK,
2689 .create = netlink_create,
2690 .owner = THIS_MODULE, /* for consistency 8) */
2693 static int __net_init netlink_net_init(struct net *net)
2695 #ifdef CONFIG_PROC_FS
2696 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2702 static void __net_exit netlink_net_exit(struct net *net)
2704 #ifdef CONFIG_PROC_FS
2705 remove_proc_entry("netlink", net->proc_net);
2709 static void __init netlink_add_usersock_entry(void)
2711 struct listeners *listeners;
2714 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2716 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2718 netlink_table_grab();
2720 nl_table[NETLINK_USERSOCK].groups = groups;
2721 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2722 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2723 nl_table[NETLINK_USERSOCK].registered = 1;
2724 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2726 netlink_table_ungrab();
2729 static struct pernet_operations __net_initdata netlink_net_ops = {
2730 .init = netlink_net_init,
2731 .exit = netlink_net_exit,
2734 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2736 const struct netlink_sock *nlk = data;
2737 struct netlink_compare_arg arg;
2739 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2740 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2743 static const struct rhashtable_params netlink_rhashtable_params = {
2744 .head_offset = offsetof(struct netlink_sock, node),
2745 .key_len = netlink_compare_arg_len,
2746 .obj_hashfn = netlink_hash,
2747 .obj_cmpfn = netlink_compare,
2748 .automatic_shrinking = true,
2751 static int __init netlink_proto_init(void)
2754 int err = proto_register(&netlink_proto, 0);
2759 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2761 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2765 for (i = 0; i < MAX_LINKS; i++) {
2766 if (rhashtable_init(&nl_table[i].hash,
2767 &netlink_rhashtable_params) < 0) {
2769 rhashtable_destroy(&nl_table[i].hash);
2775 netlink_add_usersock_entry();
2777 sock_register(&netlink_family_ops);
2778 register_pernet_subsys(&netlink_net_ops);
2779 register_pernet_subsys(&netlink_tap_net_ops);
2780 /* The netlink device handler may be needed early. */
2785 panic("netlink_init: Cannot allocate nl_table\n");
2788 core_initcall(netlink_proto_init);