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>
66 #include <linux/nospec.h>
68 #include <net/net_namespace.h>
69 #include <net/netns/generic.h>
72 #include <net/netlink.h>
74 #include "af_netlink.h"
78 unsigned long masks[0];
82 #define NETLINK_S_CONGESTED 0x0
84 static inline int netlink_is_kernel(struct sock *sk)
86 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
89 struct netlink_table *nl_table __read_mostly;
90 EXPORT_SYMBOL_GPL(nl_table);
92 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
94 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
96 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
99 "nlk_cb_mutex-USERSOCK",
100 "nlk_cb_mutex-FIREWALL",
101 "nlk_cb_mutex-SOCK_DIAG",
102 "nlk_cb_mutex-NFLOG",
104 "nlk_cb_mutex-SELINUX",
105 "nlk_cb_mutex-ISCSI",
106 "nlk_cb_mutex-AUDIT",
107 "nlk_cb_mutex-FIB_LOOKUP",
108 "nlk_cb_mutex-CONNECTOR",
109 "nlk_cb_mutex-NETFILTER",
110 "nlk_cb_mutex-IP6_FW",
111 "nlk_cb_mutex-DNRTMSG",
112 "nlk_cb_mutex-KOBJECT_UEVENT",
113 "nlk_cb_mutex-GENERIC",
115 "nlk_cb_mutex-SCSITRANSPORT",
116 "nlk_cb_mutex-ECRYPTFS",
118 "nlk_cb_mutex-CRYPTO",
129 "nlk_cb_mutex-MAX_LINKS"
132 static int netlink_dump(struct sock *sk);
134 /* nl_table locking explained:
135 * Lookup and traversal are protected with an RCU read-side lock. Insertion
136 * and removal are protected with per bucket lock while using RCU list
137 * modification primitives and may run in parallel to RCU protected lookups.
138 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
139 * been acquired * either during or after the socket has been removed from
140 * the list and after an RCU grace period.
142 DEFINE_RWLOCK(nl_table_lock);
143 EXPORT_SYMBOL_GPL(nl_table_lock);
144 static atomic_t nl_table_users = ATOMIC_INIT(0);
146 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
148 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
151 static const struct rhashtable_params netlink_rhashtable_params;
153 static inline u32 netlink_group_mask(u32 group)
155 return group ? 1 << (group - 1) : 0;
158 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
161 unsigned int len = skb_end_offset(skb);
164 new = alloc_skb(len, gfp_mask);
168 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
169 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
170 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
172 skb_put_data(new, skb->data, len);
176 static unsigned int netlink_tap_net_id;
178 struct netlink_tap_net {
179 struct list_head netlink_tap_all;
180 struct mutex netlink_tap_lock;
183 int netlink_add_tap(struct netlink_tap *nt)
185 struct net *net = dev_net(nt->dev);
186 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
188 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
191 mutex_lock(&nn->netlink_tap_lock);
192 list_add_rcu(&nt->list, &nn->netlink_tap_all);
193 mutex_unlock(&nn->netlink_tap_lock);
195 __module_get(nt->module);
199 EXPORT_SYMBOL_GPL(netlink_add_tap);
201 static int __netlink_remove_tap(struct netlink_tap *nt)
203 struct net *net = dev_net(nt->dev);
204 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
206 struct netlink_tap *tmp;
208 mutex_lock(&nn->netlink_tap_lock);
210 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
212 list_del_rcu(&nt->list);
218 pr_warn("__netlink_remove_tap: %p not found\n", nt);
220 mutex_unlock(&nn->netlink_tap_lock);
223 module_put(nt->module);
225 return found ? 0 : -ENODEV;
228 int netlink_remove_tap(struct netlink_tap *nt)
232 ret = __netlink_remove_tap(nt);
237 EXPORT_SYMBOL_GPL(netlink_remove_tap);
239 static __net_init int netlink_tap_init_net(struct net *net)
241 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
243 INIT_LIST_HEAD(&nn->netlink_tap_all);
244 mutex_init(&nn->netlink_tap_lock);
248 static void __net_exit netlink_tap_exit_net(struct net *net)
252 static struct pernet_operations netlink_tap_net_ops = {
253 .init = netlink_tap_init_net,
254 .exit = netlink_tap_exit_net,
255 .id = &netlink_tap_net_id,
256 .size = sizeof(struct netlink_tap_net),
259 static bool netlink_filter_tap(const struct sk_buff *skb)
261 struct sock *sk = skb->sk;
263 /* We take the more conservative approach and
264 * whitelist socket protocols that may pass.
266 switch (sk->sk_protocol) {
268 case NETLINK_USERSOCK:
269 case NETLINK_SOCK_DIAG:
272 case NETLINK_FIB_LOOKUP:
273 case NETLINK_NETFILTER:
274 case NETLINK_GENERIC:
281 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
282 struct net_device *dev)
284 struct sk_buff *nskb;
285 struct sock *sk = skb->sk;
288 if (!net_eq(dev_net(dev), sock_net(sk)))
293 if (is_vmalloc_addr(skb->head))
294 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
296 nskb = skb_clone(skb, GFP_ATOMIC);
299 nskb->protocol = htons((u16) sk->sk_protocol);
300 nskb->pkt_type = netlink_is_kernel(sk) ?
301 PACKET_KERNEL : PACKET_USER;
302 skb_reset_network_header(nskb);
303 ret = dev_queue_xmit(nskb);
304 if (unlikely(ret > 0))
305 ret = net_xmit_errno(ret);
312 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
315 struct netlink_tap *tmp;
317 if (!netlink_filter_tap(skb))
320 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
321 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
327 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
329 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
333 if (unlikely(!list_empty(&nn->netlink_tap_all)))
334 __netlink_deliver_tap(skb, nn);
339 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
342 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
343 netlink_deliver_tap(sock_net(dst), skb);
346 static void netlink_overrun(struct sock *sk)
348 struct netlink_sock *nlk = nlk_sk(sk);
350 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
351 if (!test_and_set_bit(NETLINK_S_CONGESTED,
352 &nlk_sk(sk)->state)) {
353 sk->sk_err = ENOBUFS;
354 sk->sk_error_report(sk);
357 atomic_inc(&sk->sk_drops);
360 static void netlink_rcv_wake(struct sock *sk)
362 struct netlink_sock *nlk = nlk_sk(sk);
364 if (skb_queue_empty(&sk->sk_receive_queue))
365 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
366 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
367 wake_up_interruptible(&nlk->wait);
370 static void netlink_skb_destructor(struct sk_buff *skb)
372 if (is_vmalloc_addr(skb->head)) {
374 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
383 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
385 WARN_ON(skb->sk != NULL);
387 skb->destructor = netlink_skb_destructor;
388 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
389 sk_mem_charge(sk, skb->truesize);
392 static void netlink_sock_destruct(struct sock *sk)
394 struct netlink_sock *nlk = nlk_sk(sk);
396 if (nlk->cb_running) {
398 nlk->cb.done(&nlk->cb);
399 module_put(nlk->cb.module);
400 kfree_skb(nlk->cb.skb);
403 skb_queue_purge(&sk->sk_receive_queue);
405 if (!sock_flag(sk, SOCK_DEAD)) {
406 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
410 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
411 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
412 WARN_ON(nlk_sk(sk)->groups);
415 static void netlink_sock_destruct_work(struct work_struct *work)
417 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
423 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
424 * SMP. Look, when several writers sleep and reader wakes them up, all but one
425 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
426 * this, _but_ remember, it adds useless work on UP machines.
429 void netlink_table_grab(void)
430 __acquires(nl_table_lock)
434 write_lock_irq(&nl_table_lock);
436 if (atomic_read(&nl_table_users)) {
437 DECLARE_WAITQUEUE(wait, current);
439 add_wait_queue_exclusive(&nl_table_wait, &wait);
441 set_current_state(TASK_UNINTERRUPTIBLE);
442 if (atomic_read(&nl_table_users) == 0)
444 write_unlock_irq(&nl_table_lock);
446 write_lock_irq(&nl_table_lock);
449 __set_current_state(TASK_RUNNING);
450 remove_wait_queue(&nl_table_wait, &wait);
454 void netlink_table_ungrab(void)
455 __releases(nl_table_lock)
457 write_unlock_irq(&nl_table_lock);
458 wake_up(&nl_table_wait);
462 netlink_lock_table(void)
466 /* read_lock() synchronizes us to netlink_table_grab */
468 read_lock_irqsave(&nl_table_lock, flags);
469 atomic_inc(&nl_table_users);
470 read_unlock_irqrestore(&nl_table_lock, flags);
474 netlink_unlock_table(void)
476 if (atomic_dec_and_test(&nl_table_users))
477 wake_up(&nl_table_wait);
480 struct netlink_compare_arg
486 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
487 #define netlink_compare_arg_len \
488 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
490 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
493 const struct netlink_compare_arg *x = arg->key;
494 const struct netlink_sock *nlk = ptr;
496 return nlk->portid != x->portid ||
497 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
500 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
501 struct net *net, u32 portid)
503 memset(arg, 0, sizeof(*arg));
504 write_pnet(&arg->pnet, net);
505 arg->portid = portid;
508 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
511 struct netlink_compare_arg arg;
513 netlink_compare_arg_init(&arg, net, portid);
514 return rhashtable_lookup_fast(&table->hash, &arg,
515 netlink_rhashtable_params);
518 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
520 struct netlink_compare_arg arg;
522 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
523 return rhashtable_lookup_insert_key(&table->hash, &arg,
525 netlink_rhashtable_params);
528 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
530 struct netlink_table *table = &nl_table[protocol];
534 sk = __netlink_lookup(table, portid, net);
542 static const struct proto_ops netlink_ops;
545 netlink_update_listeners(struct sock *sk)
547 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
550 struct listeners *listeners;
552 listeners = nl_deref_protected(tbl->listeners);
556 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
558 sk_for_each_bound(sk, &tbl->mc_list) {
559 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
560 mask |= nlk_sk(sk)->groups[i];
562 listeners->masks[i] = mask;
564 /* this function is only called with the netlink table "grabbed", which
565 * makes sure updates are visible before bind or setsockopt return. */
568 static int netlink_insert(struct sock *sk, u32 portid)
570 struct netlink_table *table = &nl_table[sk->sk_protocol];
575 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
576 if (nlk_sk(sk)->bound)
580 if (BITS_PER_LONG > 32 &&
581 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
584 nlk_sk(sk)->portid = portid;
587 err = __netlink_insert(table, sk);
589 /* In case the hashtable backend returns with -EBUSY
590 * from here, it must not escape to the caller.
592 if (unlikely(err == -EBUSY))
600 /* We need to ensure that the socket is hashed and visible. */
602 nlk_sk(sk)->bound = portid;
609 static void netlink_remove(struct sock *sk)
611 struct netlink_table *table;
613 table = &nl_table[sk->sk_protocol];
614 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
615 netlink_rhashtable_params)) {
616 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
620 netlink_table_grab();
621 if (nlk_sk(sk)->subscriptions) {
622 __sk_del_bind_node(sk);
623 netlink_update_listeners(sk);
625 if (sk->sk_protocol == NETLINK_GENERIC)
626 atomic_inc(&genl_sk_destructing_cnt);
627 netlink_table_ungrab();
630 static struct proto netlink_proto = {
632 .owner = THIS_MODULE,
633 .obj_size = sizeof(struct netlink_sock),
636 static int __netlink_create(struct net *net, struct socket *sock,
637 struct mutex *cb_mutex, int protocol,
641 struct netlink_sock *nlk;
643 sock->ops = &netlink_ops;
645 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
649 sock_init_data(sock, sk);
653 nlk->cb_mutex = cb_mutex;
655 nlk->cb_mutex = &nlk->cb_def_mutex;
656 mutex_init(nlk->cb_mutex);
657 lockdep_set_class_and_name(nlk->cb_mutex,
658 nlk_cb_mutex_keys + protocol,
659 nlk_cb_mutex_key_strings[protocol]);
661 init_waitqueue_head(&nlk->wait);
663 sk->sk_destruct = netlink_sock_destruct;
664 sk->sk_protocol = protocol;
668 static int netlink_create(struct net *net, struct socket *sock, int protocol,
671 struct module *module = NULL;
672 struct mutex *cb_mutex;
673 struct netlink_sock *nlk;
674 int (*bind)(struct net *net, int group);
675 void (*unbind)(struct net *net, int group);
678 sock->state = SS_UNCONNECTED;
680 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
681 return -ESOCKTNOSUPPORT;
683 if (protocol < 0 || protocol >= MAX_LINKS)
684 return -EPROTONOSUPPORT;
685 protocol = array_index_nospec(protocol, MAX_LINKS);
687 netlink_lock_table();
688 #ifdef CONFIG_MODULES
689 if (!nl_table[protocol].registered) {
690 netlink_unlock_table();
691 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
692 netlink_lock_table();
695 if (nl_table[protocol].registered &&
696 try_module_get(nl_table[protocol].module))
697 module = nl_table[protocol].module;
699 err = -EPROTONOSUPPORT;
700 cb_mutex = nl_table[protocol].cb_mutex;
701 bind = nl_table[protocol].bind;
702 unbind = nl_table[protocol].unbind;
703 netlink_unlock_table();
708 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
713 sock_prot_inuse_add(net, &netlink_proto, 1);
716 nlk = nlk_sk(sock->sk);
717 nlk->module = module;
718 nlk->netlink_bind = bind;
719 nlk->netlink_unbind = unbind;
728 static void deferred_put_nlk_sk(struct rcu_head *head)
730 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
731 struct sock *sk = &nlk->sk;
736 if (!refcount_dec_and_test(&sk->sk_refcnt))
739 if (nlk->cb_running && nlk->cb.done) {
740 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
741 schedule_work(&nlk->work);
748 static int netlink_release(struct socket *sock)
750 struct sock *sk = sock->sk;
751 struct netlink_sock *nlk;
761 * OK. Socket is unlinked, any packets that arrive now
765 /* must not acquire netlink_table_lock in any way again before unbind
766 * and notifying genetlink is done as otherwise it might deadlock
768 if (nlk->netlink_unbind) {
771 for (i = 0; i < nlk->ngroups; i++)
772 if (test_bit(i, nlk->groups))
773 nlk->netlink_unbind(sock_net(sk), i + 1);
775 if (sk->sk_protocol == NETLINK_GENERIC &&
776 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
777 wake_up(&genl_sk_destructing_waitq);
780 wake_up_interruptible_all(&nlk->wait);
782 skb_queue_purge(&sk->sk_write_queue);
784 if (nlk->portid && nlk->bound) {
785 struct netlink_notify n = {
787 .protocol = sk->sk_protocol,
788 .portid = nlk->portid,
790 blocking_notifier_call_chain(&netlink_chain,
791 NETLINK_URELEASE, &n);
794 module_put(nlk->module);
796 if (netlink_is_kernel(sk)) {
797 netlink_table_grab();
798 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
799 if (--nl_table[sk->sk_protocol].registered == 0) {
800 struct listeners *old;
802 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
803 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
805 nl_table[sk->sk_protocol].module = NULL;
806 nl_table[sk->sk_protocol].bind = NULL;
807 nl_table[sk->sk_protocol].unbind = NULL;
808 nl_table[sk->sk_protocol].flags = 0;
809 nl_table[sk->sk_protocol].registered = 0;
811 netlink_table_ungrab();
815 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
817 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
821 static int netlink_autobind(struct socket *sock)
823 struct sock *sk = sock->sk;
824 struct net *net = sock_net(sk);
825 struct netlink_table *table = &nl_table[sk->sk_protocol];
826 s32 portid = task_tgid_vnr(current);
834 ok = !__netlink_lookup(table, portid, net);
837 /* Bind collision, search negative portid values. */
839 /* rover will be in range [S32_MIN, -4097] */
840 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
841 else if (rover >= -4096)
847 err = netlink_insert(sk, portid);
848 if (err == -EADDRINUSE)
851 /* If 2 threads race to autobind, that is fine. */
859 * __netlink_ns_capable - General netlink message capability test
860 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
861 * @user_ns: The user namespace of the capability to use
862 * @cap: The capability to use
864 * Test to see if the opener of the socket we received the message
865 * from had when the netlink socket was created and the sender of the
866 * message has has the capability @cap in the user namespace @user_ns.
868 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
869 struct user_namespace *user_ns, int cap)
871 return ((nsp->flags & NETLINK_SKB_DST) ||
872 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
873 ns_capable(user_ns, cap);
875 EXPORT_SYMBOL(__netlink_ns_capable);
878 * netlink_ns_capable - General netlink message capability test
879 * @skb: socket buffer holding a netlink command from userspace
880 * @user_ns: The user namespace of the capability to use
881 * @cap: The capability to use
883 * Test to see if the opener of the socket we received the message
884 * from had when the netlink socket was created and the sender of the
885 * message has has the capability @cap in the user namespace @user_ns.
887 bool netlink_ns_capable(const struct sk_buff *skb,
888 struct user_namespace *user_ns, int cap)
890 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
892 EXPORT_SYMBOL(netlink_ns_capable);
895 * netlink_capable - Netlink global message capability test
896 * @skb: socket buffer holding a netlink command from userspace
897 * @cap: The capability to use
899 * Test to see if the opener of the socket we received the message
900 * from had when the netlink socket was created and the sender of the
901 * message has has the capability @cap in all user namespaces.
903 bool netlink_capable(const struct sk_buff *skb, int cap)
905 return netlink_ns_capable(skb, &init_user_ns, cap);
907 EXPORT_SYMBOL(netlink_capable);
910 * netlink_net_capable - Netlink network namespace message capability test
911 * @skb: socket buffer holding a netlink command from userspace
912 * @cap: The capability to use
914 * Test to see if the opener of the socket we received the message
915 * from had when the netlink socket was created and the sender of the
916 * message has has the capability @cap over the network namespace of
917 * the socket we received the message from.
919 bool netlink_net_capable(const struct sk_buff *skb, int cap)
921 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
923 EXPORT_SYMBOL(netlink_net_capable);
925 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
927 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
928 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
932 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
934 struct netlink_sock *nlk = nlk_sk(sk);
936 if (nlk->subscriptions && !subscriptions)
937 __sk_del_bind_node(sk);
938 else if (!nlk->subscriptions && subscriptions)
939 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
940 nlk->subscriptions = subscriptions;
943 static int netlink_realloc_groups(struct sock *sk)
945 struct netlink_sock *nlk = nlk_sk(sk);
947 unsigned long *new_groups;
950 netlink_table_grab();
952 groups = nl_table[sk->sk_protocol].groups;
953 if (!nl_table[sk->sk_protocol].registered) {
958 if (nlk->ngroups >= groups)
961 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
962 if (new_groups == NULL) {
966 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
967 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
969 nlk->groups = new_groups;
970 nlk->ngroups = groups;
972 netlink_table_ungrab();
976 static void netlink_undo_bind(int group, long unsigned int groups,
979 struct netlink_sock *nlk = nlk_sk(sk);
982 if (!nlk->netlink_unbind)
985 for (undo = 0; undo < group; undo++)
986 if (test_bit(undo, &groups))
987 nlk->netlink_unbind(sock_net(sk), undo + 1);
990 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
993 struct sock *sk = sock->sk;
994 struct net *net = sock_net(sk);
995 struct netlink_sock *nlk = nlk_sk(sk);
996 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
998 long unsigned int groups = nladdr->nl_groups;
1001 if (addr_len < sizeof(struct sockaddr_nl))
1004 if (nladdr->nl_family != AF_NETLINK)
1007 /* Only superuser is allowed to listen multicasts */
1009 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1011 err = netlink_realloc_groups(sk);
1016 if (nlk->ngroups == 0)
1018 else if (nlk->ngroups < 8*sizeof(groups))
1019 groups &= (1UL << nlk->ngroups) - 1;
1023 /* Ensure nlk->portid is up-to-date. */
1026 if (nladdr->nl_pid != nlk->portid)
1030 netlink_lock_table();
1031 if (nlk->netlink_bind && groups) {
1034 /* nl_groups is a u32, so cap the maximum groups we can bind */
1035 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1036 if (!test_bit(group, &groups))
1038 err = nlk->netlink_bind(net, group + 1);
1041 netlink_undo_bind(group, groups, sk);
1046 /* No need for barriers here as we return to user-space without
1047 * using any of the bound attributes.
1050 err = nladdr->nl_pid ?
1051 netlink_insert(sk, nladdr->nl_pid) :
1052 netlink_autobind(sock);
1054 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1059 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1061 netlink_unlock_table();
1063 netlink_table_grab();
1064 netlink_update_subscriptions(sk, nlk->subscriptions +
1066 hweight32(nlk->groups[0]));
1067 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1068 netlink_update_listeners(sk);
1069 netlink_table_ungrab();
1074 netlink_unlock_table();
1078 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1079 int alen, int flags)
1082 struct sock *sk = sock->sk;
1083 struct netlink_sock *nlk = nlk_sk(sk);
1084 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1086 if (alen < sizeof(addr->sa_family))
1089 if (addr->sa_family == AF_UNSPEC) {
1090 sk->sk_state = NETLINK_UNCONNECTED;
1091 nlk->dst_portid = 0;
1095 if (addr->sa_family != AF_NETLINK)
1098 if (alen < sizeof(struct sockaddr_nl))
1101 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1102 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1105 /* No need for barriers here as we return to user-space without
1106 * using any of the bound attributes.
1109 err = netlink_autobind(sock);
1112 sk->sk_state = NETLINK_CONNECTED;
1113 nlk->dst_portid = nladdr->nl_pid;
1114 nlk->dst_group = ffs(nladdr->nl_groups);
1120 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1123 struct sock *sk = sock->sk;
1124 struct netlink_sock *nlk = nlk_sk(sk);
1125 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1127 nladdr->nl_family = AF_NETLINK;
1131 nladdr->nl_pid = nlk->dst_portid;
1132 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1134 nladdr->nl_pid = nlk->portid;
1135 netlink_lock_table();
1136 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1137 netlink_unlock_table();
1139 return sizeof(*nladdr);
1142 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1145 /* try to hand this ioctl down to the NIC drivers.
1147 return -ENOIOCTLCMD;
1150 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1153 struct netlink_sock *nlk;
1155 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1157 return ERR_PTR(-ECONNREFUSED);
1159 /* Don't bother queuing skb if kernel socket has no input function */
1161 if (sock->sk_state == NETLINK_CONNECTED &&
1162 nlk->dst_portid != nlk_sk(ssk)->portid) {
1164 return ERR_PTR(-ECONNREFUSED);
1169 struct sock *netlink_getsockbyfilp(struct file *filp)
1171 struct inode *inode = file_inode(filp);
1174 if (!S_ISSOCK(inode->i_mode))
1175 return ERR_PTR(-ENOTSOCK);
1177 sock = SOCKET_I(inode)->sk;
1178 if (sock->sk_family != AF_NETLINK)
1179 return ERR_PTR(-EINVAL);
1185 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1188 struct sk_buff *skb;
1191 if (size <= NLMSG_GOODSIZE || broadcast)
1192 return alloc_skb(size, GFP_KERNEL);
1194 size = SKB_DATA_ALIGN(size) +
1195 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1197 data = vmalloc(size);
1201 skb = __build_skb(data, size);
1205 skb->destructor = netlink_skb_destructor;
1211 * Attach a skb to a netlink socket.
1212 * The caller must hold a reference to the destination socket. On error, the
1213 * reference is dropped. The skb is not send to the destination, just all
1214 * all error checks are performed and memory in the queue is reserved.
1216 * < 0: error. skb freed, reference to sock dropped.
1218 * 1: repeat lookup - reference dropped while waiting for socket memory.
1220 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1221 long *timeo, struct sock *ssk)
1223 struct netlink_sock *nlk;
1227 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1228 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1229 DECLARE_WAITQUEUE(wait, current);
1231 if (!ssk || netlink_is_kernel(ssk))
1232 netlink_overrun(sk);
1238 __set_current_state(TASK_INTERRUPTIBLE);
1239 add_wait_queue(&nlk->wait, &wait);
1241 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1242 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1243 !sock_flag(sk, SOCK_DEAD))
1244 *timeo = schedule_timeout(*timeo);
1246 __set_current_state(TASK_RUNNING);
1247 remove_wait_queue(&nlk->wait, &wait);
1250 if (signal_pending(current)) {
1252 return sock_intr_errno(*timeo);
1256 netlink_skb_set_owner_r(skb, sk);
1260 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1264 netlink_deliver_tap(sock_net(sk), skb);
1266 skb_queue_tail(&sk->sk_receive_queue, skb);
1267 sk->sk_data_ready(sk);
1271 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1273 int len = __netlink_sendskb(sk, skb);
1279 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1285 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1289 WARN_ON(skb->sk != NULL);
1290 delta = skb->end - skb->tail;
1291 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1294 if (skb_shared(skb)) {
1295 struct sk_buff *nskb = skb_clone(skb, allocation);
1302 pskb_expand_head(skb, 0, -delta,
1303 (allocation & ~__GFP_DIRECT_RECLAIM) |
1304 __GFP_NOWARN | __GFP_NORETRY);
1308 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1312 struct netlink_sock *nlk = nlk_sk(sk);
1314 ret = -ECONNREFUSED;
1315 if (nlk->netlink_rcv != NULL) {
1317 netlink_skb_set_owner_r(skb, sk);
1318 NETLINK_CB(skb).sk = ssk;
1319 netlink_deliver_tap_kernel(sk, ssk, skb);
1320 nlk->netlink_rcv(skb);
1329 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1330 u32 portid, int nonblock)
1336 skb = netlink_trim(skb, gfp_any());
1338 timeo = sock_sndtimeo(ssk, nonblock);
1340 sk = netlink_getsockbyportid(ssk, portid);
1345 if (netlink_is_kernel(sk))
1346 return netlink_unicast_kernel(sk, skb, ssk);
1348 if (sk_filter(sk, skb)) {
1355 err = netlink_attachskb(sk, skb, &timeo, ssk);
1361 return netlink_sendskb(sk, skb);
1363 EXPORT_SYMBOL(netlink_unicast);
1365 int netlink_has_listeners(struct sock *sk, unsigned int group)
1368 struct listeners *listeners;
1370 BUG_ON(!netlink_is_kernel(sk));
1373 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1375 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1376 res = test_bit(group - 1, listeners->masks);
1382 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1384 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1386 struct netlink_sock *nlk = nlk_sk(sk);
1388 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1389 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1390 netlink_skb_set_owner_r(skb, sk);
1391 __netlink_sendskb(sk, skb);
1392 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1397 struct netlink_broadcast_data {
1398 struct sock *exclude_sk;
1403 int delivery_failure;
1407 struct sk_buff *skb, *skb2;
1408 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1412 static void do_one_broadcast(struct sock *sk,
1413 struct netlink_broadcast_data *p)
1415 struct netlink_sock *nlk = nlk_sk(sk);
1418 if (p->exclude_sk == sk)
1421 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1422 !test_bit(p->group - 1, nlk->groups))
1425 if (!net_eq(sock_net(sk), p->net)) {
1426 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1429 if (!peernet_has_id(sock_net(sk), p->net))
1432 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1438 netlink_overrun(sk);
1443 if (p->skb2 == NULL) {
1444 if (skb_shared(p->skb)) {
1445 p->skb2 = skb_clone(p->skb, p->allocation);
1447 p->skb2 = skb_get(p->skb);
1449 * skb ownership may have been set when
1450 * delivered to a previous socket.
1452 skb_orphan(p->skb2);
1455 if (p->skb2 == NULL) {
1456 netlink_overrun(sk);
1457 /* Clone failed. Notify ALL listeners. */
1459 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1460 p->delivery_failure = 1;
1463 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1468 if (sk_filter(sk, p->skb2)) {
1473 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1474 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1475 NETLINK_CB(p->skb2).nsid_is_set = true;
1476 val = netlink_broadcast_deliver(sk, p->skb2);
1478 netlink_overrun(sk);
1479 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1480 p->delivery_failure = 1;
1482 p->congested |= val;
1490 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1491 u32 group, gfp_t allocation,
1492 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1495 struct net *net = sock_net(ssk);
1496 struct netlink_broadcast_data info;
1499 skb = netlink_trim(skb, allocation);
1501 info.exclude_sk = ssk;
1503 info.portid = portid;
1506 info.delivery_failure = 0;
1509 info.allocation = allocation;
1512 info.tx_filter = filter;
1513 info.tx_data = filter_data;
1515 /* While we sleep in clone, do not allow to change socket list */
1517 netlink_lock_table();
1519 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1520 do_one_broadcast(sk, &info);
1524 netlink_unlock_table();
1526 if (info.delivery_failure) {
1527 kfree_skb(info.skb2);
1530 consume_skb(info.skb2);
1532 if (info.delivered) {
1533 if (info.congested && gfpflags_allow_blocking(allocation))
1539 EXPORT_SYMBOL(netlink_broadcast_filtered);
1541 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1542 u32 group, gfp_t allocation)
1544 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1547 EXPORT_SYMBOL(netlink_broadcast);
1549 struct netlink_set_err_data {
1550 struct sock *exclude_sk;
1556 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1558 struct netlink_sock *nlk = nlk_sk(sk);
1561 if (sk == p->exclude_sk)
1564 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1567 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1568 !test_bit(p->group - 1, nlk->groups))
1571 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1576 sk->sk_err = p->code;
1577 sk->sk_error_report(sk);
1583 * netlink_set_err - report error to broadcast listeners
1584 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1585 * @portid: the PORTID of a process that we want to skip (if any)
1586 * @group: the broadcast group that will notice the error
1587 * @code: error code, must be negative (as usual in kernelspace)
1589 * This function returns the number of broadcast listeners that have set the
1590 * NETLINK_NO_ENOBUFS socket option.
1592 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1594 struct netlink_set_err_data info;
1598 info.exclude_sk = ssk;
1599 info.portid = portid;
1601 /* sk->sk_err wants a positive error value */
1604 read_lock(&nl_table_lock);
1606 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1607 ret += do_one_set_err(sk, &info);
1609 read_unlock(&nl_table_lock);
1612 EXPORT_SYMBOL(netlink_set_err);
1614 /* must be called with netlink table grabbed */
1615 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1619 int old, new = !!is_new, subscriptions;
1621 old = test_bit(group - 1, nlk->groups);
1622 subscriptions = nlk->subscriptions - old + new;
1624 __set_bit(group - 1, nlk->groups);
1626 __clear_bit(group - 1, nlk->groups);
1627 netlink_update_subscriptions(&nlk->sk, subscriptions);
1628 netlink_update_listeners(&nlk->sk);
1631 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1632 char __user *optval, unsigned int optlen)
1634 struct sock *sk = sock->sk;
1635 struct netlink_sock *nlk = nlk_sk(sk);
1636 unsigned int val = 0;
1639 if (level != SOL_NETLINK)
1640 return -ENOPROTOOPT;
1642 if (optlen >= sizeof(int) &&
1643 get_user(val, (unsigned int __user *)optval))
1647 case NETLINK_PKTINFO:
1649 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1651 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1654 case NETLINK_ADD_MEMBERSHIP:
1655 case NETLINK_DROP_MEMBERSHIP: {
1656 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1658 err = netlink_realloc_groups(sk);
1661 if (!val || val - 1 >= nlk->ngroups)
1663 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1664 err = nlk->netlink_bind(sock_net(sk), val);
1668 netlink_table_grab();
1669 netlink_update_socket_mc(nlk, val,
1670 optname == NETLINK_ADD_MEMBERSHIP);
1671 netlink_table_ungrab();
1672 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1673 nlk->netlink_unbind(sock_net(sk), val);
1678 case NETLINK_BROADCAST_ERROR:
1680 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1682 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1685 case NETLINK_NO_ENOBUFS:
1687 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1688 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1689 wake_up_interruptible(&nlk->wait);
1691 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1695 case NETLINK_LISTEN_ALL_NSID:
1696 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1700 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1702 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1705 case NETLINK_CAP_ACK:
1707 nlk->flags |= NETLINK_F_CAP_ACK;
1709 nlk->flags &= ~NETLINK_F_CAP_ACK;
1712 case NETLINK_EXT_ACK:
1714 nlk->flags |= NETLINK_F_EXT_ACK;
1716 nlk->flags &= ~NETLINK_F_EXT_ACK;
1725 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1726 char __user *optval, int __user *optlen)
1728 struct sock *sk = sock->sk;
1729 struct netlink_sock *nlk = nlk_sk(sk);
1732 if (level != SOL_NETLINK)
1733 return -ENOPROTOOPT;
1735 if (get_user(len, optlen))
1741 case NETLINK_PKTINFO:
1742 if (len < sizeof(int))
1745 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1746 if (put_user(len, optlen) ||
1747 put_user(val, optval))
1751 case NETLINK_BROADCAST_ERROR:
1752 if (len < sizeof(int))
1755 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1756 if (put_user(len, optlen) ||
1757 put_user(val, optval))
1761 case NETLINK_NO_ENOBUFS:
1762 if (len < sizeof(int))
1765 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1766 if (put_user(len, optlen) ||
1767 put_user(val, optval))
1771 case NETLINK_LIST_MEMBERSHIPS: {
1772 int pos, idx, shift;
1775 netlink_lock_table();
1776 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1777 if (len - pos < sizeof(u32))
1780 idx = pos / sizeof(unsigned long);
1781 shift = (pos % sizeof(unsigned long)) * 8;
1782 if (put_user((u32)(nlk->groups[idx] >> shift),
1783 (u32 __user *)(optval + pos))) {
1788 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1790 netlink_unlock_table();
1793 case NETLINK_CAP_ACK:
1794 if (len < sizeof(int))
1797 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1798 if (put_user(len, optlen) ||
1799 put_user(val, optval))
1803 case NETLINK_EXT_ACK:
1804 if (len < sizeof(int))
1807 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1808 if (put_user(len, optlen) || put_user(val, optval))
1818 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1820 struct nl_pktinfo info;
1822 info.group = NETLINK_CB(skb).dst_group;
1823 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1826 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1827 struct sk_buff *skb)
1829 if (!NETLINK_CB(skb).nsid_is_set)
1832 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1833 &NETLINK_CB(skb).nsid);
1836 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1838 struct sock *sk = sock->sk;
1839 struct netlink_sock *nlk = nlk_sk(sk);
1840 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1843 struct sk_buff *skb;
1845 struct scm_cookie scm;
1846 u32 netlink_skb_flags = 0;
1848 if (msg->msg_flags&MSG_OOB)
1851 err = scm_send(sock, msg, &scm, true);
1855 if (msg->msg_namelen) {
1857 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1859 if (addr->nl_family != AF_NETLINK)
1861 dst_portid = addr->nl_pid;
1862 dst_group = ffs(addr->nl_groups);
1864 if ((dst_group || dst_portid) &&
1865 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1867 netlink_skb_flags |= NETLINK_SKB_DST;
1869 dst_portid = nlk->dst_portid;
1870 dst_group = nlk->dst_group;
1874 err = netlink_autobind(sock);
1878 /* Ensure nlk is hashed and visible. */
1883 if (len > sk->sk_sndbuf - 32)
1886 skb = netlink_alloc_large_skb(len, dst_group);
1890 NETLINK_CB(skb).portid = nlk->portid;
1891 NETLINK_CB(skb).dst_group = dst_group;
1892 NETLINK_CB(skb).creds = scm.creds;
1893 NETLINK_CB(skb).flags = netlink_skb_flags;
1896 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1901 err = security_netlink_send(sk, skb);
1908 refcount_inc(&skb->users);
1909 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1911 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1918 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1921 struct scm_cookie scm;
1922 struct sock *sk = sock->sk;
1923 struct netlink_sock *nlk = nlk_sk(sk);
1924 int noblock = flags&MSG_DONTWAIT;
1926 struct sk_buff *skb, *data_skb;
1934 skb = skb_recv_datagram(sk, flags, noblock, &err);
1940 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1941 if (unlikely(skb_shinfo(skb)->frag_list)) {
1943 * If this skb has a frag_list, then here that means that we
1944 * will have to use the frag_list skb's data for compat tasks
1945 * and the regular skb's data for normal (non-compat) tasks.
1947 * If we need to send the compat skb, assign it to the
1948 * 'data_skb' variable so that it will be used below for data
1949 * copying. We keep 'skb' for everything else, including
1950 * freeing both later.
1952 if (flags & MSG_CMSG_COMPAT)
1953 data_skb = skb_shinfo(skb)->frag_list;
1957 /* Record the max length of recvmsg() calls for future allocations */
1958 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1959 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1960 SKB_WITH_OVERHEAD(32768));
1962 copied = data_skb->len;
1964 msg->msg_flags |= MSG_TRUNC;
1968 skb_reset_transport_header(data_skb);
1969 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1971 if (msg->msg_name) {
1972 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1973 addr->nl_family = AF_NETLINK;
1975 addr->nl_pid = NETLINK_CB(skb).portid;
1976 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1977 msg->msg_namelen = sizeof(*addr);
1980 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1981 netlink_cmsg_recv_pktinfo(msg, skb);
1982 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1983 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1985 memset(&scm, 0, sizeof(scm));
1986 scm.creds = *NETLINK_CREDS(skb);
1987 if (flags & MSG_TRUNC)
1988 copied = data_skb->len;
1990 skb_free_datagram(sk, skb);
1992 if (nlk->cb_running &&
1993 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1994 ret = netlink_dump(sk);
1997 sk->sk_error_report(sk);
2001 scm_recv(sock, msg, &scm, flags);
2003 netlink_rcv_wake(sk);
2004 return err ? : copied;
2007 static void netlink_data_ready(struct sock *sk)
2013 * We export these functions to other modules. They provide a
2014 * complete set of kernel non-blocking support for message
2019 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2020 struct netlink_kernel_cfg *cfg)
2022 struct socket *sock;
2024 struct netlink_sock *nlk;
2025 struct listeners *listeners = NULL;
2026 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2027 unsigned int groups;
2031 if (unit < 0 || unit >= MAX_LINKS)
2034 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2037 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2038 goto out_sock_release_nosk;
2042 if (!cfg || cfg->groups < 32)
2045 groups = cfg->groups;
2047 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2049 goto out_sock_release;
2051 sk->sk_data_ready = netlink_data_ready;
2052 if (cfg && cfg->input)
2053 nlk_sk(sk)->netlink_rcv = cfg->input;
2055 if (netlink_insert(sk, 0))
2056 goto out_sock_release;
2059 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2061 netlink_table_grab();
2062 if (!nl_table[unit].registered) {
2063 nl_table[unit].groups = groups;
2064 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2065 nl_table[unit].cb_mutex = cb_mutex;
2066 nl_table[unit].module = module;
2068 nl_table[unit].bind = cfg->bind;
2069 nl_table[unit].unbind = cfg->unbind;
2070 nl_table[unit].flags = cfg->flags;
2072 nl_table[unit].compare = cfg->compare;
2074 nl_table[unit].registered = 1;
2077 nl_table[unit].registered++;
2079 netlink_table_ungrab();
2084 netlink_kernel_release(sk);
2087 out_sock_release_nosk:
2091 EXPORT_SYMBOL(__netlink_kernel_create);
2094 netlink_kernel_release(struct sock *sk)
2096 if (sk == NULL || sk->sk_socket == NULL)
2099 sock_release(sk->sk_socket);
2101 EXPORT_SYMBOL(netlink_kernel_release);
2103 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2105 struct listeners *new, *old;
2106 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2111 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2112 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2115 old = nl_deref_protected(tbl->listeners);
2116 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2117 rcu_assign_pointer(tbl->listeners, new);
2119 kfree_rcu(old, rcu);
2121 tbl->groups = groups;
2127 * netlink_change_ngroups - change number of multicast groups
2129 * This changes the number of multicast groups that are available
2130 * on a certain netlink family. Note that it is not possible to
2131 * change the number of groups to below 32. Also note that it does
2132 * not implicitly call netlink_clear_multicast_users() when the
2133 * number of groups is reduced.
2135 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2136 * @groups: The new number of groups.
2138 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2142 netlink_table_grab();
2143 err = __netlink_change_ngroups(sk, groups);
2144 netlink_table_ungrab();
2149 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2152 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2154 sk_for_each_bound(sk, &tbl->mc_list)
2155 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2159 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2161 struct nlmsghdr *nlh;
2162 int size = nlmsg_msg_size(len);
2164 nlh = skb_put(skb, NLMSG_ALIGN(size));
2165 nlh->nlmsg_type = type;
2166 nlh->nlmsg_len = size;
2167 nlh->nlmsg_flags = flags;
2168 nlh->nlmsg_pid = portid;
2169 nlh->nlmsg_seq = seq;
2170 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2171 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2174 EXPORT_SYMBOL(__nlmsg_put);
2177 * It looks a bit ugly.
2178 * It would be better to create kernel thread.
2181 static int netlink_dump(struct sock *sk)
2183 struct netlink_sock *nlk = nlk_sk(sk);
2184 struct netlink_callback *cb;
2185 struct sk_buff *skb = NULL;
2186 struct nlmsghdr *nlh;
2187 struct module *module;
2192 mutex_lock(nlk->cb_mutex);
2193 if (!nlk->cb_running) {
2198 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2201 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2202 * required, but it makes sense to _attempt_ a 16K bytes allocation
2203 * to reduce number of system calls on dump operations, if user
2204 * ever provided a big enough buffer.
2207 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2209 if (alloc_min_size < nlk->max_recvmsg_len) {
2210 alloc_size = nlk->max_recvmsg_len;
2211 skb = alloc_skb(alloc_size,
2212 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2213 __GFP_NOWARN | __GFP_NORETRY);
2216 alloc_size = alloc_min_size;
2217 skb = alloc_skb(alloc_size, GFP_KERNEL);
2222 /* Trim skb to allocated size. User is expected to provide buffer as
2223 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2224 * netlink_recvmsg())). dump will pack as many smaller messages as
2225 * could fit within the allocated skb. skb is typically allocated
2226 * with larger space than required (could be as much as near 2x the
2227 * requested size with align to next power of 2 approach). Allowing
2228 * dump to use the excess space makes it difficult for a user to have a
2229 * reasonable static buffer based on the expected largest dump of a
2230 * single netdev. The outcome is MSG_TRUNC error.
2232 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2233 netlink_skb_set_owner_r(skb, sk);
2235 if (nlk->dump_done_errno > 0)
2236 nlk->dump_done_errno = cb->dump(skb, cb);
2238 if (nlk->dump_done_errno > 0 ||
2239 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2240 mutex_unlock(nlk->cb_mutex);
2242 if (sk_filter(sk, skb))
2245 __netlink_sendskb(sk, skb);
2249 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE,
2250 sizeof(nlk->dump_done_errno), NLM_F_MULTI);
2254 nl_dump_check_consistent(cb, nlh);
2256 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno,
2257 sizeof(nlk->dump_done_errno));
2259 if (sk_filter(sk, skb))
2262 __netlink_sendskb(sk, skb);
2267 nlk->cb_running = false;
2268 module = cb->module;
2270 mutex_unlock(nlk->cb_mutex);
2276 mutex_unlock(nlk->cb_mutex);
2281 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2282 const struct nlmsghdr *nlh,
2283 struct netlink_dump_control *control)
2285 struct netlink_callback *cb;
2287 struct netlink_sock *nlk;
2290 refcount_inc(&skb->users);
2292 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2294 ret = -ECONNREFUSED;
2299 mutex_lock(nlk->cb_mutex);
2300 /* A dump is in progress... */
2301 if (nlk->cb_running) {
2305 /* add reference of module which cb->dump belongs to */
2306 if (!try_module_get(control->module)) {
2307 ret = -EPROTONOSUPPORT;
2312 memset(cb, 0, sizeof(*cb));
2313 cb->dump = control->dump;
2314 cb->done = control->done;
2316 cb->data = control->data;
2317 cb->module = control->module;
2318 cb->min_dump_alloc = control->min_dump_alloc;
2321 if (control->start) {
2322 ret = control->start(cb);
2327 nlk->cb_running = true;
2328 nlk->dump_done_errno = INT_MAX;
2330 mutex_unlock(nlk->cb_mutex);
2332 ret = netlink_dump(sk);
2339 /* We successfully started a dump, by returning -EINTR we
2340 * signal not to send ACK even if it was requested.
2345 module_put(control->module);
2348 mutex_unlock(nlk->cb_mutex);
2353 EXPORT_SYMBOL(__netlink_dump_start);
2355 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2356 const struct netlink_ext_ack *extack)
2358 struct sk_buff *skb;
2359 struct nlmsghdr *rep;
2360 struct nlmsgerr *errmsg;
2361 size_t payload = sizeof(*errmsg);
2363 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2364 unsigned int flags = 0;
2365 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2367 /* Error messages get the original request appened, unless the user
2368 * requests to cap the error message, and get extra error data if
2371 if (nlk_has_extack && extack && extack->_msg)
2372 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2375 if (!(nlk->flags & NETLINK_F_CAP_ACK))
2376 payload += nlmsg_len(nlh);
2378 flags |= NLM_F_CAPPED;
2379 if (nlk_has_extack && extack && extack->bad_attr)
2380 tlvlen += nla_total_size(sizeof(u32));
2382 flags |= NLM_F_CAPPED;
2384 if (nlk_has_extack && extack && extack->cookie_len)
2385 tlvlen += nla_total_size(extack->cookie_len);
2389 flags |= NLM_F_ACK_TLVS;
2391 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2393 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2394 NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk);
2398 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2399 NLMSG_ERROR, payload, flags);
2400 errmsg = nlmsg_data(rep);
2401 errmsg->error = err;
2402 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2404 if (nlk_has_extack && extack) {
2406 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2410 if (extack->bad_attr &&
2411 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2412 (u8 *)extack->bad_attr >= in_skb->data +
2414 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2415 (u8 *)extack->bad_attr -
2418 if (extack->cookie_len)
2419 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2425 nlmsg_end(skb, rep);
2427 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2429 EXPORT_SYMBOL(netlink_ack);
2431 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2433 struct netlink_ext_ack *))
2435 struct netlink_ext_ack extack;
2436 struct nlmsghdr *nlh;
2439 while (skb->len >= nlmsg_total_size(0)) {
2442 memset(&extack, 0, sizeof(extack));
2443 nlh = nlmsg_hdr(skb);
2446 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2449 /* Only requests are handled by the kernel */
2450 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2453 /* Skip control messages */
2454 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2457 err = cb(skb, nlh, &extack);
2462 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2463 netlink_ack(skb, nlh, err, &extack);
2466 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2467 if (msglen > skb->len)
2469 skb_pull(skb, msglen);
2474 EXPORT_SYMBOL(netlink_rcv_skb);
2477 * nlmsg_notify - send a notification netlink message
2478 * @sk: netlink socket to use
2479 * @skb: notification message
2480 * @portid: destination netlink portid for reports or 0
2481 * @group: destination multicast group or 0
2482 * @report: 1 to report back, 0 to disable
2483 * @flags: allocation flags
2485 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2486 unsigned int group, int report, gfp_t flags)
2491 int exclude_portid = 0;
2494 refcount_inc(&skb->users);
2495 exclude_portid = portid;
2498 /* errors reported via destination sk->sk_err, but propagate
2499 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2500 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2508 err2 = nlmsg_unicast(sk, skb, portid);
2515 EXPORT_SYMBOL(nlmsg_notify);
2517 #ifdef CONFIG_PROC_FS
2518 struct nl_seq_iter {
2519 struct seq_net_private p;
2520 struct rhashtable_iter hti;
2524 static int netlink_walk_start(struct nl_seq_iter *iter)
2528 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti,
2531 iter->link = MAX_LINKS;
2535 rhashtable_walk_start(&iter->hti);
2540 static void netlink_walk_stop(struct nl_seq_iter *iter)
2542 rhashtable_walk_stop(&iter->hti);
2543 rhashtable_walk_exit(&iter->hti);
2546 static void *__netlink_seq_next(struct seq_file *seq)
2548 struct nl_seq_iter *iter = seq->private;
2549 struct netlink_sock *nlk;
2555 nlk = rhashtable_walk_next(&iter->hti);
2558 if (PTR_ERR(nlk) == -EAGAIN)
2567 netlink_walk_stop(iter);
2568 if (++iter->link >= MAX_LINKS)
2571 err = netlink_walk_start(iter);
2573 return ERR_PTR(err);
2575 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2580 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2582 struct nl_seq_iter *iter = seq->private;
2583 void *obj = SEQ_START_TOKEN;
2589 err = netlink_walk_start(iter);
2591 return ERR_PTR(err);
2593 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2594 obj = __netlink_seq_next(seq);
2599 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2602 return __netlink_seq_next(seq);
2605 static void netlink_seq_stop(struct seq_file *seq, void *v)
2607 struct nl_seq_iter *iter = seq->private;
2609 if (iter->link >= MAX_LINKS)
2612 netlink_walk_stop(iter);
2616 static int netlink_seq_show(struct seq_file *seq, void *v)
2618 if (v == SEQ_START_TOKEN) {
2620 "sk Eth Pid Groups "
2621 "Rmem Wmem Dump Locks Drops Inode\n");
2624 struct netlink_sock *nlk = nlk_sk(s);
2626 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8d %-8lu\n",
2630 nlk->groups ? (u32)nlk->groups[0] : 0,
2631 sk_rmem_alloc_get(s),
2632 sk_wmem_alloc_get(s),
2634 refcount_read(&s->sk_refcnt),
2635 atomic_read(&s->sk_drops),
2643 static const struct seq_operations netlink_seq_ops = {
2644 .start = netlink_seq_start,
2645 .next = netlink_seq_next,
2646 .stop = netlink_seq_stop,
2647 .show = netlink_seq_show,
2651 int netlink_register_notifier(struct notifier_block *nb)
2653 return blocking_notifier_chain_register(&netlink_chain, nb);
2655 EXPORT_SYMBOL(netlink_register_notifier);
2657 int netlink_unregister_notifier(struct notifier_block *nb)
2659 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2661 EXPORT_SYMBOL(netlink_unregister_notifier);
2663 static const struct proto_ops netlink_ops = {
2664 .family = PF_NETLINK,
2665 .owner = THIS_MODULE,
2666 .release = netlink_release,
2667 .bind = netlink_bind,
2668 .connect = netlink_connect,
2669 .socketpair = sock_no_socketpair,
2670 .accept = sock_no_accept,
2671 .getname = netlink_getname,
2672 .poll = datagram_poll,
2673 .ioctl = netlink_ioctl,
2674 .listen = sock_no_listen,
2675 .shutdown = sock_no_shutdown,
2676 .setsockopt = netlink_setsockopt,
2677 .getsockopt = netlink_getsockopt,
2678 .sendmsg = netlink_sendmsg,
2679 .recvmsg = netlink_recvmsg,
2680 .mmap = sock_no_mmap,
2681 .sendpage = sock_no_sendpage,
2684 static const struct net_proto_family netlink_family_ops = {
2685 .family = PF_NETLINK,
2686 .create = netlink_create,
2687 .owner = THIS_MODULE, /* for consistency 8) */
2690 static int __net_init netlink_net_init(struct net *net)
2692 #ifdef CONFIG_PROC_FS
2693 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2694 sizeof(struct nl_seq_iter)))
2700 static void __net_exit netlink_net_exit(struct net *net)
2702 #ifdef CONFIG_PROC_FS
2703 remove_proc_entry("netlink", net->proc_net);
2707 static void __init netlink_add_usersock_entry(void)
2709 struct listeners *listeners;
2712 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2714 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2716 netlink_table_grab();
2718 nl_table[NETLINK_USERSOCK].groups = groups;
2719 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2720 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2721 nl_table[NETLINK_USERSOCK].registered = 1;
2722 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2724 netlink_table_ungrab();
2727 static struct pernet_operations __net_initdata netlink_net_ops = {
2728 .init = netlink_net_init,
2729 .exit = netlink_net_exit,
2732 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2734 const struct netlink_sock *nlk = data;
2735 struct netlink_compare_arg arg;
2737 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2738 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2741 static const struct rhashtable_params netlink_rhashtable_params = {
2742 .head_offset = offsetof(struct netlink_sock, node),
2743 .key_len = netlink_compare_arg_len,
2744 .obj_hashfn = netlink_hash,
2745 .obj_cmpfn = netlink_compare,
2746 .automatic_shrinking = true,
2749 static int __init netlink_proto_init(void)
2752 int err = proto_register(&netlink_proto, 0);
2757 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2759 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2763 for (i = 0; i < MAX_LINKS; i++) {
2764 if (rhashtable_init(&nl_table[i].hash,
2765 &netlink_rhashtable_params) < 0) {
2767 rhashtable_destroy(&nl_table[i].hash);
2773 netlink_add_usersock_entry();
2775 sock_register(&netlink_family_ops);
2776 register_pernet_subsys(&netlink_net_ops);
2777 register_pernet_subsys(&netlink_tap_net_ops);
2778 /* The netlink device handler may be needed early. */
2783 panic("netlink_init: Cannot allocate nl_table\n");
2786 core_initcall(netlink_proto_init);