1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * NETLINK Kernel-user communication protocol.
5 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
6 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
7 * Patrick McHardy <kaber@trash.net>
9 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
10 * added netlink_proto_exit
11 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
12 * use nlk_sk, as sk->protinfo is on a diet 8)
13 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
14 * - inc module use count of module that owns
15 * the kernel socket in case userspace opens
16 * socket of same protocol
17 * - remove all module support, since netlink is
18 * mandatory if CONFIG_NET=y these days
21 #include <linux/module.h>
23 #include <linux/capability.h>
24 #include <linux/kernel.h>
25 #include <linux/init.h>
26 #include <linux/signal.h>
27 #include <linux/sched.h>
28 #include <linux/errno.h>
29 #include <linux/string.h>
30 #include <linux/stat.h>
31 #include <linux/socket.h>
33 #include <linux/fcntl.h>
34 #include <linux/termios.h>
35 #include <linux/sockios.h>
36 #include <linux/net.h>
38 #include <linux/slab.h>
39 #include <linux/uaccess.h>
40 #include <linux/skbuff.h>
41 #include <linux/netdevice.h>
42 #include <linux/rtnetlink.h>
43 #include <linux/proc_fs.h>
44 #include <linux/seq_file.h>
45 #include <linux/notifier.h>
46 #include <linux/security.h>
47 #include <linux/jhash.h>
48 #include <linux/jiffies.h>
49 #include <linux/random.h>
50 #include <linux/bitops.h>
52 #include <linux/types.h>
53 #include <linux/audit.h>
54 #include <linux/mutex.h>
55 #include <linux/vmalloc.h>
56 #include <linux/if_arp.h>
57 #include <linux/rhashtable.h>
58 #include <asm/cacheflush.h>
59 #include <linux/hash.h>
60 #include <linux/genetlink.h>
61 #include <linux/net_namespace.h>
62 #include <linux/nospec.h>
64 #include <net/net_namespace.h>
65 #include <net/netns/generic.h>
68 #include <net/netlink.h>
70 #include "af_netlink.h"
74 unsigned long masks[0];
78 #define NETLINK_S_CONGESTED 0x0
80 static inline int netlink_is_kernel(struct sock *sk)
82 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
85 struct netlink_table *nl_table __read_mostly;
86 EXPORT_SYMBOL_GPL(nl_table);
88 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
90 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
92 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
95 "nlk_cb_mutex-USERSOCK",
96 "nlk_cb_mutex-FIREWALL",
97 "nlk_cb_mutex-SOCK_DIAG",
100 "nlk_cb_mutex-SELINUX",
101 "nlk_cb_mutex-ISCSI",
102 "nlk_cb_mutex-AUDIT",
103 "nlk_cb_mutex-FIB_LOOKUP",
104 "nlk_cb_mutex-CONNECTOR",
105 "nlk_cb_mutex-NETFILTER",
106 "nlk_cb_mutex-IP6_FW",
107 "nlk_cb_mutex-DNRTMSG",
108 "nlk_cb_mutex-KOBJECT_UEVENT",
109 "nlk_cb_mutex-GENERIC",
111 "nlk_cb_mutex-SCSITRANSPORT",
112 "nlk_cb_mutex-ECRYPTFS",
114 "nlk_cb_mutex-CRYPTO",
125 "nlk_cb_mutex-MAX_LINKS"
128 static int netlink_dump(struct sock *sk);
130 /* nl_table locking explained:
131 * Lookup and traversal are protected with an RCU read-side lock. Insertion
132 * and removal are protected with per bucket lock while using RCU list
133 * modification primitives and may run in parallel to RCU protected lookups.
134 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
135 * been acquired * either during or after the socket has been removed from
136 * the list and after an RCU grace period.
138 DEFINE_RWLOCK(nl_table_lock);
139 EXPORT_SYMBOL_GPL(nl_table_lock);
140 static atomic_t nl_table_users = ATOMIC_INIT(0);
142 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
144 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
147 static const struct rhashtable_params netlink_rhashtable_params;
149 static inline u32 netlink_group_mask(u32 group)
151 return group ? 1 << (group - 1) : 0;
154 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
157 unsigned int len = skb_end_offset(skb);
160 new = alloc_skb(len, gfp_mask);
164 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
165 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
166 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
168 skb_put_data(new, skb->data, len);
172 static unsigned int netlink_tap_net_id;
174 struct netlink_tap_net {
175 struct list_head netlink_tap_all;
176 struct mutex netlink_tap_lock;
179 int netlink_add_tap(struct netlink_tap *nt)
181 struct net *net = dev_net(nt->dev);
182 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
184 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
187 mutex_lock(&nn->netlink_tap_lock);
188 list_add_rcu(&nt->list, &nn->netlink_tap_all);
189 mutex_unlock(&nn->netlink_tap_lock);
191 __module_get(nt->module);
195 EXPORT_SYMBOL_GPL(netlink_add_tap);
197 static int __netlink_remove_tap(struct netlink_tap *nt)
199 struct net *net = dev_net(nt->dev);
200 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
202 struct netlink_tap *tmp;
204 mutex_lock(&nn->netlink_tap_lock);
206 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
208 list_del_rcu(&nt->list);
214 pr_warn("__netlink_remove_tap: %p not found\n", nt);
216 mutex_unlock(&nn->netlink_tap_lock);
219 module_put(nt->module);
221 return found ? 0 : -ENODEV;
224 int netlink_remove_tap(struct netlink_tap *nt)
228 ret = __netlink_remove_tap(nt);
233 EXPORT_SYMBOL_GPL(netlink_remove_tap);
235 static __net_init int netlink_tap_init_net(struct net *net)
237 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
239 INIT_LIST_HEAD(&nn->netlink_tap_all);
240 mutex_init(&nn->netlink_tap_lock);
244 static struct pernet_operations netlink_tap_net_ops = {
245 .init = netlink_tap_init_net,
246 .id = &netlink_tap_net_id,
247 .size = sizeof(struct netlink_tap_net),
250 static bool netlink_filter_tap(const struct sk_buff *skb)
252 struct sock *sk = skb->sk;
254 /* We take the more conservative approach and
255 * whitelist socket protocols that may pass.
257 switch (sk->sk_protocol) {
259 case NETLINK_USERSOCK:
260 case NETLINK_SOCK_DIAG:
263 case NETLINK_FIB_LOOKUP:
264 case NETLINK_NETFILTER:
265 case NETLINK_GENERIC:
272 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
273 struct net_device *dev)
275 struct sk_buff *nskb;
276 struct sock *sk = skb->sk;
279 if (!net_eq(dev_net(dev), sock_net(sk)))
284 if (is_vmalloc_addr(skb->head))
285 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
287 nskb = skb_clone(skb, GFP_ATOMIC);
290 nskb->protocol = htons((u16) sk->sk_protocol);
291 nskb->pkt_type = netlink_is_kernel(sk) ?
292 PACKET_KERNEL : PACKET_USER;
293 skb_reset_network_header(nskb);
294 ret = dev_queue_xmit(nskb);
295 if (unlikely(ret > 0))
296 ret = net_xmit_errno(ret);
303 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
306 struct netlink_tap *tmp;
308 if (!netlink_filter_tap(skb))
311 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
312 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
318 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
320 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
324 if (unlikely(!list_empty(&nn->netlink_tap_all)))
325 __netlink_deliver_tap(skb, nn);
330 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
333 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
334 netlink_deliver_tap(sock_net(dst), skb);
337 static void netlink_overrun(struct sock *sk)
339 struct netlink_sock *nlk = nlk_sk(sk);
341 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
342 if (!test_and_set_bit(NETLINK_S_CONGESTED,
343 &nlk_sk(sk)->state)) {
344 sk->sk_err = ENOBUFS;
345 sk->sk_error_report(sk);
348 atomic_inc(&sk->sk_drops);
351 static void netlink_rcv_wake(struct sock *sk)
353 struct netlink_sock *nlk = nlk_sk(sk);
355 if (skb_queue_empty(&sk->sk_receive_queue))
356 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
357 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
358 wake_up_interruptible(&nlk->wait);
361 static void netlink_skb_destructor(struct sk_buff *skb)
363 if (is_vmalloc_addr(skb->head)) {
365 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
374 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
376 WARN_ON(skb->sk != NULL);
378 skb->destructor = netlink_skb_destructor;
379 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
380 sk_mem_charge(sk, skb->truesize);
383 static void netlink_sock_destruct(struct sock *sk)
385 struct netlink_sock *nlk = nlk_sk(sk);
387 if (nlk->cb_running) {
389 nlk->cb.done(&nlk->cb);
390 module_put(nlk->cb.module);
391 kfree_skb(nlk->cb.skb);
394 skb_queue_purge(&sk->sk_receive_queue);
396 if (!sock_flag(sk, SOCK_DEAD)) {
397 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
401 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
402 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
403 WARN_ON(nlk_sk(sk)->groups);
406 static void netlink_sock_destruct_work(struct work_struct *work)
408 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
414 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
415 * SMP. Look, when several writers sleep and reader wakes them up, all but one
416 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
417 * this, _but_ remember, it adds useless work on UP machines.
420 void netlink_table_grab(void)
421 __acquires(nl_table_lock)
425 write_lock_irq(&nl_table_lock);
427 if (atomic_read(&nl_table_users)) {
428 DECLARE_WAITQUEUE(wait, current);
430 add_wait_queue_exclusive(&nl_table_wait, &wait);
432 set_current_state(TASK_UNINTERRUPTIBLE);
433 if (atomic_read(&nl_table_users) == 0)
435 write_unlock_irq(&nl_table_lock);
437 write_lock_irq(&nl_table_lock);
440 __set_current_state(TASK_RUNNING);
441 remove_wait_queue(&nl_table_wait, &wait);
445 void netlink_table_ungrab(void)
446 __releases(nl_table_lock)
448 write_unlock_irq(&nl_table_lock);
449 wake_up(&nl_table_wait);
453 netlink_lock_table(void)
455 /* read_lock() synchronizes us to netlink_table_grab */
457 read_lock(&nl_table_lock);
458 atomic_inc(&nl_table_users);
459 read_unlock(&nl_table_lock);
463 netlink_unlock_table(void)
465 if (atomic_dec_and_test(&nl_table_users))
466 wake_up(&nl_table_wait);
469 struct netlink_compare_arg
475 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
476 #define netlink_compare_arg_len \
477 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
479 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
482 const struct netlink_compare_arg *x = arg->key;
483 const struct netlink_sock *nlk = ptr;
485 return nlk->portid != x->portid ||
486 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
489 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
490 struct net *net, u32 portid)
492 memset(arg, 0, sizeof(*arg));
493 write_pnet(&arg->pnet, net);
494 arg->portid = portid;
497 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
500 struct netlink_compare_arg arg;
502 netlink_compare_arg_init(&arg, net, portid);
503 return rhashtable_lookup_fast(&table->hash, &arg,
504 netlink_rhashtable_params);
507 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
509 struct netlink_compare_arg arg;
511 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
512 return rhashtable_lookup_insert_key(&table->hash, &arg,
514 netlink_rhashtable_params);
517 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
519 struct netlink_table *table = &nl_table[protocol];
523 sk = __netlink_lookup(table, portid, net);
531 static const struct proto_ops netlink_ops;
534 netlink_update_listeners(struct sock *sk)
536 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
539 struct listeners *listeners;
541 listeners = nl_deref_protected(tbl->listeners);
545 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
547 sk_for_each_bound(sk, &tbl->mc_list) {
548 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
549 mask |= nlk_sk(sk)->groups[i];
551 listeners->masks[i] = mask;
553 /* this function is only called with the netlink table "grabbed", which
554 * makes sure updates are visible before bind or setsockopt return. */
557 static int netlink_insert(struct sock *sk, u32 portid)
559 struct netlink_table *table = &nl_table[sk->sk_protocol];
564 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
565 if (nlk_sk(sk)->bound)
568 nlk_sk(sk)->portid = portid;
571 err = __netlink_insert(table, sk);
573 /* In case the hashtable backend returns with -EBUSY
574 * from here, it must not escape to the caller.
576 if (unlikely(err == -EBUSY))
584 /* We need to ensure that the socket is hashed and visible. */
586 nlk_sk(sk)->bound = portid;
593 static void netlink_remove(struct sock *sk)
595 struct netlink_table *table;
597 table = &nl_table[sk->sk_protocol];
598 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
599 netlink_rhashtable_params)) {
600 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
604 netlink_table_grab();
605 if (nlk_sk(sk)->subscriptions) {
606 __sk_del_bind_node(sk);
607 netlink_update_listeners(sk);
609 if (sk->sk_protocol == NETLINK_GENERIC)
610 atomic_inc(&genl_sk_destructing_cnt);
611 netlink_table_ungrab();
614 static struct proto netlink_proto = {
616 .owner = THIS_MODULE,
617 .obj_size = sizeof(struct netlink_sock),
620 static int __netlink_create(struct net *net, struct socket *sock,
621 struct mutex *cb_mutex, int protocol,
625 struct netlink_sock *nlk;
627 sock->ops = &netlink_ops;
629 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
633 sock_init_data(sock, sk);
637 nlk->cb_mutex = cb_mutex;
639 nlk->cb_mutex = &nlk->cb_def_mutex;
640 mutex_init(nlk->cb_mutex);
641 lockdep_set_class_and_name(nlk->cb_mutex,
642 nlk_cb_mutex_keys + protocol,
643 nlk_cb_mutex_key_strings[protocol]);
645 init_waitqueue_head(&nlk->wait);
647 sk->sk_destruct = netlink_sock_destruct;
648 sk->sk_protocol = protocol;
652 static int netlink_create(struct net *net, struct socket *sock, int protocol,
655 struct module *module = NULL;
656 struct mutex *cb_mutex;
657 struct netlink_sock *nlk;
658 int (*bind)(struct net *net, int group);
659 void (*unbind)(struct net *net, int group);
662 sock->state = SS_UNCONNECTED;
664 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
665 return -ESOCKTNOSUPPORT;
667 if (protocol < 0 || protocol >= MAX_LINKS)
668 return -EPROTONOSUPPORT;
669 protocol = array_index_nospec(protocol, MAX_LINKS);
671 netlink_lock_table();
672 #ifdef CONFIG_MODULES
673 if (!nl_table[protocol].registered) {
674 netlink_unlock_table();
675 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
676 netlink_lock_table();
679 if (nl_table[protocol].registered &&
680 try_module_get(nl_table[protocol].module))
681 module = nl_table[protocol].module;
683 err = -EPROTONOSUPPORT;
684 cb_mutex = nl_table[protocol].cb_mutex;
685 bind = nl_table[protocol].bind;
686 unbind = nl_table[protocol].unbind;
687 netlink_unlock_table();
692 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
697 sock_prot_inuse_add(net, &netlink_proto, 1);
700 nlk = nlk_sk(sock->sk);
701 nlk->module = module;
702 nlk->netlink_bind = bind;
703 nlk->netlink_unbind = unbind;
712 static void deferred_put_nlk_sk(struct rcu_head *head)
714 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
715 struct sock *sk = &nlk->sk;
720 if (!refcount_dec_and_test(&sk->sk_refcnt))
723 if (nlk->cb_running && nlk->cb.done) {
724 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
725 schedule_work(&nlk->work);
732 static int netlink_release(struct socket *sock)
734 struct sock *sk = sock->sk;
735 struct netlink_sock *nlk;
745 * OK. Socket is unlinked, any packets that arrive now
749 /* must not acquire netlink_table_lock in any way again before unbind
750 * and notifying genetlink is done as otherwise it might deadlock
752 if (nlk->netlink_unbind) {
755 for (i = 0; i < nlk->ngroups; i++)
756 if (test_bit(i, nlk->groups))
757 nlk->netlink_unbind(sock_net(sk), i + 1);
759 if (sk->sk_protocol == NETLINK_GENERIC &&
760 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
761 wake_up(&genl_sk_destructing_waitq);
764 wake_up_interruptible_all(&nlk->wait);
766 skb_queue_purge(&sk->sk_write_queue);
768 if (nlk->portid && nlk->bound) {
769 struct netlink_notify n = {
771 .protocol = sk->sk_protocol,
772 .portid = nlk->portid,
774 blocking_notifier_call_chain(&netlink_chain,
775 NETLINK_URELEASE, &n);
778 module_put(nlk->module);
780 if (netlink_is_kernel(sk)) {
781 netlink_table_grab();
782 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
783 if (--nl_table[sk->sk_protocol].registered == 0) {
784 struct listeners *old;
786 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
787 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
789 nl_table[sk->sk_protocol].module = NULL;
790 nl_table[sk->sk_protocol].bind = NULL;
791 nl_table[sk->sk_protocol].unbind = NULL;
792 nl_table[sk->sk_protocol].flags = 0;
793 nl_table[sk->sk_protocol].registered = 0;
795 netlink_table_ungrab();
799 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
801 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
805 static int netlink_autobind(struct socket *sock)
807 struct sock *sk = sock->sk;
808 struct net *net = sock_net(sk);
809 struct netlink_table *table = &nl_table[sk->sk_protocol];
810 s32 portid = task_tgid_vnr(current);
818 ok = !__netlink_lookup(table, portid, net);
821 /* Bind collision, search negative portid values. */
823 /* rover will be in range [S32_MIN, -4097] */
824 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
825 else if (rover >= -4096)
831 err = netlink_insert(sk, portid);
832 if (err == -EADDRINUSE)
835 /* If 2 threads race to autobind, that is fine. */
843 * __netlink_ns_capable - General netlink message capability test
844 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
845 * @user_ns: The user namespace of the capability to use
846 * @cap: The capability to use
848 * Test to see if the opener of the socket we received the message
849 * from had when the netlink socket was created and the sender of the
850 * message has has the capability @cap in the user namespace @user_ns.
852 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
853 struct user_namespace *user_ns, int cap)
855 return ((nsp->flags & NETLINK_SKB_DST) ||
856 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
857 ns_capable(user_ns, cap);
859 EXPORT_SYMBOL(__netlink_ns_capable);
862 * netlink_ns_capable - General netlink message capability test
863 * @skb: socket buffer holding a netlink command from userspace
864 * @user_ns: The user namespace of the capability to use
865 * @cap: The capability to use
867 * Test to see if the opener of the socket we received the message
868 * from had when the netlink socket was created and the sender of the
869 * message has has the capability @cap in the user namespace @user_ns.
871 bool netlink_ns_capable(const struct sk_buff *skb,
872 struct user_namespace *user_ns, int cap)
874 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
876 EXPORT_SYMBOL(netlink_ns_capable);
879 * netlink_capable - Netlink global message capability test
880 * @skb: socket buffer holding a netlink command from userspace
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 all user namespaces.
887 bool netlink_capable(const struct sk_buff *skb, int cap)
889 return netlink_ns_capable(skb, &init_user_ns, cap);
891 EXPORT_SYMBOL(netlink_capable);
894 * netlink_net_capable - Netlink network namespace message capability test
895 * @skb: socket buffer holding a netlink command from userspace
896 * @cap: The capability to use
898 * Test to see if the opener of the socket we received the message
899 * from had when the netlink socket was created and the sender of the
900 * message has has the capability @cap over the network namespace of
901 * the socket we received the message from.
903 bool netlink_net_capable(const struct sk_buff *skb, int cap)
905 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
907 EXPORT_SYMBOL(netlink_net_capable);
909 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
911 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
912 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
916 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
918 struct netlink_sock *nlk = nlk_sk(sk);
920 if (nlk->subscriptions && !subscriptions)
921 __sk_del_bind_node(sk);
922 else if (!nlk->subscriptions && subscriptions)
923 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
924 nlk->subscriptions = subscriptions;
927 static int netlink_realloc_groups(struct sock *sk)
929 struct netlink_sock *nlk = nlk_sk(sk);
931 unsigned long *new_groups;
934 netlink_table_grab();
936 groups = nl_table[sk->sk_protocol].groups;
937 if (!nl_table[sk->sk_protocol].registered) {
942 if (nlk->ngroups >= groups)
945 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
946 if (new_groups == NULL) {
950 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
951 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
953 nlk->groups = new_groups;
954 nlk->ngroups = groups;
956 netlink_table_ungrab();
960 static void netlink_undo_bind(int group, long unsigned int groups,
963 struct netlink_sock *nlk = nlk_sk(sk);
966 if (!nlk->netlink_unbind)
969 for (undo = 0; undo < group; undo++)
970 if (test_bit(undo, &groups))
971 nlk->netlink_unbind(sock_net(sk), undo + 1);
974 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
977 struct sock *sk = sock->sk;
978 struct net *net = sock_net(sk);
979 struct netlink_sock *nlk = nlk_sk(sk);
980 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
982 unsigned long groups;
985 if (addr_len < sizeof(struct sockaddr_nl))
988 if (nladdr->nl_family != AF_NETLINK)
990 groups = nladdr->nl_groups;
992 /* Only superuser is allowed to listen multicasts */
994 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
996 err = netlink_realloc_groups(sk);
1001 if (nlk->ngroups < BITS_PER_LONG)
1002 groups &= (1UL << nlk->ngroups) - 1;
1006 /* Ensure nlk->portid is up-to-date. */
1009 if (nladdr->nl_pid != nlk->portid)
1013 netlink_lock_table();
1014 if (nlk->netlink_bind && groups) {
1017 for (group = 0; group < nlk->ngroups; group++) {
1018 if (!test_bit(group, &groups))
1020 err = nlk->netlink_bind(net, group + 1);
1023 netlink_undo_bind(group, groups, sk);
1028 /* No need for barriers here as we return to user-space without
1029 * using any of the bound attributes.
1032 err = nladdr->nl_pid ?
1033 netlink_insert(sk, nladdr->nl_pid) :
1034 netlink_autobind(sock);
1036 netlink_undo_bind(nlk->ngroups, groups, sk);
1041 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1043 netlink_unlock_table();
1045 netlink_table_grab();
1046 netlink_update_subscriptions(sk, nlk->subscriptions +
1048 hweight32(nlk->groups[0]));
1049 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1050 netlink_update_listeners(sk);
1051 netlink_table_ungrab();
1056 netlink_unlock_table();
1060 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1061 int alen, int flags)
1064 struct sock *sk = sock->sk;
1065 struct netlink_sock *nlk = nlk_sk(sk);
1066 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1068 if (alen < sizeof(addr->sa_family))
1071 if (addr->sa_family == AF_UNSPEC) {
1072 sk->sk_state = NETLINK_UNCONNECTED;
1073 nlk->dst_portid = 0;
1077 if (addr->sa_family != AF_NETLINK)
1080 if (alen < sizeof(struct sockaddr_nl))
1083 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1084 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1087 /* No need for barriers here as we return to user-space without
1088 * using any of the bound attributes.
1091 err = netlink_autobind(sock);
1094 sk->sk_state = NETLINK_CONNECTED;
1095 nlk->dst_portid = nladdr->nl_pid;
1096 nlk->dst_group = ffs(nladdr->nl_groups);
1102 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1105 struct sock *sk = sock->sk;
1106 struct netlink_sock *nlk = nlk_sk(sk);
1107 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1109 nladdr->nl_family = AF_NETLINK;
1113 nladdr->nl_pid = nlk->dst_portid;
1114 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1116 nladdr->nl_pid = nlk->portid;
1117 netlink_lock_table();
1118 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1119 netlink_unlock_table();
1121 return sizeof(*nladdr);
1124 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1127 /* try to hand this ioctl down to the NIC drivers.
1129 return -ENOIOCTLCMD;
1132 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1135 struct netlink_sock *nlk;
1137 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1139 return ERR_PTR(-ECONNREFUSED);
1141 /* Don't bother queuing skb if kernel socket has no input function */
1143 if (sock->sk_state == NETLINK_CONNECTED &&
1144 nlk->dst_portid != nlk_sk(ssk)->portid) {
1146 return ERR_PTR(-ECONNREFUSED);
1151 struct sock *netlink_getsockbyfilp(struct file *filp)
1153 struct inode *inode = file_inode(filp);
1156 if (!S_ISSOCK(inode->i_mode))
1157 return ERR_PTR(-ENOTSOCK);
1159 sock = SOCKET_I(inode)->sk;
1160 if (sock->sk_family != AF_NETLINK)
1161 return ERR_PTR(-EINVAL);
1167 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1170 struct sk_buff *skb;
1173 if (size <= NLMSG_GOODSIZE || broadcast)
1174 return alloc_skb(size, GFP_KERNEL);
1176 size = SKB_DATA_ALIGN(size) +
1177 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1179 data = vmalloc(size);
1183 skb = __build_skb(data, size);
1187 skb->destructor = netlink_skb_destructor;
1193 * Attach a skb to a netlink socket.
1194 * The caller must hold a reference to the destination socket. On error, the
1195 * reference is dropped. The skb is not send to the destination, just all
1196 * all error checks are performed and memory in the queue is reserved.
1198 * < 0: error. skb freed, reference to sock dropped.
1200 * 1: repeat lookup - reference dropped while waiting for socket memory.
1202 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1203 long *timeo, struct sock *ssk)
1205 struct netlink_sock *nlk;
1209 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1210 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1211 DECLARE_WAITQUEUE(wait, current);
1213 if (!ssk || netlink_is_kernel(ssk))
1214 netlink_overrun(sk);
1220 __set_current_state(TASK_INTERRUPTIBLE);
1221 add_wait_queue(&nlk->wait, &wait);
1223 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1224 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1225 !sock_flag(sk, SOCK_DEAD))
1226 *timeo = schedule_timeout(*timeo);
1228 __set_current_state(TASK_RUNNING);
1229 remove_wait_queue(&nlk->wait, &wait);
1232 if (signal_pending(current)) {
1234 return sock_intr_errno(*timeo);
1238 netlink_skb_set_owner_r(skb, sk);
1242 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1246 netlink_deliver_tap(sock_net(sk), skb);
1248 skb_queue_tail(&sk->sk_receive_queue, skb);
1249 sk->sk_data_ready(sk);
1253 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1255 int len = __netlink_sendskb(sk, skb);
1261 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1267 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1271 WARN_ON(skb->sk != NULL);
1272 delta = skb->end - skb->tail;
1273 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1276 if (skb_shared(skb)) {
1277 struct sk_buff *nskb = skb_clone(skb, allocation);
1284 pskb_expand_head(skb, 0, -delta,
1285 (allocation & ~__GFP_DIRECT_RECLAIM) |
1286 __GFP_NOWARN | __GFP_NORETRY);
1290 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1294 struct netlink_sock *nlk = nlk_sk(sk);
1296 ret = -ECONNREFUSED;
1297 if (nlk->netlink_rcv != NULL) {
1299 netlink_skb_set_owner_r(skb, sk);
1300 NETLINK_CB(skb).sk = ssk;
1301 netlink_deliver_tap_kernel(sk, ssk, skb);
1302 nlk->netlink_rcv(skb);
1311 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1312 u32 portid, int nonblock)
1318 skb = netlink_trim(skb, gfp_any());
1320 timeo = sock_sndtimeo(ssk, nonblock);
1322 sk = netlink_getsockbyportid(ssk, portid);
1327 if (netlink_is_kernel(sk))
1328 return netlink_unicast_kernel(sk, skb, ssk);
1330 if (sk_filter(sk, skb)) {
1337 err = netlink_attachskb(sk, skb, &timeo, ssk);
1343 return netlink_sendskb(sk, skb);
1345 EXPORT_SYMBOL(netlink_unicast);
1347 int netlink_has_listeners(struct sock *sk, unsigned int group)
1350 struct listeners *listeners;
1352 BUG_ON(!netlink_is_kernel(sk));
1355 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1357 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1358 res = test_bit(group - 1, listeners->masks);
1364 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1366 bool netlink_strict_get_check(struct sk_buff *skb)
1368 const struct netlink_sock *nlk = nlk_sk(NETLINK_CB(skb).sk);
1370 return nlk->flags & NETLINK_F_STRICT_CHK;
1372 EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1374 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1376 struct netlink_sock *nlk = nlk_sk(sk);
1378 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1379 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1380 netlink_skb_set_owner_r(skb, sk);
1381 __netlink_sendskb(sk, skb);
1382 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1387 struct netlink_broadcast_data {
1388 struct sock *exclude_sk;
1393 int delivery_failure;
1397 struct sk_buff *skb, *skb2;
1398 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1402 static void do_one_broadcast(struct sock *sk,
1403 struct netlink_broadcast_data *p)
1405 struct netlink_sock *nlk = nlk_sk(sk);
1408 if (p->exclude_sk == sk)
1411 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1412 !test_bit(p->group - 1, nlk->groups))
1415 if (!net_eq(sock_net(sk), p->net)) {
1416 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1419 if (!peernet_has_id(sock_net(sk), p->net))
1422 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1428 netlink_overrun(sk);
1433 if (p->skb2 == NULL) {
1434 if (skb_shared(p->skb)) {
1435 p->skb2 = skb_clone(p->skb, p->allocation);
1437 p->skb2 = skb_get(p->skb);
1439 * skb ownership may have been set when
1440 * delivered to a previous socket.
1442 skb_orphan(p->skb2);
1445 if (p->skb2 == NULL) {
1446 netlink_overrun(sk);
1447 /* Clone failed. Notify ALL listeners. */
1449 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1450 p->delivery_failure = 1;
1453 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1458 if (sk_filter(sk, p->skb2)) {
1463 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1464 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1465 NETLINK_CB(p->skb2).nsid_is_set = true;
1466 val = netlink_broadcast_deliver(sk, p->skb2);
1468 netlink_overrun(sk);
1469 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1470 p->delivery_failure = 1;
1472 p->congested |= val;
1480 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1481 u32 group, gfp_t allocation,
1482 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1485 struct net *net = sock_net(ssk);
1486 struct netlink_broadcast_data info;
1489 skb = netlink_trim(skb, allocation);
1491 info.exclude_sk = ssk;
1493 info.portid = portid;
1496 info.delivery_failure = 0;
1499 info.allocation = allocation;
1502 info.tx_filter = filter;
1503 info.tx_data = filter_data;
1505 /* While we sleep in clone, do not allow to change socket list */
1507 netlink_lock_table();
1509 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1510 do_one_broadcast(sk, &info);
1514 netlink_unlock_table();
1516 if (info.delivery_failure) {
1517 kfree_skb(info.skb2);
1520 consume_skb(info.skb2);
1522 if (info.delivered) {
1523 if (info.congested && gfpflags_allow_blocking(allocation))
1529 EXPORT_SYMBOL(netlink_broadcast_filtered);
1531 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1532 u32 group, gfp_t allocation)
1534 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1537 EXPORT_SYMBOL(netlink_broadcast);
1539 struct netlink_set_err_data {
1540 struct sock *exclude_sk;
1546 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1548 struct netlink_sock *nlk = nlk_sk(sk);
1551 if (sk == p->exclude_sk)
1554 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1557 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1558 !test_bit(p->group - 1, nlk->groups))
1561 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1566 sk->sk_err = p->code;
1567 sk->sk_error_report(sk);
1573 * netlink_set_err - report error to broadcast listeners
1574 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1575 * @portid: the PORTID of a process that we want to skip (if any)
1576 * @group: the broadcast group that will notice the error
1577 * @code: error code, must be negative (as usual in kernelspace)
1579 * This function returns the number of broadcast listeners that have set the
1580 * NETLINK_NO_ENOBUFS socket option.
1582 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1584 struct netlink_set_err_data info;
1588 info.exclude_sk = ssk;
1589 info.portid = portid;
1591 /* sk->sk_err wants a positive error value */
1594 read_lock(&nl_table_lock);
1596 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1597 ret += do_one_set_err(sk, &info);
1599 read_unlock(&nl_table_lock);
1602 EXPORT_SYMBOL(netlink_set_err);
1604 /* must be called with netlink table grabbed */
1605 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1609 int old, new = !!is_new, subscriptions;
1611 old = test_bit(group - 1, nlk->groups);
1612 subscriptions = nlk->subscriptions - old + new;
1614 __set_bit(group - 1, nlk->groups);
1616 __clear_bit(group - 1, nlk->groups);
1617 netlink_update_subscriptions(&nlk->sk, subscriptions);
1618 netlink_update_listeners(&nlk->sk);
1621 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1622 char __user *optval, unsigned int optlen)
1624 struct sock *sk = sock->sk;
1625 struct netlink_sock *nlk = nlk_sk(sk);
1626 unsigned int val = 0;
1629 if (level != SOL_NETLINK)
1630 return -ENOPROTOOPT;
1632 if (optlen >= sizeof(int) &&
1633 get_user(val, (unsigned int __user *)optval))
1637 case NETLINK_PKTINFO:
1639 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1641 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1644 case NETLINK_ADD_MEMBERSHIP:
1645 case NETLINK_DROP_MEMBERSHIP: {
1646 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1648 err = netlink_realloc_groups(sk);
1651 if (!val || val - 1 >= nlk->ngroups)
1653 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1654 err = nlk->netlink_bind(sock_net(sk), val);
1658 netlink_table_grab();
1659 netlink_update_socket_mc(nlk, val,
1660 optname == NETLINK_ADD_MEMBERSHIP);
1661 netlink_table_ungrab();
1662 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1663 nlk->netlink_unbind(sock_net(sk), val);
1668 case NETLINK_BROADCAST_ERROR:
1670 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1672 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1675 case NETLINK_NO_ENOBUFS:
1677 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1678 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1679 wake_up_interruptible(&nlk->wait);
1681 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1685 case NETLINK_LISTEN_ALL_NSID:
1686 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1690 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1692 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1695 case NETLINK_CAP_ACK:
1697 nlk->flags |= NETLINK_F_CAP_ACK;
1699 nlk->flags &= ~NETLINK_F_CAP_ACK;
1702 case NETLINK_EXT_ACK:
1704 nlk->flags |= NETLINK_F_EXT_ACK;
1706 nlk->flags &= ~NETLINK_F_EXT_ACK;
1709 case NETLINK_GET_STRICT_CHK:
1711 nlk->flags |= NETLINK_F_STRICT_CHK;
1713 nlk->flags &= ~NETLINK_F_STRICT_CHK;
1722 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1723 char __user *optval, int __user *optlen)
1725 struct sock *sk = sock->sk;
1726 struct netlink_sock *nlk = nlk_sk(sk);
1729 if (level != SOL_NETLINK)
1730 return -ENOPROTOOPT;
1732 if (get_user(len, optlen))
1738 case NETLINK_PKTINFO:
1739 if (len < sizeof(int))
1742 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1743 if (put_user(len, optlen) ||
1744 put_user(val, optval))
1748 case NETLINK_BROADCAST_ERROR:
1749 if (len < sizeof(int))
1752 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1753 if (put_user(len, optlen) ||
1754 put_user(val, optval))
1758 case NETLINK_NO_ENOBUFS:
1759 if (len < sizeof(int))
1762 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1763 if (put_user(len, optlen) ||
1764 put_user(val, optval))
1768 case NETLINK_LIST_MEMBERSHIPS: {
1769 int pos, idx, shift;
1772 netlink_lock_table();
1773 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1774 if (len - pos < sizeof(u32))
1777 idx = pos / sizeof(unsigned long);
1778 shift = (pos % sizeof(unsigned long)) * 8;
1779 if (put_user((u32)(nlk->groups[idx] >> shift),
1780 (u32 __user *)(optval + pos))) {
1785 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1787 netlink_unlock_table();
1790 case NETLINK_CAP_ACK:
1791 if (len < sizeof(int))
1794 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1795 if (put_user(len, optlen) ||
1796 put_user(val, optval))
1800 case NETLINK_EXT_ACK:
1801 if (len < sizeof(int))
1804 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1805 if (put_user(len, optlen) || put_user(val, optval))
1809 case NETLINK_GET_STRICT_CHK:
1810 if (len < sizeof(int))
1813 val = nlk->flags & NETLINK_F_STRICT_CHK ? 1 : 0;
1814 if (put_user(len, optlen) || put_user(val, optval))
1824 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1826 struct nl_pktinfo info;
1828 info.group = NETLINK_CB(skb).dst_group;
1829 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1832 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1833 struct sk_buff *skb)
1835 if (!NETLINK_CB(skb).nsid_is_set)
1838 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1839 &NETLINK_CB(skb).nsid);
1842 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1844 struct sock *sk = sock->sk;
1845 struct netlink_sock *nlk = nlk_sk(sk);
1846 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1849 struct sk_buff *skb;
1851 struct scm_cookie scm;
1852 u32 netlink_skb_flags = 0;
1854 if (msg->msg_flags&MSG_OOB)
1857 err = scm_send(sock, msg, &scm, true);
1861 if (msg->msg_namelen) {
1863 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1865 if (addr->nl_family != AF_NETLINK)
1867 dst_portid = addr->nl_pid;
1868 dst_group = ffs(addr->nl_groups);
1870 if ((dst_group || dst_portid) &&
1871 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1873 netlink_skb_flags |= NETLINK_SKB_DST;
1875 dst_portid = nlk->dst_portid;
1876 dst_group = nlk->dst_group;
1880 err = netlink_autobind(sock);
1884 /* Ensure nlk is hashed and visible. */
1889 if (len > sk->sk_sndbuf - 32)
1892 skb = netlink_alloc_large_skb(len, dst_group);
1896 NETLINK_CB(skb).portid = nlk->portid;
1897 NETLINK_CB(skb).dst_group = dst_group;
1898 NETLINK_CB(skb).creds = scm.creds;
1899 NETLINK_CB(skb).flags = netlink_skb_flags;
1902 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1907 err = security_netlink_send(sk, skb);
1914 refcount_inc(&skb->users);
1915 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1917 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1924 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1927 struct scm_cookie scm;
1928 struct sock *sk = sock->sk;
1929 struct netlink_sock *nlk = nlk_sk(sk);
1930 int noblock = flags&MSG_DONTWAIT;
1932 struct sk_buff *skb, *data_skb;
1940 skb = skb_recv_datagram(sk, flags, noblock, &err);
1946 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1947 if (unlikely(skb_shinfo(skb)->frag_list)) {
1949 * If this skb has a frag_list, then here that means that we
1950 * will have to use the frag_list skb's data for compat tasks
1951 * and the regular skb's data for normal (non-compat) tasks.
1953 * If we need to send the compat skb, assign it to the
1954 * 'data_skb' variable so that it will be used below for data
1955 * copying. We keep 'skb' for everything else, including
1956 * freeing both later.
1958 if (flags & MSG_CMSG_COMPAT)
1959 data_skb = skb_shinfo(skb)->frag_list;
1963 /* Record the max length of recvmsg() calls for future allocations */
1964 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1965 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1966 SKB_WITH_OVERHEAD(32768));
1968 copied = data_skb->len;
1970 msg->msg_flags |= MSG_TRUNC;
1974 skb_reset_transport_header(data_skb);
1975 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1977 if (msg->msg_name) {
1978 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1979 addr->nl_family = AF_NETLINK;
1981 addr->nl_pid = NETLINK_CB(skb).portid;
1982 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1983 msg->msg_namelen = sizeof(*addr);
1986 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1987 netlink_cmsg_recv_pktinfo(msg, skb);
1988 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1989 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1991 memset(&scm, 0, sizeof(scm));
1992 scm.creds = *NETLINK_CREDS(skb);
1993 if (flags & MSG_TRUNC)
1994 copied = data_skb->len;
1996 skb_free_datagram(sk, skb);
1998 if (nlk->cb_running &&
1999 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2000 ret = netlink_dump(sk);
2003 sk->sk_error_report(sk);
2007 scm_recv(sock, msg, &scm, flags);
2009 netlink_rcv_wake(sk);
2010 return err ? : copied;
2013 static void netlink_data_ready(struct sock *sk)
2019 * We export these functions to other modules. They provide a
2020 * complete set of kernel non-blocking support for message
2025 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2026 struct netlink_kernel_cfg *cfg)
2028 struct socket *sock;
2030 struct netlink_sock *nlk;
2031 struct listeners *listeners = NULL;
2032 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2033 unsigned int groups;
2037 if (unit < 0 || unit >= MAX_LINKS)
2040 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2043 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2044 goto out_sock_release_nosk;
2048 if (!cfg || cfg->groups < 32)
2051 groups = cfg->groups;
2053 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2055 goto out_sock_release;
2057 sk->sk_data_ready = netlink_data_ready;
2058 if (cfg && cfg->input)
2059 nlk_sk(sk)->netlink_rcv = cfg->input;
2061 if (netlink_insert(sk, 0))
2062 goto out_sock_release;
2065 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2067 netlink_table_grab();
2068 if (!nl_table[unit].registered) {
2069 nl_table[unit].groups = groups;
2070 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2071 nl_table[unit].cb_mutex = cb_mutex;
2072 nl_table[unit].module = module;
2074 nl_table[unit].bind = cfg->bind;
2075 nl_table[unit].unbind = cfg->unbind;
2076 nl_table[unit].flags = cfg->flags;
2078 nl_table[unit].compare = cfg->compare;
2080 nl_table[unit].registered = 1;
2083 nl_table[unit].registered++;
2085 netlink_table_ungrab();
2090 netlink_kernel_release(sk);
2093 out_sock_release_nosk:
2097 EXPORT_SYMBOL(__netlink_kernel_create);
2100 netlink_kernel_release(struct sock *sk)
2102 if (sk == NULL || sk->sk_socket == NULL)
2105 sock_release(sk->sk_socket);
2107 EXPORT_SYMBOL(netlink_kernel_release);
2109 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2111 struct listeners *new, *old;
2112 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2117 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2118 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2121 old = nl_deref_protected(tbl->listeners);
2122 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2123 rcu_assign_pointer(tbl->listeners, new);
2125 kfree_rcu(old, rcu);
2127 tbl->groups = groups;
2133 * netlink_change_ngroups - change number of multicast groups
2135 * This changes the number of multicast groups that are available
2136 * on a certain netlink family. Note that it is not possible to
2137 * change the number of groups to below 32. Also note that it does
2138 * not implicitly call netlink_clear_multicast_users() when the
2139 * number of groups is reduced.
2141 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2142 * @groups: The new number of groups.
2144 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2148 netlink_table_grab();
2149 err = __netlink_change_ngroups(sk, groups);
2150 netlink_table_ungrab();
2155 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2158 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2160 sk_for_each_bound(sk, &tbl->mc_list)
2161 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2165 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2167 struct nlmsghdr *nlh;
2168 int size = nlmsg_msg_size(len);
2170 nlh = skb_put(skb, NLMSG_ALIGN(size));
2171 nlh->nlmsg_type = type;
2172 nlh->nlmsg_len = size;
2173 nlh->nlmsg_flags = flags;
2174 nlh->nlmsg_pid = portid;
2175 nlh->nlmsg_seq = seq;
2176 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2177 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2180 EXPORT_SYMBOL(__nlmsg_put);
2183 * It looks a bit ugly.
2184 * It would be better to create kernel thread.
2187 static int netlink_dump(struct sock *sk)
2189 struct netlink_sock *nlk = nlk_sk(sk);
2190 struct netlink_ext_ack extack = {};
2191 struct netlink_callback *cb;
2192 struct sk_buff *skb = NULL;
2193 struct nlmsghdr *nlh;
2194 struct module *module;
2199 mutex_lock(nlk->cb_mutex);
2200 if (!nlk->cb_running) {
2205 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2208 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2209 * required, but it makes sense to _attempt_ a 16K bytes allocation
2210 * to reduce number of system calls on dump operations, if user
2211 * ever provided a big enough buffer.
2214 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2216 if (alloc_min_size < nlk->max_recvmsg_len) {
2217 alloc_size = nlk->max_recvmsg_len;
2218 skb = alloc_skb(alloc_size,
2219 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2220 __GFP_NOWARN | __GFP_NORETRY);
2223 alloc_size = alloc_min_size;
2224 skb = alloc_skb(alloc_size, GFP_KERNEL);
2229 /* Trim skb to allocated size. User is expected to provide buffer as
2230 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2231 * netlink_recvmsg())). dump will pack as many smaller messages as
2232 * could fit within the allocated skb. skb is typically allocated
2233 * with larger space than required (could be as much as near 2x the
2234 * requested size with align to next power of 2 approach). Allowing
2235 * dump to use the excess space makes it difficult for a user to have a
2236 * reasonable static buffer based on the expected largest dump of a
2237 * single netdev. The outcome is MSG_TRUNC error.
2239 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2240 netlink_skb_set_owner_r(skb, sk);
2242 if (nlk->dump_done_errno > 0) {
2243 cb->extack = &extack;
2244 nlk->dump_done_errno = cb->dump(skb, cb);
2248 if (nlk->dump_done_errno > 0 ||
2249 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2250 mutex_unlock(nlk->cb_mutex);
2252 if (sk_filter(sk, skb))
2255 __netlink_sendskb(sk, skb);
2259 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE,
2260 sizeof(nlk->dump_done_errno),
2261 NLM_F_MULTI | cb->answer_flags);
2265 nl_dump_check_consistent(cb, nlh);
2267 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno,
2268 sizeof(nlk->dump_done_errno));
2270 if (extack._msg && nlk->flags & NETLINK_F_EXT_ACK) {
2271 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2272 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack._msg))
2273 nlmsg_end(skb, nlh);
2276 if (sk_filter(sk, skb))
2279 __netlink_sendskb(sk, skb);
2284 nlk->cb_running = false;
2285 module = cb->module;
2287 mutex_unlock(nlk->cb_mutex);
2293 mutex_unlock(nlk->cb_mutex);
2298 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2299 const struct nlmsghdr *nlh,
2300 struct netlink_dump_control *control)
2302 struct netlink_sock *nlk, *nlk2;
2303 struct netlink_callback *cb;
2307 refcount_inc(&skb->users);
2309 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2311 ret = -ECONNREFUSED;
2316 mutex_lock(nlk->cb_mutex);
2317 /* A dump is in progress... */
2318 if (nlk->cb_running) {
2322 /* add reference of module which cb->dump belongs to */
2323 if (!try_module_get(control->module)) {
2324 ret = -EPROTONOSUPPORT;
2329 memset(cb, 0, sizeof(*cb));
2330 cb->dump = control->dump;
2331 cb->done = control->done;
2333 cb->data = control->data;
2334 cb->module = control->module;
2335 cb->min_dump_alloc = control->min_dump_alloc;
2338 nlk2 = nlk_sk(NETLINK_CB(skb).sk);
2339 cb->strict_check = !!(nlk2->flags & NETLINK_F_STRICT_CHK);
2341 if (control->start) {
2342 ret = control->start(cb);
2347 nlk->cb_running = true;
2348 nlk->dump_done_errno = INT_MAX;
2350 mutex_unlock(nlk->cb_mutex);
2352 ret = netlink_dump(sk);
2359 /* We successfully started a dump, by returning -EINTR we
2360 * signal not to send ACK even if it was requested.
2365 module_put(control->module);
2368 mutex_unlock(nlk->cb_mutex);
2373 EXPORT_SYMBOL(__netlink_dump_start);
2375 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2376 const struct netlink_ext_ack *extack)
2378 struct sk_buff *skb;
2379 struct nlmsghdr *rep;
2380 struct nlmsgerr *errmsg;
2381 size_t payload = sizeof(*errmsg);
2383 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2384 unsigned int flags = 0;
2385 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2387 /* Error messages get the original request appened, unless the user
2388 * requests to cap the error message, and get extra error data if
2391 if (nlk_has_extack && extack && extack->_msg)
2392 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2395 if (!(nlk->flags & NETLINK_F_CAP_ACK))
2396 payload += nlmsg_len(nlh);
2398 flags |= NLM_F_CAPPED;
2399 if (nlk_has_extack && extack && extack->bad_attr)
2400 tlvlen += nla_total_size(sizeof(u32));
2402 flags |= NLM_F_CAPPED;
2404 if (nlk_has_extack && extack && extack->cookie_len)
2405 tlvlen += nla_total_size(extack->cookie_len);
2409 flags |= NLM_F_ACK_TLVS;
2411 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2413 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2414 NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk);
2418 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2419 NLMSG_ERROR, payload, flags);
2420 errmsg = nlmsg_data(rep);
2421 errmsg->error = err;
2422 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2424 if (nlk_has_extack && extack) {
2426 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2430 if (extack->bad_attr &&
2431 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2432 (u8 *)extack->bad_attr >= in_skb->data +
2434 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2435 (u8 *)extack->bad_attr -
2438 if (extack->cookie_len)
2439 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2445 nlmsg_end(skb, rep);
2447 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2449 EXPORT_SYMBOL(netlink_ack);
2451 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2453 struct netlink_ext_ack *))
2455 struct netlink_ext_ack extack;
2456 struct nlmsghdr *nlh;
2459 while (skb->len >= nlmsg_total_size(0)) {
2462 memset(&extack, 0, sizeof(extack));
2463 nlh = nlmsg_hdr(skb);
2466 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2469 /* Only requests are handled by the kernel */
2470 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2473 /* Skip control messages */
2474 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2477 err = cb(skb, nlh, &extack);
2482 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2483 netlink_ack(skb, nlh, err, &extack);
2486 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2487 if (msglen > skb->len)
2489 skb_pull(skb, msglen);
2494 EXPORT_SYMBOL(netlink_rcv_skb);
2497 * nlmsg_notify - send a notification netlink message
2498 * @sk: netlink socket to use
2499 * @skb: notification message
2500 * @portid: destination netlink portid for reports or 0
2501 * @group: destination multicast group or 0
2502 * @report: 1 to report back, 0 to disable
2503 * @flags: allocation flags
2505 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2506 unsigned int group, int report, gfp_t flags)
2511 int exclude_portid = 0;
2514 refcount_inc(&skb->users);
2515 exclude_portid = portid;
2518 /* errors reported via destination sk->sk_err, but propagate
2519 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2520 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2526 err2 = nlmsg_unicast(sk, skb, portid);
2527 if (!err || err == -ESRCH)
2533 EXPORT_SYMBOL(nlmsg_notify);
2535 #ifdef CONFIG_PROC_FS
2536 struct nl_seq_iter {
2537 struct seq_net_private p;
2538 struct rhashtable_iter hti;
2542 static void netlink_walk_start(struct nl_seq_iter *iter)
2544 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2545 rhashtable_walk_start(&iter->hti);
2548 static void netlink_walk_stop(struct nl_seq_iter *iter)
2550 rhashtable_walk_stop(&iter->hti);
2551 rhashtable_walk_exit(&iter->hti);
2554 static void *__netlink_seq_next(struct seq_file *seq)
2556 struct nl_seq_iter *iter = seq->private;
2557 struct netlink_sock *nlk;
2561 nlk = rhashtable_walk_next(&iter->hti);
2564 if (PTR_ERR(nlk) == -EAGAIN)
2573 netlink_walk_stop(iter);
2574 if (++iter->link >= MAX_LINKS)
2577 netlink_walk_start(iter);
2579 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2584 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2586 struct nl_seq_iter *iter = seq->private;
2587 void *obj = SEQ_START_TOKEN;
2592 netlink_walk_start(iter);
2594 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2595 obj = __netlink_seq_next(seq);
2600 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2603 return __netlink_seq_next(seq);
2606 static void netlink_seq_stop(struct seq_file *seq, void *v)
2608 struct nl_seq_iter *iter = seq->private;
2610 if (iter->link >= MAX_LINKS)
2613 netlink_walk_stop(iter);
2617 static int netlink_seq_show(struct seq_file *seq, void *v)
2619 if (v == SEQ_START_TOKEN) {
2621 "sk Eth Pid Groups "
2622 "Rmem Wmem Dump Locks Drops Inode\n");
2625 struct netlink_sock *nlk = nlk_sk(s);
2627 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2631 nlk->groups ? (u32)nlk->groups[0] : 0,
2632 sk_rmem_alloc_get(s),
2633 sk_wmem_alloc_get(s),
2635 refcount_read(&s->sk_refcnt),
2636 atomic_read(&s->sk_drops),
2644 static const struct seq_operations netlink_seq_ops = {
2645 .start = netlink_seq_start,
2646 .next = netlink_seq_next,
2647 .stop = netlink_seq_stop,
2648 .show = netlink_seq_show,
2652 int netlink_register_notifier(struct notifier_block *nb)
2654 return blocking_notifier_chain_register(&netlink_chain, nb);
2656 EXPORT_SYMBOL(netlink_register_notifier);
2658 int netlink_unregister_notifier(struct notifier_block *nb)
2660 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2662 EXPORT_SYMBOL(netlink_unregister_notifier);
2664 static const struct proto_ops netlink_ops = {
2665 .family = PF_NETLINK,
2666 .owner = THIS_MODULE,
2667 .release = netlink_release,
2668 .bind = netlink_bind,
2669 .connect = netlink_connect,
2670 .socketpair = sock_no_socketpair,
2671 .accept = sock_no_accept,
2672 .getname = netlink_getname,
2673 .poll = datagram_poll,
2674 .ioctl = netlink_ioctl,
2675 .listen = sock_no_listen,
2676 .shutdown = sock_no_shutdown,
2677 .setsockopt = netlink_setsockopt,
2678 .getsockopt = netlink_getsockopt,
2679 .sendmsg = netlink_sendmsg,
2680 .recvmsg = netlink_recvmsg,
2681 .mmap = sock_no_mmap,
2682 .sendpage = sock_no_sendpage,
2685 static const struct net_proto_family netlink_family_ops = {
2686 .family = PF_NETLINK,
2687 .create = netlink_create,
2688 .owner = THIS_MODULE, /* for consistency 8) */
2691 static int __net_init netlink_net_init(struct net *net)
2693 #ifdef CONFIG_PROC_FS
2694 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2695 sizeof(struct nl_seq_iter)))
2701 static void __net_exit netlink_net_exit(struct net *net)
2703 #ifdef CONFIG_PROC_FS
2704 remove_proc_entry("netlink", net->proc_net);
2708 static void __init netlink_add_usersock_entry(void)
2710 struct listeners *listeners;
2713 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2715 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2717 netlink_table_grab();
2719 nl_table[NETLINK_USERSOCK].groups = groups;
2720 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2721 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2722 nl_table[NETLINK_USERSOCK].registered = 1;
2723 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2725 netlink_table_ungrab();
2728 static struct pernet_operations __net_initdata netlink_net_ops = {
2729 .init = netlink_net_init,
2730 .exit = netlink_net_exit,
2733 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2735 const struct netlink_sock *nlk = data;
2736 struct netlink_compare_arg arg;
2738 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2739 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2742 static const struct rhashtable_params netlink_rhashtable_params = {
2743 .head_offset = offsetof(struct netlink_sock, node),
2744 .key_len = netlink_compare_arg_len,
2745 .obj_hashfn = netlink_hash,
2746 .obj_cmpfn = netlink_compare,
2747 .automatic_shrinking = true,
2750 static int __init netlink_proto_init(void)
2753 int err = proto_register(&netlink_proto, 0);
2758 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2760 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2764 for (i = 0; i < MAX_LINKS; i++) {
2765 if (rhashtable_init(&nl_table[i].hash,
2766 &netlink_rhashtable_params) < 0) {
2768 rhashtable_destroy(&nl_table[i].hash);
2774 netlink_add_usersock_entry();
2776 sock_register(&netlink_family_ops);
2777 register_pernet_subsys(&netlink_net_ops);
2778 register_pernet_subsys(&netlink_tap_net_ops);
2779 /* The netlink device handler may be needed early. */
2784 panic("netlink_init: Cannot allocate nl_table\n");
2787 core_initcall(netlink_proto_init);