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 void __net_exit netlink_tap_exit_net(struct net *net)
248 static struct pernet_operations netlink_tap_net_ops = {
249 .init = netlink_tap_init_net,
250 .exit = netlink_tap_exit_net,
251 .id = &netlink_tap_net_id,
252 .size = sizeof(struct netlink_tap_net),
255 static bool netlink_filter_tap(const struct sk_buff *skb)
257 struct sock *sk = skb->sk;
259 /* We take the more conservative approach and
260 * whitelist socket protocols that may pass.
262 switch (sk->sk_protocol) {
264 case NETLINK_USERSOCK:
265 case NETLINK_SOCK_DIAG:
268 case NETLINK_FIB_LOOKUP:
269 case NETLINK_NETFILTER:
270 case NETLINK_GENERIC:
277 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
278 struct net_device *dev)
280 struct sk_buff *nskb;
281 struct sock *sk = skb->sk;
284 if (!net_eq(dev_net(dev), sock_net(sk)))
289 if (is_vmalloc_addr(skb->head))
290 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
292 nskb = skb_clone(skb, GFP_ATOMIC);
295 nskb->protocol = htons((u16) sk->sk_protocol);
296 nskb->pkt_type = netlink_is_kernel(sk) ?
297 PACKET_KERNEL : PACKET_USER;
298 skb_reset_network_header(nskb);
299 ret = dev_queue_xmit(nskb);
300 if (unlikely(ret > 0))
301 ret = net_xmit_errno(ret);
308 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
311 struct netlink_tap *tmp;
313 if (!netlink_filter_tap(skb))
316 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
317 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
323 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
325 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
329 if (unlikely(!list_empty(&nn->netlink_tap_all)))
330 __netlink_deliver_tap(skb, nn);
335 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
338 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
339 netlink_deliver_tap(sock_net(dst), skb);
342 static void netlink_overrun(struct sock *sk)
344 struct netlink_sock *nlk = nlk_sk(sk);
346 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
347 if (!test_and_set_bit(NETLINK_S_CONGESTED,
348 &nlk_sk(sk)->state)) {
349 sk->sk_err = ENOBUFS;
350 sk->sk_error_report(sk);
353 atomic_inc(&sk->sk_drops);
356 static void netlink_rcv_wake(struct sock *sk)
358 struct netlink_sock *nlk = nlk_sk(sk);
360 if (skb_queue_empty(&sk->sk_receive_queue))
361 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
362 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
363 wake_up_interruptible(&nlk->wait);
366 static void netlink_skb_destructor(struct sk_buff *skb)
368 if (is_vmalloc_addr(skb->head)) {
370 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
379 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
381 WARN_ON(skb->sk != NULL);
383 skb->destructor = netlink_skb_destructor;
384 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
385 sk_mem_charge(sk, skb->truesize);
388 static void netlink_sock_destruct(struct sock *sk)
390 struct netlink_sock *nlk = nlk_sk(sk);
392 if (nlk->cb_running) {
394 nlk->cb.done(&nlk->cb);
395 module_put(nlk->cb.module);
396 kfree_skb(nlk->cb.skb);
399 skb_queue_purge(&sk->sk_receive_queue);
401 if (!sock_flag(sk, SOCK_DEAD)) {
402 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
406 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
407 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
408 WARN_ON(nlk_sk(sk)->groups);
411 static void netlink_sock_destruct_work(struct work_struct *work)
413 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
419 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
420 * SMP. Look, when several writers sleep and reader wakes them up, all but one
421 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
422 * this, _but_ remember, it adds useless work on UP machines.
425 void netlink_table_grab(void)
426 __acquires(nl_table_lock)
430 write_lock_irq(&nl_table_lock);
432 if (atomic_read(&nl_table_users)) {
433 DECLARE_WAITQUEUE(wait, current);
435 add_wait_queue_exclusive(&nl_table_wait, &wait);
437 set_current_state(TASK_UNINTERRUPTIBLE);
438 if (atomic_read(&nl_table_users) == 0)
440 write_unlock_irq(&nl_table_lock);
442 write_lock_irq(&nl_table_lock);
445 __set_current_state(TASK_RUNNING);
446 remove_wait_queue(&nl_table_wait, &wait);
450 void netlink_table_ungrab(void)
451 __releases(nl_table_lock)
453 write_unlock_irq(&nl_table_lock);
454 wake_up(&nl_table_wait);
458 netlink_lock_table(void)
460 /* read_lock() synchronizes us to netlink_table_grab */
462 read_lock(&nl_table_lock);
463 atomic_inc(&nl_table_users);
464 read_unlock(&nl_table_lock);
468 netlink_unlock_table(void)
470 if (atomic_dec_and_test(&nl_table_users))
471 wake_up(&nl_table_wait);
474 struct netlink_compare_arg
480 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
481 #define netlink_compare_arg_len \
482 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
484 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
487 const struct netlink_compare_arg *x = arg->key;
488 const struct netlink_sock *nlk = ptr;
490 return nlk->portid != x->portid ||
491 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
494 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
495 struct net *net, u32 portid)
497 memset(arg, 0, sizeof(*arg));
498 write_pnet(&arg->pnet, net);
499 arg->portid = portid;
502 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
505 struct netlink_compare_arg arg;
507 netlink_compare_arg_init(&arg, net, portid);
508 return rhashtable_lookup_fast(&table->hash, &arg,
509 netlink_rhashtable_params);
512 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
514 struct netlink_compare_arg arg;
516 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
517 return rhashtable_lookup_insert_key(&table->hash, &arg,
519 netlink_rhashtable_params);
522 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
524 struct netlink_table *table = &nl_table[protocol];
528 sk = __netlink_lookup(table, portid, net);
536 static const struct proto_ops netlink_ops;
539 netlink_update_listeners(struct sock *sk)
541 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
544 struct listeners *listeners;
546 listeners = nl_deref_protected(tbl->listeners);
550 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
552 sk_for_each_bound(sk, &tbl->mc_list) {
553 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
554 mask |= nlk_sk(sk)->groups[i];
556 listeners->masks[i] = mask;
558 /* this function is only called with the netlink table "grabbed", which
559 * makes sure updates are visible before bind or setsockopt return. */
562 static int netlink_insert(struct sock *sk, u32 portid)
564 struct netlink_table *table = &nl_table[sk->sk_protocol];
569 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
570 if (nlk_sk(sk)->bound)
573 nlk_sk(sk)->portid = portid;
576 err = __netlink_insert(table, sk);
578 /* In case the hashtable backend returns with -EBUSY
579 * from here, it must not escape to the caller.
581 if (unlikely(err == -EBUSY))
589 /* We need to ensure that the socket is hashed and visible. */
591 nlk_sk(sk)->bound = portid;
598 static void netlink_remove(struct sock *sk)
600 struct netlink_table *table;
602 table = &nl_table[sk->sk_protocol];
603 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
604 netlink_rhashtable_params)) {
605 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
609 netlink_table_grab();
610 if (nlk_sk(sk)->subscriptions) {
611 __sk_del_bind_node(sk);
612 netlink_update_listeners(sk);
614 if (sk->sk_protocol == NETLINK_GENERIC)
615 atomic_inc(&genl_sk_destructing_cnt);
616 netlink_table_ungrab();
619 static struct proto netlink_proto = {
621 .owner = THIS_MODULE,
622 .obj_size = sizeof(struct netlink_sock),
625 static int __netlink_create(struct net *net, struct socket *sock,
626 struct mutex *cb_mutex, int protocol,
630 struct netlink_sock *nlk;
632 sock->ops = &netlink_ops;
634 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
638 sock_init_data(sock, sk);
642 nlk->cb_mutex = cb_mutex;
644 nlk->cb_mutex = &nlk->cb_def_mutex;
645 mutex_init(nlk->cb_mutex);
646 lockdep_set_class_and_name(nlk->cb_mutex,
647 nlk_cb_mutex_keys + protocol,
648 nlk_cb_mutex_key_strings[protocol]);
650 init_waitqueue_head(&nlk->wait);
652 sk->sk_destruct = netlink_sock_destruct;
653 sk->sk_protocol = protocol;
657 static int netlink_create(struct net *net, struct socket *sock, int protocol,
660 struct module *module = NULL;
661 struct mutex *cb_mutex;
662 struct netlink_sock *nlk;
663 int (*bind)(struct net *net, int group);
664 void (*unbind)(struct net *net, int group);
667 sock->state = SS_UNCONNECTED;
669 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
670 return -ESOCKTNOSUPPORT;
672 if (protocol < 0 || protocol >= MAX_LINKS)
673 return -EPROTONOSUPPORT;
674 protocol = array_index_nospec(protocol, MAX_LINKS);
676 netlink_lock_table();
677 #ifdef CONFIG_MODULES
678 if (!nl_table[protocol].registered) {
679 netlink_unlock_table();
680 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
681 netlink_lock_table();
684 if (nl_table[protocol].registered &&
685 try_module_get(nl_table[protocol].module))
686 module = nl_table[protocol].module;
688 err = -EPROTONOSUPPORT;
689 cb_mutex = nl_table[protocol].cb_mutex;
690 bind = nl_table[protocol].bind;
691 unbind = nl_table[protocol].unbind;
692 netlink_unlock_table();
697 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
702 sock_prot_inuse_add(net, &netlink_proto, 1);
705 nlk = nlk_sk(sock->sk);
706 nlk->module = module;
707 nlk->netlink_bind = bind;
708 nlk->netlink_unbind = unbind;
717 static void deferred_put_nlk_sk(struct rcu_head *head)
719 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
720 struct sock *sk = &nlk->sk;
725 if (!refcount_dec_and_test(&sk->sk_refcnt))
728 if (nlk->cb_running && nlk->cb.done) {
729 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
730 schedule_work(&nlk->work);
737 static int netlink_release(struct socket *sock)
739 struct sock *sk = sock->sk;
740 struct netlink_sock *nlk;
750 * OK. Socket is unlinked, any packets that arrive now
754 /* must not acquire netlink_table_lock in any way again before unbind
755 * and notifying genetlink is done as otherwise it might deadlock
757 if (nlk->netlink_unbind) {
760 for (i = 0; i < nlk->ngroups; i++)
761 if (test_bit(i, nlk->groups))
762 nlk->netlink_unbind(sock_net(sk), i + 1);
764 if (sk->sk_protocol == NETLINK_GENERIC &&
765 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
766 wake_up(&genl_sk_destructing_waitq);
769 wake_up_interruptible_all(&nlk->wait);
771 skb_queue_purge(&sk->sk_write_queue);
773 if (nlk->portid && nlk->bound) {
774 struct netlink_notify n = {
776 .protocol = sk->sk_protocol,
777 .portid = nlk->portid,
779 blocking_notifier_call_chain(&netlink_chain,
780 NETLINK_URELEASE, &n);
783 module_put(nlk->module);
785 if (netlink_is_kernel(sk)) {
786 netlink_table_grab();
787 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
788 if (--nl_table[sk->sk_protocol].registered == 0) {
789 struct listeners *old;
791 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
792 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
794 nl_table[sk->sk_protocol].module = NULL;
795 nl_table[sk->sk_protocol].bind = NULL;
796 nl_table[sk->sk_protocol].unbind = NULL;
797 nl_table[sk->sk_protocol].flags = 0;
798 nl_table[sk->sk_protocol].registered = 0;
800 netlink_table_ungrab();
804 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
806 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
810 static int netlink_autobind(struct socket *sock)
812 struct sock *sk = sock->sk;
813 struct net *net = sock_net(sk);
814 struct netlink_table *table = &nl_table[sk->sk_protocol];
815 s32 portid = task_tgid_vnr(current);
823 ok = !__netlink_lookup(table, portid, net);
826 /* Bind collision, search negative portid values. */
828 /* rover will be in range [S32_MIN, -4097] */
829 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
830 else if (rover >= -4096)
836 err = netlink_insert(sk, portid);
837 if (err == -EADDRINUSE)
840 /* If 2 threads race to autobind, that is fine. */
848 * __netlink_ns_capable - General netlink message capability test
849 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
850 * @user_ns: The user namespace of the capability to use
851 * @cap: The capability to use
853 * Test to see if the opener of the socket we received the message
854 * from had when the netlink socket was created and the sender of the
855 * message has has the capability @cap in the user namespace @user_ns.
857 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
858 struct user_namespace *user_ns, int cap)
860 return ((nsp->flags & NETLINK_SKB_DST) ||
861 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
862 ns_capable(user_ns, cap);
864 EXPORT_SYMBOL(__netlink_ns_capable);
867 * netlink_ns_capable - General netlink message capability test
868 * @skb: socket buffer holding a netlink command from userspace
869 * @user_ns: The user namespace of the capability to use
870 * @cap: The capability to use
872 * Test to see if the opener of the socket we received the message
873 * from had when the netlink socket was created and the sender of the
874 * message has has the capability @cap in the user namespace @user_ns.
876 bool netlink_ns_capable(const struct sk_buff *skb,
877 struct user_namespace *user_ns, int cap)
879 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
881 EXPORT_SYMBOL(netlink_ns_capable);
884 * netlink_capable - Netlink global message capability test
885 * @skb: socket buffer holding a netlink command from userspace
886 * @cap: The capability to use
888 * Test to see if the opener of the socket we received the message
889 * from had when the netlink socket was created and the sender of the
890 * message has has the capability @cap in all user namespaces.
892 bool netlink_capable(const struct sk_buff *skb, int cap)
894 return netlink_ns_capable(skb, &init_user_ns, cap);
896 EXPORT_SYMBOL(netlink_capable);
899 * netlink_net_capable - Netlink network namespace message capability test
900 * @skb: socket buffer holding a netlink command from userspace
901 * @cap: The capability to use
903 * Test to see if the opener of the socket we received the message
904 * from had when the netlink socket was created and the sender of the
905 * message has has the capability @cap over the network namespace of
906 * the socket we received the message from.
908 bool netlink_net_capable(const struct sk_buff *skb, int cap)
910 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
912 EXPORT_SYMBOL(netlink_net_capable);
914 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
916 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
917 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
921 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
923 struct netlink_sock *nlk = nlk_sk(sk);
925 if (nlk->subscriptions && !subscriptions)
926 __sk_del_bind_node(sk);
927 else if (!nlk->subscriptions && subscriptions)
928 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
929 nlk->subscriptions = subscriptions;
932 static int netlink_realloc_groups(struct sock *sk)
934 struct netlink_sock *nlk = nlk_sk(sk);
936 unsigned long *new_groups;
939 netlink_table_grab();
941 groups = nl_table[sk->sk_protocol].groups;
942 if (!nl_table[sk->sk_protocol].registered) {
947 if (nlk->ngroups >= groups)
950 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
951 if (new_groups == NULL) {
955 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
956 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
958 nlk->groups = new_groups;
959 nlk->ngroups = groups;
961 netlink_table_ungrab();
965 static void netlink_undo_bind(int group, long unsigned int groups,
968 struct netlink_sock *nlk = nlk_sk(sk);
971 if (!nlk->netlink_unbind)
974 for (undo = 0; undo < group; undo++)
975 if (test_bit(undo, &groups))
976 nlk->netlink_unbind(sock_net(sk), undo + 1);
979 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
982 struct sock *sk = sock->sk;
983 struct net *net = sock_net(sk);
984 struct netlink_sock *nlk = nlk_sk(sk);
985 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
987 unsigned long groups;
990 if (addr_len < sizeof(struct sockaddr_nl))
993 if (nladdr->nl_family != AF_NETLINK)
995 groups = nladdr->nl_groups;
997 /* Only superuser is allowed to listen multicasts */
999 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1001 err = netlink_realloc_groups(sk);
1006 if (nlk->ngroups < BITS_PER_LONG)
1007 groups &= (1UL << nlk->ngroups) - 1;
1011 /* Ensure nlk->portid is up-to-date. */
1014 if (nladdr->nl_pid != nlk->portid)
1018 netlink_lock_table();
1019 if (nlk->netlink_bind && groups) {
1022 for (group = 0; group < nlk->ngroups; group++) {
1023 if (!test_bit(group, &groups))
1025 err = nlk->netlink_bind(net, group + 1);
1028 netlink_undo_bind(group, groups, sk);
1033 /* No need for barriers here as we return to user-space without
1034 * using any of the bound attributes.
1037 err = nladdr->nl_pid ?
1038 netlink_insert(sk, nladdr->nl_pid) :
1039 netlink_autobind(sock);
1041 netlink_undo_bind(nlk->ngroups, groups, sk);
1046 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1048 netlink_unlock_table();
1050 netlink_table_grab();
1051 netlink_update_subscriptions(sk, nlk->subscriptions +
1053 hweight32(nlk->groups[0]));
1054 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1055 netlink_update_listeners(sk);
1056 netlink_table_ungrab();
1061 netlink_unlock_table();
1065 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1066 int alen, int flags)
1069 struct sock *sk = sock->sk;
1070 struct netlink_sock *nlk = nlk_sk(sk);
1071 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1073 if (alen < sizeof(addr->sa_family))
1076 if (addr->sa_family == AF_UNSPEC) {
1077 sk->sk_state = NETLINK_UNCONNECTED;
1078 nlk->dst_portid = 0;
1082 if (addr->sa_family != AF_NETLINK)
1085 if (alen < sizeof(struct sockaddr_nl))
1088 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1089 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1092 /* No need for barriers here as we return to user-space without
1093 * using any of the bound attributes.
1096 err = netlink_autobind(sock);
1099 sk->sk_state = NETLINK_CONNECTED;
1100 nlk->dst_portid = nladdr->nl_pid;
1101 nlk->dst_group = ffs(nladdr->nl_groups);
1107 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1110 struct sock *sk = sock->sk;
1111 struct netlink_sock *nlk = nlk_sk(sk);
1112 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1114 nladdr->nl_family = AF_NETLINK;
1118 nladdr->nl_pid = nlk->dst_portid;
1119 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1121 nladdr->nl_pid = nlk->portid;
1122 netlink_lock_table();
1123 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1124 netlink_unlock_table();
1126 return sizeof(*nladdr);
1129 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1132 /* try to hand this ioctl down to the NIC drivers.
1134 return -ENOIOCTLCMD;
1137 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1140 struct netlink_sock *nlk;
1142 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1144 return ERR_PTR(-ECONNREFUSED);
1146 /* Don't bother queuing skb if kernel socket has no input function */
1148 if (sock->sk_state == NETLINK_CONNECTED &&
1149 nlk->dst_portid != nlk_sk(ssk)->portid) {
1151 return ERR_PTR(-ECONNREFUSED);
1156 struct sock *netlink_getsockbyfilp(struct file *filp)
1158 struct inode *inode = file_inode(filp);
1161 if (!S_ISSOCK(inode->i_mode))
1162 return ERR_PTR(-ENOTSOCK);
1164 sock = SOCKET_I(inode)->sk;
1165 if (sock->sk_family != AF_NETLINK)
1166 return ERR_PTR(-EINVAL);
1172 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1175 struct sk_buff *skb;
1178 if (size <= NLMSG_GOODSIZE || broadcast)
1179 return alloc_skb(size, GFP_KERNEL);
1181 size = SKB_DATA_ALIGN(size) +
1182 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1184 data = vmalloc(size);
1188 skb = __build_skb(data, size);
1192 skb->destructor = netlink_skb_destructor;
1198 * Attach a skb to a netlink socket.
1199 * The caller must hold a reference to the destination socket. On error, the
1200 * reference is dropped. The skb is not send to the destination, just all
1201 * all error checks are performed and memory in the queue is reserved.
1203 * < 0: error. skb freed, reference to sock dropped.
1205 * 1: repeat lookup - reference dropped while waiting for socket memory.
1207 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1208 long *timeo, struct sock *ssk)
1210 struct netlink_sock *nlk;
1214 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1215 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1216 DECLARE_WAITQUEUE(wait, current);
1218 if (!ssk || netlink_is_kernel(ssk))
1219 netlink_overrun(sk);
1225 __set_current_state(TASK_INTERRUPTIBLE);
1226 add_wait_queue(&nlk->wait, &wait);
1228 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1229 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1230 !sock_flag(sk, SOCK_DEAD))
1231 *timeo = schedule_timeout(*timeo);
1233 __set_current_state(TASK_RUNNING);
1234 remove_wait_queue(&nlk->wait, &wait);
1237 if (signal_pending(current)) {
1239 return sock_intr_errno(*timeo);
1243 netlink_skb_set_owner_r(skb, sk);
1247 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1251 netlink_deliver_tap(sock_net(sk), skb);
1253 skb_queue_tail(&sk->sk_receive_queue, skb);
1254 sk->sk_data_ready(sk);
1258 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1260 int len = __netlink_sendskb(sk, skb);
1266 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1272 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1276 WARN_ON(skb->sk != NULL);
1277 delta = skb->end - skb->tail;
1278 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1281 if (skb_shared(skb)) {
1282 struct sk_buff *nskb = skb_clone(skb, allocation);
1289 pskb_expand_head(skb, 0, -delta,
1290 (allocation & ~__GFP_DIRECT_RECLAIM) |
1291 __GFP_NOWARN | __GFP_NORETRY);
1295 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1299 struct netlink_sock *nlk = nlk_sk(sk);
1301 ret = -ECONNREFUSED;
1302 if (nlk->netlink_rcv != NULL) {
1304 netlink_skb_set_owner_r(skb, sk);
1305 NETLINK_CB(skb).sk = ssk;
1306 netlink_deliver_tap_kernel(sk, ssk, skb);
1307 nlk->netlink_rcv(skb);
1316 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1317 u32 portid, int nonblock)
1323 skb = netlink_trim(skb, gfp_any());
1325 timeo = sock_sndtimeo(ssk, nonblock);
1327 sk = netlink_getsockbyportid(ssk, portid);
1332 if (netlink_is_kernel(sk))
1333 return netlink_unicast_kernel(sk, skb, ssk);
1335 if (sk_filter(sk, skb)) {
1342 err = netlink_attachskb(sk, skb, &timeo, ssk);
1348 return netlink_sendskb(sk, skb);
1350 EXPORT_SYMBOL(netlink_unicast);
1352 int netlink_has_listeners(struct sock *sk, unsigned int group)
1355 struct listeners *listeners;
1357 BUG_ON(!netlink_is_kernel(sk));
1360 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1362 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1363 res = test_bit(group - 1, listeners->masks);
1369 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1371 bool netlink_strict_get_check(struct sk_buff *skb)
1373 const struct netlink_sock *nlk = nlk_sk(NETLINK_CB(skb).sk);
1375 return nlk->flags & NETLINK_F_STRICT_CHK;
1377 EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1379 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1381 struct netlink_sock *nlk = nlk_sk(sk);
1383 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1384 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1385 netlink_skb_set_owner_r(skb, sk);
1386 __netlink_sendskb(sk, skb);
1387 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1392 struct netlink_broadcast_data {
1393 struct sock *exclude_sk;
1398 int delivery_failure;
1402 struct sk_buff *skb, *skb2;
1403 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1407 static void do_one_broadcast(struct sock *sk,
1408 struct netlink_broadcast_data *p)
1410 struct netlink_sock *nlk = nlk_sk(sk);
1413 if (p->exclude_sk == sk)
1416 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1417 !test_bit(p->group - 1, nlk->groups))
1420 if (!net_eq(sock_net(sk), p->net)) {
1421 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1424 if (!peernet_has_id(sock_net(sk), p->net))
1427 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1433 netlink_overrun(sk);
1438 if (p->skb2 == NULL) {
1439 if (skb_shared(p->skb)) {
1440 p->skb2 = skb_clone(p->skb, p->allocation);
1442 p->skb2 = skb_get(p->skb);
1444 * skb ownership may have been set when
1445 * delivered to a previous socket.
1447 skb_orphan(p->skb2);
1450 if (p->skb2 == NULL) {
1451 netlink_overrun(sk);
1452 /* Clone failed. Notify ALL listeners. */
1454 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1455 p->delivery_failure = 1;
1458 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1463 if (sk_filter(sk, p->skb2)) {
1468 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1469 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1470 NETLINK_CB(p->skb2).nsid_is_set = true;
1471 val = netlink_broadcast_deliver(sk, p->skb2);
1473 netlink_overrun(sk);
1474 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1475 p->delivery_failure = 1;
1477 p->congested |= val;
1485 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1486 u32 group, gfp_t allocation,
1487 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1490 struct net *net = sock_net(ssk);
1491 struct netlink_broadcast_data info;
1494 skb = netlink_trim(skb, allocation);
1496 info.exclude_sk = ssk;
1498 info.portid = portid;
1501 info.delivery_failure = 0;
1504 info.allocation = allocation;
1507 info.tx_filter = filter;
1508 info.tx_data = filter_data;
1510 /* While we sleep in clone, do not allow to change socket list */
1512 netlink_lock_table();
1514 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1515 do_one_broadcast(sk, &info);
1519 netlink_unlock_table();
1521 if (info.delivery_failure) {
1522 kfree_skb(info.skb2);
1525 consume_skb(info.skb2);
1527 if (info.delivered) {
1528 if (info.congested && gfpflags_allow_blocking(allocation))
1534 EXPORT_SYMBOL(netlink_broadcast_filtered);
1536 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1537 u32 group, gfp_t allocation)
1539 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1542 EXPORT_SYMBOL(netlink_broadcast);
1544 struct netlink_set_err_data {
1545 struct sock *exclude_sk;
1551 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1553 struct netlink_sock *nlk = nlk_sk(sk);
1556 if (sk == p->exclude_sk)
1559 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1562 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1563 !test_bit(p->group - 1, nlk->groups))
1566 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1571 sk->sk_err = p->code;
1572 sk->sk_error_report(sk);
1578 * netlink_set_err - report error to broadcast listeners
1579 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1580 * @portid: the PORTID of a process that we want to skip (if any)
1581 * @group: the broadcast group that will notice the error
1582 * @code: error code, must be negative (as usual in kernelspace)
1584 * This function returns the number of broadcast listeners that have set the
1585 * NETLINK_NO_ENOBUFS socket option.
1587 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1589 struct netlink_set_err_data info;
1593 info.exclude_sk = ssk;
1594 info.portid = portid;
1596 /* sk->sk_err wants a positive error value */
1599 read_lock(&nl_table_lock);
1601 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1602 ret += do_one_set_err(sk, &info);
1604 read_unlock(&nl_table_lock);
1607 EXPORT_SYMBOL(netlink_set_err);
1609 /* must be called with netlink table grabbed */
1610 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1614 int old, new = !!is_new, subscriptions;
1616 old = test_bit(group - 1, nlk->groups);
1617 subscriptions = nlk->subscriptions - old + new;
1619 __set_bit(group - 1, nlk->groups);
1621 __clear_bit(group - 1, nlk->groups);
1622 netlink_update_subscriptions(&nlk->sk, subscriptions);
1623 netlink_update_listeners(&nlk->sk);
1626 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1627 char __user *optval, unsigned int optlen)
1629 struct sock *sk = sock->sk;
1630 struct netlink_sock *nlk = nlk_sk(sk);
1631 unsigned int val = 0;
1634 if (level != SOL_NETLINK)
1635 return -ENOPROTOOPT;
1637 if (optlen >= sizeof(int) &&
1638 get_user(val, (unsigned int __user *)optval))
1642 case NETLINK_PKTINFO:
1644 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1646 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1649 case NETLINK_ADD_MEMBERSHIP:
1650 case NETLINK_DROP_MEMBERSHIP: {
1651 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1653 err = netlink_realloc_groups(sk);
1656 if (!val || val - 1 >= nlk->ngroups)
1658 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1659 err = nlk->netlink_bind(sock_net(sk), val);
1663 netlink_table_grab();
1664 netlink_update_socket_mc(nlk, val,
1665 optname == NETLINK_ADD_MEMBERSHIP);
1666 netlink_table_ungrab();
1667 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1668 nlk->netlink_unbind(sock_net(sk), val);
1673 case NETLINK_BROADCAST_ERROR:
1675 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1677 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1680 case NETLINK_NO_ENOBUFS:
1682 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1683 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1684 wake_up_interruptible(&nlk->wait);
1686 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1690 case NETLINK_LISTEN_ALL_NSID:
1691 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1695 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1697 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1700 case NETLINK_CAP_ACK:
1702 nlk->flags |= NETLINK_F_CAP_ACK;
1704 nlk->flags &= ~NETLINK_F_CAP_ACK;
1707 case NETLINK_EXT_ACK:
1709 nlk->flags |= NETLINK_F_EXT_ACK;
1711 nlk->flags &= ~NETLINK_F_EXT_ACK;
1714 case NETLINK_GET_STRICT_CHK:
1716 nlk->flags |= NETLINK_F_STRICT_CHK;
1718 nlk->flags &= ~NETLINK_F_STRICT_CHK;
1727 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1728 char __user *optval, int __user *optlen)
1730 struct sock *sk = sock->sk;
1731 struct netlink_sock *nlk = nlk_sk(sk);
1734 if (level != SOL_NETLINK)
1735 return -ENOPROTOOPT;
1737 if (get_user(len, optlen))
1743 case NETLINK_PKTINFO:
1744 if (len < sizeof(int))
1747 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1748 if (put_user(len, optlen) ||
1749 put_user(val, optval))
1753 case NETLINK_BROADCAST_ERROR:
1754 if (len < sizeof(int))
1757 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1758 if (put_user(len, optlen) ||
1759 put_user(val, optval))
1763 case NETLINK_NO_ENOBUFS:
1764 if (len < sizeof(int))
1767 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1768 if (put_user(len, optlen) ||
1769 put_user(val, optval))
1773 case NETLINK_LIST_MEMBERSHIPS: {
1774 int pos, idx, shift;
1777 netlink_lock_table();
1778 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1779 if (len - pos < sizeof(u32))
1782 idx = pos / sizeof(unsigned long);
1783 shift = (pos % sizeof(unsigned long)) * 8;
1784 if (put_user((u32)(nlk->groups[idx] >> shift),
1785 (u32 __user *)(optval + pos))) {
1790 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1792 netlink_unlock_table();
1795 case NETLINK_CAP_ACK:
1796 if (len < sizeof(int))
1799 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1800 if (put_user(len, optlen) ||
1801 put_user(val, optval))
1805 case NETLINK_EXT_ACK:
1806 if (len < sizeof(int))
1809 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1810 if (put_user(len, optlen) || put_user(val, optval))
1814 case NETLINK_GET_STRICT_CHK:
1815 if (len < sizeof(int))
1818 val = nlk->flags & NETLINK_F_STRICT_CHK ? 1 : 0;
1819 if (put_user(len, optlen) || put_user(val, optval))
1829 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1831 struct nl_pktinfo info;
1833 info.group = NETLINK_CB(skb).dst_group;
1834 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1837 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1838 struct sk_buff *skb)
1840 if (!NETLINK_CB(skb).nsid_is_set)
1843 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1844 &NETLINK_CB(skb).nsid);
1847 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1849 struct sock *sk = sock->sk;
1850 struct netlink_sock *nlk = nlk_sk(sk);
1851 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1854 struct sk_buff *skb;
1856 struct scm_cookie scm;
1857 u32 netlink_skb_flags = 0;
1859 if (msg->msg_flags&MSG_OOB)
1862 err = scm_send(sock, msg, &scm, true);
1866 if (msg->msg_namelen) {
1868 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1870 if (addr->nl_family != AF_NETLINK)
1872 dst_portid = addr->nl_pid;
1873 dst_group = ffs(addr->nl_groups);
1875 if ((dst_group || dst_portid) &&
1876 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1878 netlink_skb_flags |= NETLINK_SKB_DST;
1880 dst_portid = nlk->dst_portid;
1881 dst_group = nlk->dst_group;
1885 err = netlink_autobind(sock);
1889 /* Ensure nlk is hashed and visible. */
1894 if (len > sk->sk_sndbuf - 32)
1897 skb = netlink_alloc_large_skb(len, dst_group);
1901 NETLINK_CB(skb).portid = nlk->portid;
1902 NETLINK_CB(skb).dst_group = dst_group;
1903 NETLINK_CB(skb).creds = scm.creds;
1904 NETLINK_CB(skb).flags = netlink_skb_flags;
1907 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1912 err = security_netlink_send(sk, skb);
1919 refcount_inc(&skb->users);
1920 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1922 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1929 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1932 struct scm_cookie scm;
1933 struct sock *sk = sock->sk;
1934 struct netlink_sock *nlk = nlk_sk(sk);
1935 int noblock = flags&MSG_DONTWAIT;
1937 struct sk_buff *skb, *data_skb;
1945 skb = skb_recv_datagram(sk, flags, noblock, &err);
1951 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1952 if (unlikely(skb_shinfo(skb)->frag_list)) {
1954 * If this skb has a frag_list, then here that means that we
1955 * will have to use the frag_list skb's data for compat tasks
1956 * and the regular skb's data for normal (non-compat) tasks.
1958 * If we need to send the compat skb, assign it to the
1959 * 'data_skb' variable so that it will be used below for data
1960 * copying. We keep 'skb' for everything else, including
1961 * freeing both later.
1963 if (flags & MSG_CMSG_COMPAT)
1964 data_skb = skb_shinfo(skb)->frag_list;
1968 /* Record the max length of recvmsg() calls for future allocations */
1969 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1970 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1971 SKB_WITH_OVERHEAD(32768));
1973 copied = data_skb->len;
1975 msg->msg_flags |= MSG_TRUNC;
1979 skb_reset_transport_header(data_skb);
1980 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1982 if (msg->msg_name) {
1983 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1984 addr->nl_family = AF_NETLINK;
1986 addr->nl_pid = NETLINK_CB(skb).portid;
1987 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1988 msg->msg_namelen = sizeof(*addr);
1991 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1992 netlink_cmsg_recv_pktinfo(msg, skb);
1993 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1994 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1996 memset(&scm, 0, sizeof(scm));
1997 scm.creds = *NETLINK_CREDS(skb);
1998 if (flags & MSG_TRUNC)
1999 copied = data_skb->len;
2001 skb_free_datagram(sk, skb);
2003 if (nlk->cb_running &&
2004 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2005 ret = netlink_dump(sk);
2008 sk->sk_error_report(sk);
2012 scm_recv(sock, msg, &scm, flags);
2014 netlink_rcv_wake(sk);
2015 return err ? : copied;
2018 static void netlink_data_ready(struct sock *sk)
2024 * We export these functions to other modules. They provide a
2025 * complete set of kernel non-blocking support for message
2030 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2031 struct netlink_kernel_cfg *cfg)
2033 struct socket *sock;
2035 struct netlink_sock *nlk;
2036 struct listeners *listeners = NULL;
2037 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2038 unsigned int groups;
2042 if (unit < 0 || unit >= MAX_LINKS)
2045 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2048 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2049 goto out_sock_release_nosk;
2053 if (!cfg || cfg->groups < 32)
2056 groups = cfg->groups;
2058 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2060 goto out_sock_release;
2062 sk->sk_data_ready = netlink_data_ready;
2063 if (cfg && cfg->input)
2064 nlk_sk(sk)->netlink_rcv = cfg->input;
2066 if (netlink_insert(sk, 0))
2067 goto out_sock_release;
2070 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2072 netlink_table_grab();
2073 if (!nl_table[unit].registered) {
2074 nl_table[unit].groups = groups;
2075 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2076 nl_table[unit].cb_mutex = cb_mutex;
2077 nl_table[unit].module = module;
2079 nl_table[unit].bind = cfg->bind;
2080 nl_table[unit].unbind = cfg->unbind;
2081 nl_table[unit].flags = cfg->flags;
2083 nl_table[unit].compare = cfg->compare;
2085 nl_table[unit].registered = 1;
2088 nl_table[unit].registered++;
2090 netlink_table_ungrab();
2095 netlink_kernel_release(sk);
2098 out_sock_release_nosk:
2102 EXPORT_SYMBOL(__netlink_kernel_create);
2105 netlink_kernel_release(struct sock *sk)
2107 if (sk == NULL || sk->sk_socket == NULL)
2110 sock_release(sk->sk_socket);
2112 EXPORT_SYMBOL(netlink_kernel_release);
2114 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2116 struct listeners *new, *old;
2117 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2122 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2123 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2126 old = nl_deref_protected(tbl->listeners);
2127 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2128 rcu_assign_pointer(tbl->listeners, new);
2130 kfree_rcu(old, rcu);
2132 tbl->groups = groups;
2138 * netlink_change_ngroups - change number of multicast groups
2140 * This changes the number of multicast groups that are available
2141 * on a certain netlink family. Note that it is not possible to
2142 * change the number of groups to below 32. Also note that it does
2143 * not implicitly call netlink_clear_multicast_users() when the
2144 * number of groups is reduced.
2146 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2147 * @groups: The new number of groups.
2149 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2153 netlink_table_grab();
2154 err = __netlink_change_ngroups(sk, groups);
2155 netlink_table_ungrab();
2160 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2163 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2165 sk_for_each_bound(sk, &tbl->mc_list)
2166 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2170 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2172 struct nlmsghdr *nlh;
2173 int size = nlmsg_msg_size(len);
2175 nlh = skb_put(skb, NLMSG_ALIGN(size));
2176 nlh->nlmsg_type = type;
2177 nlh->nlmsg_len = size;
2178 nlh->nlmsg_flags = flags;
2179 nlh->nlmsg_pid = portid;
2180 nlh->nlmsg_seq = seq;
2181 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2182 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2185 EXPORT_SYMBOL(__nlmsg_put);
2188 * It looks a bit ugly.
2189 * It would be better to create kernel thread.
2192 static int netlink_dump(struct sock *sk)
2194 struct netlink_sock *nlk = nlk_sk(sk);
2195 struct netlink_ext_ack extack = {};
2196 struct netlink_callback *cb;
2197 struct sk_buff *skb = NULL;
2198 struct nlmsghdr *nlh;
2199 struct module *module;
2204 mutex_lock(nlk->cb_mutex);
2205 if (!nlk->cb_running) {
2210 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2213 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2214 * required, but it makes sense to _attempt_ a 16K bytes allocation
2215 * to reduce number of system calls on dump operations, if user
2216 * ever provided a big enough buffer.
2219 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2221 if (alloc_min_size < nlk->max_recvmsg_len) {
2222 alloc_size = nlk->max_recvmsg_len;
2223 skb = alloc_skb(alloc_size,
2224 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2225 __GFP_NOWARN | __GFP_NORETRY);
2228 alloc_size = alloc_min_size;
2229 skb = alloc_skb(alloc_size, GFP_KERNEL);
2234 /* Trim skb to allocated size. User is expected to provide buffer as
2235 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2236 * netlink_recvmsg())). dump will pack as many smaller messages as
2237 * could fit within the allocated skb. skb is typically allocated
2238 * with larger space than required (could be as much as near 2x the
2239 * requested size with align to next power of 2 approach). Allowing
2240 * dump to use the excess space makes it difficult for a user to have a
2241 * reasonable static buffer based on the expected largest dump of a
2242 * single netdev. The outcome is MSG_TRUNC error.
2244 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2245 netlink_skb_set_owner_r(skb, sk);
2247 if (nlk->dump_done_errno > 0) {
2248 cb->extack = &extack;
2249 nlk->dump_done_errno = cb->dump(skb, cb);
2253 if (nlk->dump_done_errno > 0 ||
2254 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2255 mutex_unlock(nlk->cb_mutex);
2257 if (sk_filter(sk, skb))
2260 __netlink_sendskb(sk, skb);
2264 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE,
2265 sizeof(nlk->dump_done_errno),
2266 NLM_F_MULTI | cb->answer_flags);
2270 nl_dump_check_consistent(cb, nlh);
2272 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno,
2273 sizeof(nlk->dump_done_errno));
2275 if (extack._msg && nlk->flags & NETLINK_F_EXT_ACK) {
2276 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2277 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack._msg))
2278 nlmsg_end(skb, nlh);
2281 if (sk_filter(sk, skb))
2284 __netlink_sendskb(sk, skb);
2289 nlk->cb_running = false;
2290 module = cb->module;
2292 mutex_unlock(nlk->cb_mutex);
2298 mutex_unlock(nlk->cb_mutex);
2303 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2304 const struct nlmsghdr *nlh,
2305 struct netlink_dump_control *control)
2307 struct netlink_sock *nlk, *nlk2;
2308 struct netlink_callback *cb;
2312 refcount_inc(&skb->users);
2314 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2316 ret = -ECONNREFUSED;
2321 mutex_lock(nlk->cb_mutex);
2322 /* A dump is in progress... */
2323 if (nlk->cb_running) {
2327 /* add reference of module which cb->dump belongs to */
2328 if (!try_module_get(control->module)) {
2329 ret = -EPROTONOSUPPORT;
2334 memset(cb, 0, sizeof(*cb));
2335 cb->dump = control->dump;
2336 cb->done = control->done;
2338 cb->data = control->data;
2339 cb->module = control->module;
2340 cb->min_dump_alloc = control->min_dump_alloc;
2343 nlk2 = nlk_sk(NETLINK_CB(skb).sk);
2344 cb->strict_check = !!(nlk2->flags & NETLINK_F_STRICT_CHK);
2346 if (control->start) {
2347 ret = control->start(cb);
2352 nlk->cb_running = true;
2353 nlk->dump_done_errno = INT_MAX;
2355 mutex_unlock(nlk->cb_mutex);
2357 ret = netlink_dump(sk);
2364 /* We successfully started a dump, by returning -EINTR we
2365 * signal not to send ACK even if it was requested.
2370 module_put(control->module);
2373 mutex_unlock(nlk->cb_mutex);
2378 EXPORT_SYMBOL(__netlink_dump_start);
2380 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2381 const struct netlink_ext_ack *extack)
2383 struct sk_buff *skb;
2384 struct nlmsghdr *rep;
2385 struct nlmsgerr *errmsg;
2386 size_t payload = sizeof(*errmsg);
2388 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2389 unsigned int flags = 0;
2390 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2392 /* Error messages get the original request appened, unless the user
2393 * requests to cap the error message, and get extra error data if
2396 if (nlk_has_extack && extack && extack->_msg)
2397 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2400 if (!(nlk->flags & NETLINK_F_CAP_ACK))
2401 payload += nlmsg_len(nlh);
2403 flags |= NLM_F_CAPPED;
2404 if (nlk_has_extack && extack && extack->bad_attr)
2405 tlvlen += nla_total_size(sizeof(u32));
2407 flags |= NLM_F_CAPPED;
2409 if (nlk_has_extack && extack && extack->cookie_len)
2410 tlvlen += nla_total_size(extack->cookie_len);
2414 flags |= NLM_F_ACK_TLVS;
2416 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2418 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2419 NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk);
2423 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2424 NLMSG_ERROR, payload, flags);
2425 errmsg = nlmsg_data(rep);
2426 errmsg->error = err;
2427 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2429 if (nlk_has_extack && extack) {
2431 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2435 if (extack->bad_attr &&
2436 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2437 (u8 *)extack->bad_attr >= in_skb->data +
2439 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2440 (u8 *)extack->bad_attr -
2443 if (extack->cookie_len)
2444 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2450 nlmsg_end(skb, rep);
2452 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2454 EXPORT_SYMBOL(netlink_ack);
2456 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2458 struct netlink_ext_ack *))
2460 struct netlink_ext_ack extack;
2461 struct nlmsghdr *nlh;
2464 while (skb->len >= nlmsg_total_size(0)) {
2467 memset(&extack, 0, sizeof(extack));
2468 nlh = nlmsg_hdr(skb);
2471 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2474 /* Only requests are handled by the kernel */
2475 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2478 /* Skip control messages */
2479 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2482 err = cb(skb, nlh, &extack);
2487 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2488 netlink_ack(skb, nlh, err, &extack);
2491 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2492 if (msglen > skb->len)
2494 skb_pull(skb, msglen);
2499 EXPORT_SYMBOL(netlink_rcv_skb);
2502 * nlmsg_notify - send a notification netlink message
2503 * @sk: netlink socket to use
2504 * @skb: notification message
2505 * @portid: destination netlink portid for reports or 0
2506 * @group: destination multicast group or 0
2507 * @report: 1 to report back, 0 to disable
2508 * @flags: allocation flags
2510 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2511 unsigned int group, int report, gfp_t flags)
2516 int exclude_portid = 0;
2519 refcount_inc(&skb->users);
2520 exclude_portid = portid;
2523 /* errors reported via destination sk->sk_err, but propagate
2524 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2525 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2531 err2 = nlmsg_unicast(sk, skb, portid);
2532 if (!err || err == -ESRCH)
2538 EXPORT_SYMBOL(nlmsg_notify);
2540 #ifdef CONFIG_PROC_FS
2541 struct nl_seq_iter {
2542 struct seq_net_private p;
2543 struct rhashtable_iter hti;
2547 static int netlink_walk_start(struct nl_seq_iter *iter)
2549 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2550 rhashtable_walk_start(&iter->hti);
2555 static void netlink_walk_stop(struct nl_seq_iter *iter)
2557 rhashtable_walk_stop(&iter->hti);
2558 rhashtable_walk_exit(&iter->hti);
2561 static void *__netlink_seq_next(struct seq_file *seq)
2563 struct nl_seq_iter *iter = seq->private;
2564 struct netlink_sock *nlk;
2570 nlk = rhashtable_walk_next(&iter->hti);
2573 if (PTR_ERR(nlk) == -EAGAIN)
2582 netlink_walk_stop(iter);
2583 if (++iter->link >= MAX_LINKS)
2586 err = netlink_walk_start(iter);
2588 return ERR_PTR(err);
2590 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2595 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2597 struct nl_seq_iter *iter = seq->private;
2598 void *obj = SEQ_START_TOKEN;
2604 err = netlink_walk_start(iter);
2606 return ERR_PTR(err);
2608 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2609 obj = __netlink_seq_next(seq);
2614 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2617 return __netlink_seq_next(seq);
2620 static void netlink_seq_stop(struct seq_file *seq, void *v)
2622 struct nl_seq_iter *iter = seq->private;
2624 if (iter->link >= MAX_LINKS)
2627 netlink_walk_stop(iter);
2631 static int netlink_seq_show(struct seq_file *seq, void *v)
2633 if (v == SEQ_START_TOKEN) {
2635 "sk Eth Pid Groups "
2636 "Rmem Wmem Dump Locks Drops Inode\n");
2639 struct netlink_sock *nlk = nlk_sk(s);
2641 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2645 nlk->groups ? (u32)nlk->groups[0] : 0,
2646 sk_rmem_alloc_get(s),
2647 sk_wmem_alloc_get(s),
2649 refcount_read(&s->sk_refcnt),
2650 atomic_read(&s->sk_drops),
2658 static const struct seq_operations netlink_seq_ops = {
2659 .start = netlink_seq_start,
2660 .next = netlink_seq_next,
2661 .stop = netlink_seq_stop,
2662 .show = netlink_seq_show,
2666 int netlink_register_notifier(struct notifier_block *nb)
2668 return blocking_notifier_chain_register(&netlink_chain, nb);
2670 EXPORT_SYMBOL(netlink_register_notifier);
2672 int netlink_unregister_notifier(struct notifier_block *nb)
2674 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2676 EXPORT_SYMBOL(netlink_unregister_notifier);
2678 static const struct proto_ops netlink_ops = {
2679 .family = PF_NETLINK,
2680 .owner = THIS_MODULE,
2681 .release = netlink_release,
2682 .bind = netlink_bind,
2683 .connect = netlink_connect,
2684 .socketpair = sock_no_socketpair,
2685 .accept = sock_no_accept,
2686 .getname = netlink_getname,
2687 .poll = datagram_poll,
2688 .ioctl = netlink_ioctl,
2689 .listen = sock_no_listen,
2690 .shutdown = sock_no_shutdown,
2691 .setsockopt = netlink_setsockopt,
2692 .getsockopt = netlink_getsockopt,
2693 .sendmsg = netlink_sendmsg,
2694 .recvmsg = netlink_recvmsg,
2695 .mmap = sock_no_mmap,
2696 .sendpage = sock_no_sendpage,
2699 static const struct net_proto_family netlink_family_ops = {
2700 .family = PF_NETLINK,
2701 .create = netlink_create,
2702 .owner = THIS_MODULE, /* for consistency 8) */
2705 static int __net_init netlink_net_init(struct net *net)
2707 #ifdef CONFIG_PROC_FS
2708 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2709 sizeof(struct nl_seq_iter)))
2715 static void __net_exit netlink_net_exit(struct net *net)
2717 #ifdef CONFIG_PROC_FS
2718 remove_proc_entry("netlink", net->proc_net);
2722 static void __init netlink_add_usersock_entry(void)
2724 struct listeners *listeners;
2727 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2729 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2731 netlink_table_grab();
2733 nl_table[NETLINK_USERSOCK].groups = groups;
2734 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2735 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2736 nl_table[NETLINK_USERSOCK].registered = 1;
2737 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2739 netlink_table_ungrab();
2742 static struct pernet_operations __net_initdata netlink_net_ops = {
2743 .init = netlink_net_init,
2744 .exit = netlink_net_exit,
2747 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2749 const struct netlink_sock *nlk = data;
2750 struct netlink_compare_arg arg;
2752 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2753 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2756 static const struct rhashtable_params netlink_rhashtable_params = {
2757 .head_offset = offsetof(struct netlink_sock, node),
2758 .key_len = netlink_compare_arg_len,
2759 .obj_hashfn = netlink_hash,
2760 .obj_cmpfn = netlink_compare,
2761 .automatic_shrinking = true,
2764 static int __init netlink_proto_init(void)
2767 int err = proto_register(&netlink_proto, 0);
2772 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2774 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2778 for (i = 0; i < MAX_LINKS; i++) {
2779 if (rhashtable_init(&nl_table[i].hash,
2780 &netlink_rhashtable_params) < 0) {
2782 rhashtable_destroy(&nl_table[i].hash);
2788 netlink_add_usersock_entry();
2790 sock_register(&netlink_family_ops);
2791 register_pernet_subsys(&netlink_net_ops);
2792 register_pernet_subsys(&netlink_tap_net_ops);
2793 /* The netlink device handler may be needed early. */
2798 panic("netlink_init: Cannot allocate nl_table\n");
2801 core_initcall(netlink_proto_init);