2 * NETLINK Kernel-user communication protocol.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6 * Patrick McHardy <kaber@trash.net>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
13 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
14 * added netlink_proto_exit
15 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
16 * use nlk_sk, as sk->protinfo is on a diet 8)
17 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
18 * - inc module use count of module that owns
19 * the kernel socket in case userspace opens
20 * socket of same protocol
21 * - remove all module support, since netlink is
22 * mandatory if CONFIG_NET=y these days
25 #include <linux/module.h>
27 #include <linux/capability.h>
28 #include <linux/kernel.h>
29 #include <linux/init.h>
30 #include <linux/signal.h>
31 #include <linux/sched.h>
32 #include <linux/errno.h>
33 #include <linux/string.h>
34 #include <linux/stat.h>
35 #include <linux/socket.h>
37 #include <linux/fcntl.h>
38 #include <linux/termios.h>
39 #include <linux/sockios.h>
40 #include <linux/net.h>
42 #include <linux/slab.h>
43 #include <linux/uaccess.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/rtnetlink.h>
47 #include <linux/proc_fs.h>
48 #include <linux/seq_file.h>
49 #include <linux/notifier.h>
50 #include <linux/security.h>
51 #include <linux/jhash.h>
52 #include <linux/jiffies.h>
53 #include <linux/random.h>
54 #include <linux/bitops.h>
56 #include <linux/types.h>
57 #include <linux/audit.h>
58 #include <linux/mutex.h>
59 #include <linux/vmalloc.h>
60 #include <linux/if_arp.h>
61 #include <linux/rhashtable.h>
62 #include <asm/cacheflush.h>
63 #include <linux/hash.h>
64 #include <linux/genetlink.h>
65 #include <linux/net_namespace.h>
66 #include <linux/nospec.h>
68 #include <net/net_namespace.h>
69 #include <net/netns/generic.h>
72 #include <net/netlink.h>
74 #include "af_netlink.h"
78 unsigned long masks[0];
82 #define NETLINK_S_CONGESTED 0x0
84 static inline int netlink_is_kernel(struct sock *sk)
86 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
89 struct netlink_table *nl_table __read_mostly;
90 EXPORT_SYMBOL_GPL(nl_table);
92 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
94 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
96 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
99 "nlk_cb_mutex-USERSOCK",
100 "nlk_cb_mutex-FIREWALL",
101 "nlk_cb_mutex-SOCK_DIAG",
102 "nlk_cb_mutex-NFLOG",
104 "nlk_cb_mutex-SELINUX",
105 "nlk_cb_mutex-ISCSI",
106 "nlk_cb_mutex-AUDIT",
107 "nlk_cb_mutex-FIB_LOOKUP",
108 "nlk_cb_mutex-CONNECTOR",
109 "nlk_cb_mutex-NETFILTER",
110 "nlk_cb_mutex-IP6_FW",
111 "nlk_cb_mutex-DNRTMSG",
112 "nlk_cb_mutex-KOBJECT_UEVENT",
113 "nlk_cb_mutex-GENERIC",
115 "nlk_cb_mutex-SCSITRANSPORT",
116 "nlk_cb_mutex-ECRYPTFS",
118 "nlk_cb_mutex-CRYPTO",
129 "nlk_cb_mutex-MAX_LINKS"
132 static int netlink_dump(struct sock *sk);
134 /* nl_table locking explained:
135 * Lookup and traversal are protected with an RCU read-side lock. Insertion
136 * and removal are protected with per bucket lock while using RCU list
137 * modification primitives and may run in parallel to RCU protected lookups.
138 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
139 * been acquired * either during or after the socket has been removed from
140 * the list and after an RCU grace period.
142 DEFINE_RWLOCK(nl_table_lock);
143 EXPORT_SYMBOL_GPL(nl_table_lock);
144 static atomic_t nl_table_users = ATOMIC_INIT(0);
146 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
148 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
151 static const struct rhashtable_params netlink_rhashtable_params;
153 static inline u32 netlink_group_mask(u32 group)
157 return group ? 1 << (group - 1) : 0;
160 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
163 unsigned int len = skb_end_offset(skb);
166 new = alloc_skb(len, gfp_mask);
170 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
171 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
172 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
174 skb_put_data(new, skb->data, len);
178 static unsigned int netlink_tap_net_id;
180 struct netlink_tap_net {
181 struct list_head netlink_tap_all;
182 struct mutex netlink_tap_lock;
185 int netlink_add_tap(struct netlink_tap *nt)
187 struct net *net = dev_net(nt->dev);
188 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
190 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
193 mutex_lock(&nn->netlink_tap_lock);
194 list_add_rcu(&nt->list, &nn->netlink_tap_all);
195 mutex_unlock(&nn->netlink_tap_lock);
197 __module_get(nt->module);
201 EXPORT_SYMBOL_GPL(netlink_add_tap);
203 static int __netlink_remove_tap(struct netlink_tap *nt)
205 struct net *net = dev_net(nt->dev);
206 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
208 struct netlink_tap *tmp;
210 mutex_lock(&nn->netlink_tap_lock);
212 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
214 list_del_rcu(&nt->list);
220 pr_warn("__netlink_remove_tap: %p not found\n", nt);
222 mutex_unlock(&nn->netlink_tap_lock);
225 module_put(nt->module);
227 return found ? 0 : -ENODEV;
230 int netlink_remove_tap(struct netlink_tap *nt)
234 ret = __netlink_remove_tap(nt);
239 EXPORT_SYMBOL_GPL(netlink_remove_tap);
241 static __net_init int netlink_tap_init_net(struct net *net)
243 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
245 INIT_LIST_HEAD(&nn->netlink_tap_all);
246 mutex_init(&nn->netlink_tap_lock);
250 static void __net_exit netlink_tap_exit_net(struct net *net)
254 static struct pernet_operations netlink_tap_net_ops = {
255 .init = netlink_tap_init_net,
256 .exit = netlink_tap_exit_net,
257 .id = &netlink_tap_net_id,
258 .size = sizeof(struct netlink_tap_net),
261 static bool netlink_filter_tap(const struct sk_buff *skb)
263 struct sock *sk = skb->sk;
265 /* We take the more conservative approach and
266 * whitelist socket protocols that may pass.
268 switch (sk->sk_protocol) {
270 case NETLINK_USERSOCK:
271 case NETLINK_SOCK_DIAG:
274 case NETLINK_FIB_LOOKUP:
275 case NETLINK_NETFILTER:
276 case NETLINK_GENERIC:
283 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
284 struct net_device *dev)
286 struct sk_buff *nskb;
287 struct sock *sk = skb->sk;
290 if (!net_eq(dev_net(dev), sock_net(sk)))
295 if (is_vmalloc_addr(skb->head))
296 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
298 nskb = skb_clone(skb, GFP_ATOMIC);
301 nskb->protocol = htons((u16) sk->sk_protocol);
302 nskb->pkt_type = netlink_is_kernel(sk) ?
303 PACKET_KERNEL : PACKET_USER;
304 skb_reset_network_header(nskb);
305 ret = dev_queue_xmit(nskb);
306 if (unlikely(ret > 0))
307 ret = net_xmit_errno(ret);
314 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
317 struct netlink_tap *tmp;
319 if (!netlink_filter_tap(skb))
322 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
323 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
329 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
331 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
335 if (unlikely(!list_empty(&nn->netlink_tap_all)))
336 __netlink_deliver_tap(skb, nn);
341 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
344 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
345 netlink_deliver_tap(sock_net(dst), skb);
348 static void netlink_overrun(struct sock *sk)
350 struct netlink_sock *nlk = nlk_sk(sk);
352 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
353 if (!test_and_set_bit(NETLINK_S_CONGESTED,
354 &nlk_sk(sk)->state)) {
355 sk->sk_err = ENOBUFS;
356 sk->sk_error_report(sk);
359 atomic_inc(&sk->sk_drops);
362 static void netlink_rcv_wake(struct sock *sk)
364 struct netlink_sock *nlk = nlk_sk(sk);
366 if (skb_queue_empty(&sk->sk_receive_queue))
367 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
368 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
369 wake_up_interruptible(&nlk->wait);
372 static void netlink_skb_destructor(struct sk_buff *skb)
374 if (is_vmalloc_addr(skb->head)) {
376 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
385 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
387 WARN_ON(skb->sk != NULL);
389 skb->destructor = netlink_skb_destructor;
390 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
391 sk_mem_charge(sk, skb->truesize);
394 static void netlink_sock_destruct(struct sock *sk)
396 struct netlink_sock *nlk = nlk_sk(sk);
398 if (nlk->cb_running) {
400 nlk->cb.done(&nlk->cb);
401 module_put(nlk->cb.module);
402 kfree_skb(nlk->cb.skb);
405 skb_queue_purge(&sk->sk_receive_queue);
407 if (!sock_flag(sk, SOCK_DEAD)) {
408 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
412 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
413 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
414 WARN_ON(nlk_sk(sk)->groups);
417 static void netlink_sock_destruct_work(struct work_struct *work)
419 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
425 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
426 * SMP. Look, when several writers sleep and reader wakes them up, all but one
427 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
428 * this, _but_ remember, it adds useless work on UP machines.
431 void netlink_table_grab(void)
432 __acquires(nl_table_lock)
436 write_lock_irq(&nl_table_lock);
438 if (atomic_read(&nl_table_users)) {
439 DECLARE_WAITQUEUE(wait, current);
441 add_wait_queue_exclusive(&nl_table_wait, &wait);
443 set_current_state(TASK_UNINTERRUPTIBLE);
444 if (atomic_read(&nl_table_users) == 0)
446 write_unlock_irq(&nl_table_lock);
448 write_lock_irq(&nl_table_lock);
451 __set_current_state(TASK_RUNNING);
452 remove_wait_queue(&nl_table_wait, &wait);
456 void netlink_table_ungrab(void)
457 __releases(nl_table_lock)
459 write_unlock_irq(&nl_table_lock);
460 wake_up(&nl_table_wait);
464 netlink_lock_table(void)
468 /* read_lock() synchronizes us to netlink_table_grab */
470 read_lock_irqsave(&nl_table_lock, flags);
471 atomic_inc(&nl_table_users);
472 read_unlock_irqrestore(&nl_table_lock, flags);
476 netlink_unlock_table(void)
478 if (atomic_dec_and_test(&nl_table_users))
479 wake_up(&nl_table_wait);
482 struct netlink_compare_arg
488 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
489 #define netlink_compare_arg_len \
490 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
492 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
495 const struct netlink_compare_arg *x = arg->key;
496 const struct netlink_sock *nlk = ptr;
498 return nlk->portid != x->portid ||
499 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
502 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
503 struct net *net, u32 portid)
505 memset(arg, 0, sizeof(*arg));
506 write_pnet(&arg->pnet, net);
507 arg->portid = portid;
510 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
513 struct netlink_compare_arg arg;
515 netlink_compare_arg_init(&arg, net, portid);
516 return rhashtable_lookup_fast(&table->hash, &arg,
517 netlink_rhashtable_params);
520 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
522 struct netlink_compare_arg arg;
524 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
525 return rhashtable_lookup_insert_key(&table->hash, &arg,
527 netlink_rhashtable_params);
530 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
532 struct netlink_table *table = &nl_table[protocol];
536 sk = __netlink_lookup(table, portid, net);
544 static const struct proto_ops netlink_ops;
547 netlink_update_listeners(struct sock *sk)
549 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
552 struct listeners *listeners;
554 listeners = nl_deref_protected(tbl->listeners);
558 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
560 sk_for_each_bound(sk, &tbl->mc_list) {
561 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
562 mask |= nlk_sk(sk)->groups[i];
564 listeners->masks[i] = mask;
566 /* this function is only called with the netlink table "grabbed", which
567 * makes sure updates are visible before bind or setsockopt return. */
570 static int netlink_insert(struct sock *sk, u32 portid)
572 struct netlink_table *table = &nl_table[sk->sk_protocol];
577 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
578 if (nlk_sk(sk)->bound)
582 if (BITS_PER_LONG > 32 &&
583 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
586 nlk_sk(sk)->portid = portid;
589 err = __netlink_insert(table, sk);
591 /* In case the hashtable backend returns with -EBUSY
592 * from here, it must not escape to the caller.
594 if (unlikely(err == -EBUSY))
602 /* We need to ensure that the socket is hashed and visible. */
604 /* Paired with lockless reads from netlink_bind(),
605 * netlink_connect() and netlink_sendmsg().
607 WRITE_ONCE(nlk_sk(sk)->bound, portid);
614 static void netlink_remove(struct sock *sk)
616 struct netlink_table *table;
618 table = &nl_table[sk->sk_protocol];
619 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
620 netlink_rhashtable_params)) {
621 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
625 netlink_table_grab();
626 if (nlk_sk(sk)->subscriptions) {
627 __sk_del_bind_node(sk);
628 netlink_update_listeners(sk);
630 if (sk->sk_protocol == NETLINK_GENERIC)
631 atomic_inc(&genl_sk_destructing_cnt);
632 netlink_table_ungrab();
635 static struct proto netlink_proto = {
637 .owner = THIS_MODULE,
638 .obj_size = sizeof(struct netlink_sock),
641 static int __netlink_create(struct net *net, struct socket *sock,
642 struct mutex *cb_mutex, int protocol,
646 struct netlink_sock *nlk;
648 sock->ops = &netlink_ops;
650 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
654 sock_init_data(sock, sk);
658 nlk->cb_mutex = cb_mutex;
660 nlk->cb_mutex = &nlk->cb_def_mutex;
661 mutex_init(nlk->cb_mutex);
662 lockdep_set_class_and_name(nlk->cb_mutex,
663 nlk_cb_mutex_keys + protocol,
664 nlk_cb_mutex_key_strings[protocol]);
666 init_waitqueue_head(&nlk->wait);
668 sk->sk_destruct = netlink_sock_destruct;
669 sk->sk_protocol = protocol;
673 static int netlink_create(struct net *net, struct socket *sock, int protocol,
676 struct module *module = NULL;
677 struct mutex *cb_mutex;
678 struct netlink_sock *nlk;
679 int (*bind)(struct net *net, int group);
680 void (*unbind)(struct net *net, int group);
683 sock->state = SS_UNCONNECTED;
685 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
686 return -ESOCKTNOSUPPORT;
688 if (protocol < 0 || protocol >= MAX_LINKS)
689 return -EPROTONOSUPPORT;
690 protocol = array_index_nospec(protocol, MAX_LINKS);
692 netlink_lock_table();
693 #ifdef CONFIG_MODULES
694 if (!nl_table[protocol].registered) {
695 netlink_unlock_table();
696 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
697 netlink_lock_table();
700 if (nl_table[protocol].registered &&
701 try_module_get(nl_table[protocol].module))
702 module = nl_table[protocol].module;
704 err = -EPROTONOSUPPORT;
705 cb_mutex = nl_table[protocol].cb_mutex;
706 bind = nl_table[protocol].bind;
707 unbind = nl_table[protocol].unbind;
708 netlink_unlock_table();
713 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
718 sock_prot_inuse_add(net, &netlink_proto, 1);
721 nlk = nlk_sk(sock->sk);
722 nlk->module = module;
723 nlk->netlink_bind = bind;
724 nlk->netlink_unbind = unbind;
733 static void deferred_put_nlk_sk(struct rcu_head *head)
735 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
736 struct sock *sk = &nlk->sk;
741 if (!refcount_dec_and_test(&sk->sk_refcnt))
744 if (nlk->cb_running && nlk->cb.done) {
745 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
746 schedule_work(&nlk->work);
753 static int netlink_release(struct socket *sock)
755 struct sock *sk = sock->sk;
756 struct netlink_sock *nlk;
766 * OK. Socket is unlinked, any packets that arrive now
770 /* must not acquire netlink_table_lock in any way again before unbind
771 * and notifying genetlink is done as otherwise it might deadlock
773 if (nlk->netlink_unbind) {
776 for (i = 0; i < nlk->ngroups; i++)
777 if (test_bit(i, nlk->groups))
778 nlk->netlink_unbind(sock_net(sk), i + 1);
780 if (sk->sk_protocol == NETLINK_GENERIC &&
781 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
782 wake_up(&genl_sk_destructing_waitq);
785 wake_up_interruptible_all(&nlk->wait);
787 skb_queue_purge(&sk->sk_write_queue);
789 if (nlk->portid && nlk->bound) {
790 struct netlink_notify n = {
792 .protocol = sk->sk_protocol,
793 .portid = nlk->portid,
795 blocking_notifier_call_chain(&netlink_chain,
796 NETLINK_URELEASE, &n);
799 module_put(nlk->module);
801 if (netlink_is_kernel(sk)) {
802 netlink_table_grab();
803 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
804 if (--nl_table[sk->sk_protocol].registered == 0) {
805 struct listeners *old;
807 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
808 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
810 nl_table[sk->sk_protocol].module = NULL;
811 nl_table[sk->sk_protocol].bind = NULL;
812 nl_table[sk->sk_protocol].unbind = NULL;
813 nl_table[sk->sk_protocol].flags = 0;
814 nl_table[sk->sk_protocol].registered = 0;
816 netlink_table_ungrab();
820 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
822 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
826 static int netlink_autobind(struct socket *sock)
828 struct sock *sk = sock->sk;
829 struct net *net = sock_net(sk);
830 struct netlink_table *table = &nl_table[sk->sk_protocol];
831 s32 portid = task_tgid_vnr(current);
839 ok = !__netlink_lookup(table, portid, net);
842 /* Bind collision, search negative portid values. */
844 /* rover will be in range [S32_MIN, -4097] */
845 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
846 else if (rover >= -4096)
852 err = netlink_insert(sk, portid);
853 if (err == -EADDRINUSE)
856 /* If 2 threads race to autobind, that is fine. */
864 * __netlink_ns_capable - General netlink message capability test
865 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
866 * @user_ns: The user namespace of the capability to use
867 * @cap: The capability to use
869 * Test to see if the opener of the socket we received the message
870 * from had when the netlink socket was created and the sender of the
871 * message has has the capability @cap in the user namespace @user_ns.
873 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
874 struct user_namespace *user_ns, int cap)
876 return ((nsp->flags & NETLINK_SKB_DST) ||
877 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
878 ns_capable(user_ns, cap);
880 EXPORT_SYMBOL(__netlink_ns_capable);
883 * netlink_ns_capable - General netlink message capability test
884 * @skb: socket buffer holding a netlink command from userspace
885 * @user_ns: The user namespace of the capability to use
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 the user namespace @user_ns.
892 bool netlink_ns_capable(const struct sk_buff *skb,
893 struct user_namespace *user_ns, int cap)
895 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
897 EXPORT_SYMBOL(netlink_ns_capable);
900 * netlink_capable - Netlink global message capability test
901 * @skb: socket buffer holding a netlink command from userspace
902 * @cap: The capability to use
904 * Test to see if the opener of the socket we received the message
905 * from had when the netlink socket was created and the sender of the
906 * message has has the capability @cap in all user namespaces.
908 bool netlink_capable(const struct sk_buff *skb, int cap)
910 return netlink_ns_capable(skb, &init_user_ns, cap);
912 EXPORT_SYMBOL(netlink_capable);
915 * netlink_net_capable - Netlink network namespace message capability test
916 * @skb: socket buffer holding a netlink command from userspace
917 * @cap: The capability to use
919 * Test to see if the opener of the socket we received the message
920 * from had when the netlink socket was created and the sender of the
921 * message has has the capability @cap over the network namespace of
922 * the socket we received the message from.
924 bool netlink_net_capable(const struct sk_buff *skb, int cap)
926 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
928 EXPORT_SYMBOL(netlink_net_capable);
930 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
932 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
933 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
937 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
939 struct netlink_sock *nlk = nlk_sk(sk);
941 if (nlk->subscriptions && !subscriptions)
942 __sk_del_bind_node(sk);
943 else if (!nlk->subscriptions && subscriptions)
944 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
945 nlk->subscriptions = subscriptions;
948 static int netlink_realloc_groups(struct sock *sk)
950 struct netlink_sock *nlk = nlk_sk(sk);
952 unsigned long *new_groups;
955 netlink_table_grab();
957 groups = nl_table[sk->sk_protocol].groups;
958 if (!nl_table[sk->sk_protocol].registered) {
963 if (nlk->ngroups >= groups)
966 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
967 if (new_groups == NULL) {
971 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
972 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
974 nlk->groups = new_groups;
975 nlk->ngroups = groups;
977 netlink_table_ungrab();
981 static void netlink_undo_bind(int group, long unsigned int groups,
984 struct netlink_sock *nlk = nlk_sk(sk);
987 if (!nlk->netlink_unbind)
990 for (undo = 0; undo < group; undo++)
991 if (test_bit(undo, &groups))
992 nlk->netlink_unbind(sock_net(sk), undo + 1);
995 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
998 struct sock *sk = sock->sk;
999 struct net *net = sock_net(sk);
1000 struct netlink_sock *nlk = nlk_sk(sk);
1001 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1003 long unsigned int groups = nladdr->nl_groups;
1006 if (addr_len < sizeof(struct sockaddr_nl))
1009 if (nladdr->nl_family != AF_NETLINK)
1012 /* Only superuser is allowed to listen multicasts */
1014 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1016 err = netlink_realloc_groups(sk);
1021 if (nlk->ngroups == 0)
1023 else if (nlk->ngroups < 8*sizeof(groups))
1024 groups &= (1UL << nlk->ngroups) - 1;
1026 /* Paired with WRITE_ONCE() in netlink_insert() */
1027 bound = READ_ONCE(nlk->bound);
1029 /* Ensure nlk->portid is up-to-date. */
1032 if (nladdr->nl_pid != nlk->portid)
1036 netlink_lock_table();
1037 if (nlk->netlink_bind && groups) {
1040 /* nl_groups is a u32, so cap the maximum groups we can bind */
1041 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1042 if (!test_bit(group, &groups))
1044 err = nlk->netlink_bind(net, group + 1);
1047 netlink_undo_bind(group, groups, sk);
1052 /* No need for barriers here as we return to user-space without
1053 * using any of the bound attributes.
1056 err = nladdr->nl_pid ?
1057 netlink_insert(sk, nladdr->nl_pid) :
1058 netlink_autobind(sock);
1060 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1065 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1067 netlink_unlock_table();
1069 netlink_table_grab();
1070 netlink_update_subscriptions(sk, nlk->subscriptions +
1072 hweight32(nlk->groups[0]));
1073 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1074 netlink_update_listeners(sk);
1075 netlink_table_ungrab();
1080 netlink_unlock_table();
1084 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1085 int alen, int flags)
1088 struct sock *sk = sock->sk;
1089 struct netlink_sock *nlk = nlk_sk(sk);
1090 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1092 if (alen < sizeof(addr->sa_family))
1095 if (addr->sa_family == AF_UNSPEC) {
1096 sk->sk_state = NETLINK_UNCONNECTED;
1097 nlk->dst_portid = 0;
1101 if (addr->sa_family != AF_NETLINK)
1104 if (alen < sizeof(struct sockaddr_nl))
1107 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1108 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1111 /* No need for barriers here as we return to user-space without
1112 * using any of the bound attributes.
1113 * Paired with WRITE_ONCE() in netlink_insert().
1115 if (!READ_ONCE(nlk->bound))
1116 err = netlink_autobind(sock);
1119 sk->sk_state = NETLINK_CONNECTED;
1120 nlk->dst_portid = nladdr->nl_pid;
1121 nlk->dst_group = ffs(nladdr->nl_groups);
1127 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1130 struct sock *sk = sock->sk;
1131 struct netlink_sock *nlk = nlk_sk(sk);
1132 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1134 nladdr->nl_family = AF_NETLINK;
1138 nladdr->nl_pid = nlk->dst_portid;
1139 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1141 nladdr->nl_pid = nlk->portid;
1142 netlink_lock_table();
1143 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1144 netlink_unlock_table();
1146 return sizeof(*nladdr);
1149 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1152 /* try to hand this ioctl down to the NIC drivers.
1154 return -ENOIOCTLCMD;
1157 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1160 struct netlink_sock *nlk;
1162 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1164 return ERR_PTR(-ECONNREFUSED);
1166 /* Don't bother queuing skb if kernel socket has no input function */
1168 if (sock->sk_state == NETLINK_CONNECTED &&
1169 nlk->dst_portid != nlk_sk(ssk)->portid) {
1171 return ERR_PTR(-ECONNREFUSED);
1176 struct sock *netlink_getsockbyfilp(struct file *filp)
1178 struct inode *inode = file_inode(filp);
1181 if (!S_ISSOCK(inode->i_mode))
1182 return ERR_PTR(-ENOTSOCK);
1184 sock = SOCKET_I(inode)->sk;
1185 if (sock->sk_family != AF_NETLINK)
1186 return ERR_PTR(-EINVAL);
1192 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1195 struct sk_buff *skb;
1198 if (size <= NLMSG_GOODSIZE || broadcast)
1199 return alloc_skb(size, GFP_KERNEL);
1201 size = SKB_DATA_ALIGN(size) +
1202 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1204 data = vmalloc(size);
1208 skb = __build_skb(data, size);
1212 skb->destructor = netlink_skb_destructor;
1218 * Attach a skb to a netlink socket.
1219 * The caller must hold a reference to the destination socket. On error, the
1220 * reference is dropped. The skb is not send to the destination, just all
1221 * all error checks are performed and memory in the queue is reserved.
1223 * < 0: error. skb freed, reference to sock dropped.
1225 * 1: repeat lookup - reference dropped while waiting for socket memory.
1227 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1228 long *timeo, struct sock *ssk)
1230 struct netlink_sock *nlk;
1234 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1235 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1236 DECLARE_WAITQUEUE(wait, current);
1238 if (!ssk || netlink_is_kernel(ssk))
1239 netlink_overrun(sk);
1245 __set_current_state(TASK_INTERRUPTIBLE);
1246 add_wait_queue(&nlk->wait, &wait);
1248 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1249 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1250 !sock_flag(sk, SOCK_DEAD))
1251 *timeo = schedule_timeout(*timeo);
1253 __set_current_state(TASK_RUNNING);
1254 remove_wait_queue(&nlk->wait, &wait);
1257 if (signal_pending(current)) {
1259 return sock_intr_errno(*timeo);
1263 netlink_skb_set_owner_r(skb, sk);
1267 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1271 netlink_deliver_tap(sock_net(sk), skb);
1273 skb_queue_tail(&sk->sk_receive_queue, skb);
1274 sk->sk_data_ready(sk);
1278 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1280 int len = __netlink_sendskb(sk, skb);
1286 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1292 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1296 WARN_ON(skb->sk != NULL);
1297 delta = skb->end - skb->tail;
1298 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1301 if (skb_shared(skb)) {
1302 struct sk_buff *nskb = skb_clone(skb, allocation);
1309 pskb_expand_head(skb, 0, -delta,
1310 (allocation & ~__GFP_DIRECT_RECLAIM) |
1311 __GFP_NOWARN | __GFP_NORETRY);
1315 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1319 struct netlink_sock *nlk = nlk_sk(sk);
1321 ret = -ECONNREFUSED;
1322 if (nlk->netlink_rcv != NULL) {
1324 netlink_skb_set_owner_r(skb, sk);
1325 NETLINK_CB(skb).sk = ssk;
1326 netlink_deliver_tap_kernel(sk, ssk, skb);
1327 nlk->netlink_rcv(skb);
1336 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1337 u32 portid, int nonblock)
1343 skb = netlink_trim(skb, gfp_any());
1345 timeo = sock_sndtimeo(ssk, nonblock);
1347 sk = netlink_getsockbyportid(ssk, portid);
1352 if (netlink_is_kernel(sk))
1353 return netlink_unicast_kernel(sk, skb, ssk);
1355 if (sk_filter(sk, skb)) {
1362 err = netlink_attachskb(sk, skb, &timeo, ssk);
1368 return netlink_sendskb(sk, skb);
1370 EXPORT_SYMBOL(netlink_unicast);
1372 int netlink_has_listeners(struct sock *sk, unsigned int group)
1375 struct listeners *listeners;
1377 BUG_ON(!netlink_is_kernel(sk));
1380 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1382 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1383 res = test_bit(group - 1, listeners->masks);
1389 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1391 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1393 struct netlink_sock *nlk = nlk_sk(sk);
1395 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1396 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1397 netlink_skb_set_owner_r(skb, sk);
1398 __netlink_sendskb(sk, skb);
1399 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1404 struct netlink_broadcast_data {
1405 struct sock *exclude_sk;
1410 int delivery_failure;
1414 struct sk_buff *skb, *skb2;
1415 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1419 static void do_one_broadcast(struct sock *sk,
1420 struct netlink_broadcast_data *p)
1422 struct netlink_sock *nlk = nlk_sk(sk);
1425 if (p->exclude_sk == sk)
1428 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1429 !test_bit(p->group - 1, nlk->groups))
1432 if (!net_eq(sock_net(sk), p->net)) {
1433 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1436 if (!peernet_has_id(sock_net(sk), p->net))
1439 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1445 netlink_overrun(sk);
1450 if (p->skb2 == NULL) {
1451 if (skb_shared(p->skb)) {
1452 p->skb2 = skb_clone(p->skb, p->allocation);
1454 p->skb2 = skb_get(p->skb);
1456 * skb ownership may have been set when
1457 * delivered to a previous socket.
1459 skb_orphan(p->skb2);
1462 if (p->skb2 == NULL) {
1463 netlink_overrun(sk);
1464 /* Clone failed. Notify ALL listeners. */
1466 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1467 p->delivery_failure = 1;
1470 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1475 if (sk_filter(sk, p->skb2)) {
1480 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1481 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1482 NETLINK_CB(p->skb2).nsid_is_set = true;
1483 val = netlink_broadcast_deliver(sk, p->skb2);
1485 netlink_overrun(sk);
1486 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1487 p->delivery_failure = 1;
1489 p->congested |= val;
1497 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1498 u32 group, gfp_t allocation,
1499 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1502 struct net *net = sock_net(ssk);
1503 struct netlink_broadcast_data info;
1506 skb = netlink_trim(skb, allocation);
1508 info.exclude_sk = ssk;
1510 info.portid = portid;
1513 info.delivery_failure = 0;
1516 info.allocation = allocation;
1519 info.tx_filter = filter;
1520 info.tx_data = filter_data;
1522 /* While we sleep in clone, do not allow to change socket list */
1524 netlink_lock_table();
1526 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1527 do_one_broadcast(sk, &info);
1531 netlink_unlock_table();
1533 if (info.delivery_failure) {
1534 kfree_skb(info.skb2);
1537 consume_skb(info.skb2);
1539 if (info.delivered) {
1540 if (info.congested && gfpflags_allow_blocking(allocation))
1546 EXPORT_SYMBOL(netlink_broadcast_filtered);
1548 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1549 u32 group, gfp_t allocation)
1551 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1554 EXPORT_SYMBOL(netlink_broadcast);
1556 struct netlink_set_err_data {
1557 struct sock *exclude_sk;
1563 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1565 struct netlink_sock *nlk = nlk_sk(sk);
1568 if (sk == p->exclude_sk)
1571 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1574 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1575 !test_bit(p->group - 1, nlk->groups))
1578 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1583 sk->sk_err = p->code;
1584 sk->sk_error_report(sk);
1590 * netlink_set_err - report error to broadcast listeners
1591 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1592 * @portid: the PORTID of a process that we want to skip (if any)
1593 * @group: the broadcast group that will notice the error
1594 * @code: error code, must be negative (as usual in kernelspace)
1596 * This function returns the number of broadcast listeners that have set the
1597 * NETLINK_NO_ENOBUFS socket option.
1599 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1601 struct netlink_set_err_data info;
1605 info.exclude_sk = ssk;
1606 info.portid = portid;
1608 /* sk->sk_err wants a positive error value */
1611 read_lock(&nl_table_lock);
1613 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1614 ret += do_one_set_err(sk, &info);
1616 read_unlock(&nl_table_lock);
1619 EXPORT_SYMBOL(netlink_set_err);
1621 /* must be called with netlink table grabbed */
1622 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1626 int old, new = !!is_new, subscriptions;
1628 old = test_bit(group - 1, nlk->groups);
1629 subscriptions = nlk->subscriptions - old + new;
1631 __set_bit(group - 1, nlk->groups);
1633 __clear_bit(group - 1, nlk->groups);
1634 netlink_update_subscriptions(&nlk->sk, subscriptions);
1635 netlink_update_listeners(&nlk->sk);
1638 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1639 char __user *optval, unsigned int optlen)
1641 struct sock *sk = sock->sk;
1642 struct netlink_sock *nlk = nlk_sk(sk);
1643 unsigned int val = 0;
1646 if (level != SOL_NETLINK)
1647 return -ENOPROTOOPT;
1649 if (optlen >= sizeof(int) &&
1650 get_user(val, (unsigned int __user *)optval))
1654 case NETLINK_PKTINFO:
1656 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1658 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1661 case NETLINK_ADD_MEMBERSHIP:
1662 case NETLINK_DROP_MEMBERSHIP: {
1663 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1665 err = netlink_realloc_groups(sk);
1668 if (!val || val - 1 >= nlk->ngroups)
1670 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1671 err = nlk->netlink_bind(sock_net(sk), val);
1675 netlink_table_grab();
1676 netlink_update_socket_mc(nlk, val,
1677 optname == NETLINK_ADD_MEMBERSHIP);
1678 netlink_table_ungrab();
1679 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1680 nlk->netlink_unbind(sock_net(sk), val);
1685 case NETLINK_BROADCAST_ERROR:
1687 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1689 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1692 case NETLINK_NO_ENOBUFS:
1694 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1695 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1696 wake_up_interruptible(&nlk->wait);
1698 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1702 case NETLINK_LISTEN_ALL_NSID:
1703 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1707 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1709 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1712 case NETLINK_CAP_ACK:
1714 nlk->flags |= NETLINK_F_CAP_ACK;
1716 nlk->flags &= ~NETLINK_F_CAP_ACK;
1719 case NETLINK_EXT_ACK:
1721 nlk->flags |= NETLINK_F_EXT_ACK;
1723 nlk->flags &= ~NETLINK_F_EXT_ACK;
1732 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1733 char __user *optval, int __user *optlen)
1735 struct sock *sk = sock->sk;
1736 struct netlink_sock *nlk = nlk_sk(sk);
1739 if (level != SOL_NETLINK)
1740 return -ENOPROTOOPT;
1742 if (get_user(len, optlen))
1748 case NETLINK_PKTINFO:
1749 if (len < sizeof(int))
1752 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1753 if (put_user(len, optlen) ||
1754 put_user(val, optval))
1758 case NETLINK_BROADCAST_ERROR:
1759 if (len < sizeof(int))
1762 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1763 if (put_user(len, optlen) ||
1764 put_user(val, optval))
1768 case NETLINK_NO_ENOBUFS:
1769 if (len < sizeof(int))
1772 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1773 if (put_user(len, optlen) ||
1774 put_user(val, optval))
1778 case NETLINK_LIST_MEMBERSHIPS: {
1779 int pos, idx, shift;
1782 netlink_lock_table();
1783 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1784 if (len - pos < sizeof(u32))
1787 idx = pos / sizeof(unsigned long);
1788 shift = (pos % sizeof(unsigned long)) * 8;
1789 if (put_user((u32)(nlk->groups[idx] >> shift),
1790 (u32 __user *)(optval + pos))) {
1795 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1797 netlink_unlock_table();
1800 case NETLINK_CAP_ACK:
1801 if (len < sizeof(int))
1804 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1805 if (put_user(len, optlen) ||
1806 put_user(val, optval))
1810 case NETLINK_EXT_ACK:
1811 if (len < sizeof(int))
1814 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1815 if (put_user(len, optlen) || put_user(val, optval))
1825 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1827 struct nl_pktinfo info;
1829 info.group = NETLINK_CB(skb).dst_group;
1830 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1833 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1834 struct sk_buff *skb)
1836 if (!NETLINK_CB(skb).nsid_is_set)
1839 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1840 &NETLINK_CB(skb).nsid);
1843 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1845 struct sock *sk = sock->sk;
1846 struct netlink_sock *nlk = nlk_sk(sk);
1847 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1850 struct sk_buff *skb;
1852 struct scm_cookie scm;
1853 u32 netlink_skb_flags = 0;
1855 if (msg->msg_flags&MSG_OOB)
1859 pr_warn_once("Zero length message leads to an empty skb\n");
1863 err = scm_send(sock, msg, &scm, true);
1867 if (msg->msg_namelen) {
1869 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1871 if (addr->nl_family != AF_NETLINK)
1873 dst_portid = addr->nl_pid;
1874 dst_group = ffs(addr->nl_groups);
1876 if ((dst_group || dst_portid) &&
1877 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1879 netlink_skb_flags |= NETLINK_SKB_DST;
1881 dst_portid = nlk->dst_portid;
1882 dst_group = nlk->dst_group;
1885 /* Paired with WRITE_ONCE() in netlink_insert() */
1886 if (!READ_ONCE(nlk->bound)) {
1887 err = netlink_autobind(sock);
1891 /* Ensure nlk is hashed and visible. */
1896 if (len > sk->sk_sndbuf - 32)
1899 skb = netlink_alloc_large_skb(len, dst_group);
1903 NETLINK_CB(skb).portid = nlk->portid;
1904 NETLINK_CB(skb).dst_group = dst_group;
1905 NETLINK_CB(skb).creds = scm.creds;
1906 NETLINK_CB(skb).flags = netlink_skb_flags;
1909 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1914 err = security_netlink_send(sk, skb);
1921 refcount_inc(&skb->users);
1922 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1924 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1931 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1934 struct scm_cookie scm;
1935 struct sock *sk = sock->sk;
1936 struct netlink_sock *nlk = nlk_sk(sk);
1937 int noblock = flags&MSG_DONTWAIT;
1939 struct sk_buff *skb, *data_skb;
1947 skb = skb_recv_datagram(sk, flags, noblock, &err);
1953 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1954 if (unlikely(skb_shinfo(skb)->frag_list)) {
1956 * If this skb has a frag_list, then here that means that we
1957 * will have to use the frag_list skb's data for compat tasks
1958 * and the regular skb's data for normal (non-compat) tasks.
1960 * If we need to send the compat skb, assign it to the
1961 * 'data_skb' variable so that it will be used below for data
1962 * copying. We keep 'skb' for everything else, including
1963 * freeing both later.
1965 if (flags & MSG_CMSG_COMPAT)
1966 data_skb = skb_shinfo(skb)->frag_list;
1970 /* Record the max length of recvmsg() calls for future allocations */
1971 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1972 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1973 SKB_WITH_OVERHEAD(32768));
1975 copied = data_skb->len;
1977 msg->msg_flags |= MSG_TRUNC;
1981 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1983 if (msg->msg_name) {
1984 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1985 addr->nl_family = AF_NETLINK;
1987 addr->nl_pid = NETLINK_CB(skb).portid;
1988 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1989 msg->msg_namelen = sizeof(*addr);
1992 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1993 netlink_cmsg_recv_pktinfo(msg, skb);
1994 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1995 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1997 memset(&scm, 0, sizeof(scm));
1998 scm.creds = *NETLINK_CREDS(skb);
1999 if (flags & MSG_TRUNC)
2000 copied = data_skb->len;
2002 skb_free_datagram(sk, skb);
2004 if (nlk->cb_running &&
2005 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2006 ret = netlink_dump(sk);
2009 sk->sk_error_report(sk);
2013 scm_recv(sock, msg, &scm, flags);
2015 netlink_rcv_wake(sk);
2016 return err ? : copied;
2019 static void netlink_data_ready(struct sock *sk)
2025 * We export these functions to other modules. They provide a
2026 * complete set of kernel non-blocking support for message
2031 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2032 struct netlink_kernel_cfg *cfg)
2034 struct socket *sock;
2036 struct netlink_sock *nlk;
2037 struct listeners *listeners = NULL;
2038 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2039 unsigned int groups;
2043 if (unit < 0 || unit >= MAX_LINKS)
2046 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2049 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2050 goto out_sock_release_nosk;
2054 if (!cfg || cfg->groups < 32)
2057 groups = cfg->groups;
2059 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2061 goto out_sock_release;
2063 sk->sk_data_ready = netlink_data_ready;
2064 if (cfg && cfg->input)
2065 nlk_sk(sk)->netlink_rcv = cfg->input;
2067 if (netlink_insert(sk, 0))
2068 goto out_sock_release;
2071 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2073 netlink_table_grab();
2074 if (!nl_table[unit].registered) {
2075 nl_table[unit].groups = groups;
2076 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2077 nl_table[unit].cb_mutex = cb_mutex;
2078 nl_table[unit].module = module;
2080 nl_table[unit].bind = cfg->bind;
2081 nl_table[unit].unbind = cfg->unbind;
2082 nl_table[unit].flags = cfg->flags;
2084 nl_table[unit].compare = cfg->compare;
2086 nl_table[unit].registered = 1;
2089 nl_table[unit].registered++;
2091 netlink_table_ungrab();
2096 netlink_kernel_release(sk);
2099 out_sock_release_nosk:
2103 EXPORT_SYMBOL(__netlink_kernel_create);
2106 netlink_kernel_release(struct sock *sk)
2108 if (sk == NULL || sk->sk_socket == NULL)
2111 sock_release(sk->sk_socket);
2113 EXPORT_SYMBOL(netlink_kernel_release);
2115 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2117 struct listeners *new, *old;
2118 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2123 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2124 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2127 old = nl_deref_protected(tbl->listeners);
2128 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2129 rcu_assign_pointer(tbl->listeners, new);
2131 kfree_rcu(old, rcu);
2133 tbl->groups = groups;
2139 * netlink_change_ngroups - change number of multicast groups
2141 * This changes the number of multicast groups that are available
2142 * on a certain netlink family. Note that it is not possible to
2143 * change the number of groups to below 32. Also note that it does
2144 * not implicitly call netlink_clear_multicast_users() when the
2145 * number of groups is reduced.
2147 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2148 * @groups: The new number of groups.
2150 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2154 netlink_table_grab();
2155 err = __netlink_change_ngroups(sk, groups);
2156 netlink_table_ungrab();
2161 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2164 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2166 sk_for_each_bound(sk, &tbl->mc_list)
2167 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2171 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2173 struct nlmsghdr *nlh;
2174 int size = nlmsg_msg_size(len);
2176 nlh = skb_put(skb, NLMSG_ALIGN(size));
2177 nlh->nlmsg_type = type;
2178 nlh->nlmsg_len = size;
2179 nlh->nlmsg_flags = flags;
2180 nlh->nlmsg_pid = portid;
2181 nlh->nlmsg_seq = seq;
2182 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2183 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2186 EXPORT_SYMBOL(__nlmsg_put);
2189 * It looks a bit ugly.
2190 * It would be better to create kernel thread.
2193 static int netlink_dump(struct sock *sk)
2195 struct netlink_sock *nlk = nlk_sk(sk);
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);
2246 /* Make sure malicious BPF programs can not read unitialized memory
2247 * from skb->head -> skb->data
2249 skb_reset_network_header(skb);
2250 skb_reset_mac_header(skb);
2252 netlink_skb_set_owner_r(skb, sk);
2254 if (nlk->dump_done_errno > 0)
2255 nlk->dump_done_errno = cb->dump(skb, cb);
2257 if (nlk->dump_done_errno > 0 ||
2258 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2259 mutex_unlock(nlk->cb_mutex);
2261 if (sk_filter(sk, skb))
2264 __netlink_sendskb(sk, skb);
2268 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE,
2269 sizeof(nlk->dump_done_errno), NLM_F_MULTI);
2273 nl_dump_check_consistent(cb, nlh);
2275 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno,
2276 sizeof(nlk->dump_done_errno));
2278 if (sk_filter(sk, skb))
2281 __netlink_sendskb(sk, skb);
2286 nlk->cb_running = false;
2287 module = cb->module;
2289 mutex_unlock(nlk->cb_mutex);
2295 mutex_unlock(nlk->cb_mutex);
2300 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2301 const struct nlmsghdr *nlh,
2302 struct netlink_dump_control *control)
2304 struct netlink_callback *cb;
2306 struct netlink_sock *nlk;
2309 refcount_inc(&skb->users);
2311 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2313 ret = -ECONNREFUSED;
2318 mutex_lock(nlk->cb_mutex);
2319 /* A dump is in progress... */
2320 if (nlk->cb_running) {
2324 /* add reference of module which cb->dump belongs to */
2325 if (!try_module_get(control->module)) {
2326 ret = -EPROTONOSUPPORT;
2331 memset(cb, 0, sizeof(*cb));
2332 cb->dump = control->dump;
2333 cb->done = control->done;
2335 cb->data = control->data;
2336 cb->module = control->module;
2337 cb->min_dump_alloc = control->min_dump_alloc;
2340 if (control->start) {
2341 ret = control->start(cb);
2346 nlk->cb_running = true;
2347 nlk->dump_done_errno = INT_MAX;
2349 mutex_unlock(nlk->cb_mutex);
2351 ret = netlink_dump(sk);
2358 /* We successfully started a dump, by returning -EINTR we
2359 * signal not to send ACK even if it was requested.
2364 module_put(control->module);
2367 mutex_unlock(nlk->cb_mutex);
2372 EXPORT_SYMBOL(__netlink_dump_start);
2374 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2375 const struct netlink_ext_ack *extack)
2377 struct sk_buff *skb;
2378 struct nlmsghdr *rep;
2379 struct nlmsgerr *errmsg;
2380 size_t payload = sizeof(*errmsg);
2382 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2383 unsigned int flags = 0;
2384 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2386 /* Error messages get the original request appened, unless the user
2387 * requests to cap the error message, and get extra error data if
2390 if (nlk_has_extack && extack && extack->_msg)
2391 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2394 if (!(nlk->flags & NETLINK_F_CAP_ACK))
2395 payload += nlmsg_len(nlh);
2397 flags |= NLM_F_CAPPED;
2398 if (nlk_has_extack && extack && extack->bad_attr)
2399 tlvlen += nla_total_size(sizeof(u32));
2401 flags |= NLM_F_CAPPED;
2403 if (nlk_has_extack && extack && extack->cookie_len)
2404 tlvlen += nla_total_size(extack->cookie_len);
2408 flags |= NLM_F_ACK_TLVS;
2410 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2412 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2413 NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk);
2417 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2418 NLMSG_ERROR, payload, flags);
2419 errmsg = nlmsg_data(rep);
2420 errmsg->error = err;
2421 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2423 if (nlk_has_extack && extack) {
2425 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2429 if (extack->bad_attr &&
2430 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2431 (u8 *)extack->bad_attr >= in_skb->data +
2433 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2434 (u8 *)extack->bad_attr -
2437 if (extack->cookie_len)
2438 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2444 nlmsg_end(skb, rep);
2446 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2448 EXPORT_SYMBOL(netlink_ack);
2450 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2452 struct netlink_ext_ack *))
2454 struct netlink_ext_ack extack;
2455 struct nlmsghdr *nlh;
2458 while (skb->len >= nlmsg_total_size(0)) {
2461 memset(&extack, 0, sizeof(extack));
2462 nlh = nlmsg_hdr(skb);
2465 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2468 /* Only requests are handled by the kernel */
2469 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2472 /* Skip control messages */
2473 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2476 err = cb(skb, nlh, &extack);
2481 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2482 netlink_ack(skb, nlh, err, &extack);
2485 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2486 if (msglen > skb->len)
2488 skb_pull(skb, msglen);
2493 EXPORT_SYMBOL(netlink_rcv_skb);
2496 * nlmsg_notify - send a notification netlink message
2497 * @sk: netlink socket to use
2498 * @skb: notification message
2499 * @portid: destination netlink portid for reports or 0
2500 * @group: destination multicast group or 0
2501 * @report: 1 to report back, 0 to disable
2502 * @flags: allocation flags
2504 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2505 unsigned int group, int report, gfp_t flags)
2510 int exclude_portid = 0;
2513 refcount_inc(&skb->users);
2514 exclude_portid = portid;
2517 /* errors reported via destination sk->sk_err, but propagate
2518 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2519 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2527 err2 = nlmsg_unicast(sk, skb, portid);
2534 EXPORT_SYMBOL(nlmsg_notify);
2536 #ifdef CONFIG_PROC_FS
2537 struct nl_seq_iter {
2538 struct seq_net_private p;
2539 struct rhashtable_iter hti;
2543 static int netlink_walk_start(struct nl_seq_iter *iter)
2547 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti,
2550 iter->link = MAX_LINKS;
2554 rhashtable_walk_start(&iter->hti);
2559 static void netlink_walk_stop(struct nl_seq_iter *iter)
2561 rhashtable_walk_stop(&iter->hti);
2562 rhashtable_walk_exit(&iter->hti);
2565 static void *__netlink_seq_next(struct seq_file *seq)
2567 struct nl_seq_iter *iter = seq->private;
2568 struct netlink_sock *nlk;
2574 nlk = rhashtable_walk_next(&iter->hti);
2577 if (PTR_ERR(nlk) == -EAGAIN)
2586 netlink_walk_stop(iter);
2587 if (++iter->link >= MAX_LINKS)
2590 err = netlink_walk_start(iter);
2592 return ERR_PTR(err);
2594 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2599 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2601 struct nl_seq_iter *iter = seq->private;
2602 void *obj = SEQ_START_TOKEN;
2608 err = netlink_walk_start(iter);
2610 return ERR_PTR(err);
2612 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2613 obj = __netlink_seq_next(seq);
2618 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2621 return __netlink_seq_next(seq);
2624 static void netlink_seq_stop(struct seq_file *seq, void *v)
2626 struct nl_seq_iter *iter = seq->private;
2628 if (iter->link >= MAX_LINKS)
2631 netlink_walk_stop(iter);
2635 static int netlink_seq_show(struct seq_file *seq, void *v)
2637 if (v == SEQ_START_TOKEN) {
2639 "sk Eth Pid Groups "
2640 "Rmem Wmem Dump Locks Drops Inode\n");
2643 struct netlink_sock *nlk = nlk_sk(s);
2645 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8d %-8lu\n",
2649 nlk->groups ? (u32)nlk->groups[0] : 0,
2650 sk_rmem_alloc_get(s),
2651 sk_wmem_alloc_get(s),
2653 refcount_read(&s->sk_refcnt),
2654 atomic_read(&s->sk_drops),
2662 static const struct seq_operations netlink_seq_ops = {
2663 .start = netlink_seq_start,
2664 .next = netlink_seq_next,
2665 .stop = netlink_seq_stop,
2666 .show = netlink_seq_show,
2670 int netlink_register_notifier(struct notifier_block *nb)
2672 return blocking_notifier_chain_register(&netlink_chain, nb);
2674 EXPORT_SYMBOL(netlink_register_notifier);
2676 int netlink_unregister_notifier(struct notifier_block *nb)
2678 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2680 EXPORT_SYMBOL(netlink_unregister_notifier);
2682 static const struct proto_ops netlink_ops = {
2683 .family = PF_NETLINK,
2684 .owner = THIS_MODULE,
2685 .release = netlink_release,
2686 .bind = netlink_bind,
2687 .connect = netlink_connect,
2688 .socketpair = sock_no_socketpair,
2689 .accept = sock_no_accept,
2690 .getname = netlink_getname,
2691 .poll = datagram_poll,
2692 .ioctl = netlink_ioctl,
2693 .listen = sock_no_listen,
2694 .shutdown = sock_no_shutdown,
2695 .setsockopt = netlink_setsockopt,
2696 .getsockopt = netlink_getsockopt,
2697 .sendmsg = netlink_sendmsg,
2698 .recvmsg = netlink_recvmsg,
2699 .mmap = sock_no_mmap,
2700 .sendpage = sock_no_sendpage,
2703 static const struct net_proto_family netlink_family_ops = {
2704 .family = PF_NETLINK,
2705 .create = netlink_create,
2706 .owner = THIS_MODULE, /* for consistency 8) */
2709 static int __net_init netlink_net_init(struct net *net)
2711 #ifdef CONFIG_PROC_FS
2712 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2713 sizeof(struct nl_seq_iter)))
2719 static void __net_exit netlink_net_exit(struct net *net)
2721 #ifdef CONFIG_PROC_FS
2722 remove_proc_entry("netlink", net->proc_net);
2726 static void __init netlink_add_usersock_entry(void)
2728 struct listeners *listeners;
2731 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2733 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2735 netlink_table_grab();
2737 nl_table[NETLINK_USERSOCK].groups = groups;
2738 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2739 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2740 nl_table[NETLINK_USERSOCK].registered = 1;
2741 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2743 netlink_table_ungrab();
2746 static struct pernet_operations __net_initdata netlink_net_ops = {
2747 .init = netlink_net_init,
2748 .exit = netlink_net_exit,
2751 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2753 const struct netlink_sock *nlk = data;
2754 struct netlink_compare_arg arg;
2756 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2757 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2760 static const struct rhashtable_params netlink_rhashtable_params = {
2761 .head_offset = offsetof(struct netlink_sock, node),
2762 .key_len = netlink_compare_arg_len,
2763 .obj_hashfn = netlink_hash,
2764 .obj_cmpfn = netlink_compare,
2765 .automatic_shrinking = true,
2768 static int __init netlink_proto_init(void)
2771 int err = proto_register(&netlink_proto, 0);
2776 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2778 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2782 for (i = 0; i < MAX_LINKS; i++) {
2783 if (rhashtable_init(&nl_table[i].hash,
2784 &netlink_rhashtable_params) < 0) {
2786 rhashtable_destroy(&nl_table[i].hash);
2792 netlink_add_usersock_entry();
2794 sock_register(&netlink_family_ops);
2795 register_pernet_subsys(&netlink_net_ops);
2796 register_pernet_subsys(&netlink_tap_net_ops);
2797 /* The netlink device handler may be needed early. */
2802 panic("netlink_init: Cannot allocate nl_table\n");
2805 core_initcall(netlink_proto_init);