2 * NETLINK Kernel-user communication protocol.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6 * Patrick McHardy <kaber@trash.net>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
13 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
14 * added netlink_proto_exit
15 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
16 * use nlk_sk, as sk->protinfo is on a diet 8)
17 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
18 * - inc module use count of module that owns
19 * the kernel socket in case userspace opens
20 * socket of same protocol
21 * - remove all module support, since netlink is
22 * mandatory if CONFIG_NET=y these days
25 #include <linux/module.h>
27 #include <linux/capability.h>
28 #include <linux/kernel.h>
29 #include <linux/init.h>
30 #include <linux/signal.h>
31 #include <linux/sched.h>
32 #include <linux/errno.h>
33 #include <linux/string.h>
34 #include <linux/stat.h>
35 #include <linux/socket.h>
37 #include <linux/fcntl.h>
38 #include <linux/termios.h>
39 #include <linux/sockios.h>
40 #include <linux/net.h>
42 #include <linux/slab.h>
43 #include <linux/uaccess.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/rtnetlink.h>
47 #include <linux/proc_fs.h>
48 #include <linux/seq_file.h>
49 #include <linux/notifier.h>
50 #include <linux/security.h>
51 #include <linux/jhash.h>
52 #include <linux/jiffies.h>
53 #include <linux/random.h>
54 #include <linux/bitops.h>
56 #include <linux/types.h>
57 #include <linux/audit.h>
58 #include <linux/mutex.h>
59 #include <linux/vmalloc.h>
60 #include <linux/if_arp.h>
61 #include <linux/rhashtable.h>
62 #include <asm/cacheflush.h>
63 #include <linux/hash.h>
64 #include <linux/genetlink.h>
65 #include <linux/net_namespace.h>
66 #include <linux/nospec.h>
68 #include <net/net_namespace.h>
69 #include <net/netns/generic.h>
72 #include <net/netlink.h>
74 #include "af_netlink.h"
78 unsigned long masks[0];
82 #define NETLINK_S_CONGESTED 0x0
84 static inline int netlink_is_kernel(struct sock *sk)
86 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
89 struct netlink_table *nl_table __read_mostly;
90 EXPORT_SYMBOL_GPL(nl_table);
92 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
94 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
96 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
99 "nlk_cb_mutex-USERSOCK",
100 "nlk_cb_mutex-FIREWALL",
101 "nlk_cb_mutex-SOCK_DIAG",
102 "nlk_cb_mutex-NFLOG",
104 "nlk_cb_mutex-SELINUX",
105 "nlk_cb_mutex-ISCSI",
106 "nlk_cb_mutex-AUDIT",
107 "nlk_cb_mutex-FIB_LOOKUP",
108 "nlk_cb_mutex-CONNECTOR",
109 "nlk_cb_mutex-NETFILTER",
110 "nlk_cb_mutex-IP6_FW",
111 "nlk_cb_mutex-DNRTMSG",
112 "nlk_cb_mutex-KOBJECT_UEVENT",
113 "nlk_cb_mutex-GENERIC",
115 "nlk_cb_mutex-SCSITRANSPORT",
116 "nlk_cb_mutex-ECRYPTFS",
118 "nlk_cb_mutex-CRYPTO",
129 "nlk_cb_mutex-MAX_LINKS"
132 static int netlink_dump(struct sock *sk);
134 /* nl_table locking explained:
135 * Lookup and traversal are protected with an RCU read-side lock. Insertion
136 * and removal are protected with per bucket lock while using RCU list
137 * modification primitives and may run in parallel to RCU protected lookups.
138 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
139 * been acquired * either during or after the socket has been removed from
140 * the list and after an RCU grace period.
142 DEFINE_RWLOCK(nl_table_lock);
143 EXPORT_SYMBOL_GPL(nl_table_lock);
144 static atomic_t nl_table_users = ATOMIC_INIT(0);
146 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
148 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
151 static const struct rhashtable_params netlink_rhashtable_params;
153 static inline u32 netlink_group_mask(u32 group)
155 return group ? 1 << (group - 1) : 0;
158 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
161 unsigned int len = skb_end_offset(skb);
164 new = alloc_skb(len, gfp_mask);
168 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
169 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
170 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
172 skb_put_data(new, skb->data, len);
176 static unsigned int netlink_tap_net_id;
178 struct netlink_tap_net {
179 struct list_head netlink_tap_all;
180 struct mutex netlink_tap_lock;
183 int netlink_add_tap(struct netlink_tap *nt)
185 struct net *net = dev_net(nt->dev);
186 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
188 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
191 mutex_lock(&nn->netlink_tap_lock);
192 list_add_rcu(&nt->list, &nn->netlink_tap_all);
193 mutex_unlock(&nn->netlink_tap_lock);
195 __module_get(nt->module);
199 EXPORT_SYMBOL_GPL(netlink_add_tap);
201 static int __netlink_remove_tap(struct netlink_tap *nt)
203 struct net *net = dev_net(nt->dev);
204 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
206 struct netlink_tap *tmp;
208 mutex_lock(&nn->netlink_tap_lock);
210 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
212 list_del_rcu(&nt->list);
218 pr_warn("__netlink_remove_tap: %p not found\n", nt);
220 mutex_unlock(&nn->netlink_tap_lock);
223 module_put(nt->module);
225 return found ? 0 : -ENODEV;
228 int netlink_remove_tap(struct netlink_tap *nt)
232 ret = __netlink_remove_tap(nt);
237 EXPORT_SYMBOL_GPL(netlink_remove_tap);
239 static __net_init int netlink_tap_init_net(struct net *net)
241 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
243 INIT_LIST_HEAD(&nn->netlink_tap_all);
244 mutex_init(&nn->netlink_tap_lock);
248 static void __net_exit netlink_tap_exit_net(struct net *net)
252 static struct pernet_operations netlink_tap_net_ops = {
253 .init = netlink_tap_init_net,
254 .exit = netlink_tap_exit_net,
255 .id = &netlink_tap_net_id,
256 .size = sizeof(struct netlink_tap_net),
259 static bool netlink_filter_tap(const struct sk_buff *skb)
261 struct sock *sk = skb->sk;
263 /* We take the more conservative approach and
264 * whitelist socket protocols that may pass.
266 switch (sk->sk_protocol) {
268 case NETLINK_USERSOCK:
269 case NETLINK_SOCK_DIAG:
272 case NETLINK_FIB_LOOKUP:
273 case NETLINK_NETFILTER:
274 case NETLINK_GENERIC:
281 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
282 struct net_device *dev)
284 struct sk_buff *nskb;
285 struct sock *sk = skb->sk;
288 if (!net_eq(dev_net(dev), sock_net(sk)))
293 if (is_vmalloc_addr(skb->head))
294 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
296 nskb = skb_clone(skb, GFP_ATOMIC);
299 nskb->protocol = htons((u16) sk->sk_protocol);
300 nskb->pkt_type = netlink_is_kernel(sk) ?
301 PACKET_KERNEL : PACKET_USER;
302 skb_reset_network_header(nskb);
303 ret = dev_queue_xmit(nskb);
304 if (unlikely(ret > 0))
305 ret = net_xmit_errno(ret);
312 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
315 struct netlink_tap *tmp;
317 if (!netlink_filter_tap(skb))
320 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
321 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
327 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
329 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
333 if (unlikely(!list_empty(&nn->netlink_tap_all)))
334 __netlink_deliver_tap(skb, nn);
339 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
342 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
343 netlink_deliver_tap(sock_net(dst), skb);
346 static void netlink_overrun(struct sock *sk)
348 struct netlink_sock *nlk = nlk_sk(sk);
350 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
351 if (!test_and_set_bit(NETLINK_S_CONGESTED,
352 &nlk_sk(sk)->state)) {
353 sk->sk_err = ENOBUFS;
354 sk->sk_error_report(sk);
357 atomic_inc(&sk->sk_drops);
360 static void netlink_rcv_wake(struct sock *sk)
362 struct netlink_sock *nlk = nlk_sk(sk);
364 if (skb_queue_empty(&sk->sk_receive_queue))
365 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
366 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
367 wake_up_interruptible(&nlk->wait);
370 static void netlink_skb_destructor(struct sk_buff *skb)
372 if (is_vmalloc_addr(skb->head)) {
374 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
383 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
385 WARN_ON(skb->sk != NULL);
387 skb->destructor = netlink_skb_destructor;
388 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
389 sk_mem_charge(sk, skb->truesize);
392 static void netlink_sock_destruct(struct sock *sk)
394 struct netlink_sock *nlk = nlk_sk(sk);
396 if (nlk->cb_running) {
398 nlk->cb.done(&nlk->cb);
399 module_put(nlk->cb.module);
400 kfree_skb(nlk->cb.skb);
403 skb_queue_purge(&sk->sk_receive_queue);
405 if (!sock_flag(sk, SOCK_DEAD)) {
406 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
410 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
411 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
412 WARN_ON(nlk_sk(sk)->groups);
415 static void netlink_sock_destruct_work(struct work_struct *work)
417 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
423 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
424 * SMP. Look, when several writers sleep and reader wakes them up, all but one
425 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
426 * this, _but_ remember, it adds useless work on UP machines.
429 void netlink_table_grab(void)
430 __acquires(nl_table_lock)
434 write_lock_irq(&nl_table_lock);
436 if (atomic_read(&nl_table_users)) {
437 DECLARE_WAITQUEUE(wait, current);
439 add_wait_queue_exclusive(&nl_table_wait, &wait);
441 set_current_state(TASK_UNINTERRUPTIBLE);
442 if (atomic_read(&nl_table_users) == 0)
444 write_unlock_irq(&nl_table_lock);
446 write_lock_irq(&nl_table_lock);
449 __set_current_state(TASK_RUNNING);
450 remove_wait_queue(&nl_table_wait, &wait);
454 void netlink_table_ungrab(void)
455 __releases(nl_table_lock)
457 write_unlock_irq(&nl_table_lock);
458 wake_up(&nl_table_wait);
462 netlink_lock_table(void)
464 /* read_lock() synchronizes us to netlink_table_grab */
466 read_lock(&nl_table_lock);
467 atomic_inc(&nl_table_users);
468 read_unlock(&nl_table_lock);
472 netlink_unlock_table(void)
474 if (atomic_dec_and_test(&nl_table_users))
475 wake_up(&nl_table_wait);
478 struct netlink_compare_arg
484 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
485 #define netlink_compare_arg_len \
486 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
488 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
491 const struct netlink_compare_arg *x = arg->key;
492 const struct netlink_sock *nlk = ptr;
494 return nlk->portid != x->portid ||
495 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
498 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
499 struct net *net, u32 portid)
501 memset(arg, 0, sizeof(*arg));
502 write_pnet(&arg->pnet, net);
503 arg->portid = portid;
506 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
509 struct netlink_compare_arg arg;
511 netlink_compare_arg_init(&arg, net, portid);
512 return rhashtable_lookup_fast(&table->hash, &arg,
513 netlink_rhashtable_params);
516 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
518 struct netlink_compare_arg arg;
520 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
521 return rhashtable_lookup_insert_key(&table->hash, &arg,
523 netlink_rhashtable_params);
526 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
528 struct netlink_table *table = &nl_table[protocol];
532 sk = __netlink_lookup(table, portid, net);
540 static const struct proto_ops netlink_ops;
543 netlink_update_listeners(struct sock *sk)
545 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
548 struct listeners *listeners;
550 listeners = nl_deref_protected(tbl->listeners);
554 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
556 sk_for_each_bound(sk, &tbl->mc_list) {
557 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
558 mask |= nlk_sk(sk)->groups[i];
560 listeners->masks[i] = mask;
562 /* this function is only called with the netlink table "grabbed", which
563 * makes sure updates are visible before bind or setsockopt return. */
566 static int netlink_insert(struct sock *sk, u32 portid)
568 struct netlink_table *table = &nl_table[sk->sk_protocol];
573 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
574 if (nlk_sk(sk)->bound)
577 nlk_sk(sk)->portid = portid;
580 err = __netlink_insert(table, sk);
582 /* In case the hashtable backend returns with -EBUSY
583 * from here, it must not escape to the caller.
585 if (unlikely(err == -EBUSY))
593 /* We need to ensure that the socket is hashed and visible. */
595 nlk_sk(sk)->bound = portid;
602 static void netlink_remove(struct sock *sk)
604 struct netlink_table *table;
606 table = &nl_table[sk->sk_protocol];
607 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
608 netlink_rhashtable_params)) {
609 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
613 netlink_table_grab();
614 if (nlk_sk(sk)->subscriptions) {
615 __sk_del_bind_node(sk);
616 netlink_update_listeners(sk);
618 if (sk->sk_protocol == NETLINK_GENERIC)
619 atomic_inc(&genl_sk_destructing_cnt);
620 netlink_table_ungrab();
623 static struct proto netlink_proto = {
625 .owner = THIS_MODULE,
626 .obj_size = sizeof(struct netlink_sock),
629 static int __netlink_create(struct net *net, struct socket *sock,
630 struct mutex *cb_mutex, int protocol,
634 struct netlink_sock *nlk;
636 sock->ops = &netlink_ops;
638 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
642 sock_init_data(sock, sk);
646 nlk->cb_mutex = cb_mutex;
648 nlk->cb_mutex = &nlk->cb_def_mutex;
649 mutex_init(nlk->cb_mutex);
650 lockdep_set_class_and_name(nlk->cb_mutex,
651 nlk_cb_mutex_keys + protocol,
652 nlk_cb_mutex_key_strings[protocol]);
654 init_waitqueue_head(&nlk->wait);
656 sk->sk_destruct = netlink_sock_destruct;
657 sk->sk_protocol = protocol;
661 static int netlink_create(struct net *net, struct socket *sock, int protocol,
664 struct module *module = NULL;
665 struct mutex *cb_mutex;
666 struct netlink_sock *nlk;
667 int (*bind)(struct net *net, int group);
668 void (*unbind)(struct net *net, int group);
671 sock->state = SS_UNCONNECTED;
673 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
674 return -ESOCKTNOSUPPORT;
676 if (protocol < 0 || protocol >= MAX_LINKS)
677 return -EPROTONOSUPPORT;
678 protocol = array_index_nospec(protocol, MAX_LINKS);
680 netlink_lock_table();
681 #ifdef CONFIG_MODULES
682 if (!nl_table[protocol].registered) {
683 netlink_unlock_table();
684 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
685 netlink_lock_table();
688 if (nl_table[protocol].registered &&
689 try_module_get(nl_table[protocol].module))
690 module = nl_table[protocol].module;
692 err = -EPROTONOSUPPORT;
693 cb_mutex = nl_table[protocol].cb_mutex;
694 bind = nl_table[protocol].bind;
695 unbind = nl_table[protocol].unbind;
696 netlink_unlock_table();
701 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
706 sock_prot_inuse_add(net, &netlink_proto, 1);
709 nlk = nlk_sk(sock->sk);
710 nlk->module = module;
711 nlk->netlink_bind = bind;
712 nlk->netlink_unbind = unbind;
721 static void deferred_put_nlk_sk(struct rcu_head *head)
723 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
724 struct sock *sk = &nlk->sk;
729 if (!refcount_dec_and_test(&sk->sk_refcnt))
732 if (nlk->cb_running && nlk->cb.done) {
733 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
734 schedule_work(&nlk->work);
741 static int netlink_release(struct socket *sock)
743 struct sock *sk = sock->sk;
744 struct netlink_sock *nlk;
754 * OK. Socket is unlinked, any packets that arrive now
758 /* must not acquire netlink_table_lock in any way again before unbind
759 * and notifying genetlink is done as otherwise it might deadlock
761 if (nlk->netlink_unbind) {
764 for (i = 0; i < nlk->ngroups; i++)
765 if (test_bit(i, nlk->groups))
766 nlk->netlink_unbind(sock_net(sk), i + 1);
768 if (sk->sk_protocol == NETLINK_GENERIC &&
769 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
770 wake_up(&genl_sk_destructing_waitq);
773 wake_up_interruptible_all(&nlk->wait);
775 skb_queue_purge(&sk->sk_write_queue);
777 if (nlk->portid && nlk->bound) {
778 struct netlink_notify n = {
780 .protocol = sk->sk_protocol,
781 .portid = nlk->portid,
783 blocking_notifier_call_chain(&netlink_chain,
784 NETLINK_URELEASE, &n);
787 module_put(nlk->module);
789 if (netlink_is_kernel(sk)) {
790 netlink_table_grab();
791 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
792 if (--nl_table[sk->sk_protocol].registered == 0) {
793 struct listeners *old;
795 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
796 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
798 nl_table[sk->sk_protocol].module = NULL;
799 nl_table[sk->sk_protocol].bind = NULL;
800 nl_table[sk->sk_protocol].unbind = NULL;
801 nl_table[sk->sk_protocol].flags = 0;
802 nl_table[sk->sk_protocol].registered = 0;
804 netlink_table_ungrab();
808 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
810 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
814 static int netlink_autobind(struct socket *sock)
816 struct sock *sk = sock->sk;
817 struct net *net = sock_net(sk);
818 struct netlink_table *table = &nl_table[sk->sk_protocol];
819 s32 portid = task_tgid_vnr(current);
827 ok = !__netlink_lookup(table, portid, net);
830 /* Bind collision, search negative portid values. */
832 /* rover will be in range [S32_MIN, -4097] */
833 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
834 else if (rover >= -4096)
840 err = netlink_insert(sk, portid);
841 if (err == -EADDRINUSE)
844 /* If 2 threads race to autobind, that is fine. */
852 * __netlink_ns_capable - General netlink message capability test
853 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
854 * @user_ns: The user namespace of the capability to use
855 * @cap: The capability to use
857 * Test to see if the opener of the socket we received the message
858 * from had when the netlink socket was created and the sender of the
859 * message has has the capability @cap in the user namespace @user_ns.
861 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
862 struct user_namespace *user_ns, int cap)
864 return ((nsp->flags & NETLINK_SKB_DST) ||
865 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
866 ns_capable(user_ns, cap);
868 EXPORT_SYMBOL(__netlink_ns_capable);
871 * netlink_ns_capable - General netlink message capability test
872 * @skb: socket buffer holding a netlink command from userspace
873 * @user_ns: The user namespace of the capability to use
874 * @cap: The capability to use
876 * Test to see if the opener of the socket we received the message
877 * from had when the netlink socket was created and the sender of the
878 * message has has the capability @cap in the user namespace @user_ns.
880 bool netlink_ns_capable(const struct sk_buff *skb,
881 struct user_namespace *user_ns, int cap)
883 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
885 EXPORT_SYMBOL(netlink_ns_capable);
888 * netlink_capable - Netlink global message capability test
889 * @skb: socket buffer holding a netlink command from userspace
890 * @cap: The capability to use
892 * Test to see if the opener of the socket we received the message
893 * from had when the netlink socket was created and the sender of the
894 * message has has the capability @cap in all user namespaces.
896 bool netlink_capable(const struct sk_buff *skb, int cap)
898 return netlink_ns_capable(skb, &init_user_ns, cap);
900 EXPORT_SYMBOL(netlink_capable);
903 * netlink_net_capable - Netlink network namespace message capability test
904 * @skb: socket buffer holding a netlink command from userspace
905 * @cap: The capability to use
907 * Test to see if the opener of the socket we received the message
908 * from had when the netlink socket was created and the sender of the
909 * message has has the capability @cap over the network namespace of
910 * the socket we received the message from.
912 bool netlink_net_capable(const struct sk_buff *skb, int cap)
914 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
916 EXPORT_SYMBOL(netlink_net_capable);
918 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
920 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
921 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
925 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
927 struct netlink_sock *nlk = nlk_sk(sk);
929 if (nlk->subscriptions && !subscriptions)
930 __sk_del_bind_node(sk);
931 else if (!nlk->subscriptions && subscriptions)
932 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
933 nlk->subscriptions = subscriptions;
936 static int netlink_realloc_groups(struct sock *sk)
938 struct netlink_sock *nlk = nlk_sk(sk);
940 unsigned long *new_groups;
943 netlink_table_grab();
945 groups = nl_table[sk->sk_protocol].groups;
946 if (!nl_table[sk->sk_protocol].registered) {
951 if (nlk->ngroups >= groups)
954 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
955 if (new_groups == NULL) {
959 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
960 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
962 nlk->groups = new_groups;
963 nlk->ngroups = groups;
965 netlink_table_ungrab();
969 static void netlink_undo_bind(int group, long unsigned int groups,
972 struct netlink_sock *nlk = nlk_sk(sk);
975 if (!nlk->netlink_unbind)
978 for (undo = 0; undo < group; undo++)
979 if (test_bit(undo, &groups))
980 nlk->netlink_unbind(sock_net(sk), undo + 1);
983 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
986 struct sock *sk = sock->sk;
987 struct net *net = sock_net(sk);
988 struct netlink_sock *nlk = nlk_sk(sk);
989 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
991 unsigned long groups = nladdr->nl_groups;
994 if (addr_len < sizeof(struct sockaddr_nl))
997 if (nladdr->nl_family != AF_NETLINK)
1000 /* Only superuser is allowed to listen multicasts */
1002 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1004 err = netlink_realloc_groups(sk);
1009 if (nlk->ngroups < BITS_PER_LONG)
1010 groups &= (1UL << nlk->ngroups) - 1;
1014 /* Ensure nlk->portid is up-to-date. */
1017 if (nladdr->nl_pid != nlk->portid)
1021 netlink_lock_table();
1022 if (nlk->netlink_bind && groups) {
1025 for (group = 0; group < nlk->ngroups; group++) {
1026 if (!test_bit(group, &groups))
1028 err = nlk->netlink_bind(net, group + 1);
1031 netlink_undo_bind(group, groups, sk);
1036 /* No need for barriers here as we return to user-space without
1037 * using any of the bound attributes.
1040 err = nladdr->nl_pid ?
1041 netlink_insert(sk, nladdr->nl_pid) :
1042 netlink_autobind(sock);
1044 netlink_undo_bind(nlk->ngroups, groups, sk);
1049 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1051 netlink_unlock_table();
1053 netlink_table_grab();
1054 netlink_update_subscriptions(sk, nlk->subscriptions +
1056 hweight32(nlk->groups[0]));
1057 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1058 netlink_update_listeners(sk);
1059 netlink_table_ungrab();
1064 netlink_unlock_table();
1068 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1069 int alen, int flags)
1072 struct sock *sk = sock->sk;
1073 struct netlink_sock *nlk = nlk_sk(sk);
1074 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1076 if (alen < sizeof(addr->sa_family))
1079 if (addr->sa_family == AF_UNSPEC) {
1080 sk->sk_state = NETLINK_UNCONNECTED;
1081 nlk->dst_portid = 0;
1085 if (addr->sa_family != AF_NETLINK)
1088 if (alen < sizeof(struct sockaddr_nl))
1091 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1092 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1095 /* No need for barriers here as we return to user-space without
1096 * using any of the bound attributes.
1099 err = netlink_autobind(sock);
1102 sk->sk_state = NETLINK_CONNECTED;
1103 nlk->dst_portid = nladdr->nl_pid;
1104 nlk->dst_group = ffs(nladdr->nl_groups);
1110 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1113 struct sock *sk = sock->sk;
1114 struct netlink_sock *nlk = nlk_sk(sk);
1115 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1117 nladdr->nl_family = AF_NETLINK;
1121 nladdr->nl_pid = nlk->dst_portid;
1122 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1124 nladdr->nl_pid = nlk->portid;
1125 netlink_lock_table();
1126 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1127 netlink_unlock_table();
1129 return sizeof(*nladdr);
1132 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1135 /* try to hand this ioctl down to the NIC drivers.
1137 return -ENOIOCTLCMD;
1140 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1143 struct netlink_sock *nlk;
1145 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1147 return ERR_PTR(-ECONNREFUSED);
1149 /* Don't bother queuing skb if kernel socket has no input function */
1151 if (sock->sk_state == NETLINK_CONNECTED &&
1152 nlk->dst_portid != nlk_sk(ssk)->portid) {
1154 return ERR_PTR(-ECONNREFUSED);
1159 struct sock *netlink_getsockbyfilp(struct file *filp)
1161 struct inode *inode = file_inode(filp);
1164 if (!S_ISSOCK(inode->i_mode))
1165 return ERR_PTR(-ENOTSOCK);
1167 sock = SOCKET_I(inode)->sk;
1168 if (sock->sk_family != AF_NETLINK)
1169 return ERR_PTR(-EINVAL);
1175 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1178 struct sk_buff *skb;
1181 if (size <= NLMSG_GOODSIZE || broadcast)
1182 return alloc_skb(size, GFP_KERNEL);
1184 size = SKB_DATA_ALIGN(size) +
1185 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1187 data = vmalloc(size);
1191 skb = __build_skb(data, size);
1195 skb->destructor = netlink_skb_destructor;
1201 * Attach a skb to a netlink socket.
1202 * The caller must hold a reference to the destination socket. On error, the
1203 * reference is dropped. The skb is not send to the destination, just all
1204 * all error checks are performed and memory in the queue is reserved.
1206 * < 0: error. skb freed, reference to sock dropped.
1208 * 1: repeat lookup - reference dropped while waiting for socket memory.
1210 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1211 long *timeo, struct sock *ssk)
1213 struct netlink_sock *nlk;
1217 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1218 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1219 DECLARE_WAITQUEUE(wait, current);
1221 if (!ssk || netlink_is_kernel(ssk))
1222 netlink_overrun(sk);
1228 __set_current_state(TASK_INTERRUPTIBLE);
1229 add_wait_queue(&nlk->wait, &wait);
1231 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1232 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1233 !sock_flag(sk, SOCK_DEAD))
1234 *timeo = schedule_timeout(*timeo);
1236 __set_current_state(TASK_RUNNING);
1237 remove_wait_queue(&nlk->wait, &wait);
1240 if (signal_pending(current)) {
1242 return sock_intr_errno(*timeo);
1246 netlink_skb_set_owner_r(skb, sk);
1250 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1254 netlink_deliver_tap(sock_net(sk), skb);
1256 skb_queue_tail(&sk->sk_receive_queue, skb);
1257 sk->sk_data_ready(sk);
1261 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1263 int len = __netlink_sendskb(sk, skb);
1269 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1275 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1279 WARN_ON(skb->sk != NULL);
1280 delta = skb->end - skb->tail;
1281 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1284 if (skb_shared(skb)) {
1285 struct sk_buff *nskb = skb_clone(skb, allocation);
1292 pskb_expand_head(skb, 0, -delta,
1293 (allocation & ~__GFP_DIRECT_RECLAIM) |
1294 __GFP_NOWARN | __GFP_NORETRY);
1298 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1302 struct netlink_sock *nlk = nlk_sk(sk);
1304 ret = -ECONNREFUSED;
1305 if (nlk->netlink_rcv != NULL) {
1307 netlink_skb_set_owner_r(skb, sk);
1308 NETLINK_CB(skb).sk = ssk;
1309 netlink_deliver_tap_kernel(sk, ssk, skb);
1310 nlk->netlink_rcv(skb);
1319 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1320 u32 portid, int nonblock)
1326 skb = netlink_trim(skb, gfp_any());
1328 timeo = sock_sndtimeo(ssk, nonblock);
1330 sk = netlink_getsockbyportid(ssk, portid);
1335 if (netlink_is_kernel(sk))
1336 return netlink_unicast_kernel(sk, skb, ssk);
1338 if (sk_filter(sk, skb)) {
1345 err = netlink_attachskb(sk, skb, &timeo, ssk);
1351 return netlink_sendskb(sk, skb);
1353 EXPORT_SYMBOL(netlink_unicast);
1355 int netlink_has_listeners(struct sock *sk, unsigned int group)
1358 struct listeners *listeners;
1360 BUG_ON(!netlink_is_kernel(sk));
1363 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1365 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1366 res = test_bit(group - 1, listeners->masks);
1372 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1374 bool netlink_strict_get_check(struct sk_buff *skb)
1376 const struct netlink_sock *nlk = nlk_sk(NETLINK_CB(skb).sk);
1378 return nlk->flags & NETLINK_F_STRICT_CHK;
1380 EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1382 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1384 struct netlink_sock *nlk = nlk_sk(sk);
1386 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1387 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1388 netlink_skb_set_owner_r(skb, sk);
1389 __netlink_sendskb(sk, skb);
1390 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1395 struct netlink_broadcast_data {
1396 struct sock *exclude_sk;
1401 int delivery_failure;
1405 struct sk_buff *skb, *skb2;
1406 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1410 static void do_one_broadcast(struct sock *sk,
1411 struct netlink_broadcast_data *p)
1413 struct netlink_sock *nlk = nlk_sk(sk);
1416 if (p->exclude_sk == sk)
1419 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1420 !test_bit(p->group - 1, nlk->groups))
1423 if (!net_eq(sock_net(sk), p->net)) {
1424 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1427 if (!peernet_has_id(sock_net(sk), p->net))
1430 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1436 netlink_overrun(sk);
1441 if (p->skb2 == NULL) {
1442 if (skb_shared(p->skb)) {
1443 p->skb2 = skb_clone(p->skb, p->allocation);
1445 p->skb2 = skb_get(p->skb);
1447 * skb ownership may have been set when
1448 * delivered to a previous socket.
1450 skb_orphan(p->skb2);
1453 if (p->skb2 == NULL) {
1454 netlink_overrun(sk);
1455 /* Clone failed. Notify ALL listeners. */
1457 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1458 p->delivery_failure = 1;
1461 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1466 if (sk_filter(sk, p->skb2)) {
1471 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1472 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1473 NETLINK_CB(p->skb2).nsid_is_set = true;
1474 val = netlink_broadcast_deliver(sk, p->skb2);
1476 netlink_overrun(sk);
1477 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1478 p->delivery_failure = 1;
1480 p->congested |= val;
1488 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1489 u32 group, gfp_t allocation,
1490 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1493 struct net *net = sock_net(ssk);
1494 struct netlink_broadcast_data info;
1497 skb = netlink_trim(skb, allocation);
1499 info.exclude_sk = ssk;
1501 info.portid = portid;
1504 info.delivery_failure = 0;
1507 info.allocation = allocation;
1510 info.tx_filter = filter;
1511 info.tx_data = filter_data;
1513 /* While we sleep in clone, do not allow to change socket list */
1515 netlink_lock_table();
1517 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1518 do_one_broadcast(sk, &info);
1522 netlink_unlock_table();
1524 if (info.delivery_failure) {
1525 kfree_skb(info.skb2);
1528 consume_skb(info.skb2);
1530 if (info.delivered) {
1531 if (info.congested && gfpflags_allow_blocking(allocation))
1537 EXPORT_SYMBOL(netlink_broadcast_filtered);
1539 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1540 u32 group, gfp_t allocation)
1542 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1545 EXPORT_SYMBOL(netlink_broadcast);
1547 struct netlink_set_err_data {
1548 struct sock *exclude_sk;
1554 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1556 struct netlink_sock *nlk = nlk_sk(sk);
1559 if (sk == p->exclude_sk)
1562 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1565 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1566 !test_bit(p->group - 1, nlk->groups))
1569 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1574 sk->sk_err = p->code;
1575 sk->sk_error_report(sk);
1581 * netlink_set_err - report error to broadcast listeners
1582 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1583 * @portid: the PORTID of a process that we want to skip (if any)
1584 * @group: the broadcast group that will notice the error
1585 * @code: error code, must be negative (as usual in kernelspace)
1587 * This function returns the number of broadcast listeners that have set the
1588 * NETLINK_NO_ENOBUFS socket option.
1590 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1592 struct netlink_set_err_data info;
1596 info.exclude_sk = ssk;
1597 info.portid = portid;
1599 /* sk->sk_err wants a positive error value */
1602 read_lock(&nl_table_lock);
1604 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1605 ret += do_one_set_err(sk, &info);
1607 read_unlock(&nl_table_lock);
1610 EXPORT_SYMBOL(netlink_set_err);
1612 /* must be called with netlink table grabbed */
1613 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1617 int old, new = !!is_new, subscriptions;
1619 old = test_bit(group - 1, nlk->groups);
1620 subscriptions = nlk->subscriptions - old + new;
1622 __set_bit(group - 1, nlk->groups);
1624 __clear_bit(group - 1, nlk->groups);
1625 netlink_update_subscriptions(&nlk->sk, subscriptions);
1626 netlink_update_listeners(&nlk->sk);
1629 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1630 char __user *optval, unsigned int optlen)
1632 struct sock *sk = sock->sk;
1633 struct netlink_sock *nlk = nlk_sk(sk);
1634 unsigned int val = 0;
1637 if (level != SOL_NETLINK)
1638 return -ENOPROTOOPT;
1640 if (optlen >= sizeof(int) &&
1641 get_user(val, (unsigned int __user *)optval))
1645 case NETLINK_PKTINFO:
1647 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1649 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1652 case NETLINK_ADD_MEMBERSHIP:
1653 case NETLINK_DROP_MEMBERSHIP: {
1654 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1656 err = netlink_realloc_groups(sk);
1659 if (!val || val - 1 >= nlk->ngroups)
1661 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1662 err = nlk->netlink_bind(sock_net(sk), val);
1666 netlink_table_grab();
1667 netlink_update_socket_mc(nlk, val,
1668 optname == NETLINK_ADD_MEMBERSHIP);
1669 netlink_table_ungrab();
1670 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1671 nlk->netlink_unbind(sock_net(sk), val);
1676 case NETLINK_BROADCAST_ERROR:
1678 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1680 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1683 case NETLINK_NO_ENOBUFS:
1685 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1686 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1687 wake_up_interruptible(&nlk->wait);
1689 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1693 case NETLINK_LISTEN_ALL_NSID:
1694 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1698 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1700 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1703 case NETLINK_CAP_ACK:
1705 nlk->flags |= NETLINK_F_CAP_ACK;
1707 nlk->flags &= ~NETLINK_F_CAP_ACK;
1710 case NETLINK_EXT_ACK:
1712 nlk->flags |= NETLINK_F_EXT_ACK;
1714 nlk->flags &= ~NETLINK_F_EXT_ACK;
1717 case NETLINK_GET_STRICT_CHK:
1719 nlk->flags |= NETLINK_F_STRICT_CHK;
1721 nlk->flags &= ~NETLINK_F_STRICT_CHK;
1730 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1731 char __user *optval, int __user *optlen)
1733 struct sock *sk = sock->sk;
1734 struct netlink_sock *nlk = nlk_sk(sk);
1737 if (level != SOL_NETLINK)
1738 return -ENOPROTOOPT;
1740 if (get_user(len, optlen))
1746 case NETLINK_PKTINFO:
1747 if (len < sizeof(int))
1750 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1751 if (put_user(len, optlen) ||
1752 put_user(val, optval))
1756 case NETLINK_BROADCAST_ERROR:
1757 if (len < sizeof(int))
1760 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1761 if (put_user(len, optlen) ||
1762 put_user(val, optval))
1766 case NETLINK_NO_ENOBUFS:
1767 if (len < sizeof(int))
1770 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1771 if (put_user(len, optlen) ||
1772 put_user(val, optval))
1776 case NETLINK_LIST_MEMBERSHIPS: {
1777 int pos, idx, shift;
1780 netlink_lock_table();
1781 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1782 if (len - pos < sizeof(u32))
1785 idx = pos / sizeof(unsigned long);
1786 shift = (pos % sizeof(unsigned long)) * 8;
1787 if (put_user((u32)(nlk->groups[idx] >> shift),
1788 (u32 __user *)(optval + pos))) {
1793 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1795 netlink_unlock_table();
1798 case NETLINK_CAP_ACK:
1799 if (len < sizeof(int))
1802 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1803 if (put_user(len, optlen) ||
1804 put_user(val, optval))
1808 case NETLINK_EXT_ACK:
1809 if (len < sizeof(int))
1812 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1813 if (put_user(len, optlen) || put_user(val, optval))
1817 case NETLINK_GET_STRICT_CHK:
1818 if (len < sizeof(int))
1821 val = nlk->flags & NETLINK_F_STRICT_CHK ? 1 : 0;
1822 if (put_user(len, optlen) || put_user(val, optval))
1832 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1834 struct nl_pktinfo info;
1836 info.group = NETLINK_CB(skb).dst_group;
1837 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1840 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1841 struct sk_buff *skb)
1843 if (!NETLINK_CB(skb).nsid_is_set)
1846 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1847 &NETLINK_CB(skb).nsid);
1850 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1852 struct sock *sk = sock->sk;
1853 struct netlink_sock *nlk = nlk_sk(sk);
1854 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1857 struct sk_buff *skb;
1859 struct scm_cookie scm;
1860 u32 netlink_skb_flags = 0;
1862 if (msg->msg_flags&MSG_OOB)
1865 err = scm_send(sock, msg, &scm, true);
1869 if (msg->msg_namelen) {
1871 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1873 if (addr->nl_family != AF_NETLINK)
1875 dst_portid = addr->nl_pid;
1876 dst_group = ffs(addr->nl_groups);
1878 if ((dst_group || dst_portid) &&
1879 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1881 netlink_skb_flags |= NETLINK_SKB_DST;
1883 dst_portid = nlk->dst_portid;
1884 dst_group = nlk->dst_group;
1888 err = netlink_autobind(sock);
1892 /* Ensure nlk is hashed and visible. */
1897 if (len > sk->sk_sndbuf - 32)
1900 skb = netlink_alloc_large_skb(len, dst_group);
1904 NETLINK_CB(skb).portid = nlk->portid;
1905 NETLINK_CB(skb).dst_group = dst_group;
1906 NETLINK_CB(skb).creds = scm.creds;
1907 NETLINK_CB(skb).flags = netlink_skb_flags;
1910 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1915 err = security_netlink_send(sk, skb);
1922 refcount_inc(&skb->users);
1923 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1925 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1932 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1935 struct scm_cookie scm;
1936 struct sock *sk = sock->sk;
1937 struct netlink_sock *nlk = nlk_sk(sk);
1938 int noblock = flags&MSG_DONTWAIT;
1940 struct sk_buff *skb, *data_skb;
1948 skb = skb_recv_datagram(sk, flags, noblock, &err);
1954 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1955 if (unlikely(skb_shinfo(skb)->frag_list)) {
1957 * If this skb has a frag_list, then here that means that we
1958 * will have to use the frag_list skb's data for compat tasks
1959 * and the regular skb's data for normal (non-compat) tasks.
1961 * If we need to send the compat skb, assign it to the
1962 * 'data_skb' variable so that it will be used below for data
1963 * copying. We keep 'skb' for everything else, including
1964 * freeing both later.
1966 if (flags & MSG_CMSG_COMPAT)
1967 data_skb = skb_shinfo(skb)->frag_list;
1971 /* Record the max length of recvmsg() calls for future allocations */
1972 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1973 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1974 SKB_WITH_OVERHEAD(32768));
1976 copied = data_skb->len;
1978 msg->msg_flags |= MSG_TRUNC;
1982 skb_reset_transport_header(data_skb);
1983 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1985 if (msg->msg_name) {
1986 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1987 addr->nl_family = AF_NETLINK;
1989 addr->nl_pid = NETLINK_CB(skb).portid;
1990 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1991 msg->msg_namelen = sizeof(*addr);
1994 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1995 netlink_cmsg_recv_pktinfo(msg, skb);
1996 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1997 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1999 memset(&scm, 0, sizeof(scm));
2000 scm.creds = *NETLINK_CREDS(skb);
2001 if (flags & MSG_TRUNC)
2002 copied = data_skb->len;
2004 skb_free_datagram(sk, skb);
2006 if (nlk->cb_running &&
2007 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2008 ret = netlink_dump(sk);
2011 sk->sk_error_report(sk);
2015 scm_recv(sock, msg, &scm, flags);
2017 netlink_rcv_wake(sk);
2018 return err ? : copied;
2021 static void netlink_data_ready(struct sock *sk)
2027 * We export these functions to other modules. They provide a
2028 * complete set of kernel non-blocking support for message
2033 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2034 struct netlink_kernel_cfg *cfg)
2036 struct socket *sock;
2038 struct netlink_sock *nlk;
2039 struct listeners *listeners = NULL;
2040 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2041 unsigned int groups;
2045 if (unit < 0 || unit >= MAX_LINKS)
2048 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2051 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2052 goto out_sock_release_nosk;
2056 if (!cfg || cfg->groups < 32)
2059 groups = cfg->groups;
2061 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2063 goto out_sock_release;
2065 sk->sk_data_ready = netlink_data_ready;
2066 if (cfg && cfg->input)
2067 nlk_sk(sk)->netlink_rcv = cfg->input;
2069 if (netlink_insert(sk, 0))
2070 goto out_sock_release;
2073 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2075 netlink_table_grab();
2076 if (!nl_table[unit].registered) {
2077 nl_table[unit].groups = groups;
2078 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2079 nl_table[unit].cb_mutex = cb_mutex;
2080 nl_table[unit].module = module;
2082 nl_table[unit].bind = cfg->bind;
2083 nl_table[unit].unbind = cfg->unbind;
2084 nl_table[unit].flags = cfg->flags;
2086 nl_table[unit].compare = cfg->compare;
2088 nl_table[unit].registered = 1;
2091 nl_table[unit].registered++;
2093 netlink_table_ungrab();
2098 netlink_kernel_release(sk);
2101 out_sock_release_nosk:
2105 EXPORT_SYMBOL(__netlink_kernel_create);
2108 netlink_kernel_release(struct sock *sk)
2110 if (sk == NULL || sk->sk_socket == NULL)
2113 sock_release(sk->sk_socket);
2115 EXPORT_SYMBOL(netlink_kernel_release);
2117 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2119 struct listeners *new, *old;
2120 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2125 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2126 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2129 old = nl_deref_protected(tbl->listeners);
2130 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2131 rcu_assign_pointer(tbl->listeners, new);
2133 kfree_rcu(old, rcu);
2135 tbl->groups = groups;
2141 * netlink_change_ngroups - change number of multicast groups
2143 * This changes the number of multicast groups that are available
2144 * on a certain netlink family. Note that it is not possible to
2145 * change the number of groups to below 32. Also note that it does
2146 * not implicitly call netlink_clear_multicast_users() when the
2147 * number of groups is reduced.
2149 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2150 * @groups: The new number of groups.
2152 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2156 netlink_table_grab();
2157 err = __netlink_change_ngroups(sk, groups);
2158 netlink_table_ungrab();
2163 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2166 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2168 sk_for_each_bound(sk, &tbl->mc_list)
2169 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2173 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2175 struct nlmsghdr *nlh;
2176 int size = nlmsg_msg_size(len);
2178 nlh = skb_put(skb, NLMSG_ALIGN(size));
2179 nlh->nlmsg_type = type;
2180 nlh->nlmsg_len = size;
2181 nlh->nlmsg_flags = flags;
2182 nlh->nlmsg_pid = portid;
2183 nlh->nlmsg_seq = seq;
2184 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2185 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2188 EXPORT_SYMBOL(__nlmsg_put);
2191 * It looks a bit ugly.
2192 * It would be better to create kernel thread.
2195 static int netlink_dump(struct sock *sk)
2197 struct netlink_sock *nlk = nlk_sk(sk);
2198 struct netlink_ext_ack extack = {};
2199 struct netlink_callback *cb;
2200 struct sk_buff *skb = NULL;
2201 struct nlmsghdr *nlh;
2202 struct module *module;
2207 mutex_lock(nlk->cb_mutex);
2208 if (!nlk->cb_running) {
2213 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2216 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2217 * required, but it makes sense to _attempt_ a 16K bytes allocation
2218 * to reduce number of system calls on dump operations, if user
2219 * ever provided a big enough buffer.
2222 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2224 if (alloc_min_size < nlk->max_recvmsg_len) {
2225 alloc_size = nlk->max_recvmsg_len;
2226 skb = alloc_skb(alloc_size,
2227 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2228 __GFP_NOWARN | __GFP_NORETRY);
2231 alloc_size = alloc_min_size;
2232 skb = alloc_skb(alloc_size, GFP_KERNEL);
2237 /* Trim skb to allocated size. User is expected to provide buffer as
2238 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2239 * netlink_recvmsg())). dump will pack as many smaller messages as
2240 * could fit within the allocated skb. skb is typically allocated
2241 * with larger space than required (could be as much as near 2x the
2242 * requested size with align to next power of 2 approach). Allowing
2243 * dump to use the excess space makes it difficult for a user to have a
2244 * reasonable static buffer based on the expected largest dump of a
2245 * single netdev. The outcome is MSG_TRUNC error.
2247 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2248 netlink_skb_set_owner_r(skb, sk);
2250 if (nlk->dump_done_errno > 0) {
2251 cb->extack = &extack;
2252 nlk->dump_done_errno = cb->dump(skb, cb);
2256 if (nlk->dump_done_errno > 0 ||
2257 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2258 mutex_unlock(nlk->cb_mutex);
2260 if (sk_filter(sk, skb))
2263 __netlink_sendskb(sk, skb);
2267 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE,
2268 sizeof(nlk->dump_done_errno),
2269 NLM_F_MULTI | cb->answer_flags);
2273 nl_dump_check_consistent(cb, nlh);
2275 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno,
2276 sizeof(nlk->dump_done_errno));
2278 if (extack._msg && nlk->flags & NETLINK_F_EXT_ACK) {
2279 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2280 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack._msg))
2281 nlmsg_end(skb, nlh);
2284 if (sk_filter(sk, skb))
2287 __netlink_sendskb(sk, skb);
2292 nlk->cb_running = false;
2293 module = cb->module;
2295 mutex_unlock(nlk->cb_mutex);
2301 mutex_unlock(nlk->cb_mutex);
2306 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2307 const struct nlmsghdr *nlh,
2308 struct netlink_dump_control *control)
2310 struct netlink_sock *nlk, *nlk2;
2311 struct netlink_callback *cb;
2315 refcount_inc(&skb->users);
2317 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2319 ret = -ECONNREFUSED;
2324 mutex_lock(nlk->cb_mutex);
2325 /* A dump is in progress... */
2326 if (nlk->cb_running) {
2330 /* add reference of module which cb->dump belongs to */
2331 if (!try_module_get(control->module)) {
2332 ret = -EPROTONOSUPPORT;
2337 memset(cb, 0, sizeof(*cb));
2338 cb->dump = control->dump;
2339 cb->done = control->done;
2341 cb->data = control->data;
2342 cb->module = control->module;
2343 cb->min_dump_alloc = control->min_dump_alloc;
2346 nlk2 = nlk_sk(NETLINK_CB(skb).sk);
2347 cb->strict_check = !!(nlk2->flags & NETLINK_F_STRICT_CHK);
2349 if (control->start) {
2350 ret = control->start(cb);
2355 nlk->cb_running = true;
2356 nlk->dump_done_errno = INT_MAX;
2358 mutex_unlock(nlk->cb_mutex);
2360 ret = netlink_dump(sk);
2367 /* We successfully started a dump, by returning -EINTR we
2368 * signal not to send ACK even if it was requested.
2373 module_put(control->module);
2376 mutex_unlock(nlk->cb_mutex);
2381 EXPORT_SYMBOL(__netlink_dump_start);
2383 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2384 const struct netlink_ext_ack *extack)
2386 struct sk_buff *skb;
2387 struct nlmsghdr *rep;
2388 struct nlmsgerr *errmsg;
2389 size_t payload = sizeof(*errmsg);
2391 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2392 unsigned int flags = 0;
2393 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2395 /* Error messages get the original request appened, unless the user
2396 * requests to cap the error message, and get extra error data if
2399 if (nlk_has_extack && extack && extack->_msg)
2400 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2403 if (!(nlk->flags & NETLINK_F_CAP_ACK))
2404 payload += nlmsg_len(nlh);
2406 flags |= NLM_F_CAPPED;
2407 if (nlk_has_extack && extack && extack->bad_attr)
2408 tlvlen += nla_total_size(sizeof(u32));
2410 flags |= NLM_F_CAPPED;
2412 if (nlk_has_extack && extack && extack->cookie_len)
2413 tlvlen += nla_total_size(extack->cookie_len);
2417 flags |= NLM_F_ACK_TLVS;
2419 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2421 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2422 NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk);
2426 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2427 NLMSG_ERROR, payload, flags);
2428 errmsg = nlmsg_data(rep);
2429 errmsg->error = err;
2430 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2432 if (nlk_has_extack && extack) {
2434 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2438 if (extack->bad_attr &&
2439 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2440 (u8 *)extack->bad_attr >= in_skb->data +
2442 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2443 (u8 *)extack->bad_attr -
2446 if (extack->cookie_len)
2447 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2453 nlmsg_end(skb, rep);
2455 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2457 EXPORT_SYMBOL(netlink_ack);
2459 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2461 struct netlink_ext_ack *))
2463 struct netlink_ext_ack extack;
2464 struct nlmsghdr *nlh;
2467 while (skb->len >= nlmsg_total_size(0)) {
2470 memset(&extack, 0, sizeof(extack));
2471 nlh = nlmsg_hdr(skb);
2474 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2477 /* Only requests are handled by the kernel */
2478 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2481 /* Skip control messages */
2482 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2485 err = cb(skb, nlh, &extack);
2490 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2491 netlink_ack(skb, nlh, err, &extack);
2494 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2495 if (msglen > skb->len)
2497 skb_pull(skb, msglen);
2502 EXPORT_SYMBOL(netlink_rcv_skb);
2505 * nlmsg_notify - send a notification netlink message
2506 * @sk: netlink socket to use
2507 * @skb: notification message
2508 * @portid: destination netlink portid for reports or 0
2509 * @group: destination multicast group or 0
2510 * @report: 1 to report back, 0 to disable
2511 * @flags: allocation flags
2513 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2514 unsigned int group, int report, gfp_t flags)
2519 int exclude_portid = 0;
2522 refcount_inc(&skb->users);
2523 exclude_portid = portid;
2526 /* errors reported via destination sk->sk_err, but propagate
2527 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2528 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2534 err2 = nlmsg_unicast(sk, skb, portid);
2535 if (!err || err == -ESRCH)
2541 EXPORT_SYMBOL(nlmsg_notify);
2543 #ifdef CONFIG_PROC_FS
2544 struct nl_seq_iter {
2545 struct seq_net_private p;
2546 struct rhashtable_iter hti;
2550 static int netlink_walk_start(struct nl_seq_iter *iter)
2552 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2553 rhashtable_walk_start(&iter->hti);
2558 static void netlink_walk_stop(struct nl_seq_iter *iter)
2560 rhashtable_walk_stop(&iter->hti);
2561 rhashtable_walk_exit(&iter->hti);
2564 static void *__netlink_seq_next(struct seq_file *seq)
2566 struct nl_seq_iter *iter = seq->private;
2567 struct netlink_sock *nlk;
2573 nlk = rhashtable_walk_next(&iter->hti);
2576 if (PTR_ERR(nlk) == -EAGAIN)
2585 netlink_walk_stop(iter);
2586 if (++iter->link >= MAX_LINKS)
2589 err = netlink_walk_start(iter);
2591 return ERR_PTR(err);
2593 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2598 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2600 struct nl_seq_iter *iter = seq->private;
2601 void *obj = SEQ_START_TOKEN;
2607 err = netlink_walk_start(iter);
2609 return ERR_PTR(err);
2611 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2612 obj = __netlink_seq_next(seq);
2617 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2620 return __netlink_seq_next(seq);
2623 static void netlink_seq_stop(struct seq_file *seq, void *v)
2625 struct nl_seq_iter *iter = seq->private;
2627 if (iter->link >= MAX_LINKS)
2630 netlink_walk_stop(iter);
2634 static int netlink_seq_show(struct seq_file *seq, void *v)
2636 if (v == SEQ_START_TOKEN) {
2638 "sk Eth Pid Groups "
2639 "Rmem Wmem Dump Locks Drops Inode\n");
2642 struct netlink_sock *nlk = nlk_sk(s);
2644 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8d %-8lu\n",
2648 nlk->groups ? (u32)nlk->groups[0] : 0,
2649 sk_rmem_alloc_get(s),
2650 sk_wmem_alloc_get(s),
2652 refcount_read(&s->sk_refcnt),
2653 atomic_read(&s->sk_drops),
2661 static const struct seq_operations netlink_seq_ops = {
2662 .start = netlink_seq_start,
2663 .next = netlink_seq_next,
2664 .stop = netlink_seq_stop,
2665 .show = netlink_seq_show,
2669 int netlink_register_notifier(struct notifier_block *nb)
2671 return blocking_notifier_chain_register(&netlink_chain, nb);
2673 EXPORT_SYMBOL(netlink_register_notifier);
2675 int netlink_unregister_notifier(struct notifier_block *nb)
2677 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2679 EXPORT_SYMBOL(netlink_unregister_notifier);
2681 static const struct proto_ops netlink_ops = {
2682 .family = PF_NETLINK,
2683 .owner = THIS_MODULE,
2684 .release = netlink_release,
2685 .bind = netlink_bind,
2686 .connect = netlink_connect,
2687 .socketpair = sock_no_socketpair,
2688 .accept = sock_no_accept,
2689 .getname = netlink_getname,
2690 .poll = datagram_poll,
2691 .ioctl = netlink_ioctl,
2692 .listen = sock_no_listen,
2693 .shutdown = sock_no_shutdown,
2694 .setsockopt = netlink_setsockopt,
2695 .getsockopt = netlink_getsockopt,
2696 .sendmsg = netlink_sendmsg,
2697 .recvmsg = netlink_recvmsg,
2698 .mmap = sock_no_mmap,
2699 .sendpage = sock_no_sendpage,
2702 static const struct net_proto_family netlink_family_ops = {
2703 .family = PF_NETLINK,
2704 .create = netlink_create,
2705 .owner = THIS_MODULE, /* for consistency 8) */
2708 static int __net_init netlink_net_init(struct net *net)
2710 #ifdef CONFIG_PROC_FS
2711 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2712 sizeof(struct nl_seq_iter)))
2718 static void __net_exit netlink_net_exit(struct net *net)
2720 #ifdef CONFIG_PROC_FS
2721 remove_proc_entry("netlink", net->proc_net);
2725 static void __init netlink_add_usersock_entry(void)
2727 struct listeners *listeners;
2730 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2732 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2734 netlink_table_grab();
2736 nl_table[NETLINK_USERSOCK].groups = groups;
2737 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2738 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2739 nl_table[NETLINK_USERSOCK].registered = 1;
2740 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2742 netlink_table_ungrab();
2745 static struct pernet_operations __net_initdata netlink_net_ops = {
2746 .init = netlink_net_init,
2747 .exit = netlink_net_exit,
2750 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2752 const struct netlink_sock *nlk = data;
2753 struct netlink_compare_arg arg;
2755 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2756 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2759 static const struct rhashtable_params netlink_rhashtable_params = {
2760 .head_offset = offsetof(struct netlink_sock, node),
2761 .key_len = netlink_compare_arg_len,
2762 .obj_hashfn = netlink_hash,
2763 .obj_cmpfn = netlink_compare,
2764 .automatic_shrinking = true,
2767 static int __init netlink_proto_init(void)
2770 int err = proto_register(&netlink_proto, 0);
2775 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2777 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2781 for (i = 0; i < MAX_LINKS; i++) {
2782 if (rhashtable_init(&nl_table[i].hash,
2783 &netlink_rhashtable_params) < 0) {
2785 rhashtable_destroy(&nl_table[i].hash);
2791 netlink_add_usersock_entry();
2793 sock_register(&netlink_family_ops);
2794 register_pernet_subsys(&netlink_net_ops);
2795 register_pernet_subsys(&netlink_tap_net_ops);
2796 /* The netlink device handler may be needed early. */
2801 panic("netlink_init: Cannot allocate nl_table\n");
2804 core_initcall(netlink_proto_init);