2 * NETLINK Kernel-user communication protocol.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
12 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
13 * added netlink_proto_exit
14 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
15 * use nlk_sk, as sk->protinfo is on a diet 8)
16 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
17 * - inc module use count of module that owns
18 * the kernel socket in case userspace opens
19 * socket of same protocol
20 * - remove all module support, since netlink is
21 * mandatory if CONFIG_NET=y these days
24 #include <linux/module.h>
26 #include <linux/capability.h>
27 #include <linux/kernel.h>
28 #include <linux/init.h>
29 #include <linux/signal.h>
30 #include <linux/sched.h>
31 #include <linux/errno.h>
32 #include <linux/string.h>
33 #include <linux/stat.h>
34 #include <linux/socket.h>
36 #include <linux/fcntl.h>
37 #include <linux/termios.h>
38 #include <linux/sockios.h>
39 #include <linux/net.h>
41 #include <linux/slab.h>
42 #include <asm/uaccess.h>
43 #include <linux/skbuff.h>
44 #include <linux/netdevice.h>
45 #include <linux/rtnetlink.h>
46 #include <linux/proc_fs.h>
47 #include <linux/seq_file.h>
48 #include <linux/notifier.h>
49 #include <linux/security.h>
50 #include <linux/jhash.h>
51 #include <linux/jiffies.h>
52 #include <linux/random.h>
53 #include <linux/bitops.h>
55 #include <linux/types.h>
56 #include <linux/audit.h>
57 #include <linux/mutex.h>
59 #include <net/net_namespace.h>
62 #include <net/netlink.h>
64 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
65 #define NLGRPLONGS(x) (NLGRPSZ(x)/sizeof(unsigned long))
68 /* struct sock has to be the first member of netlink_sock */
76 unsigned long *groups;
78 wait_queue_head_t wait;
79 struct netlink_callback *cb;
80 struct mutex *cb_mutex;
81 struct mutex cb_def_mutex;
82 void (*netlink_rcv)(struct sk_buff *skb);
83 struct module *module;
86 #define NETLINK_KERNEL_SOCKET 0x1
87 #define NETLINK_RECV_PKTINFO 0x2
89 static inline struct netlink_sock *nlk_sk(struct sock *sk)
91 return container_of(sk, struct netlink_sock, sk);
94 static inline int netlink_is_kernel(struct sock *sk)
96 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
100 struct hlist_head *table;
101 unsigned long rehash_time;
106 unsigned int entries;
107 unsigned int max_shift;
112 struct netlink_table {
113 struct nl_pid_hash hash;
114 struct hlist_head mc_list;
115 unsigned long *listeners;
116 unsigned int nl_nonroot;
118 struct mutex *cb_mutex;
119 struct module *module;
123 static struct netlink_table *nl_table;
125 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
127 static int netlink_dump(struct sock *sk);
128 static void netlink_destroy_callback(struct netlink_callback *cb);
130 static DEFINE_RWLOCK(nl_table_lock);
131 static atomic_t nl_table_users = ATOMIC_INIT(0);
133 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
135 static u32 netlink_group_mask(u32 group)
137 return group ? 1 << (group - 1) : 0;
140 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
142 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
145 static void netlink_sock_destruct(struct sock *sk)
147 struct netlink_sock *nlk = nlk_sk(sk);
151 nlk->cb->done(nlk->cb);
152 netlink_destroy_callback(nlk->cb);
155 skb_queue_purge(&sk->sk_receive_queue);
157 if (!sock_flag(sk, SOCK_DEAD)) {
158 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
162 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
163 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
164 WARN_ON(nlk_sk(sk)->groups);
167 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
168 * SMP. Look, when several writers sleep and reader wakes them up, all but one
169 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
170 * this, _but_ remember, it adds useless work on UP machines.
173 static void netlink_table_grab(void)
174 __acquires(nl_table_lock)
176 write_lock_irq(&nl_table_lock);
178 if (atomic_read(&nl_table_users)) {
179 DECLARE_WAITQUEUE(wait, current);
181 add_wait_queue_exclusive(&nl_table_wait, &wait);
183 set_current_state(TASK_UNINTERRUPTIBLE);
184 if (atomic_read(&nl_table_users) == 0)
186 write_unlock_irq(&nl_table_lock);
188 write_lock_irq(&nl_table_lock);
191 __set_current_state(TASK_RUNNING);
192 remove_wait_queue(&nl_table_wait, &wait);
196 static void netlink_table_ungrab(void)
197 __releases(nl_table_lock)
199 write_unlock_irq(&nl_table_lock);
200 wake_up(&nl_table_wait);
204 netlink_lock_table(void)
206 /* read_lock() synchronizes us to netlink_table_grab */
208 read_lock(&nl_table_lock);
209 atomic_inc(&nl_table_users);
210 read_unlock(&nl_table_lock);
214 netlink_unlock_table(void)
216 if (atomic_dec_and_test(&nl_table_users))
217 wake_up(&nl_table_wait);
220 static inline struct sock *netlink_lookup(struct net *net, int protocol,
223 struct nl_pid_hash *hash = &nl_table[protocol].hash;
224 struct hlist_head *head;
226 struct hlist_node *node;
228 read_lock(&nl_table_lock);
229 head = nl_pid_hashfn(hash, pid);
230 sk_for_each(sk, node, head) {
231 if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) {
238 read_unlock(&nl_table_lock);
242 static inline struct hlist_head *nl_pid_hash_zalloc(size_t size)
244 if (size <= PAGE_SIZE)
245 return kzalloc(size, GFP_ATOMIC);
247 return (struct hlist_head *)
248 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
252 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
254 if (size <= PAGE_SIZE)
257 free_pages((unsigned long)table, get_order(size));
260 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
262 unsigned int omask, mask, shift;
264 struct hlist_head *otable, *table;
267 omask = mask = hash->mask;
268 osize = size = (mask + 1) * sizeof(*table);
272 if (++shift > hash->max_shift)
278 table = nl_pid_hash_zalloc(size);
282 otable = hash->table;
286 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
288 for (i = 0; i <= omask; i++) {
290 struct hlist_node *node, *tmp;
292 sk_for_each_safe(sk, node, tmp, &otable[i])
293 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
296 nl_pid_hash_free(otable, osize);
297 hash->rehash_time = jiffies + 10 * 60 * HZ;
301 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
303 int avg = hash->entries >> hash->shift;
305 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
308 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
309 nl_pid_hash_rehash(hash, 0);
316 static const struct proto_ops netlink_ops;
319 netlink_update_listeners(struct sock *sk)
321 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
322 struct hlist_node *node;
326 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
328 sk_for_each_bound(sk, node, &tbl->mc_list) {
329 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
330 mask |= nlk_sk(sk)->groups[i];
332 tbl->listeners[i] = mask;
334 /* this function is only called with the netlink table "grabbed", which
335 * makes sure updates are visible before bind or setsockopt return. */
338 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
340 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
341 struct hlist_head *head;
342 int err = -EADDRINUSE;
344 struct hlist_node *node;
347 netlink_table_grab();
348 head = nl_pid_hashfn(hash, pid);
350 sk_for_each(osk, node, head) {
351 if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid))
363 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
366 if (len && nl_pid_hash_dilute(hash, len))
367 head = nl_pid_hashfn(hash, pid);
369 nlk_sk(sk)->pid = pid;
370 sk_add_node(sk, head);
374 netlink_table_ungrab();
378 static void netlink_remove(struct sock *sk)
380 netlink_table_grab();
381 if (sk_del_node_init(sk))
382 nl_table[sk->sk_protocol].hash.entries--;
383 if (nlk_sk(sk)->subscriptions)
384 __sk_del_bind_node(sk);
385 netlink_table_ungrab();
388 static struct proto netlink_proto = {
390 .owner = THIS_MODULE,
391 .obj_size = sizeof(struct netlink_sock),
394 static int __netlink_create(struct net *net, struct socket *sock,
395 struct mutex *cb_mutex, int protocol)
398 struct netlink_sock *nlk;
400 sock->ops = &netlink_ops;
402 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
406 sock_init_data(sock, sk);
410 nlk->cb_mutex = cb_mutex;
412 nlk->cb_mutex = &nlk->cb_def_mutex;
413 mutex_init(nlk->cb_mutex);
415 init_waitqueue_head(&nlk->wait);
417 sk->sk_destruct = netlink_sock_destruct;
418 sk->sk_protocol = protocol;
422 static int netlink_create(struct net *net, struct socket *sock, int protocol)
424 struct module *module = NULL;
425 struct mutex *cb_mutex;
426 struct netlink_sock *nlk;
429 sock->state = SS_UNCONNECTED;
431 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
432 return -ESOCKTNOSUPPORT;
434 if (protocol < 0 || protocol >= MAX_LINKS)
435 return -EPROTONOSUPPORT;
437 netlink_lock_table();
438 #ifdef CONFIG_MODULES
439 if (!nl_table[protocol].registered) {
440 netlink_unlock_table();
441 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
442 netlink_lock_table();
445 if (nl_table[protocol].registered &&
446 try_module_get(nl_table[protocol].module))
447 module = nl_table[protocol].module;
448 cb_mutex = nl_table[protocol].cb_mutex;
449 netlink_unlock_table();
451 err = __netlink_create(net, sock, cb_mutex, protocol);
455 nlk = nlk_sk(sock->sk);
456 nlk->module = module;
465 static int netlink_release(struct socket *sock)
467 struct sock *sk = sock->sk;
468 struct netlink_sock *nlk;
478 * OK. Socket is unlinked, any packets that arrive now
483 wake_up_interruptible_all(&nlk->wait);
485 skb_queue_purge(&sk->sk_write_queue);
487 if (nlk->pid && !nlk->subscriptions) {
488 struct netlink_notify n = {
490 .protocol = sk->sk_protocol,
493 atomic_notifier_call_chain(&netlink_chain,
494 NETLINK_URELEASE, &n);
497 module_put(nlk->module);
499 netlink_table_grab();
500 if (netlink_is_kernel(sk)) {
501 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
502 if (--nl_table[sk->sk_protocol].registered == 0) {
503 kfree(nl_table[sk->sk_protocol].listeners);
504 nl_table[sk->sk_protocol].module = NULL;
505 nl_table[sk->sk_protocol].registered = 0;
507 } else if (nlk->subscriptions)
508 netlink_update_listeners(sk);
509 netlink_table_ungrab();
518 static int netlink_autobind(struct socket *sock)
520 struct sock *sk = sock->sk;
521 struct net *net = sock_net(sk);
522 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
523 struct hlist_head *head;
525 struct hlist_node *node;
526 s32 pid = current->tgid;
528 static s32 rover = -4097;
532 netlink_table_grab();
533 head = nl_pid_hashfn(hash, pid);
534 sk_for_each(osk, node, head) {
535 if (!net_eq(sock_net(osk), net))
537 if (nlk_sk(osk)->pid == pid) {
538 /* Bind collision, search negative pid values. */
542 netlink_table_ungrab();
546 netlink_table_ungrab();
548 err = netlink_insert(sk, net, pid);
549 if (err == -EADDRINUSE)
552 /* If 2 threads race to autobind, that is fine. */
559 static inline int netlink_capable(struct socket *sock, unsigned int flag)
561 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
562 capable(CAP_NET_ADMIN);
566 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
568 struct netlink_sock *nlk = nlk_sk(sk);
570 if (nlk->subscriptions && !subscriptions)
571 __sk_del_bind_node(sk);
572 else if (!nlk->subscriptions && subscriptions)
573 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
574 nlk->subscriptions = subscriptions;
577 static int netlink_realloc_groups(struct sock *sk)
579 struct netlink_sock *nlk = nlk_sk(sk);
581 unsigned long *new_groups;
584 netlink_table_grab();
586 groups = nl_table[sk->sk_protocol].groups;
587 if (!nl_table[sk->sk_protocol].registered) {
592 if (nlk->ngroups >= groups)
595 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
596 if (new_groups == NULL) {
600 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
601 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
603 nlk->groups = new_groups;
604 nlk->ngroups = groups;
606 netlink_table_ungrab();
610 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
613 struct sock *sk = sock->sk;
614 struct net *net = sock_net(sk);
615 struct netlink_sock *nlk = nlk_sk(sk);
616 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
619 if (nladdr->nl_family != AF_NETLINK)
622 /* Only superuser is allowed to listen multicasts */
623 if (nladdr->nl_groups) {
624 if (!netlink_capable(sock, NL_NONROOT_RECV))
626 err = netlink_realloc_groups(sk);
632 if (nladdr->nl_pid != nlk->pid)
635 err = nladdr->nl_pid ?
636 netlink_insert(sk, net, nladdr->nl_pid) :
637 netlink_autobind(sock);
642 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
645 netlink_table_grab();
646 netlink_update_subscriptions(sk, nlk->subscriptions +
647 hweight32(nladdr->nl_groups) -
648 hweight32(nlk->groups[0]));
649 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
650 netlink_update_listeners(sk);
651 netlink_table_ungrab();
656 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
660 struct sock *sk = sock->sk;
661 struct netlink_sock *nlk = nlk_sk(sk);
662 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
664 if (addr->sa_family == AF_UNSPEC) {
665 sk->sk_state = NETLINK_UNCONNECTED;
670 if (addr->sa_family != AF_NETLINK)
673 /* Only superuser is allowed to send multicasts */
674 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
678 err = netlink_autobind(sock);
681 sk->sk_state = NETLINK_CONNECTED;
682 nlk->dst_pid = nladdr->nl_pid;
683 nlk->dst_group = ffs(nladdr->nl_groups);
689 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
690 int *addr_len, int peer)
692 struct sock *sk = sock->sk;
693 struct netlink_sock *nlk = nlk_sk(sk);
694 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
696 nladdr->nl_family = AF_NETLINK;
698 *addr_len = sizeof(*nladdr);
701 nladdr->nl_pid = nlk->dst_pid;
702 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
704 nladdr->nl_pid = nlk->pid;
705 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
710 static void netlink_overrun(struct sock *sk)
712 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
713 sk->sk_err = ENOBUFS;
714 sk->sk_error_report(sk);
718 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
721 struct netlink_sock *nlk;
723 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
725 return ERR_PTR(-ECONNREFUSED);
727 /* Don't bother queuing skb if kernel socket has no input function */
729 if (sock->sk_state == NETLINK_CONNECTED &&
730 nlk->dst_pid != nlk_sk(ssk)->pid) {
732 return ERR_PTR(-ECONNREFUSED);
737 struct sock *netlink_getsockbyfilp(struct file *filp)
739 struct inode *inode = filp->f_path.dentry->d_inode;
742 if (!S_ISSOCK(inode->i_mode))
743 return ERR_PTR(-ENOTSOCK);
745 sock = SOCKET_I(inode)->sk;
746 if (sock->sk_family != AF_NETLINK)
747 return ERR_PTR(-EINVAL);
754 * Attach a skb to a netlink socket.
755 * The caller must hold a reference to the destination socket. On error, the
756 * reference is dropped. The skb is not send to the destination, just all
757 * all error checks are performed and memory in the queue is reserved.
759 * < 0: error. skb freed, reference to sock dropped.
761 * 1: repeat lookup - reference dropped while waiting for socket memory.
763 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
764 long *timeo, struct sock *ssk)
766 struct netlink_sock *nlk;
770 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
771 test_bit(0, &nlk->state)) {
772 DECLARE_WAITQUEUE(wait, current);
774 if (!ssk || netlink_is_kernel(ssk))
781 __set_current_state(TASK_INTERRUPTIBLE);
782 add_wait_queue(&nlk->wait, &wait);
784 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
785 test_bit(0, &nlk->state)) &&
786 !sock_flag(sk, SOCK_DEAD))
787 *timeo = schedule_timeout(*timeo);
789 __set_current_state(TASK_RUNNING);
790 remove_wait_queue(&nlk->wait, &wait);
793 if (signal_pending(current)) {
795 return sock_intr_errno(*timeo);
799 skb_set_owner_r(skb, sk);
803 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
807 skb_queue_tail(&sk->sk_receive_queue, skb);
808 sk->sk_data_ready(sk, len);
813 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
819 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
826 delta = skb->end - skb->tail;
827 if (delta * 2 < skb->truesize)
830 if (skb_shared(skb)) {
831 struct sk_buff *nskb = skb_clone(skb, allocation);
838 if (!pskb_expand_head(skb, 0, -delta, allocation))
839 skb->truesize -= delta;
844 static inline void netlink_rcv_wake(struct sock *sk)
846 struct netlink_sock *nlk = nlk_sk(sk);
848 if (skb_queue_empty(&sk->sk_receive_queue))
849 clear_bit(0, &nlk->state);
850 if (!test_bit(0, &nlk->state))
851 wake_up_interruptible(&nlk->wait);
854 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
857 struct netlink_sock *nlk = nlk_sk(sk);
860 if (nlk->netlink_rcv != NULL) {
862 skb_set_owner_r(skb, sk);
863 nlk->netlink_rcv(skb);
870 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
871 u32 pid, int nonblock)
877 skb = netlink_trim(skb, gfp_any());
879 timeo = sock_sndtimeo(ssk, nonblock);
881 sk = netlink_getsockbypid(ssk, pid);
886 if (netlink_is_kernel(sk))
887 return netlink_unicast_kernel(sk, skb);
889 if (sk_filter(sk, skb)) {
896 err = netlink_attachskb(sk, skb, &timeo, ssk);
902 return netlink_sendskb(sk, skb);
904 EXPORT_SYMBOL(netlink_unicast);
906 int netlink_has_listeners(struct sock *sk, unsigned int group)
909 unsigned long *listeners;
911 BUG_ON(!netlink_is_kernel(sk));
914 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
916 if (group - 1 < nl_table[sk->sk_protocol].groups)
917 res = test_bit(group - 1, listeners);
923 EXPORT_SYMBOL_GPL(netlink_has_listeners);
925 static inline int netlink_broadcast_deliver(struct sock *sk,
928 struct netlink_sock *nlk = nlk_sk(sk);
930 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
931 !test_bit(0, &nlk->state)) {
932 skb_set_owner_r(skb, sk);
933 skb_queue_tail(&sk->sk_receive_queue, skb);
934 sk->sk_data_ready(sk, skb->len);
935 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
940 struct netlink_broadcast_data {
941 struct sock *exclude_sk;
949 struct sk_buff *skb, *skb2;
952 static inline int do_one_broadcast(struct sock *sk,
953 struct netlink_broadcast_data *p)
955 struct netlink_sock *nlk = nlk_sk(sk);
958 if (p->exclude_sk == sk)
961 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
962 !test_bit(p->group - 1, nlk->groups))
965 if (!net_eq(sock_net(sk), p->net))
974 if (p->skb2 == NULL) {
975 if (skb_shared(p->skb)) {
976 p->skb2 = skb_clone(p->skb, p->allocation);
978 p->skb2 = skb_get(p->skb);
980 * skb ownership may have been set when
981 * delivered to a previous socket.
986 if (p->skb2 == NULL) {
988 /* Clone failed. Notify ALL listeners. */
990 } else if (sk_filter(sk, p->skb2)) {
993 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1006 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1007 u32 group, gfp_t allocation)
1009 struct net *net = sock_net(ssk);
1010 struct netlink_broadcast_data info;
1011 struct hlist_node *node;
1014 skb = netlink_trim(skb, allocation);
1016 info.exclude_sk = ssk;
1023 info.allocation = allocation;
1027 /* While we sleep in clone, do not allow to change socket list */
1029 netlink_lock_table();
1031 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1032 do_one_broadcast(sk, &info);
1036 netlink_unlock_table();
1039 kfree_skb(info.skb2);
1041 if (info.delivered) {
1042 if (info.congested && (allocation & __GFP_WAIT))
1050 EXPORT_SYMBOL(netlink_broadcast);
1052 struct netlink_set_err_data {
1053 struct sock *exclude_sk;
1059 static inline int do_one_set_err(struct sock *sk,
1060 struct netlink_set_err_data *p)
1062 struct netlink_sock *nlk = nlk_sk(sk);
1064 if (sk == p->exclude_sk)
1067 if (sock_net(sk) != sock_net(p->exclude_sk))
1070 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1071 !test_bit(p->group - 1, nlk->groups))
1074 sk->sk_err = p->code;
1075 sk->sk_error_report(sk);
1080 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1082 struct netlink_set_err_data info;
1083 struct hlist_node *node;
1086 info.exclude_sk = ssk;
1091 read_lock(&nl_table_lock);
1093 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1094 do_one_set_err(sk, &info);
1096 read_unlock(&nl_table_lock);
1099 /* must be called with netlink table grabbed */
1100 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1104 int old, new = !!is_new, subscriptions;
1106 old = test_bit(group - 1, nlk->groups);
1107 subscriptions = nlk->subscriptions - old + new;
1109 __set_bit(group - 1, nlk->groups);
1111 __clear_bit(group - 1, nlk->groups);
1112 netlink_update_subscriptions(&nlk->sk, subscriptions);
1113 netlink_update_listeners(&nlk->sk);
1116 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1117 char __user *optval, int optlen)
1119 struct sock *sk = sock->sk;
1120 struct netlink_sock *nlk = nlk_sk(sk);
1121 unsigned int val = 0;
1124 if (level != SOL_NETLINK)
1125 return -ENOPROTOOPT;
1127 if (optlen >= sizeof(int) &&
1128 get_user(val, (unsigned int __user *)optval))
1132 case NETLINK_PKTINFO:
1134 nlk->flags |= NETLINK_RECV_PKTINFO;
1136 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1139 case NETLINK_ADD_MEMBERSHIP:
1140 case NETLINK_DROP_MEMBERSHIP: {
1141 if (!netlink_capable(sock, NL_NONROOT_RECV))
1143 err = netlink_realloc_groups(sk);
1146 if (!val || val - 1 >= nlk->ngroups)
1148 netlink_table_grab();
1149 netlink_update_socket_mc(nlk, val,
1150 optname == NETLINK_ADD_MEMBERSHIP);
1151 netlink_table_ungrab();
1161 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1162 char __user *optval, int __user *optlen)
1164 struct sock *sk = sock->sk;
1165 struct netlink_sock *nlk = nlk_sk(sk);
1168 if (level != SOL_NETLINK)
1169 return -ENOPROTOOPT;
1171 if (get_user(len, optlen))
1177 case NETLINK_PKTINFO:
1178 if (len < sizeof(int))
1181 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1182 if (put_user(len, optlen) ||
1183 put_user(val, optval))
1193 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1195 struct nl_pktinfo info;
1197 info.group = NETLINK_CB(skb).dst_group;
1198 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1201 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1202 struct msghdr *msg, size_t len)
1204 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1205 struct sock *sk = sock->sk;
1206 struct netlink_sock *nlk = nlk_sk(sk);
1207 struct sockaddr_nl *addr = msg->msg_name;
1210 struct sk_buff *skb;
1212 struct scm_cookie scm;
1214 if (msg->msg_flags&MSG_OOB)
1217 if (NULL == siocb->scm)
1219 err = scm_send(sock, msg, siocb->scm);
1223 if (msg->msg_namelen) {
1224 if (addr->nl_family != AF_NETLINK)
1226 dst_pid = addr->nl_pid;
1227 dst_group = ffs(addr->nl_groups);
1228 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1231 dst_pid = nlk->dst_pid;
1232 dst_group = nlk->dst_group;
1236 err = netlink_autobind(sock);
1242 if (len > sk->sk_sndbuf - 32)
1245 skb = alloc_skb(len, GFP_KERNEL);
1249 NETLINK_CB(skb).pid = nlk->pid;
1250 NETLINK_CB(skb).dst_group = dst_group;
1251 NETLINK_CB(skb).loginuid = audit_get_loginuid(current);
1252 NETLINK_CB(skb).sessionid = audit_get_sessionid(current);
1253 security_task_getsecid(current, &(NETLINK_CB(skb).sid));
1254 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1256 /* What can I do? Netlink is asynchronous, so that
1257 we will have to save current capabilities to
1258 check them, when this message will be delivered
1259 to corresponding kernel module. --ANK (980802)
1263 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1268 err = security_netlink_send(sk, skb);
1275 atomic_inc(&skb->users);
1276 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1278 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1284 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1285 struct msghdr *msg, size_t len,
1288 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1289 struct scm_cookie scm;
1290 struct sock *sk = sock->sk;
1291 struct netlink_sock *nlk = nlk_sk(sk);
1292 int noblock = flags&MSG_DONTWAIT;
1294 struct sk_buff *skb;
1302 skb = skb_recv_datagram(sk, flags, noblock, &err);
1306 msg->msg_namelen = 0;
1310 msg->msg_flags |= MSG_TRUNC;
1314 skb_reset_transport_header(skb);
1315 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1317 if (msg->msg_name) {
1318 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1319 addr->nl_family = AF_NETLINK;
1321 addr->nl_pid = NETLINK_CB(skb).pid;
1322 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1323 msg->msg_namelen = sizeof(*addr);
1326 if (nlk->flags & NETLINK_RECV_PKTINFO)
1327 netlink_cmsg_recv_pktinfo(msg, skb);
1329 if (NULL == siocb->scm) {
1330 memset(&scm, 0, sizeof(scm));
1333 siocb->scm->creds = *NETLINK_CREDS(skb);
1334 if (flags & MSG_TRUNC)
1336 skb_free_datagram(sk, skb);
1338 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1341 scm_recv(sock, msg, siocb->scm, flags);
1343 netlink_rcv_wake(sk);
1344 return err ? : copied;
1347 static void netlink_data_ready(struct sock *sk, int len)
1353 * We export these functions to other modules. They provide a
1354 * complete set of kernel non-blocking support for message
1359 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1360 void (*input)(struct sk_buff *skb),
1361 struct mutex *cb_mutex, struct module *module)
1363 struct socket *sock;
1365 struct netlink_sock *nlk;
1366 unsigned long *listeners = NULL;
1370 if (unit < 0 || unit >= MAX_LINKS)
1373 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1377 * We have to just have a reference on the net from sk, but don't
1378 * get_net it. Besides, we cannot get and then put the net here.
1379 * So we create one inside init_net and the move it to net.
1382 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1383 goto out_sock_release_nosk;
1386 sk_change_net(sk, net);
1391 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1393 goto out_sock_release;
1395 sk->sk_data_ready = netlink_data_ready;
1397 nlk_sk(sk)->netlink_rcv = input;
1399 if (netlink_insert(sk, net, 0))
1400 goto out_sock_release;
1403 nlk->flags |= NETLINK_KERNEL_SOCKET;
1405 netlink_table_grab();
1406 if (!nl_table[unit].registered) {
1407 nl_table[unit].groups = groups;
1408 nl_table[unit].listeners = listeners;
1409 nl_table[unit].cb_mutex = cb_mutex;
1410 nl_table[unit].module = module;
1411 nl_table[unit].registered = 1;
1414 nl_table[unit].registered++;
1416 netlink_table_ungrab();
1421 netlink_kernel_release(sk);
1424 out_sock_release_nosk:
1428 EXPORT_SYMBOL(netlink_kernel_create);
1432 netlink_kernel_release(struct sock *sk)
1434 sk_release_kernel(sk);
1436 EXPORT_SYMBOL(netlink_kernel_release);
1440 * netlink_change_ngroups - change number of multicast groups
1442 * This changes the number of multicast groups that are available
1443 * on a certain netlink family. Note that it is not possible to
1444 * change the number of groups to below 32. Also note that it does
1445 * not implicitly call netlink_clear_multicast_users() when the
1446 * number of groups is reduced.
1448 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1449 * @groups: The new number of groups.
1451 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1453 unsigned long *listeners, *old = NULL;
1454 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1460 netlink_table_grab();
1461 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1462 listeners = kzalloc(NLGRPSZ(groups), GFP_ATOMIC);
1467 old = tbl->listeners;
1468 memcpy(listeners, old, NLGRPSZ(tbl->groups));
1469 rcu_assign_pointer(tbl->listeners, listeners);
1471 tbl->groups = groups;
1474 netlink_table_ungrab();
1479 EXPORT_SYMBOL(netlink_change_ngroups);
1482 * netlink_clear_multicast_users - kick off multicast listeners
1484 * This function removes all listeners from the given group.
1485 * @ksk: The kernel netlink socket, as returned by
1486 * netlink_kernel_create().
1487 * @group: The multicast group to clear.
1489 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1492 struct hlist_node *node;
1493 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1495 netlink_table_grab();
1497 sk_for_each_bound(sk, node, &tbl->mc_list)
1498 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1500 netlink_table_ungrab();
1502 EXPORT_SYMBOL(netlink_clear_multicast_users);
1504 void netlink_set_nonroot(int protocol, unsigned int flags)
1506 if ((unsigned int)protocol < MAX_LINKS)
1507 nl_table[protocol].nl_nonroot = flags;
1509 EXPORT_SYMBOL(netlink_set_nonroot);
1511 static void netlink_destroy_callback(struct netlink_callback *cb)
1519 * It looks a bit ugly.
1520 * It would be better to create kernel thread.
1523 static int netlink_dump(struct sock *sk)
1525 struct netlink_sock *nlk = nlk_sk(sk);
1526 struct netlink_callback *cb;
1527 struct sk_buff *skb;
1528 struct nlmsghdr *nlh;
1529 int len, err = -ENOBUFS;
1531 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1535 mutex_lock(nlk->cb_mutex);
1543 len = cb->dump(skb, cb);
1546 mutex_unlock(nlk->cb_mutex);
1548 if (sk_filter(sk, skb))
1551 skb_queue_tail(&sk->sk_receive_queue, skb);
1552 sk->sk_data_ready(sk, skb->len);
1557 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1561 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1563 if (sk_filter(sk, skb))
1566 skb_queue_tail(&sk->sk_receive_queue, skb);
1567 sk->sk_data_ready(sk, skb->len);
1573 mutex_unlock(nlk->cb_mutex);
1575 netlink_destroy_callback(cb);
1579 mutex_unlock(nlk->cb_mutex);
1585 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1586 struct nlmsghdr *nlh,
1587 int (*dump)(struct sk_buff *skb,
1588 struct netlink_callback *),
1589 int (*done)(struct netlink_callback *))
1591 struct netlink_callback *cb;
1593 struct netlink_sock *nlk;
1595 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1602 atomic_inc(&skb->users);
1605 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1607 netlink_destroy_callback(cb);
1608 return -ECONNREFUSED;
1611 /* A dump is in progress... */
1612 mutex_lock(nlk->cb_mutex);
1614 mutex_unlock(nlk->cb_mutex);
1615 netlink_destroy_callback(cb);
1620 mutex_unlock(nlk->cb_mutex);
1625 /* We successfully started a dump, by returning -EINTR we
1626 * signal not to send ACK even if it was requested.
1630 EXPORT_SYMBOL(netlink_dump_start);
1632 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1634 struct sk_buff *skb;
1635 struct nlmsghdr *rep;
1636 struct nlmsgerr *errmsg;
1637 size_t payload = sizeof(*errmsg);
1639 /* error messages get the original request appened */
1641 payload += nlmsg_len(nlh);
1643 skb = nlmsg_new(payload, GFP_KERNEL);
1647 sk = netlink_lookup(sock_net(in_skb->sk),
1648 in_skb->sk->sk_protocol,
1649 NETLINK_CB(in_skb).pid);
1651 sk->sk_err = ENOBUFS;
1652 sk->sk_error_report(sk);
1658 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1659 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1660 errmsg = nlmsg_data(rep);
1661 errmsg->error = err;
1662 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1663 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1665 EXPORT_SYMBOL(netlink_ack);
1667 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1670 struct nlmsghdr *nlh;
1673 while (skb->len >= nlmsg_total_size(0)) {
1676 nlh = nlmsg_hdr(skb);
1679 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1682 /* Only requests are handled by the kernel */
1683 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1686 /* Skip control messages */
1687 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1695 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1696 netlink_ack(skb, nlh, err);
1699 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1700 if (msglen > skb->len)
1702 skb_pull(skb, msglen);
1707 EXPORT_SYMBOL(netlink_rcv_skb);
1710 * nlmsg_notify - send a notification netlink message
1711 * @sk: netlink socket to use
1712 * @skb: notification message
1713 * @pid: destination netlink pid for reports or 0
1714 * @group: destination multicast group or 0
1715 * @report: 1 to report back, 0 to disable
1716 * @flags: allocation flags
1718 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1719 unsigned int group, int report, gfp_t flags)
1724 int exclude_pid = 0;
1727 atomic_inc(&skb->users);
1731 /* errors reported via destination sk->sk_err */
1732 nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1736 err = nlmsg_unicast(sk, skb, pid);
1740 EXPORT_SYMBOL(nlmsg_notify);
1742 #ifdef CONFIG_PROC_FS
1743 struct nl_seq_iter {
1744 struct seq_net_private p;
1749 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1751 struct nl_seq_iter *iter = seq->private;
1754 struct hlist_node *node;
1757 for (i = 0; i < MAX_LINKS; i++) {
1758 struct nl_pid_hash *hash = &nl_table[i].hash;
1760 for (j = 0; j <= hash->mask; j++) {
1761 sk_for_each(s, node, &hash->table[j]) {
1762 if (sock_net(s) != seq_file_net(seq))
1776 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1777 __acquires(nl_table_lock)
1779 read_lock(&nl_table_lock);
1780 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1783 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1786 struct nl_seq_iter *iter;
1791 if (v == SEQ_START_TOKEN)
1792 return netlink_seq_socket_idx(seq, 0);
1794 iter = seq->private;
1798 } while (s && sock_net(s) != seq_file_net(seq));
1803 j = iter->hash_idx + 1;
1806 struct nl_pid_hash *hash = &nl_table[i].hash;
1808 for (; j <= hash->mask; j++) {
1809 s = sk_head(&hash->table[j]);
1810 while (s && sock_net(s) != seq_file_net(seq))
1820 } while (++i < MAX_LINKS);
1825 static void netlink_seq_stop(struct seq_file *seq, void *v)
1826 __releases(nl_table_lock)
1828 read_unlock(&nl_table_lock);
1832 static int netlink_seq_show(struct seq_file *seq, void *v)
1834 if (v == SEQ_START_TOKEN)
1836 "sk Eth Pid Groups "
1837 "Rmem Wmem Dump Locks\n");
1840 struct netlink_sock *nlk = nlk_sk(s);
1842 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1846 nlk->groups ? (u32)nlk->groups[0] : 0,
1847 atomic_read(&s->sk_rmem_alloc),
1848 atomic_read(&s->sk_wmem_alloc),
1850 atomic_read(&s->sk_refcnt)
1857 static const struct seq_operations netlink_seq_ops = {
1858 .start = netlink_seq_start,
1859 .next = netlink_seq_next,
1860 .stop = netlink_seq_stop,
1861 .show = netlink_seq_show,
1865 static int netlink_seq_open(struct inode *inode, struct file *file)
1867 return seq_open_net(inode, file, &netlink_seq_ops,
1868 sizeof(struct nl_seq_iter));
1871 static const struct file_operations netlink_seq_fops = {
1872 .owner = THIS_MODULE,
1873 .open = netlink_seq_open,
1875 .llseek = seq_lseek,
1876 .release = seq_release_net,
1881 int netlink_register_notifier(struct notifier_block *nb)
1883 return atomic_notifier_chain_register(&netlink_chain, nb);
1885 EXPORT_SYMBOL(netlink_register_notifier);
1887 int netlink_unregister_notifier(struct notifier_block *nb)
1889 return atomic_notifier_chain_unregister(&netlink_chain, nb);
1891 EXPORT_SYMBOL(netlink_unregister_notifier);
1893 static const struct proto_ops netlink_ops = {
1894 .family = PF_NETLINK,
1895 .owner = THIS_MODULE,
1896 .release = netlink_release,
1897 .bind = netlink_bind,
1898 .connect = netlink_connect,
1899 .socketpair = sock_no_socketpair,
1900 .accept = sock_no_accept,
1901 .getname = netlink_getname,
1902 .poll = datagram_poll,
1903 .ioctl = sock_no_ioctl,
1904 .listen = sock_no_listen,
1905 .shutdown = sock_no_shutdown,
1906 .setsockopt = netlink_setsockopt,
1907 .getsockopt = netlink_getsockopt,
1908 .sendmsg = netlink_sendmsg,
1909 .recvmsg = netlink_recvmsg,
1910 .mmap = sock_no_mmap,
1911 .sendpage = sock_no_sendpage,
1914 static struct net_proto_family netlink_family_ops = {
1915 .family = PF_NETLINK,
1916 .create = netlink_create,
1917 .owner = THIS_MODULE, /* for consistency 8) */
1920 static int __net_init netlink_net_init(struct net *net)
1922 #ifdef CONFIG_PROC_FS
1923 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
1929 static void __net_exit netlink_net_exit(struct net *net)
1931 #ifdef CONFIG_PROC_FS
1932 proc_net_remove(net, "netlink");
1936 static struct pernet_operations __net_initdata netlink_net_ops = {
1937 .init = netlink_net_init,
1938 .exit = netlink_net_exit,
1941 static int __init netlink_proto_init(void)
1943 struct sk_buff *dummy_skb;
1945 unsigned long limit;
1947 int err = proto_register(&netlink_proto, 0);
1952 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1954 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1958 if (num_physpages >= (128 * 1024))
1959 limit = num_physpages >> (21 - PAGE_SHIFT);
1961 limit = num_physpages >> (23 - PAGE_SHIFT);
1963 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
1964 limit = (1UL << order) / sizeof(struct hlist_head);
1965 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
1967 for (i = 0; i < MAX_LINKS; i++) {
1968 struct nl_pid_hash *hash = &nl_table[i].hash;
1970 hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
1973 nl_pid_hash_free(nl_table[i].hash.table,
1974 1 * sizeof(*hash->table));
1978 hash->max_shift = order;
1981 hash->rehash_time = jiffies;
1984 sock_register(&netlink_family_ops);
1985 register_pernet_subsys(&netlink_net_ops);
1986 /* The netlink device handler may be needed early. */
1991 panic("netlink_init: Cannot allocate nl_table\n");
1994 core_initcall(netlink_proto_init);