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);
456 sock_prot_inuse_add(net, &netlink_proto, 1);
459 nlk = nlk_sk(sock->sk);
460 nlk->module = module;
469 static int netlink_release(struct socket *sock)
471 struct sock *sk = sock->sk;
472 struct netlink_sock *nlk;
482 * OK. Socket is unlinked, any packets that arrive now
487 wake_up_interruptible_all(&nlk->wait);
489 skb_queue_purge(&sk->sk_write_queue);
491 if (nlk->pid && !nlk->subscriptions) {
492 struct netlink_notify n = {
494 .protocol = sk->sk_protocol,
497 atomic_notifier_call_chain(&netlink_chain,
498 NETLINK_URELEASE, &n);
501 module_put(nlk->module);
503 netlink_table_grab();
504 if (netlink_is_kernel(sk)) {
505 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
506 if (--nl_table[sk->sk_protocol].registered == 0) {
507 kfree(nl_table[sk->sk_protocol].listeners);
508 nl_table[sk->sk_protocol].module = NULL;
509 nl_table[sk->sk_protocol].registered = 0;
511 } else if (nlk->subscriptions)
512 netlink_update_listeners(sk);
513 netlink_table_ungrab();
518 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
523 static int netlink_autobind(struct socket *sock)
525 struct sock *sk = sock->sk;
526 struct net *net = sock_net(sk);
527 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
528 struct hlist_head *head;
530 struct hlist_node *node;
531 s32 pid = current->tgid;
533 static s32 rover = -4097;
537 netlink_table_grab();
538 head = nl_pid_hashfn(hash, pid);
539 sk_for_each(osk, node, head) {
540 if (!net_eq(sock_net(osk), net))
542 if (nlk_sk(osk)->pid == pid) {
543 /* Bind collision, search negative pid values. */
547 netlink_table_ungrab();
551 netlink_table_ungrab();
553 err = netlink_insert(sk, net, pid);
554 if (err == -EADDRINUSE)
557 /* If 2 threads race to autobind, that is fine. */
564 static inline int netlink_capable(struct socket *sock, unsigned int flag)
566 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
567 capable(CAP_NET_ADMIN);
571 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
573 struct netlink_sock *nlk = nlk_sk(sk);
575 if (nlk->subscriptions && !subscriptions)
576 __sk_del_bind_node(sk);
577 else if (!nlk->subscriptions && subscriptions)
578 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
579 nlk->subscriptions = subscriptions;
582 static int netlink_realloc_groups(struct sock *sk)
584 struct netlink_sock *nlk = nlk_sk(sk);
586 unsigned long *new_groups;
589 netlink_table_grab();
591 groups = nl_table[sk->sk_protocol].groups;
592 if (!nl_table[sk->sk_protocol].registered) {
597 if (nlk->ngroups >= groups)
600 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
601 if (new_groups == NULL) {
605 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
606 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
608 nlk->groups = new_groups;
609 nlk->ngroups = groups;
611 netlink_table_ungrab();
615 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
618 struct sock *sk = sock->sk;
619 struct net *net = sock_net(sk);
620 struct netlink_sock *nlk = nlk_sk(sk);
621 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
624 if (nladdr->nl_family != AF_NETLINK)
627 /* Only superuser is allowed to listen multicasts */
628 if (nladdr->nl_groups) {
629 if (!netlink_capable(sock, NL_NONROOT_RECV))
631 err = netlink_realloc_groups(sk);
637 if (nladdr->nl_pid != nlk->pid)
640 err = nladdr->nl_pid ?
641 netlink_insert(sk, net, nladdr->nl_pid) :
642 netlink_autobind(sock);
647 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
650 netlink_table_grab();
651 netlink_update_subscriptions(sk, nlk->subscriptions +
652 hweight32(nladdr->nl_groups) -
653 hweight32(nlk->groups[0]));
654 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
655 netlink_update_listeners(sk);
656 netlink_table_ungrab();
661 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
665 struct sock *sk = sock->sk;
666 struct netlink_sock *nlk = nlk_sk(sk);
667 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
669 if (addr->sa_family == AF_UNSPEC) {
670 sk->sk_state = NETLINK_UNCONNECTED;
675 if (addr->sa_family != AF_NETLINK)
678 /* Only superuser is allowed to send multicasts */
679 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
683 err = netlink_autobind(sock);
686 sk->sk_state = NETLINK_CONNECTED;
687 nlk->dst_pid = nladdr->nl_pid;
688 nlk->dst_group = ffs(nladdr->nl_groups);
694 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
695 int *addr_len, int peer)
697 struct sock *sk = sock->sk;
698 struct netlink_sock *nlk = nlk_sk(sk);
699 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
701 nladdr->nl_family = AF_NETLINK;
703 *addr_len = sizeof(*nladdr);
706 nladdr->nl_pid = nlk->dst_pid;
707 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
709 nladdr->nl_pid = nlk->pid;
710 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
715 static void netlink_overrun(struct sock *sk)
717 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
718 sk->sk_err = ENOBUFS;
719 sk->sk_error_report(sk);
723 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
726 struct netlink_sock *nlk;
728 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
730 return ERR_PTR(-ECONNREFUSED);
732 /* Don't bother queuing skb if kernel socket has no input function */
734 if (sock->sk_state == NETLINK_CONNECTED &&
735 nlk->dst_pid != nlk_sk(ssk)->pid) {
737 return ERR_PTR(-ECONNREFUSED);
742 struct sock *netlink_getsockbyfilp(struct file *filp)
744 struct inode *inode = filp->f_path.dentry->d_inode;
747 if (!S_ISSOCK(inode->i_mode))
748 return ERR_PTR(-ENOTSOCK);
750 sock = SOCKET_I(inode)->sk;
751 if (sock->sk_family != AF_NETLINK)
752 return ERR_PTR(-EINVAL);
759 * Attach a skb to a netlink socket.
760 * The caller must hold a reference to the destination socket. On error, the
761 * reference is dropped. The skb is not send to the destination, just all
762 * all error checks are performed and memory in the queue is reserved.
764 * < 0: error. skb freed, reference to sock dropped.
766 * 1: repeat lookup - reference dropped while waiting for socket memory.
768 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
769 long *timeo, struct sock *ssk)
771 struct netlink_sock *nlk;
775 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
776 test_bit(0, &nlk->state)) {
777 DECLARE_WAITQUEUE(wait, current);
779 if (!ssk || netlink_is_kernel(ssk))
786 __set_current_state(TASK_INTERRUPTIBLE);
787 add_wait_queue(&nlk->wait, &wait);
789 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
790 test_bit(0, &nlk->state)) &&
791 !sock_flag(sk, SOCK_DEAD))
792 *timeo = schedule_timeout(*timeo);
794 __set_current_state(TASK_RUNNING);
795 remove_wait_queue(&nlk->wait, &wait);
798 if (signal_pending(current)) {
800 return sock_intr_errno(*timeo);
804 skb_set_owner_r(skb, sk);
808 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
812 skb_queue_tail(&sk->sk_receive_queue, skb);
813 sk->sk_data_ready(sk, len);
818 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
824 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
831 delta = skb->end - skb->tail;
832 if (delta * 2 < skb->truesize)
835 if (skb_shared(skb)) {
836 struct sk_buff *nskb = skb_clone(skb, allocation);
843 if (!pskb_expand_head(skb, 0, -delta, allocation))
844 skb->truesize -= delta;
849 static inline void netlink_rcv_wake(struct sock *sk)
851 struct netlink_sock *nlk = nlk_sk(sk);
853 if (skb_queue_empty(&sk->sk_receive_queue))
854 clear_bit(0, &nlk->state);
855 if (!test_bit(0, &nlk->state))
856 wake_up_interruptible(&nlk->wait);
859 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
862 struct netlink_sock *nlk = nlk_sk(sk);
865 if (nlk->netlink_rcv != NULL) {
867 skb_set_owner_r(skb, sk);
868 nlk->netlink_rcv(skb);
875 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
876 u32 pid, int nonblock)
882 skb = netlink_trim(skb, gfp_any());
884 timeo = sock_sndtimeo(ssk, nonblock);
886 sk = netlink_getsockbypid(ssk, pid);
891 if (netlink_is_kernel(sk))
892 return netlink_unicast_kernel(sk, skb);
894 if (sk_filter(sk, skb)) {
901 err = netlink_attachskb(sk, skb, &timeo, ssk);
907 return netlink_sendskb(sk, skb);
909 EXPORT_SYMBOL(netlink_unicast);
911 int netlink_has_listeners(struct sock *sk, unsigned int group)
914 unsigned long *listeners;
916 BUG_ON(!netlink_is_kernel(sk));
919 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
921 if (group - 1 < nl_table[sk->sk_protocol].groups)
922 res = test_bit(group - 1, listeners);
928 EXPORT_SYMBOL_GPL(netlink_has_listeners);
930 static inline int netlink_broadcast_deliver(struct sock *sk,
933 struct netlink_sock *nlk = nlk_sk(sk);
935 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
936 !test_bit(0, &nlk->state)) {
937 skb_set_owner_r(skb, sk);
938 skb_queue_tail(&sk->sk_receive_queue, skb);
939 sk->sk_data_ready(sk, skb->len);
940 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
945 struct netlink_broadcast_data {
946 struct sock *exclude_sk;
954 struct sk_buff *skb, *skb2;
957 static inline int do_one_broadcast(struct sock *sk,
958 struct netlink_broadcast_data *p)
960 struct netlink_sock *nlk = nlk_sk(sk);
963 if (p->exclude_sk == sk)
966 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
967 !test_bit(p->group - 1, nlk->groups))
970 if (!net_eq(sock_net(sk), p->net))
979 if (p->skb2 == NULL) {
980 if (skb_shared(p->skb)) {
981 p->skb2 = skb_clone(p->skb, p->allocation);
983 p->skb2 = skb_get(p->skb);
985 * skb ownership may have been set when
986 * delivered to a previous socket.
991 if (p->skb2 == NULL) {
993 /* Clone failed. Notify ALL listeners. */
995 } else if (sk_filter(sk, p->skb2)) {
998 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1001 p->congested |= val;
1011 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1012 u32 group, gfp_t allocation)
1014 struct net *net = sock_net(ssk);
1015 struct netlink_broadcast_data info;
1016 struct hlist_node *node;
1019 skb = netlink_trim(skb, allocation);
1021 info.exclude_sk = ssk;
1028 info.allocation = allocation;
1032 /* While we sleep in clone, do not allow to change socket list */
1034 netlink_lock_table();
1036 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1037 do_one_broadcast(sk, &info);
1041 netlink_unlock_table();
1044 kfree_skb(info.skb2);
1046 if (info.delivered) {
1047 if (info.congested && (allocation & __GFP_WAIT))
1055 EXPORT_SYMBOL(netlink_broadcast);
1057 struct netlink_set_err_data {
1058 struct sock *exclude_sk;
1064 static inline int do_one_set_err(struct sock *sk,
1065 struct netlink_set_err_data *p)
1067 struct netlink_sock *nlk = nlk_sk(sk);
1069 if (sk == p->exclude_sk)
1072 if (sock_net(sk) != sock_net(p->exclude_sk))
1075 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1076 !test_bit(p->group - 1, nlk->groups))
1079 sk->sk_err = p->code;
1080 sk->sk_error_report(sk);
1085 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1087 struct netlink_set_err_data info;
1088 struct hlist_node *node;
1091 info.exclude_sk = ssk;
1096 read_lock(&nl_table_lock);
1098 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1099 do_one_set_err(sk, &info);
1101 read_unlock(&nl_table_lock);
1104 /* must be called with netlink table grabbed */
1105 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1109 int old, new = !!is_new, subscriptions;
1111 old = test_bit(group - 1, nlk->groups);
1112 subscriptions = nlk->subscriptions - old + new;
1114 __set_bit(group - 1, nlk->groups);
1116 __clear_bit(group - 1, nlk->groups);
1117 netlink_update_subscriptions(&nlk->sk, subscriptions);
1118 netlink_update_listeners(&nlk->sk);
1121 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1122 char __user *optval, int optlen)
1124 struct sock *sk = sock->sk;
1125 struct netlink_sock *nlk = nlk_sk(sk);
1126 unsigned int val = 0;
1129 if (level != SOL_NETLINK)
1130 return -ENOPROTOOPT;
1132 if (optlen >= sizeof(int) &&
1133 get_user(val, (unsigned int __user *)optval))
1137 case NETLINK_PKTINFO:
1139 nlk->flags |= NETLINK_RECV_PKTINFO;
1141 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1144 case NETLINK_ADD_MEMBERSHIP:
1145 case NETLINK_DROP_MEMBERSHIP: {
1146 if (!netlink_capable(sock, NL_NONROOT_RECV))
1148 err = netlink_realloc_groups(sk);
1151 if (!val || val - 1 >= nlk->ngroups)
1153 netlink_table_grab();
1154 netlink_update_socket_mc(nlk, val,
1155 optname == NETLINK_ADD_MEMBERSHIP);
1156 netlink_table_ungrab();
1166 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1167 char __user *optval, int __user *optlen)
1169 struct sock *sk = sock->sk;
1170 struct netlink_sock *nlk = nlk_sk(sk);
1173 if (level != SOL_NETLINK)
1174 return -ENOPROTOOPT;
1176 if (get_user(len, optlen))
1182 case NETLINK_PKTINFO:
1183 if (len < sizeof(int))
1186 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1187 if (put_user(len, optlen) ||
1188 put_user(val, optval))
1198 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1200 struct nl_pktinfo info;
1202 info.group = NETLINK_CB(skb).dst_group;
1203 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1206 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1207 struct msghdr *msg, size_t len)
1209 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1210 struct sock *sk = sock->sk;
1211 struct netlink_sock *nlk = nlk_sk(sk);
1212 struct sockaddr_nl *addr = msg->msg_name;
1215 struct sk_buff *skb;
1217 struct scm_cookie scm;
1219 if (msg->msg_flags&MSG_OOB)
1222 if (NULL == siocb->scm)
1224 err = scm_send(sock, msg, siocb->scm);
1228 if (msg->msg_namelen) {
1229 if (addr->nl_family != AF_NETLINK)
1231 dst_pid = addr->nl_pid;
1232 dst_group = ffs(addr->nl_groups);
1233 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1236 dst_pid = nlk->dst_pid;
1237 dst_group = nlk->dst_group;
1241 err = netlink_autobind(sock);
1247 if (len > sk->sk_sndbuf - 32)
1250 skb = alloc_skb(len, GFP_KERNEL);
1254 NETLINK_CB(skb).pid = nlk->pid;
1255 NETLINK_CB(skb).dst_group = dst_group;
1256 NETLINK_CB(skb).loginuid = audit_get_loginuid(current);
1257 NETLINK_CB(skb).sessionid = audit_get_sessionid(current);
1258 security_task_getsecid(current, &(NETLINK_CB(skb).sid));
1259 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1261 /* What can I do? Netlink is asynchronous, so that
1262 we will have to save current capabilities to
1263 check them, when this message will be delivered
1264 to corresponding kernel module. --ANK (980802)
1268 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1273 err = security_netlink_send(sk, skb);
1280 atomic_inc(&skb->users);
1281 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1283 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1289 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1290 struct msghdr *msg, size_t len,
1293 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1294 struct scm_cookie scm;
1295 struct sock *sk = sock->sk;
1296 struct netlink_sock *nlk = nlk_sk(sk);
1297 int noblock = flags&MSG_DONTWAIT;
1299 struct sk_buff *skb;
1307 skb = skb_recv_datagram(sk, flags, noblock, &err);
1311 msg->msg_namelen = 0;
1315 msg->msg_flags |= MSG_TRUNC;
1319 skb_reset_transport_header(skb);
1320 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1322 if (msg->msg_name) {
1323 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1324 addr->nl_family = AF_NETLINK;
1326 addr->nl_pid = NETLINK_CB(skb).pid;
1327 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1328 msg->msg_namelen = sizeof(*addr);
1331 if (nlk->flags & NETLINK_RECV_PKTINFO)
1332 netlink_cmsg_recv_pktinfo(msg, skb);
1334 if (NULL == siocb->scm) {
1335 memset(&scm, 0, sizeof(scm));
1338 siocb->scm->creds = *NETLINK_CREDS(skb);
1339 if (flags & MSG_TRUNC)
1341 skb_free_datagram(sk, skb);
1343 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1346 scm_recv(sock, msg, siocb->scm, flags);
1348 netlink_rcv_wake(sk);
1349 return err ? : copied;
1352 static void netlink_data_ready(struct sock *sk, int len)
1358 * We export these functions to other modules. They provide a
1359 * complete set of kernel non-blocking support for message
1364 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1365 void (*input)(struct sk_buff *skb),
1366 struct mutex *cb_mutex, struct module *module)
1368 struct socket *sock;
1370 struct netlink_sock *nlk;
1371 unsigned long *listeners = NULL;
1375 if (unit < 0 || unit >= MAX_LINKS)
1378 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1382 * We have to just have a reference on the net from sk, but don't
1383 * get_net it. Besides, we cannot get and then put the net here.
1384 * So we create one inside init_net and the move it to net.
1387 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1388 goto out_sock_release_nosk;
1391 sk_change_net(sk, net);
1396 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1398 goto out_sock_release;
1400 sk->sk_data_ready = netlink_data_ready;
1402 nlk_sk(sk)->netlink_rcv = input;
1404 if (netlink_insert(sk, net, 0))
1405 goto out_sock_release;
1408 nlk->flags |= NETLINK_KERNEL_SOCKET;
1410 netlink_table_grab();
1411 if (!nl_table[unit].registered) {
1412 nl_table[unit].groups = groups;
1413 nl_table[unit].listeners = listeners;
1414 nl_table[unit].cb_mutex = cb_mutex;
1415 nl_table[unit].module = module;
1416 nl_table[unit].registered = 1;
1419 nl_table[unit].registered++;
1421 netlink_table_ungrab();
1426 netlink_kernel_release(sk);
1429 out_sock_release_nosk:
1433 EXPORT_SYMBOL(netlink_kernel_create);
1437 netlink_kernel_release(struct sock *sk)
1439 sk_release_kernel(sk);
1441 EXPORT_SYMBOL(netlink_kernel_release);
1445 * netlink_change_ngroups - change number of multicast groups
1447 * This changes the number of multicast groups that are available
1448 * on a certain netlink family. Note that it is not possible to
1449 * change the number of groups to below 32. Also note that it does
1450 * not implicitly call netlink_clear_multicast_users() when the
1451 * number of groups is reduced.
1453 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1454 * @groups: The new number of groups.
1456 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1458 unsigned long *listeners, *old = NULL;
1459 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1465 netlink_table_grab();
1466 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1467 listeners = kzalloc(NLGRPSZ(groups), GFP_ATOMIC);
1472 old = tbl->listeners;
1473 memcpy(listeners, old, NLGRPSZ(tbl->groups));
1474 rcu_assign_pointer(tbl->listeners, listeners);
1476 tbl->groups = groups;
1479 netlink_table_ungrab();
1484 EXPORT_SYMBOL(netlink_change_ngroups);
1487 * netlink_clear_multicast_users - kick off multicast listeners
1489 * This function removes all listeners from the given group.
1490 * @ksk: The kernel netlink socket, as returned by
1491 * netlink_kernel_create().
1492 * @group: The multicast group to clear.
1494 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1497 struct hlist_node *node;
1498 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1500 netlink_table_grab();
1502 sk_for_each_bound(sk, node, &tbl->mc_list)
1503 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1505 netlink_table_ungrab();
1507 EXPORT_SYMBOL(netlink_clear_multicast_users);
1509 void netlink_set_nonroot(int protocol, unsigned int flags)
1511 if ((unsigned int)protocol < MAX_LINKS)
1512 nl_table[protocol].nl_nonroot = flags;
1514 EXPORT_SYMBOL(netlink_set_nonroot);
1516 static void netlink_destroy_callback(struct netlink_callback *cb)
1524 * It looks a bit ugly.
1525 * It would be better to create kernel thread.
1528 static int netlink_dump(struct sock *sk)
1530 struct netlink_sock *nlk = nlk_sk(sk);
1531 struct netlink_callback *cb;
1532 struct sk_buff *skb;
1533 struct nlmsghdr *nlh;
1534 int len, err = -ENOBUFS;
1536 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1540 mutex_lock(nlk->cb_mutex);
1548 len = cb->dump(skb, cb);
1551 mutex_unlock(nlk->cb_mutex);
1553 if (sk_filter(sk, skb))
1556 skb_queue_tail(&sk->sk_receive_queue, skb);
1557 sk->sk_data_ready(sk, skb->len);
1562 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1566 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1568 if (sk_filter(sk, skb))
1571 skb_queue_tail(&sk->sk_receive_queue, skb);
1572 sk->sk_data_ready(sk, skb->len);
1578 mutex_unlock(nlk->cb_mutex);
1580 netlink_destroy_callback(cb);
1584 mutex_unlock(nlk->cb_mutex);
1590 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1591 struct nlmsghdr *nlh,
1592 int (*dump)(struct sk_buff *skb,
1593 struct netlink_callback *),
1594 int (*done)(struct netlink_callback *))
1596 struct netlink_callback *cb;
1598 struct netlink_sock *nlk;
1600 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1607 atomic_inc(&skb->users);
1610 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1612 netlink_destroy_callback(cb);
1613 return -ECONNREFUSED;
1616 /* A dump is in progress... */
1617 mutex_lock(nlk->cb_mutex);
1619 mutex_unlock(nlk->cb_mutex);
1620 netlink_destroy_callback(cb);
1625 mutex_unlock(nlk->cb_mutex);
1630 /* We successfully started a dump, by returning -EINTR we
1631 * signal not to send ACK even if it was requested.
1635 EXPORT_SYMBOL(netlink_dump_start);
1637 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1639 struct sk_buff *skb;
1640 struct nlmsghdr *rep;
1641 struct nlmsgerr *errmsg;
1642 size_t payload = sizeof(*errmsg);
1644 /* error messages get the original request appened */
1646 payload += nlmsg_len(nlh);
1648 skb = nlmsg_new(payload, GFP_KERNEL);
1652 sk = netlink_lookup(sock_net(in_skb->sk),
1653 in_skb->sk->sk_protocol,
1654 NETLINK_CB(in_skb).pid);
1656 sk->sk_err = ENOBUFS;
1657 sk->sk_error_report(sk);
1663 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1664 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1665 errmsg = nlmsg_data(rep);
1666 errmsg->error = err;
1667 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1668 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1670 EXPORT_SYMBOL(netlink_ack);
1672 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1675 struct nlmsghdr *nlh;
1678 while (skb->len >= nlmsg_total_size(0)) {
1681 nlh = nlmsg_hdr(skb);
1684 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1687 /* Only requests are handled by the kernel */
1688 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1691 /* Skip control messages */
1692 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1700 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1701 netlink_ack(skb, nlh, err);
1704 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1705 if (msglen > skb->len)
1707 skb_pull(skb, msglen);
1712 EXPORT_SYMBOL(netlink_rcv_skb);
1715 * nlmsg_notify - send a notification netlink message
1716 * @sk: netlink socket to use
1717 * @skb: notification message
1718 * @pid: destination netlink pid for reports or 0
1719 * @group: destination multicast group or 0
1720 * @report: 1 to report back, 0 to disable
1721 * @flags: allocation flags
1723 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1724 unsigned int group, int report, gfp_t flags)
1729 int exclude_pid = 0;
1732 atomic_inc(&skb->users);
1736 /* errors reported via destination sk->sk_err */
1737 nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1741 err = nlmsg_unicast(sk, skb, pid);
1745 EXPORT_SYMBOL(nlmsg_notify);
1747 #ifdef CONFIG_PROC_FS
1748 struct nl_seq_iter {
1749 struct seq_net_private p;
1754 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1756 struct nl_seq_iter *iter = seq->private;
1759 struct hlist_node *node;
1762 for (i = 0; i < MAX_LINKS; i++) {
1763 struct nl_pid_hash *hash = &nl_table[i].hash;
1765 for (j = 0; j <= hash->mask; j++) {
1766 sk_for_each(s, node, &hash->table[j]) {
1767 if (sock_net(s) != seq_file_net(seq))
1781 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1782 __acquires(nl_table_lock)
1784 read_lock(&nl_table_lock);
1785 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1788 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1791 struct nl_seq_iter *iter;
1796 if (v == SEQ_START_TOKEN)
1797 return netlink_seq_socket_idx(seq, 0);
1799 iter = seq->private;
1803 } while (s && sock_net(s) != seq_file_net(seq));
1808 j = iter->hash_idx + 1;
1811 struct nl_pid_hash *hash = &nl_table[i].hash;
1813 for (; j <= hash->mask; j++) {
1814 s = sk_head(&hash->table[j]);
1815 while (s && sock_net(s) != seq_file_net(seq))
1825 } while (++i < MAX_LINKS);
1830 static void netlink_seq_stop(struct seq_file *seq, void *v)
1831 __releases(nl_table_lock)
1833 read_unlock(&nl_table_lock);
1837 static int netlink_seq_show(struct seq_file *seq, void *v)
1839 if (v == SEQ_START_TOKEN)
1841 "sk Eth Pid Groups "
1842 "Rmem Wmem Dump Locks\n");
1845 struct netlink_sock *nlk = nlk_sk(s);
1847 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1851 nlk->groups ? (u32)nlk->groups[0] : 0,
1852 atomic_read(&s->sk_rmem_alloc),
1853 atomic_read(&s->sk_wmem_alloc),
1855 atomic_read(&s->sk_refcnt)
1862 static const struct seq_operations netlink_seq_ops = {
1863 .start = netlink_seq_start,
1864 .next = netlink_seq_next,
1865 .stop = netlink_seq_stop,
1866 .show = netlink_seq_show,
1870 static int netlink_seq_open(struct inode *inode, struct file *file)
1872 return seq_open_net(inode, file, &netlink_seq_ops,
1873 sizeof(struct nl_seq_iter));
1876 static const struct file_operations netlink_seq_fops = {
1877 .owner = THIS_MODULE,
1878 .open = netlink_seq_open,
1880 .llseek = seq_lseek,
1881 .release = seq_release_net,
1886 int netlink_register_notifier(struct notifier_block *nb)
1888 return atomic_notifier_chain_register(&netlink_chain, nb);
1890 EXPORT_SYMBOL(netlink_register_notifier);
1892 int netlink_unregister_notifier(struct notifier_block *nb)
1894 return atomic_notifier_chain_unregister(&netlink_chain, nb);
1896 EXPORT_SYMBOL(netlink_unregister_notifier);
1898 static const struct proto_ops netlink_ops = {
1899 .family = PF_NETLINK,
1900 .owner = THIS_MODULE,
1901 .release = netlink_release,
1902 .bind = netlink_bind,
1903 .connect = netlink_connect,
1904 .socketpair = sock_no_socketpair,
1905 .accept = sock_no_accept,
1906 .getname = netlink_getname,
1907 .poll = datagram_poll,
1908 .ioctl = sock_no_ioctl,
1909 .listen = sock_no_listen,
1910 .shutdown = sock_no_shutdown,
1911 .setsockopt = netlink_setsockopt,
1912 .getsockopt = netlink_getsockopt,
1913 .sendmsg = netlink_sendmsg,
1914 .recvmsg = netlink_recvmsg,
1915 .mmap = sock_no_mmap,
1916 .sendpage = sock_no_sendpage,
1919 static struct net_proto_family netlink_family_ops = {
1920 .family = PF_NETLINK,
1921 .create = netlink_create,
1922 .owner = THIS_MODULE, /* for consistency 8) */
1925 static int __net_init netlink_net_init(struct net *net)
1927 #ifdef CONFIG_PROC_FS
1928 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
1934 static void __net_exit netlink_net_exit(struct net *net)
1936 #ifdef CONFIG_PROC_FS
1937 proc_net_remove(net, "netlink");
1941 static struct pernet_operations __net_initdata netlink_net_ops = {
1942 .init = netlink_net_init,
1943 .exit = netlink_net_exit,
1946 static int __init netlink_proto_init(void)
1948 struct sk_buff *dummy_skb;
1950 unsigned long limit;
1952 int err = proto_register(&netlink_proto, 0);
1957 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1959 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1963 if (num_physpages >= (128 * 1024))
1964 limit = num_physpages >> (21 - PAGE_SHIFT);
1966 limit = num_physpages >> (23 - PAGE_SHIFT);
1968 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
1969 limit = (1UL << order) / sizeof(struct hlist_head);
1970 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
1972 for (i = 0; i < MAX_LINKS; i++) {
1973 struct nl_pid_hash *hash = &nl_table[i].hash;
1975 hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
1978 nl_pid_hash_free(nl_table[i].hash.table,
1979 1 * sizeof(*hash->table));
1983 hash->max_shift = order;
1986 hash->rehash_time = jiffies;
1989 sock_register(&netlink_family_ops);
1990 register_pernet_subsys(&netlink_net_ops);
1991 /* The netlink device handler may be needed early. */
1996 panic("netlink_init: Cannot allocate nl_table\n");
1999 core_initcall(netlink_proto_init);