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();
519 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
525 static int netlink_autobind(struct socket *sock)
527 struct sock *sk = sock->sk;
528 struct net *net = sock_net(sk);
529 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
530 struct hlist_head *head;
532 struct hlist_node *node;
533 s32 pid = current->tgid;
535 static s32 rover = -4097;
539 netlink_table_grab();
540 head = nl_pid_hashfn(hash, pid);
541 sk_for_each(osk, node, head) {
542 if (!net_eq(sock_net(osk), net))
544 if (nlk_sk(osk)->pid == pid) {
545 /* Bind collision, search negative pid values. */
549 netlink_table_ungrab();
553 netlink_table_ungrab();
555 err = netlink_insert(sk, net, pid);
556 if (err == -EADDRINUSE)
559 /* If 2 threads race to autobind, that is fine. */
566 static inline int netlink_capable(struct socket *sock, unsigned int flag)
568 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
569 capable(CAP_NET_ADMIN);
573 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
575 struct netlink_sock *nlk = nlk_sk(sk);
577 if (nlk->subscriptions && !subscriptions)
578 __sk_del_bind_node(sk);
579 else if (!nlk->subscriptions && subscriptions)
580 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
581 nlk->subscriptions = subscriptions;
584 static int netlink_realloc_groups(struct sock *sk)
586 struct netlink_sock *nlk = nlk_sk(sk);
588 unsigned long *new_groups;
591 netlink_table_grab();
593 groups = nl_table[sk->sk_protocol].groups;
594 if (!nl_table[sk->sk_protocol].registered) {
599 if (nlk->ngroups >= groups)
602 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
603 if (new_groups == NULL) {
607 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
608 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
610 nlk->groups = new_groups;
611 nlk->ngroups = groups;
613 netlink_table_ungrab();
617 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
620 struct sock *sk = sock->sk;
621 struct net *net = sock_net(sk);
622 struct netlink_sock *nlk = nlk_sk(sk);
623 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
626 if (nladdr->nl_family != AF_NETLINK)
629 /* Only superuser is allowed to listen multicasts */
630 if (nladdr->nl_groups) {
631 if (!netlink_capable(sock, NL_NONROOT_RECV))
633 err = netlink_realloc_groups(sk);
639 if (nladdr->nl_pid != nlk->pid)
642 err = nladdr->nl_pid ?
643 netlink_insert(sk, net, nladdr->nl_pid) :
644 netlink_autobind(sock);
649 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
652 netlink_table_grab();
653 netlink_update_subscriptions(sk, nlk->subscriptions +
654 hweight32(nladdr->nl_groups) -
655 hweight32(nlk->groups[0]));
656 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
657 netlink_update_listeners(sk);
658 netlink_table_ungrab();
663 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
667 struct sock *sk = sock->sk;
668 struct netlink_sock *nlk = nlk_sk(sk);
669 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
671 if (addr->sa_family == AF_UNSPEC) {
672 sk->sk_state = NETLINK_UNCONNECTED;
677 if (addr->sa_family != AF_NETLINK)
680 /* Only superuser is allowed to send multicasts */
681 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
685 err = netlink_autobind(sock);
688 sk->sk_state = NETLINK_CONNECTED;
689 nlk->dst_pid = nladdr->nl_pid;
690 nlk->dst_group = ffs(nladdr->nl_groups);
696 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
697 int *addr_len, int peer)
699 struct sock *sk = sock->sk;
700 struct netlink_sock *nlk = nlk_sk(sk);
701 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
703 nladdr->nl_family = AF_NETLINK;
705 *addr_len = sizeof(*nladdr);
708 nladdr->nl_pid = nlk->dst_pid;
709 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
711 nladdr->nl_pid = nlk->pid;
712 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
717 static void netlink_overrun(struct sock *sk)
719 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
720 sk->sk_err = ENOBUFS;
721 sk->sk_error_report(sk);
725 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
728 struct netlink_sock *nlk;
730 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
732 return ERR_PTR(-ECONNREFUSED);
734 /* Don't bother queuing skb if kernel socket has no input function */
736 if (sock->sk_state == NETLINK_CONNECTED &&
737 nlk->dst_pid != nlk_sk(ssk)->pid) {
739 return ERR_PTR(-ECONNREFUSED);
744 struct sock *netlink_getsockbyfilp(struct file *filp)
746 struct inode *inode = filp->f_path.dentry->d_inode;
749 if (!S_ISSOCK(inode->i_mode))
750 return ERR_PTR(-ENOTSOCK);
752 sock = SOCKET_I(inode)->sk;
753 if (sock->sk_family != AF_NETLINK)
754 return ERR_PTR(-EINVAL);
761 * Attach a skb to a netlink socket.
762 * The caller must hold a reference to the destination socket. On error, the
763 * reference is dropped. The skb is not send to the destination, just all
764 * all error checks are performed and memory in the queue is reserved.
766 * < 0: error. skb freed, reference to sock dropped.
768 * 1: repeat lookup - reference dropped while waiting for socket memory.
770 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
771 long *timeo, struct sock *ssk)
773 struct netlink_sock *nlk;
777 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
778 test_bit(0, &nlk->state)) {
779 DECLARE_WAITQUEUE(wait, current);
781 if (!ssk || netlink_is_kernel(ssk))
788 __set_current_state(TASK_INTERRUPTIBLE);
789 add_wait_queue(&nlk->wait, &wait);
791 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
792 test_bit(0, &nlk->state)) &&
793 !sock_flag(sk, SOCK_DEAD))
794 *timeo = schedule_timeout(*timeo);
796 __set_current_state(TASK_RUNNING);
797 remove_wait_queue(&nlk->wait, &wait);
800 if (signal_pending(current)) {
802 return sock_intr_errno(*timeo);
806 skb_set_owner_r(skb, sk);
810 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
814 skb_queue_tail(&sk->sk_receive_queue, skb);
815 sk->sk_data_ready(sk, len);
820 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
826 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
833 delta = skb->end - skb->tail;
834 if (delta * 2 < skb->truesize)
837 if (skb_shared(skb)) {
838 struct sk_buff *nskb = skb_clone(skb, allocation);
845 if (!pskb_expand_head(skb, 0, -delta, allocation))
846 skb->truesize -= delta;
851 static inline void netlink_rcv_wake(struct sock *sk)
853 struct netlink_sock *nlk = nlk_sk(sk);
855 if (skb_queue_empty(&sk->sk_receive_queue))
856 clear_bit(0, &nlk->state);
857 if (!test_bit(0, &nlk->state))
858 wake_up_interruptible(&nlk->wait);
861 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
864 struct netlink_sock *nlk = nlk_sk(sk);
867 if (nlk->netlink_rcv != NULL) {
869 skb_set_owner_r(skb, sk);
870 nlk->netlink_rcv(skb);
877 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
878 u32 pid, int nonblock)
884 skb = netlink_trim(skb, gfp_any());
886 timeo = sock_sndtimeo(ssk, nonblock);
888 sk = netlink_getsockbypid(ssk, pid);
893 if (netlink_is_kernel(sk))
894 return netlink_unicast_kernel(sk, skb);
896 if (sk_filter(sk, skb)) {
903 err = netlink_attachskb(sk, skb, &timeo, ssk);
909 return netlink_sendskb(sk, skb);
911 EXPORT_SYMBOL(netlink_unicast);
913 int netlink_has_listeners(struct sock *sk, unsigned int group)
916 unsigned long *listeners;
918 BUG_ON(!netlink_is_kernel(sk));
921 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
923 if (group - 1 < nl_table[sk->sk_protocol].groups)
924 res = test_bit(group - 1, listeners);
930 EXPORT_SYMBOL_GPL(netlink_has_listeners);
932 static inline int netlink_broadcast_deliver(struct sock *sk,
935 struct netlink_sock *nlk = nlk_sk(sk);
937 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
938 !test_bit(0, &nlk->state)) {
939 skb_set_owner_r(skb, sk);
940 skb_queue_tail(&sk->sk_receive_queue, skb);
941 sk->sk_data_ready(sk, skb->len);
942 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
947 struct netlink_broadcast_data {
948 struct sock *exclude_sk;
956 struct sk_buff *skb, *skb2;
959 static inline int do_one_broadcast(struct sock *sk,
960 struct netlink_broadcast_data *p)
962 struct netlink_sock *nlk = nlk_sk(sk);
965 if (p->exclude_sk == sk)
968 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
969 !test_bit(p->group - 1, nlk->groups))
972 if (!net_eq(sock_net(sk), p->net))
981 if (p->skb2 == NULL) {
982 if (skb_shared(p->skb)) {
983 p->skb2 = skb_clone(p->skb, p->allocation);
985 p->skb2 = skb_get(p->skb);
987 * skb ownership may have been set when
988 * delivered to a previous socket.
993 if (p->skb2 == NULL) {
995 /* Clone failed. Notify ALL listeners. */
997 } else if (sk_filter(sk, p->skb2)) {
1000 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1001 netlink_overrun(sk);
1003 p->congested |= val;
1013 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1014 u32 group, gfp_t allocation)
1016 struct net *net = sock_net(ssk);
1017 struct netlink_broadcast_data info;
1018 struct hlist_node *node;
1021 skb = netlink_trim(skb, allocation);
1023 info.exclude_sk = ssk;
1030 info.allocation = allocation;
1034 /* While we sleep in clone, do not allow to change socket list */
1036 netlink_lock_table();
1038 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1039 do_one_broadcast(sk, &info);
1043 netlink_unlock_table();
1046 kfree_skb(info.skb2);
1048 if (info.delivered) {
1049 if (info.congested && (allocation & __GFP_WAIT))
1057 EXPORT_SYMBOL(netlink_broadcast);
1059 struct netlink_set_err_data {
1060 struct sock *exclude_sk;
1066 static inline int do_one_set_err(struct sock *sk,
1067 struct netlink_set_err_data *p)
1069 struct netlink_sock *nlk = nlk_sk(sk);
1071 if (sk == p->exclude_sk)
1074 if (sock_net(sk) != sock_net(p->exclude_sk))
1077 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1078 !test_bit(p->group - 1, nlk->groups))
1081 sk->sk_err = p->code;
1082 sk->sk_error_report(sk);
1087 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1089 struct netlink_set_err_data info;
1090 struct hlist_node *node;
1093 info.exclude_sk = ssk;
1098 read_lock(&nl_table_lock);
1100 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1101 do_one_set_err(sk, &info);
1103 read_unlock(&nl_table_lock);
1106 /* must be called with netlink table grabbed */
1107 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1111 int old, new = !!is_new, subscriptions;
1113 old = test_bit(group - 1, nlk->groups);
1114 subscriptions = nlk->subscriptions - old + new;
1116 __set_bit(group - 1, nlk->groups);
1118 __clear_bit(group - 1, nlk->groups);
1119 netlink_update_subscriptions(&nlk->sk, subscriptions);
1120 netlink_update_listeners(&nlk->sk);
1123 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1124 char __user *optval, int optlen)
1126 struct sock *sk = sock->sk;
1127 struct netlink_sock *nlk = nlk_sk(sk);
1128 unsigned int val = 0;
1131 if (level != SOL_NETLINK)
1132 return -ENOPROTOOPT;
1134 if (optlen >= sizeof(int) &&
1135 get_user(val, (unsigned int __user *)optval))
1139 case NETLINK_PKTINFO:
1141 nlk->flags |= NETLINK_RECV_PKTINFO;
1143 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1146 case NETLINK_ADD_MEMBERSHIP:
1147 case NETLINK_DROP_MEMBERSHIP: {
1148 if (!netlink_capable(sock, NL_NONROOT_RECV))
1150 err = netlink_realloc_groups(sk);
1153 if (!val || val - 1 >= nlk->ngroups)
1155 netlink_table_grab();
1156 netlink_update_socket_mc(nlk, val,
1157 optname == NETLINK_ADD_MEMBERSHIP);
1158 netlink_table_ungrab();
1168 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1169 char __user *optval, int __user *optlen)
1171 struct sock *sk = sock->sk;
1172 struct netlink_sock *nlk = nlk_sk(sk);
1175 if (level != SOL_NETLINK)
1176 return -ENOPROTOOPT;
1178 if (get_user(len, optlen))
1184 case NETLINK_PKTINFO:
1185 if (len < sizeof(int))
1188 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1189 if (put_user(len, optlen) ||
1190 put_user(val, optval))
1200 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1202 struct nl_pktinfo info;
1204 info.group = NETLINK_CB(skb).dst_group;
1205 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1208 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1209 struct msghdr *msg, size_t len)
1211 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1212 struct sock *sk = sock->sk;
1213 struct netlink_sock *nlk = nlk_sk(sk);
1214 struct sockaddr_nl *addr = msg->msg_name;
1217 struct sk_buff *skb;
1219 struct scm_cookie scm;
1221 if (msg->msg_flags&MSG_OOB)
1224 if (NULL == siocb->scm)
1226 err = scm_send(sock, msg, siocb->scm);
1230 if (msg->msg_namelen) {
1231 if (addr->nl_family != AF_NETLINK)
1233 dst_pid = addr->nl_pid;
1234 dst_group = ffs(addr->nl_groups);
1235 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1238 dst_pid = nlk->dst_pid;
1239 dst_group = nlk->dst_group;
1243 err = netlink_autobind(sock);
1249 if (len > sk->sk_sndbuf - 32)
1252 skb = alloc_skb(len, GFP_KERNEL);
1256 NETLINK_CB(skb).pid = nlk->pid;
1257 NETLINK_CB(skb).dst_group = dst_group;
1258 NETLINK_CB(skb).loginuid = audit_get_loginuid(current);
1259 NETLINK_CB(skb).sessionid = audit_get_sessionid(current);
1260 security_task_getsecid(current, &(NETLINK_CB(skb).sid));
1261 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1263 /* What can I do? Netlink is asynchronous, so that
1264 we will have to save current capabilities to
1265 check them, when this message will be delivered
1266 to corresponding kernel module. --ANK (980802)
1270 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1275 err = security_netlink_send(sk, skb);
1282 atomic_inc(&skb->users);
1283 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1285 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1291 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1292 struct msghdr *msg, size_t len,
1295 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1296 struct scm_cookie scm;
1297 struct sock *sk = sock->sk;
1298 struct netlink_sock *nlk = nlk_sk(sk);
1299 int noblock = flags&MSG_DONTWAIT;
1301 struct sk_buff *skb;
1309 skb = skb_recv_datagram(sk, flags, noblock, &err);
1313 msg->msg_namelen = 0;
1317 msg->msg_flags |= MSG_TRUNC;
1321 skb_reset_transport_header(skb);
1322 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1324 if (msg->msg_name) {
1325 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1326 addr->nl_family = AF_NETLINK;
1328 addr->nl_pid = NETLINK_CB(skb).pid;
1329 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1330 msg->msg_namelen = sizeof(*addr);
1333 if (nlk->flags & NETLINK_RECV_PKTINFO)
1334 netlink_cmsg_recv_pktinfo(msg, skb);
1336 if (NULL == siocb->scm) {
1337 memset(&scm, 0, sizeof(scm));
1340 siocb->scm->creds = *NETLINK_CREDS(skb);
1341 if (flags & MSG_TRUNC)
1343 skb_free_datagram(sk, skb);
1345 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1348 scm_recv(sock, msg, siocb->scm, flags);
1350 netlink_rcv_wake(sk);
1351 return err ? : copied;
1354 static void netlink_data_ready(struct sock *sk, int len)
1360 * We export these functions to other modules. They provide a
1361 * complete set of kernel non-blocking support for message
1366 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1367 void (*input)(struct sk_buff *skb),
1368 struct mutex *cb_mutex, struct module *module)
1370 struct socket *sock;
1372 struct netlink_sock *nlk;
1373 unsigned long *listeners = NULL;
1377 if (unit < 0 || unit >= MAX_LINKS)
1380 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1384 * We have to just have a reference on the net from sk, but don't
1385 * get_net it. Besides, we cannot get and then put the net here.
1386 * So we create one inside init_net and the move it to net.
1389 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1390 goto out_sock_release_nosk;
1393 sk_change_net(sk, net);
1398 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1400 goto out_sock_release;
1402 sk->sk_data_ready = netlink_data_ready;
1404 nlk_sk(sk)->netlink_rcv = input;
1406 if (netlink_insert(sk, net, 0))
1407 goto out_sock_release;
1410 nlk->flags |= NETLINK_KERNEL_SOCKET;
1412 netlink_table_grab();
1413 if (!nl_table[unit].registered) {
1414 nl_table[unit].groups = groups;
1415 nl_table[unit].listeners = listeners;
1416 nl_table[unit].cb_mutex = cb_mutex;
1417 nl_table[unit].module = module;
1418 nl_table[unit].registered = 1;
1421 nl_table[unit].registered++;
1423 netlink_table_ungrab();
1428 netlink_kernel_release(sk);
1431 out_sock_release_nosk:
1435 EXPORT_SYMBOL(netlink_kernel_create);
1439 netlink_kernel_release(struct sock *sk)
1441 sk_release_kernel(sk);
1443 EXPORT_SYMBOL(netlink_kernel_release);
1447 * netlink_change_ngroups - change number of multicast groups
1449 * This changes the number of multicast groups that are available
1450 * on a certain netlink family. Note that it is not possible to
1451 * change the number of groups to below 32. Also note that it does
1452 * not implicitly call netlink_clear_multicast_users() when the
1453 * number of groups is reduced.
1455 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1456 * @groups: The new number of groups.
1458 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1460 unsigned long *listeners, *old = NULL;
1461 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1467 netlink_table_grab();
1468 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1469 listeners = kzalloc(NLGRPSZ(groups), GFP_ATOMIC);
1474 old = tbl->listeners;
1475 memcpy(listeners, old, NLGRPSZ(tbl->groups));
1476 rcu_assign_pointer(tbl->listeners, listeners);
1478 tbl->groups = groups;
1481 netlink_table_ungrab();
1486 EXPORT_SYMBOL(netlink_change_ngroups);
1489 * netlink_clear_multicast_users - kick off multicast listeners
1491 * This function removes all listeners from the given group.
1492 * @ksk: The kernel netlink socket, as returned by
1493 * netlink_kernel_create().
1494 * @group: The multicast group to clear.
1496 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1499 struct hlist_node *node;
1500 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1502 netlink_table_grab();
1504 sk_for_each_bound(sk, node, &tbl->mc_list)
1505 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1507 netlink_table_ungrab();
1509 EXPORT_SYMBOL(netlink_clear_multicast_users);
1511 void netlink_set_nonroot(int protocol, unsigned int flags)
1513 if ((unsigned int)protocol < MAX_LINKS)
1514 nl_table[protocol].nl_nonroot = flags;
1516 EXPORT_SYMBOL(netlink_set_nonroot);
1518 static void netlink_destroy_callback(struct netlink_callback *cb)
1526 * It looks a bit ugly.
1527 * It would be better to create kernel thread.
1530 static int netlink_dump(struct sock *sk)
1532 struct netlink_sock *nlk = nlk_sk(sk);
1533 struct netlink_callback *cb;
1534 struct sk_buff *skb;
1535 struct nlmsghdr *nlh;
1536 int len, err = -ENOBUFS;
1538 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1542 mutex_lock(nlk->cb_mutex);
1550 len = cb->dump(skb, cb);
1553 mutex_unlock(nlk->cb_mutex);
1555 if (sk_filter(sk, skb))
1558 skb_queue_tail(&sk->sk_receive_queue, skb);
1559 sk->sk_data_ready(sk, skb->len);
1564 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1568 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1570 if (sk_filter(sk, skb))
1573 skb_queue_tail(&sk->sk_receive_queue, skb);
1574 sk->sk_data_ready(sk, skb->len);
1580 mutex_unlock(nlk->cb_mutex);
1582 netlink_destroy_callback(cb);
1586 mutex_unlock(nlk->cb_mutex);
1592 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1593 struct nlmsghdr *nlh,
1594 int (*dump)(struct sk_buff *skb,
1595 struct netlink_callback *),
1596 int (*done)(struct netlink_callback *))
1598 struct netlink_callback *cb;
1600 struct netlink_sock *nlk;
1602 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1609 atomic_inc(&skb->users);
1612 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1614 netlink_destroy_callback(cb);
1615 return -ECONNREFUSED;
1618 /* A dump is in progress... */
1619 mutex_lock(nlk->cb_mutex);
1621 mutex_unlock(nlk->cb_mutex);
1622 netlink_destroy_callback(cb);
1627 mutex_unlock(nlk->cb_mutex);
1632 /* We successfully started a dump, by returning -EINTR we
1633 * signal not to send ACK even if it was requested.
1637 EXPORT_SYMBOL(netlink_dump_start);
1639 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1641 struct sk_buff *skb;
1642 struct nlmsghdr *rep;
1643 struct nlmsgerr *errmsg;
1644 size_t payload = sizeof(*errmsg);
1646 /* error messages get the original request appened */
1648 payload += nlmsg_len(nlh);
1650 skb = nlmsg_new(payload, GFP_KERNEL);
1654 sk = netlink_lookup(sock_net(in_skb->sk),
1655 in_skb->sk->sk_protocol,
1656 NETLINK_CB(in_skb).pid);
1658 sk->sk_err = ENOBUFS;
1659 sk->sk_error_report(sk);
1665 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1666 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1667 errmsg = nlmsg_data(rep);
1668 errmsg->error = err;
1669 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1670 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1672 EXPORT_SYMBOL(netlink_ack);
1674 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1677 struct nlmsghdr *nlh;
1680 while (skb->len >= nlmsg_total_size(0)) {
1683 nlh = nlmsg_hdr(skb);
1686 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1689 /* Only requests are handled by the kernel */
1690 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1693 /* Skip control messages */
1694 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1702 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1703 netlink_ack(skb, nlh, err);
1706 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1707 if (msglen > skb->len)
1709 skb_pull(skb, msglen);
1714 EXPORT_SYMBOL(netlink_rcv_skb);
1717 * nlmsg_notify - send a notification netlink message
1718 * @sk: netlink socket to use
1719 * @skb: notification message
1720 * @pid: destination netlink pid for reports or 0
1721 * @group: destination multicast group or 0
1722 * @report: 1 to report back, 0 to disable
1723 * @flags: allocation flags
1725 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1726 unsigned int group, int report, gfp_t flags)
1731 int exclude_pid = 0;
1734 atomic_inc(&skb->users);
1738 /* errors reported via destination sk->sk_err */
1739 nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1743 err = nlmsg_unicast(sk, skb, pid);
1747 EXPORT_SYMBOL(nlmsg_notify);
1749 #ifdef CONFIG_PROC_FS
1750 struct nl_seq_iter {
1751 struct seq_net_private p;
1756 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1758 struct nl_seq_iter *iter = seq->private;
1761 struct hlist_node *node;
1764 for (i = 0; i < MAX_LINKS; i++) {
1765 struct nl_pid_hash *hash = &nl_table[i].hash;
1767 for (j = 0; j <= hash->mask; j++) {
1768 sk_for_each(s, node, &hash->table[j]) {
1769 if (sock_net(s) != seq_file_net(seq))
1783 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1784 __acquires(nl_table_lock)
1786 read_lock(&nl_table_lock);
1787 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1790 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1793 struct nl_seq_iter *iter;
1798 if (v == SEQ_START_TOKEN)
1799 return netlink_seq_socket_idx(seq, 0);
1801 iter = seq->private;
1805 } while (s && sock_net(s) != seq_file_net(seq));
1810 j = iter->hash_idx + 1;
1813 struct nl_pid_hash *hash = &nl_table[i].hash;
1815 for (; j <= hash->mask; j++) {
1816 s = sk_head(&hash->table[j]);
1817 while (s && sock_net(s) != seq_file_net(seq))
1827 } while (++i < MAX_LINKS);
1832 static void netlink_seq_stop(struct seq_file *seq, void *v)
1833 __releases(nl_table_lock)
1835 read_unlock(&nl_table_lock);
1839 static int netlink_seq_show(struct seq_file *seq, void *v)
1841 if (v == SEQ_START_TOKEN)
1843 "sk Eth Pid Groups "
1844 "Rmem Wmem Dump Locks\n");
1847 struct netlink_sock *nlk = nlk_sk(s);
1849 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1853 nlk->groups ? (u32)nlk->groups[0] : 0,
1854 atomic_read(&s->sk_rmem_alloc),
1855 atomic_read(&s->sk_wmem_alloc),
1857 atomic_read(&s->sk_refcnt)
1864 static const struct seq_operations netlink_seq_ops = {
1865 .start = netlink_seq_start,
1866 .next = netlink_seq_next,
1867 .stop = netlink_seq_stop,
1868 .show = netlink_seq_show,
1872 static int netlink_seq_open(struct inode *inode, struct file *file)
1874 return seq_open_net(inode, file, &netlink_seq_ops,
1875 sizeof(struct nl_seq_iter));
1878 static const struct file_operations netlink_seq_fops = {
1879 .owner = THIS_MODULE,
1880 .open = netlink_seq_open,
1882 .llseek = seq_lseek,
1883 .release = seq_release_net,
1888 int netlink_register_notifier(struct notifier_block *nb)
1890 return atomic_notifier_chain_register(&netlink_chain, nb);
1892 EXPORT_SYMBOL(netlink_register_notifier);
1894 int netlink_unregister_notifier(struct notifier_block *nb)
1896 return atomic_notifier_chain_unregister(&netlink_chain, nb);
1898 EXPORT_SYMBOL(netlink_unregister_notifier);
1900 static const struct proto_ops netlink_ops = {
1901 .family = PF_NETLINK,
1902 .owner = THIS_MODULE,
1903 .release = netlink_release,
1904 .bind = netlink_bind,
1905 .connect = netlink_connect,
1906 .socketpair = sock_no_socketpair,
1907 .accept = sock_no_accept,
1908 .getname = netlink_getname,
1909 .poll = datagram_poll,
1910 .ioctl = sock_no_ioctl,
1911 .listen = sock_no_listen,
1912 .shutdown = sock_no_shutdown,
1913 .setsockopt = netlink_setsockopt,
1914 .getsockopt = netlink_getsockopt,
1915 .sendmsg = netlink_sendmsg,
1916 .recvmsg = netlink_recvmsg,
1917 .mmap = sock_no_mmap,
1918 .sendpage = sock_no_sendpage,
1921 static struct net_proto_family netlink_family_ops = {
1922 .family = PF_NETLINK,
1923 .create = netlink_create,
1924 .owner = THIS_MODULE, /* for consistency 8) */
1927 static int __net_init netlink_net_init(struct net *net)
1929 #ifdef CONFIG_PROC_FS
1930 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
1936 static void __net_exit netlink_net_exit(struct net *net)
1938 #ifdef CONFIG_PROC_FS
1939 proc_net_remove(net, "netlink");
1943 static struct pernet_operations __net_initdata netlink_net_ops = {
1944 .init = netlink_net_init,
1945 .exit = netlink_net_exit,
1948 static int __init netlink_proto_init(void)
1950 struct sk_buff *dummy_skb;
1952 unsigned long limit;
1954 int err = proto_register(&netlink_proto, 0);
1959 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1961 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1965 if (num_physpages >= (128 * 1024))
1966 limit = num_physpages >> (21 - PAGE_SHIFT);
1968 limit = num_physpages >> (23 - PAGE_SHIFT);
1970 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
1971 limit = (1UL << order) / sizeof(struct hlist_head);
1972 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
1974 for (i = 0; i < MAX_LINKS; i++) {
1975 struct nl_pid_hash *hash = &nl_table[i].hash;
1977 hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
1980 nl_pid_hash_free(nl_table[i].hash.table,
1981 1 * sizeof(*hash->table));
1985 hash->max_shift = order;
1988 hash->rehash_time = jiffies;
1991 sock_register(&netlink_family_ops);
1992 register_pernet_subsys(&netlink_net_ops);
1993 /* The netlink device handler may be needed early. */
1998 panic("netlink_init: Cannot allocate nl_table\n");
2001 core_initcall(netlink_proto_init);