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 void (*netlink_bind)(int group);
84 struct module *module;
89 unsigned long masks[0];
92 #define NETLINK_KERNEL_SOCKET 0x1
93 #define NETLINK_RECV_PKTINFO 0x2
94 #define NETLINK_BROADCAST_SEND_ERROR 0x4
95 #define NETLINK_RECV_NO_ENOBUFS 0x8
97 static inline struct netlink_sock *nlk_sk(struct sock *sk)
99 return container_of(sk, struct netlink_sock, sk);
102 static inline int netlink_is_kernel(struct sock *sk)
104 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
108 struct hlist_head *table;
109 unsigned long rehash_time;
114 unsigned int entries;
115 unsigned int max_shift;
120 struct netlink_table {
121 struct nl_pid_hash hash;
122 struct hlist_head mc_list;
123 struct listeners __rcu *listeners;
124 unsigned int nl_nonroot;
126 struct mutex *cb_mutex;
127 struct module *module;
128 void (*bind)(int group);
132 static struct netlink_table *nl_table;
134 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
136 static int netlink_dump(struct sock *sk);
138 static DEFINE_RWLOCK(nl_table_lock);
139 static atomic_t nl_table_users = ATOMIC_INIT(0);
141 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
143 static inline u32 netlink_group_mask(u32 group)
145 return group ? 1 << (group - 1) : 0;
148 static inline struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
150 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
153 static void netlink_destroy_callback(struct netlink_callback *cb)
159 static void netlink_consume_callback(struct netlink_callback *cb)
161 consume_skb(cb->skb);
165 static void netlink_sock_destruct(struct sock *sk)
167 struct netlink_sock *nlk = nlk_sk(sk);
171 nlk->cb->done(nlk->cb);
172 netlink_destroy_callback(nlk->cb);
175 skb_queue_purge(&sk->sk_receive_queue);
177 if (!sock_flag(sk, SOCK_DEAD)) {
178 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
182 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
183 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
184 WARN_ON(nlk_sk(sk)->groups);
187 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
188 * SMP. Look, when several writers sleep and reader wakes them up, all but one
189 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
190 * this, _but_ remember, it adds useless work on UP machines.
193 void netlink_table_grab(void)
194 __acquires(nl_table_lock)
198 write_lock_irq(&nl_table_lock);
200 if (atomic_read(&nl_table_users)) {
201 DECLARE_WAITQUEUE(wait, current);
203 add_wait_queue_exclusive(&nl_table_wait, &wait);
205 set_current_state(TASK_UNINTERRUPTIBLE);
206 if (atomic_read(&nl_table_users) == 0)
208 write_unlock_irq(&nl_table_lock);
210 write_lock_irq(&nl_table_lock);
213 __set_current_state(TASK_RUNNING);
214 remove_wait_queue(&nl_table_wait, &wait);
218 void netlink_table_ungrab(void)
219 __releases(nl_table_lock)
221 write_unlock_irq(&nl_table_lock);
222 wake_up(&nl_table_wait);
226 netlink_lock_table(void)
228 /* read_lock() synchronizes us to netlink_table_grab */
230 read_lock(&nl_table_lock);
231 atomic_inc(&nl_table_users);
232 read_unlock(&nl_table_lock);
236 netlink_unlock_table(void)
238 if (atomic_dec_and_test(&nl_table_users))
239 wake_up(&nl_table_wait);
242 static struct sock *netlink_lookup(struct net *net, int protocol, u32 pid)
244 struct nl_pid_hash *hash = &nl_table[protocol].hash;
245 struct hlist_head *head;
247 struct hlist_node *node;
249 read_lock(&nl_table_lock);
250 head = nl_pid_hashfn(hash, pid);
251 sk_for_each(sk, node, head) {
252 if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) {
259 read_unlock(&nl_table_lock);
263 static struct hlist_head *nl_pid_hash_zalloc(size_t size)
265 if (size <= PAGE_SIZE)
266 return kzalloc(size, GFP_ATOMIC);
268 return (struct hlist_head *)
269 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
273 static void nl_pid_hash_free(struct hlist_head *table, size_t size)
275 if (size <= PAGE_SIZE)
278 free_pages((unsigned long)table, get_order(size));
281 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
283 unsigned int omask, mask, shift;
285 struct hlist_head *otable, *table;
288 omask = mask = hash->mask;
289 osize = size = (mask + 1) * sizeof(*table);
293 if (++shift > hash->max_shift)
299 table = nl_pid_hash_zalloc(size);
303 otable = hash->table;
307 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
309 for (i = 0; i <= omask; i++) {
311 struct hlist_node *node, *tmp;
313 sk_for_each_safe(sk, node, tmp, &otable[i])
314 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
317 nl_pid_hash_free(otable, osize);
318 hash->rehash_time = jiffies + 10 * 60 * HZ;
322 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
324 int avg = hash->entries >> hash->shift;
326 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
329 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
330 nl_pid_hash_rehash(hash, 0);
337 static const struct proto_ops netlink_ops;
340 netlink_update_listeners(struct sock *sk)
342 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
343 struct hlist_node *node;
347 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
349 sk_for_each_bound(sk, node, &tbl->mc_list) {
350 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
351 mask |= nlk_sk(sk)->groups[i];
353 tbl->listeners->masks[i] = mask;
355 /* this function is only called with the netlink table "grabbed", which
356 * makes sure updates are visible before bind or setsockopt return. */
359 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
361 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
362 struct hlist_head *head;
363 int err = -EADDRINUSE;
365 struct hlist_node *node;
368 netlink_table_grab();
369 head = nl_pid_hashfn(hash, pid);
371 sk_for_each(osk, node, head) {
372 if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid))
384 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
387 if (len && nl_pid_hash_dilute(hash, len))
388 head = nl_pid_hashfn(hash, pid);
390 nlk_sk(sk)->pid = pid;
391 sk_add_node(sk, head);
395 netlink_table_ungrab();
399 static void netlink_remove(struct sock *sk)
401 netlink_table_grab();
402 if (sk_del_node_init(sk))
403 nl_table[sk->sk_protocol].hash.entries--;
404 if (nlk_sk(sk)->subscriptions)
405 __sk_del_bind_node(sk);
406 netlink_table_ungrab();
409 static struct proto netlink_proto = {
411 .owner = THIS_MODULE,
412 .obj_size = sizeof(struct netlink_sock),
415 static int __netlink_create(struct net *net, struct socket *sock,
416 struct mutex *cb_mutex, int protocol)
419 struct netlink_sock *nlk;
421 sock->ops = &netlink_ops;
423 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
427 sock_init_data(sock, sk);
431 nlk->cb_mutex = cb_mutex;
433 nlk->cb_mutex = &nlk->cb_def_mutex;
434 mutex_init(nlk->cb_mutex);
436 init_waitqueue_head(&nlk->wait);
438 sk->sk_destruct = netlink_sock_destruct;
439 sk->sk_protocol = protocol;
443 static int netlink_create(struct net *net, struct socket *sock, int protocol,
446 struct module *module = NULL;
447 struct mutex *cb_mutex;
448 struct netlink_sock *nlk;
449 void (*bind)(int group);
452 sock->state = SS_UNCONNECTED;
454 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
455 return -ESOCKTNOSUPPORT;
457 if (protocol < 0 || protocol >= MAX_LINKS)
458 return -EPROTONOSUPPORT;
460 netlink_lock_table();
461 #ifdef CONFIG_MODULES
462 if (!nl_table[protocol].registered) {
463 netlink_unlock_table();
464 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
465 netlink_lock_table();
468 if (nl_table[protocol].registered &&
469 try_module_get(nl_table[protocol].module))
470 module = nl_table[protocol].module;
472 err = -EPROTONOSUPPORT;
473 cb_mutex = nl_table[protocol].cb_mutex;
474 bind = nl_table[protocol].bind;
475 netlink_unlock_table();
480 err = __netlink_create(net, sock, cb_mutex, protocol);
485 sock_prot_inuse_add(net, &netlink_proto, 1);
488 nlk = nlk_sk(sock->sk);
489 nlk->module = module;
490 nlk->netlink_bind = bind;
499 static int netlink_release(struct socket *sock)
501 struct sock *sk = sock->sk;
502 struct netlink_sock *nlk;
512 * OK. Socket is unlinked, any packets that arrive now
517 wake_up_interruptible_all(&nlk->wait);
519 skb_queue_purge(&sk->sk_write_queue);
522 struct netlink_notify n = {
524 .protocol = sk->sk_protocol,
527 atomic_notifier_call_chain(&netlink_chain,
528 NETLINK_URELEASE, &n);
531 module_put(nlk->module);
533 netlink_table_grab();
534 if (netlink_is_kernel(sk)) {
535 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
536 if (--nl_table[sk->sk_protocol].registered == 0) {
537 kfree(nl_table[sk->sk_protocol].listeners);
538 nl_table[sk->sk_protocol].module = NULL;
539 nl_table[sk->sk_protocol].registered = 0;
541 } else if (nlk->subscriptions) {
542 netlink_update_listeners(sk);
544 netlink_table_ungrab();
550 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
556 static int netlink_autobind(struct socket *sock)
558 struct sock *sk = sock->sk;
559 struct net *net = sock_net(sk);
560 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
561 struct hlist_head *head;
563 struct hlist_node *node;
564 s32 pid = task_tgid_vnr(current);
566 static s32 rover = -4097;
570 netlink_table_grab();
571 head = nl_pid_hashfn(hash, pid);
572 sk_for_each(osk, node, head) {
573 if (!net_eq(sock_net(osk), net))
575 if (nlk_sk(osk)->pid == pid) {
576 /* Bind collision, search negative pid values. */
580 netlink_table_ungrab();
584 netlink_table_ungrab();
586 err = netlink_insert(sk, net, pid);
587 if (err == -EADDRINUSE)
590 /* If 2 threads race to autobind, that is fine. */
597 static inline int netlink_capable(const struct socket *sock, unsigned int flag)
599 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
600 capable(CAP_NET_ADMIN);
604 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
606 struct netlink_sock *nlk = nlk_sk(sk);
608 if (nlk->subscriptions && !subscriptions)
609 __sk_del_bind_node(sk);
610 else if (!nlk->subscriptions && subscriptions)
611 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
612 nlk->subscriptions = subscriptions;
615 static int netlink_realloc_groups(struct sock *sk)
617 struct netlink_sock *nlk = nlk_sk(sk);
619 unsigned long *new_groups;
622 netlink_table_grab();
624 groups = nl_table[sk->sk_protocol].groups;
625 if (!nl_table[sk->sk_protocol].registered) {
630 if (nlk->ngroups >= groups)
633 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
634 if (new_groups == NULL) {
638 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
639 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
641 nlk->groups = new_groups;
642 nlk->ngroups = groups;
644 netlink_table_ungrab();
648 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
651 struct sock *sk = sock->sk;
652 struct net *net = sock_net(sk);
653 struct netlink_sock *nlk = nlk_sk(sk);
654 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
657 if (nladdr->nl_family != AF_NETLINK)
660 /* Only superuser is allowed to listen multicasts */
661 if (nladdr->nl_groups) {
662 if (!netlink_capable(sock, NL_NONROOT_RECV))
664 err = netlink_realloc_groups(sk);
670 if (nladdr->nl_pid != nlk->pid)
673 err = nladdr->nl_pid ?
674 netlink_insert(sk, net, nladdr->nl_pid) :
675 netlink_autobind(sock);
680 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
683 netlink_table_grab();
684 netlink_update_subscriptions(sk, nlk->subscriptions +
685 hweight32(nladdr->nl_groups) -
686 hweight32(nlk->groups[0]));
687 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
688 netlink_update_listeners(sk);
689 netlink_table_ungrab();
691 if (nlk->netlink_bind && nlk->groups[0]) {
694 for (i=0; i<nlk->ngroups; i++) {
695 if (test_bit(i, nlk->groups))
696 nlk->netlink_bind(i);
703 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
707 struct sock *sk = sock->sk;
708 struct netlink_sock *nlk = nlk_sk(sk);
709 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
711 if (alen < sizeof(addr->sa_family))
714 if (addr->sa_family == AF_UNSPEC) {
715 sk->sk_state = NETLINK_UNCONNECTED;
720 if (addr->sa_family != AF_NETLINK)
723 /* Only superuser is allowed to send multicasts */
724 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
728 err = netlink_autobind(sock);
731 sk->sk_state = NETLINK_CONNECTED;
732 nlk->dst_pid = nladdr->nl_pid;
733 nlk->dst_group = ffs(nladdr->nl_groups);
739 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
740 int *addr_len, int peer)
742 struct sock *sk = sock->sk;
743 struct netlink_sock *nlk = nlk_sk(sk);
744 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
746 nladdr->nl_family = AF_NETLINK;
748 *addr_len = sizeof(*nladdr);
751 nladdr->nl_pid = nlk->dst_pid;
752 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
754 nladdr->nl_pid = nlk->pid;
755 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
760 static void netlink_overrun(struct sock *sk)
762 struct netlink_sock *nlk = nlk_sk(sk);
764 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
765 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
766 sk->sk_err = ENOBUFS;
767 sk->sk_error_report(sk);
770 atomic_inc(&sk->sk_drops);
773 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
776 struct netlink_sock *nlk;
778 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
780 return ERR_PTR(-ECONNREFUSED);
782 /* Don't bother queuing skb if kernel socket has no input function */
784 if (sock->sk_state == NETLINK_CONNECTED &&
785 nlk->dst_pid != nlk_sk(ssk)->pid) {
787 return ERR_PTR(-ECONNREFUSED);
792 struct sock *netlink_getsockbyfilp(struct file *filp)
794 struct inode *inode = filp->f_path.dentry->d_inode;
797 if (!S_ISSOCK(inode->i_mode))
798 return ERR_PTR(-ENOTSOCK);
800 sock = SOCKET_I(inode)->sk;
801 if (sock->sk_family != AF_NETLINK)
802 return ERR_PTR(-EINVAL);
809 * Attach a skb to a netlink socket.
810 * The caller must hold a reference to the destination socket. On error, the
811 * reference is dropped. The skb is not send to the destination, just all
812 * all error checks are performed and memory in the queue is reserved.
814 * < 0: error. skb freed, reference to sock dropped.
816 * 1: repeat lookup - reference dropped while waiting for socket memory.
818 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
819 long *timeo, struct sock *ssk)
821 struct netlink_sock *nlk;
825 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
826 test_bit(0, &nlk->state)) {
827 DECLARE_WAITQUEUE(wait, current);
829 if (!ssk || netlink_is_kernel(ssk))
836 __set_current_state(TASK_INTERRUPTIBLE);
837 add_wait_queue(&nlk->wait, &wait);
839 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
840 test_bit(0, &nlk->state)) &&
841 !sock_flag(sk, SOCK_DEAD))
842 *timeo = schedule_timeout(*timeo);
844 __set_current_state(TASK_RUNNING);
845 remove_wait_queue(&nlk->wait, &wait);
848 if (signal_pending(current)) {
850 return sock_intr_errno(*timeo);
854 skb_set_owner_r(skb, sk);
858 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
862 skb_queue_tail(&sk->sk_receive_queue, skb);
863 sk->sk_data_ready(sk, len);
867 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
869 int len = __netlink_sendskb(sk, skb);
875 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
881 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
887 delta = skb->end - skb->tail;
888 if (delta * 2 < skb->truesize)
891 if (skb_shared(skb)) {
892 struct sk_buff *nskb = skb_clone(skb, allocation);
899 if (!pskb_expand_head(skb, 0, -delta, allocation))
900 skb->truesize -= delta;
905 static void netlink_rcv_wake(struct sock *sk)
907 struct netlink_sock *nlk = nlk_sk(sk);
909 if (skb_queue_empty(&sk->sk_receive_queue))
910 clear_bit(0, &nlk->state);
911 if (!test_bit(0, &nlk->state))
912 wake_up_interruptible(&nlk->wait);
915 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
919 struct netlink_sock *nlk = nlk_sk(sk);
922 if (nlk->netlink_rcv != NULL) {
924 skb_set_owner_r(skb, sk);
925 NETLINK_CB(skb).ssk = ssk;
926 nlk->netlink_rcv(skb);
935 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
936 u32 pid, int nonblock)
942 skb = netlink_trim(skb, gfp_any());
944 timeo = sock_sndtimeo(ssk, nonblock);
946 sk = netlink_getsockbypid(ssk, pid);
951 if (netlink_is_kernel(sk))
952 return netlink_unicast_kernel(sk, skb, ssk);
954 if (sk_filter(sk, skb)) {
961 err = netlink_attachskb(sk, skb, &timeo, ssk);
967 return netlink_sendskb(sk, skb);
969 EXPORT_SYMBOL(netlink_unicast);
971 int netlink_has_listeners(struct sock *sk, unsigned int group)
974 struct listeners *listeners;
976 BUG_ON(!netlink_is_kernel(sk));
979 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
981 if (group - 1 < nl_table[sk->sk_protocol].groups)
982 res = test_bit(group - 1, listeners->masks);
988 EXPORT_SYMBOL_GPL(netlink_has_listeners);
990 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
992 struct netlink_sock *nlk = nlk_sk(sk);
994 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
995 !test_bit(0, &nlk->state)) {
996 skb_set_owner_r(skb, sk);
997 __netlink_sendskb(sk, skb);
998 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1003 struct netlink_broadcast_data {
1004 struct sock *exclude_sk;
1009 int delivery_failure;
1013 struct sk_buff *skb, *skb2;
1014 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1018 static int do_one_broadcast(struct sock *sk,
1019 struct netlink_broadcast_data *p)
1021 struct netlink_sock *nlk = nlk_sk(sk);
1024 if (p->exclude_sk == sk)
1027 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1028 !test_bit(p->group - 1, nlk->groups))
1031 if (!net_eq(sock_net(sk), p->net))
1035 netlink_overrun(sk);
1040 if (p->skb2 == NULL) {
1041 if (skb_shared(p->skb)) {
1042 p->skb2 = skb_clone(p->skb, p->allocation);
1044 p->skb2 = skb_get(p->skb);
1046 * skb ownership may have been set when
1047 * delivered to a previous socket.
1049 skb_orphan(p->skb2);
1052 if (p->skb2 == NULL) {
1053 netlink_overrun(sk);
1054 /* Clone failed. Notify ALL listeners. */
1056 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1057 p->delivery_failure = 1;
1058 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1061 } else if (sk_filter(sk, p->skb2)) {
1064 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1065 netlink_overrun(sk);
1066 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1067 p->delivery_failure = 1;
1069 p->congested |= val;
1079 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 pid,
1080 u32 group, gfp_t allocation,
1081 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1084 struct net *net = sock_net(ssk);
1085 struct netlink_broadcast_data info;
1086 struct hlist_node *node;
1089 skb = netlink_trim(skb, allocation);
1091 info.exclude_sk = ssk;
1096 info.delivery_failure = 0;
1099 info.allocation = allocation;
1102 info.tx_filter = filter;
1103 info.tx_data = filter_data;
1105 /* While we sleep in clone, do not allow to change socket list */
1107 netlink_lock_table();
1109 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1110 do_one_broadcast(sk, &info);
1114 netlink_unlock_table();
1116 if (info.delivery_failure) {
1117 kfree_skb(info.skb2);
1120 consume_skb(info.skb2);
1122 if (info.delivered) {
1123 if (info.congested && (allocation & __GFP_WAIT))
1129 EXPORT_SYMBOL(netlink_broadcast_filtered);
1131 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1132 u32 group, gfp_t allocation)
1134 return netlink_broadcast_filtered(ssk, skb, pid, group, allocation,
1137 EXPORT_SYMBOL(netlink_broadcast);
1139 struct netlink_set_err_data {
1140 struct sock *exclude_sk;
1146 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1148 struct netlink_sock *nlk = nlk_sk(sk);
1151 if (sk == p->exclude_sk)
1154 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1157 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1158 !test_bit(p->group - 1, nlk->groups))
1161 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
1166 sk->sk_err = p->code;
1167 sk->sk_error_report(sk);
1173 * netlink_set_err - report error to broadcast listeners
1174 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1175 * @pid: the PID of a process that we want to skip (if any)
1176 * @groups: the broadcast group that will notice the error
1177 * @code: error code, must be negative (as usual in kernelspace)
1179 * This function returns the number of broadcast listeners that have set the
1180 * NETLINK_RECV_NO_ENOBUFS socket option.
1182 int netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1184 struct netlink_set_err_data info;
1185 struct hlist_node *node;
1189 info.exclude_sk = ssk;
1192 /* sk->sk_err wants a positive error value */
1195 read_lock(&nl_table_lock);
1197 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1198 ret += do_one_set_err(sk, &info);
1200 read_unlock(&nl_table_lock);
1203 EXPORT_SYMBOL(netlink_set_err);
1205 /* must be called with netlink table grabbed */
1206 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1210 int old, new = !!is_new, subscriptions;
1212 old = test_bit(group - 1, nlk->groups);
1213 subscriptions = nlk->subscriptions - old + new;
1215 __set_bit(group - 1, nlk->groups);
1217 __clear_bit(group - 1, nlk->groups);
1218 netlink_update_subscriptions(&nlk->sk, subscriptions);
1219 netlink_update_listeners(&nlk->sk);
1222 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1223 char __user *optval, unsigned int optlen)
1225 struct sock *sk = sock->sk;
1226 struct netlink_sock *nlk = nlk_sk(sk);
1227 unsigned int val = 0;
1230 if (level != SOL_NETLINK)
1231 return -ENOPROTOOPT;
1233 if (optlen >= sizeof(int) &&
1234 get_user(val, (unsigned int __user *)optval))
1238 case NETLINK_PKTINFO:
1240 nlk->flags |= NETLINK_RECV_PKTINFO;
1242 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1245 case NETLINK_ADD_MEMBERSHIP:
1246 case NETLINK_DROP_MEMBERSHIP: {
1247 if (!netlink_capable(sock, NL_NONROOT_RECV))
1249 err = netlink_realloc_groups(sk);
1252 if (!val || val - 1 >= nlk->ngroups)
1254 netlink_table_grab();
1255 netlink_update_socket_mc(nlk, val,
1256 optname == NETLINK_ADD_MEMBERSHIP);
1257 netlink_table_ungrab();
1259 if (nlk->netlink_bind)
1260 nlk->netlink_bind(val);
1265 case NETLINK_BROADCAST_ERROR:
1267 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1269 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1272 case NETLINK_NO_ENOBUFS:
1274 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1275 clear_bit(0, &nlk->state);
1276 wake_up_interruptible(&nlk->wait);
1278 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
1288 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1289 char __user *optval, int __user *optlen)
1291 struct sock *sk = sock->sk;
1292 struct netlink_sock *nlk = nlk_sk(sk);
1295 if (level != SOL_NETLINK)
1296 return -ENOPROTOOPT;
1298 if (get_user(len, optlen))
1304 case NETLINK_PKTINFO:
1305 if (len < sizeof(int))
1308 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1309 if (put_user(len, optlen) ||
1310 put_user(val, optval))
1314 case NETLINK_BROADCAST_ERROR:
1315 if (len < sizeof(int))
1318 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
1319 if (put_user(len, optlen) ||
1320 put_user(val, optval))
1324 case NETLINK_NO_ENOBUFS:
1325 if (len < sizeof(int))
1328 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
1329 if (put_user(len, optlen) ||
1330 put_user(val, optval))
1340 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1342 struct nl_pktinfo info;
1344 info.group = NETLINK_CB(skb).dst_group;
1345 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1348 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1349 struct msghdr *msg, size_t len)
1351 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1352 struct sock *sk = sock->sk;
1353 struct netlink_sock *nlk = nlk_sk(sk);
1354 struct sockaddr_nl *addr = msg->msg_name;
1357 struct sk_buff *skb;
1359 struct scm_cookie scm;
1361 if (msg->msg_flags&MSG_OOB)
1364 if (NULL == siocb->scm)
1367 err = scm_send(sock, msg, siocb->scm, true);
1371 if (msg->msg_namelen) {
1373 if (addr->nl_family != AF_NETLINK)
1375 dst_pid = addr->nl_pid;
1376 dst_group = ffs(addr->nl_groups);
1378 if ((dst_group || dst_pid) &&
1379 !netlink_capable(sock, NL_NONROOT_SEND))
1382 dst_pid = nlk->dst_pid;
1383 dst_group = nlk->dst_group;
1387 err = netlink_autobind(sock);
1393 if (len > sk->sk_sndbuf - 32)
1396 skb = alloc_skb(len, GFP_KERNEL);
1400 NETLINK_CB(skb).pid = nlk->pid;
1401 NETLINK_CB(skb).dst_group = dst_group;
1402 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1405 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1410 err = security_netlink_send(sk, skb);
1417 atomic_inc(&skb->users);
1418 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1420 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1423 scm_destroy(siocb->scm);
1427 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1428 struct msghdr *msg, size_t len,
1431 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1432 struct scm_cookie scm;
1433 struct sock *sk = sock->sk;
1434 struct netlink_sock *nlk = nlk_sk(sk);
1435 int noblock = flags&MSG_DONTWAIT;
1437 struct sk_buff *skb, *data_skb;
1445 skb = skb_recv_datagram(sk, flags, noblock, &err);
1451 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1452 if (unlikely(skb_shinfo(skb)->frag_list)) {
1454 * If this skb has a frag_list, then here that means that we
1455 * will have to use the frag_list skb's data for compat tasks
1456 * and the regular skb's data for normal (non-compat) tasks.
1458 * If we need to send the compat skb, assign it to the
1459 * 'data_skb' variable so that it will be used below for data
1460 * copying. We keep 'skb' for everything else, including
1461 * freeing both later.
1463 if (flags & MSG_CMSG_COMPAT)
1464 data_skb = skb_shinfo(skb)->frag_list;
1468 msg->msg_namelen = 0;
1470 copied = data_skb->len;
1472 msg->msg_flags |= MSG_TRUNC;
1476 skb_reset_transport_header(data_skb);
1477 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
1479 if (msg->msg_name) {
1480 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1481 addr->nl_family = AF_NETLINK;
1483 addr->nl_pid = NETLINK_CB(skb).pid;
1484 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1485 msg->msg_namelen = sizeof(*addr);
1488 if (nlk->flags & NETLINK_RECV_PKTINFO)
1489 netlink_cmsg_recv_pktinfo(msg, skb);
1491 if (NULL == siocb->scm) {
1492 memset(&scm, 0, sizeof(scm));
1495 siocb->scm->creds = *NETLINK_CREDS(skb);
1496 if (flags & MSG_TRUNC)
1497 copied = data_skb->len;
1499 skb_free_datagram(sk, skb);
1501 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1502 ret = netlink_dump(sk);
1505 sk->sk_error_report(sk);
1509 scm_recv(sock, msg, siocb->scm, flags);
1511 netlink_rcv_wake(sk);
1512 return err ? : copied;
1515 static void netlink_data_ready(struct sock *sk, int len)
1521 * We export these functions to other modules. They provide a
1522 * complete set of kernel non-blocking support for message
1527 netlink_kernel_create(struct net *net, int unit,
1528 struct module *module,
1529 struct netlink_kernel_cfg *cfg)
1531 struct socket *sock;
1533 struct netlink_sock *nlk;
1534 struct listeners *listeners = NULL;
1535 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1536 unsigned int groups;
1540 if (unit < 0 || unit >= MAX_LINKS)
1543 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1547 * We have to just have a reference on the net from sk, but don't
1548 * get_net it. Besides, we cannot get and then put the net here.
1549 * So we create one inside init_net and the move it to net.
1552 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1553 goto out_sock_release_nosk;
1556 sk_change_net(sk, net);
1558 if (!cfg || cfg->groups < 32)
1561 groups = cfg->groups;
1563 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1565 goto out_sock_release;
1567 sk->sk_data_ready = netlink_data_ready;
1568 if (cfg && cfg->input)
1569 nlk_sk(sk)->netlink_rcv = cfg->input;
1571 if (netlink_insert(sk, net, 0))
1572 goto out_sock_release;
1575 nlk->flags |= NETLINK_KERNEL_SOCKET;
1577 netlink_table_grab();
1578 if (!nl_table[unit].registered) {
1579 nl_table[unit].groups = groups;
1580 rcu_assign_pointer(nl_table[unit].listeners, listeners);
1581 nl_table[unit].cb_mutex = cb_mutex;
1582 nl_table[unit].module = module;
1583 nl_table[unit].bind = cfg ? cfg->bind : NULL;
1584 nl_table[unit].registered = 1;
1587 nl_table[unit].registered++;
1589 netlink_table_ungrab();
1594 netlink_kernel_release(sk);
1597 out_sock_release_nosk:
1601 EXPORT_SYMBOL(netlink_kernel_create);
1605 netlink_kernel_release(struct sock *sk)
1607 sk_release_kernel(sk);
1609 EXPORT_SYMBOL(netlink_kernel_release);
1611 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1613 struct listeners *new, *old;
1614 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1619 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1620 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
1623 old = rcu_dereference_protected(tbl->listeners, 1);
1624 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
1625 rcu_assign_pointer(tbl->listeners, new);
1627 kfree_rcu(old, rcu);
1629 tbl->groups = groups;
1635 * netlink_change_ngroups - change number of multicast groups
1637 * This changes the number of multicast groups that are available
1638 * on a certain netlink family. Note that it is not possible to
1639 * change the number of groups to below 32. Also note that it does
1640 * not implicitly call netlink_clear_multicast_users() when the
1641 * number of groups is reduced.
1643 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1644 * @groups: The new number of groups.
1646 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1650 netlink_table_grab();
1651 err = __netlink_change_ngroups(sk, groups);
1652 netlink_table_ungrab();
1657 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1660 struct hlist_node *node;
1661 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1663 sk_for_each_bound(sk, node, &tbl->mc_list)
1664 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1668 * netlink_clear_multicast_users - kick off multicast listeners
1670 * This function removes all listeners from the given group.
1671 * @ksk: The kernel netlink socket, as returned by
1672 * netlink_kernel_create().
1673 * @group: The multicast group to clear.
1675 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1677 netlink_table_grab();
1678 __netlink_clear_multicast_users(ksk, group);
1679 netlink_table_ungrab();
1682 void netlink_set_nonroot(int protocol, unsigned int flags)
1684 if ((unsigned int)protocol < MAX_LINKS)
1685 nl_table[protocol].nl_nonroot = flags;
1687 EXPORT_SYMBOL(netlink_set_nonroot);
1690 __nlmsg_put(struct sk_buff *skb, u32 pid, u32 seq, int type, int len, int flags)
1692 struct nlmsghdr *nlh;
1693 int size = NLMSG_LENGTH(len);
1695 nlh = (struct nlmsghdr*)skb_put(skb, NLMSG_ALIGN(size));
1696 nlh->nlmsg_type = type;
1697 nlh->nlmsg_len = size;
1698 nlh->nlmsg_flags = flags;
1699 nlh->nlmsg_pid = pid;
1700 nlh->nlmsg_seq = seq;
1701 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
1702 memset(NLMSG_DATA(nlh) + len, 0, NLMSG_ALIGN(size) - size);
1705 EXPORT_SYMBOL(__nlmsg_put);
1708 * It looks a bit ugly.
1709 * It would be better to create kernel thread.
1712 static int netlink_dump(struct sock *sk)
1714 struct netlink_sock *nlk = nlk_sk(sk);
1715 struct netlink_callback *cb;
1716 struct sk_buff *skb = NULL;
1717 struct nlmsghdr *nlh;
1718 int len, err = -ENOBUFS;
1721 mutex_lock(nlk->cb_mutex);
1729 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
1731 skb = sock_rmalloc(sk, alloc_size, 0, GFP_KERNEL);
1735 len = cb->dump(skb, cb);
1738 mutex_unlock(nlk->cb_mutex);
1740 if (sk_filter(sk, skb))
1743 __netlink_sendskb(sk, skb);
1747 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1751 nl_dump_check_consistent(cb, nlh);
1753 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1755 if (sk_filter(sk, skb))
1758 __netlink_sendskb(sk, skb);
1763 mutex_unlock(nlk->cb_mutex);
1765 netlink_consume_callback(cb);
1769 mutex_unlock(nlk->cb_mutex);
1774 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1775 const struct nlmsghdr *nlh,
1776 struct netlink_dump_control *control)
1778 struct netlink_callback *cb;
1780 struct netlink_sock *nlk;
1783 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1787 cb->dump = control->dump;
1788 cb->done = control->done;
1790 cb->data = control->data;
1791 cb->min_dump_alloc = control->min_dump_alloc;
1792 atomic_inc(&skb->users);
1795 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1797 netlink_destroy_callback(cb);
1798 return -ECONNREFUSED;
1801 /* A dump is in progress... */
1802 mutex_lock(nlk->cb_mutex);
1804 mutex_unlock(nlk->cb_mutex);
1805 netlink_destroy_callback(cb);
1810 mutex_unlock(nlk->cb_mutex);
1812 ret = netlink_dump(sk);
1819 /* We successfully started a dump, by returning -EINTR we
1820 * signal not to send ACK even if it was requested.
1824 EXPORT_SYMBOL(netlink_dump_start);
1826 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1828 struct sk_buff *skb;
1829 struct nlmsghdr *rep;
1830 struct nlmsgerr *errmsg;
1831 size_t payload = sizeof(*errmsg);
1833 /* error messages get the original request appened */
1835 payload += nlmsg_len(nlh);
1837 skb = nlmsg_new(payload, GFP_KERNEL);
1841 sk = netlink_lookup(sock_net(in_skb->sk),
1842 in_skb->sk->sk_protocol,
1843 NETLINK_CB(in_skb).pid);
1845 sk->sk_err = ENOBUFS;
1846 sk->sk_error_report(sk);
1852 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1853 NLMSG_ERROR, payload, 0);
1854 errmsg = nlmsg_data(rep);
1855 errmsg->error = err;
1856 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1857 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1859 EXPORT_SYMBOL(netlink_ack);
1861 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1864 struct nlmsghdr *nlh;
1867 while (skb->len >= nlmsg_total_size(0)) {
1870 nlh = nlmsg_hdr(skb);
1873 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1876 /* Only requests are handled by the kernel */
1877 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1880 /* Skip control messages */
1881 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1889 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1890 netlink_ack(skb, nlh, err);
1893 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1894 if (msglen > skb->len)
1896 skb_pull(skb, msglen);
1901 EXPORT_SYMBOL(netlink_rcv_skb);
1904 * nlmsg_notify - send a notification netlink message
1905 * @sk: netlink socket to use
1906 * @skb: notification message
1907 * @pid: destination netlink pid for reports or 0
1908 * @group: destination multicast group or 0
1909 * @report: 1 to report back, 0 to disable
1910 * @flags: allocation flags
1912 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1913 unsigned int group, int report, gfp_t flags)
1918 int exclude_pid = 0;
1921 atomic_inc(&skb->users);
1925 /* errors reported via destination sk->sk_err, but propagate
1926 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
1927 err = nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1933 err2 = nlmsg_unicast(sk, skb, pid);
1934 if (!err || err == -ESRCH)
1940 EXPORT_SYMBOL(nlmsg_notify);
1942 #ifdef CONFIG_PROC_FS
1943 struct nl_seq_iter {
1944 struct seq_net_private p;
1949 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1951 struct nl_seq_iter *iter = seq->private;
1954 struct hlist_node *node;
1957 for (i = 0; i < MAX_LINKS; i++) {
1958 struct nl_pid_hash *hash = &nl_table[i].hash;
1960 for (j = 0; j <= hash->mask; j++) {
1961 sk_for_each(s, node, &hash->table[j]) {
1962 if (sock_net(s) != seq_file_net(seq))
1976 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1977 __acquires(nl_table_lock)
1979 read_lock(&nl_table_lock);
1980 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1983 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1986 struct nl_seq_iter *iter;
1991 if (v == SEQ_START_TOKEN)
1992 return netlink_seq_socket_idx(seq, 0);
1994 iter = seq->private;
1998 } while (s && sock_net(s) != seq_file_net(seq));
2003 j = iter->hash_idx + 1;
2006 struct nl_pid_hash *hash = &nl_table[i].hash;
2008 for (; j <= hash->mask; j++) {
2009 s = sk_head(&hash->table[j]);
2010 while (s && sock_net(s) != seq_file_net(seq))
2020 } while (++i < MAX_LINKS);
2025 static void netlink_seq_stop(struct seq_file *seq, void *v)
2026 __releases(nl_table_lock)
2028 read_unlock(&nl_table_lock);
2032 static int netlink_seq_show(struct seq_file *seq, void *v)
2034 if (v == SEQ_START_TOKEN) {
2036 "sk Eth Pid Groups "
2037 "Rmem Wmem Dump Locks Drops Inode\n");
2040 struct netlink_sock *nlk = nlk_sk(s);
2042 seq_printf(seq, "%pK %-3d %-6d %08x %-8d %-8d %pK %-8d %-8d %-8lu\n",
2046 nlk->groups ? (u32)nlk->groups[0] : 0,
2047 sk_rmem_alloc_get(s),
2048 sk_wmem_alloc_get(s),
2050 atomic_read(&s->sk_refcnt),
2051 atomic_read(&s->sk_drops),
2059 static const struct seq_operations netlink_seq_ops = {
2060 .start = netlink_seq_start,
2061 .next = netlink_seq_next,
2062 .stop = netlink_seq_stop,
2063 .show = netlink_seq_show,
2067 static int netlink_seq_open(struct inode *inode, struct file *file)
2069 return seq_open_net(inode, file, &netlink_seq_ops,
2070 sizeof(struct nl_seq_iter));
2073 static const struct file_operations netlink_seq_fops = {
2074 .owner = THIS_MODULE,
2075 .open = netlink_seq_open,
2077 .llseek = seq_lseek,
2078 .release = seq_release_net,
2083 int netlink_register_notifier(struct notifier_block *nb)
2085 return atomic_notifier_chain_register(&netlink_chain, nb);
2087 EXPORT_SYMBOL(netlink_register_notifier);
2089 int netlink_unregister_notifier(struct notifier_block *nb)
2091 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2093 EXPORT_SYMBOL(netlink_unregister_notifier);
2095 static const struct proto_ops netlink_ops = {
2096 .family = PF_NETLINK,
2097 .owner = THIS_MODULE,
2098 .release = netlink_release,
2099 .bind = netlink_bind,
2100 .connect = netlink_connect,
2101 .socketpair = sock_no_socketpair,
2102 .accept = sock_no_accept,
2103 .getname = netlink_getname,
2104 .poll = datagram_poll,
2105 .ioctl = sock_no_ioctl,
2106 .listen = sock_no_listen,
2107 .shutdown = sock_no_shutdown,
2108 .setsockopt = netlink_setsockopt,
2109 .getsockopt = netlink_getsockopt,
2110 .sendmsg = netlink_sendmsg,
2111 .recvmsg = netlink_recvmsg,
2112 .mmap = sock_no_mmap,
2113 .sendpage = sock_no_sendpage,
2116 static const struct net_proto_family netlink_family_ops = {
2117 .family = PF_NETLINK,
2118 .create = netlink_create,
2119 .owner = THIS_MODULE, /* for consistency 8) */
2122 static int __net_init netlink_net_init(struct net *net)
2124 #ifdef CONFIG_PROC_FS
2125 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
2131 static void __net_exit netlink_net_exit(struct net *net)
2133 #ifdef CONFIG_PROC_FS
2134 proc_net_remove(net, "netlink");
2138 static void __init netlink_add_usersock_entry(void)
2140 struct listeners *listeners;
2143 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2145 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2147 netlink_table_grab();
2149 nl_table[NETLINK_USERSOCK].groups = groups;
2150 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2151 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2152 nl_table[NETLINK_USERSOCK].registered = 1;
2153 nl_table[NETLINK_USERSOCK].nl_nonroot = NL_NONROOT_SEND;
2155 netlink_table_ungrab();
2158 static struct pernet_operations __net_initdata netlink_net_ops = {
2159 .init = netlink_net_init,
2160 .exit = netlink_net_exit,
2163 static int __init netlink_proto_init(void)
2165 struct sk_buff *dummy_skb;
2167 unsigned long limit;
2169 int err = proto_register(&netlink_proto, 0);
2174 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
2176 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2180 if (totalram_pages >= (128 * 1024))
2181 limit = totalram_pages >> (21 - PAGE_SHIFT);
2183 limit = totalram_pages >> (23 - PAGE_SHIFT);
2185 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2186 limit = (1UL << order) / sizeof(struct hlist_head);
2187 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2189 for (i = 0; i < MAX_LINKS; i++) {
2190 struct nl_pid_hash *hash = &nl_table[i].hash;
2192 hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
2195 nl_pid_hash_free(nl_table[i].hash.table,
2196 1 * sizeof(*hash->table));
2200 hash->max_shift = order;
2203 hash->rehash_time = jiffies;
2206 netlink_add_usersock_entry();
2208 sock_register(&netlink_family_ops);
2209 register_pernet_subsys(&netlink_net_ops);
2210 /* The netlink device handler may be needed early. */
2215 panic("netlink_init: Cannot allocate nl_table\n");
2218 core_initcall(netlink_proto_init);